The Organization of Behavior Over Time:
Insights from Mid-Session Reversal

Rebecca M. Rayburn-Reeves and Robert G. Cook
Tufts University

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

What are the mechanisms by which behavior is organized sequentially over time? The recently developed mid-session reversal (MSR) task offers new insights into this fundamental question. The typical MSR task is arranged to have a single reversed discrimination occurring in a consistent location within each session and across sessions. In this task, we examine the relevance of time, reinforcement, and other factors as the switching cue in the sequential modulation of control in MSR. New analyses also highlight some of the potential mechanisms underlying this serially organized behavior. MSR provides new evidence about how cues interact to compete for the control of behavior within and across sessions. We suggest that MSR is an excellent preparation for studying the competition among psychological states and their resolution toward action.

Keywords: mid-session reversal, behavioral organization, timing, anticipation, switching cue, discriminative cue

Author Note: Dr. Rebecca M. Rayburn-Reeves, Department of Psychology, Tufts University, 490 Boston Ave., Medford, MA 02155.

Correspondence concerning this article should be addressed Dr. Rayburn-Reeves at beckyreeves02@gmail.com.

Acknowledgments: This research and its preparation were supported by a grant from the National Eye Institute (#RO1EY022655) to RGC. We thank the members of the Cook lab (Ali Qadri, Dan Brooks, Ashlynn Keller, and Suzanne Gray) for the many stimulating conversations about MSR and its possible mechanisms during the preparation of this paper. We would also like to extend an additional thanks to Ali and Ashlynn for their assistance with the manuscript and citations. E-mail: Robert.Cook@tufts.edu. Home page: www.pigeon.psy.tufts.edu.

---

The Sequential and Temporal Organization of Behavior

How do different behaviors come to be organized or sequenced as observed in the natural world? What processes ultimately determine which specific behaviors are exhibited in any next moment? How is behavior organized such that all of its global and local components unfold in the right order and at the right time to optimize an adaptive response? Understanding the answers to such questions is as salient today as when Lashley wrote his seminal paper on these general issues (Lashley, 1951). He suggested that the temporal integration of behavioral sequences was “. . . both the most important and also the most neglected problem of cerebral physiology” (p. 112). More than 60 years later, we have made progress in our understanding of the correlates and mechanisms of specific behaviors, but find ourselves still asking the same general questions about how complex behavioral sequences are learned and organized.

The psychological and neural processes that govern the order and timing of complex behavior, be it making coffee or engaging in a conversation or attracting a mate, remain poorly understood. While there has been a historical focus on humans, especially in regard to the complexities of learning and processing language and its syntax, animals face the same problems of selecting and organizing their behavior with respect to time and order. Performing a mating ritual in the wrong order or at the wrong time, for example, is maladaptive, even if all of the behavioral elements are present. Such information is also critical to understanding and treating humans exhibiting disorganized or maladaptive behavioral patterns (e.g., addiction, OCD, ADHD, schizophrenia, depression). As an example, research has shown that patients with schizophrenia exhibit losses in temporal continuity, where the subjective experience of events in time becomes fragmented or disordered (Andreasen, Paradiso, & O’Leary, 1998; Martin, Giersch, Huron, & van Wassenhove, 2013). In a similar vein, impairments in time perception based on interactions with working memory and inhibitory processes have been documented in patients with ADHD (Barkley, 1997). Therefore, understanding the mechanisms by which evolutionarily adaptive behaviors are selected and governed by temporal and sequential regularities is critical to a complete theory of psychology.

In any complex environment there are often multiple cues that could be used to select among competing behaviors at any given moment. Adaptive behavior relies on the ability of animals to attend to the most appropriate cue or cues in any context and to flexibly switch to new or changing cues depending on their relative utility across time. The additional ability to temporally anticipate which cues and behaviors will lead to profitable outcomes would be especially valuable in a world where temporal regularities exist. Antle and Silver (2009) argue that anticipation in both cognitive and noncognitive systems is critical to successfully navigating dynamic and complex environments. Because many environmental events have temporal periodicities occurring on different time scales (e.g., daily, annual, lunar), it is perhaps not surprising then that many animals have evolved specialized physiological and cognitive mechanisms that seem to take advantage of these temporal regularities. The highly visible seasonal migrations of many species, as well as the strongly organized circadian behavior by virtually all species, are well-known examples of this type of temporal organization. In both of these domains, a combination of endogenous (e.g., hormone levels, circadian clocks, etc.) and exogenous (e.g., day length, temperature, etc.) factors within these larger temporal structures have been identified to support the organization and sequencing of such behaviors (Buhusi & Meck, 2005).

Aside from the systems that have evolved to regulate behavior at these larger time scales, processing the passage of time itself would be a highly valuable and informative cue for predicting environmental changes that have regular and repetitive temporal properties. A timing system, whether consciously available to the organism or not, could provide information about when significant events will recur. Such temporal perception allows animals to anticipate and predict significant environmental changes and make appropriate responses at the right time. The study of associative learning can be characterized as observing what types of behaviors are exhibited in reaction to recent stimuli or environmental signals in the temporal stream (Miller & Barnet, 1993). Traditionally, time perception has most often been examined by assessing how well an animal can judge the passage of time, such as in time estimation tasks (Buhusi & Meck, 2005; Roberts, 1981) or discriminate between two different temporal durations, such as in time discrimination tasks (Meck & Church, 1983). Thus, time perception research of this type examines how specific behaviors are mapped onto specific time durations (e.g., pecking a red key after 4 s and a blue key after 8 s) as tested on individual trials. Such studies have provided a wealth of knowledge about how human and nonhuman animals perceive the passage of time within a range of seconds to minutes, a phenomenon known as interval timing (Buhusi & Meck, 2005; Cheng & Roberts, 1989, 1991; Matell & Meck, 2000; McMillan & Roberts, 2013; Meck & Church, 1983; Staddon & Higa, 1999).

While we have made considerable progress at understanding the perception of time on macro-level (e.g., daily, annual) and micro-level (e.g., milliseconds to minutes) time scales, the organization of behavior across a series of events that occur over more intermediate time scales (many minutes to hours) has received far less attention (Buhusi & Meck, 2005). One reason for this lacuna has been a lack of experimental procedures that permit looking at this level of behavioral organization. In humans, it is often presumed that this type of behavioral organization requires integrating a number of discrete events across time (Lewis & Miall, 2006; Vicario, 2013). How nonhuman animals utilize, or are influenced by, the passage of time in the regulation of their behavior over a long series of events is still unclear. Recently, a new procedure has been developed that offers insights into how such extended behavioral patterns are learned and regulated within the course of a session.

This review paper summarizes and examines recent empirical research on the mid-session reversal (MSR) task (Cook & Rosen, 2010; Rayburn-Reeves, Molet, & Zentall, 2011). We believe it offers new theoretical insights into how animals organize their behavior at such intermediate time scales (i.e., over a session). We further suggest that MSR is an excellent preparation for studying the competition among various psychological states and the mechanisms of their resolution that eventually leads to a specific action or series of actions. We begin by introducing this type of reversal task and presenting data from different studies to illustrate the consistent organization of behavior that emerges in this task. We then discuss the various types of cues available to inform behavioral choice within a session and how they modulate the behavioral patterns observed. We end by considering several new analyses that reveal more about the possible mechanisms for how competing task choices are selected and organized within individual sessions. We believe this research approach provides a new window into how animals learn, maintain, and organize the serial nature of their complex behaviors.

The Mid-Session Reversal Task

Discrimination reversal learning has a long history as a means for studying cognition in nonhuman and human animals (Bitterman, 1965, 1975; Mackintosh, 1974; Tinklepaugh, 1928). Reversal training consists of successfully teaching an animal to discriminate between a set of stimuli and then reversing the reward values of the alternatives and observing the subsequent learning of the new contingencies. In serial reversal learning, each time the new discrimination is learned to some performance criterion, the contingencies are reversed (Bitterman, 1975). Basing the reversal on the animal’s performance makes the time to any subsequent reversal unpredictable, making the changed reinforcement contingencies from recent trials the most reliable cue an animal can use to shift its discriminative behavior.

Mid-session reversal is different from serial reversal learning tasks in that each session contains a discrimination reversal that occurs at a consistent point in the session, most typically after a fixed number of trials. As a result, it offers a new way to assess how animals come to manage the transition between the two task contingencies. In an MSR discrimination task each session starts with one task contingency, and then halfway through the session the contingency is reversed. For instance, an animal could be tested with a simultaneous visual discrimination in which choice responses to a red stimulus are rewarded for the first half of the session (red+/green–), and then choice responses to a green stimulus would be rewarded for the second half of the session (red–/green+). Responses to incorrect stimuli typically result in a short time-out. The reversal’s regular occurrence after a specific time or trial and the resulting predictability introduces a number of cues in MSR tasks that could be used to reliably predict the reversal. Besides counting the number of trials until the reversal, for example, animals could keep track of elapsed time, since the timing of events constituting each trial is also quite regular. The specifics of this important latter issue are considered more fully in the next section following a brief outline of the main features of MSR behavior.

Studies using the MSR procedure have revealed a highly consistent pattern of choice behavior across a session. Shown in Figure 1 are the post-acquisition results from three separate investigations examining dissimilar types of visual discriminations with pigeons using an MSR procedure (Cook & Rosen, 2010; McMillan, Sturdy, & Spetch, 2015; Rayburn-Reeves et al., 2011). Within each investigation, the same stimuli were consistently ordered with respect to time across trials within each session. The only thing that changed was the identity of the correct stimulus. This was always reversed, or switched, midway through a session. Identical to the way psychophysical functions are shown, each curve depicts the proportion of the birds’ responses to one stimulus choice relative to the alternate stimulus choices. Thus, choice behavior is plotted as the percentage choice of the correct stimulus of the first task as a function of trial number across a session. As a result, successful choice behavior for both tasks is reflected by high values at the beginning (performing Task 1) and low values at the end (performing Task 2) of each session, respectively.

The top panel shows the average results from pigeons performing related conditional discriminations across the two segments within each session (Cook & Rosen, 2010). In the first half of each session, pigeons were rewarded for performing matching-to-sample responses (e.g., if red, choose red; if cyan, choose cyan) and in the second half they were rewarded for oddity-from-sample responses (e.g., if red, choose cyan; if cyan, choose red). The middle panel shows the results of pigeons performing a simultaneous, two-alternative simple choice discrimination using red and green stimuli (Rayburn-Reeves et al., 2011). Here in the first half of a session, choice responses to the red stimulus were rewarded, while in the second half, choice responses to the green stimulus were rewarded. The bottom panel shows the results of pigeons performing a successive go/no-go discrimination using red and green stimuli (McMillan et al., 2015). In this task, responses to the red stimulus were rewarded when presented in the first half of the session, while responses to the green stimulus were rewarded when presented in the second half. Despite the different task organizations of these discriminations, at asymptotic performance the pigeons exhibit highly similar behavioral characteristics over the course of a session, suggesting that the same mechanisms are likely regulating the observed behavioral pattern across these experiments.

The common result is that, regardless of the task, all pigeons learned to accurately discriminate and order each of the competing task contingencies (e.g., matching to oddity; red to green). This switching function reflects that the predictability of the reversal at the midpoint of a session, as well as the consistent ordering of reinforcement contingencies associated with the discriminative stimuli, allows the pigeons to learn and respond to the appropriate task during the correct portion of a session. Figure 1 also reveals that this highly organized and consistent reinforcement structure over training results in a similar pattern of choice behaviors across a session, regardless of the nature of the stimulus contingency (simple or conditional) or the type of discrimination (simultaneous or successive). In addition to the high accuracy at both terminal points of each session, there is a smooth, sigmoidal transition seen in all three panels. When averaged over sessions, it appears the pigeons gradually and probabilistically stop performing the first discrimination and start performing the second one. This gradual transition results in a function depicting a systematic reduction in choice accuracy based on proximity to the reversal. For each of the discriminations in Figure 1, a logistic function has been fitted to each result. Three parameters control the shape of the function. These parameters include the asymptote (a), inflection point (x₀), and rate of change (b) of the curve. This function produces excellent fits (R²s > 99%) to all three discriminations. The excellent baseline performance for each of the tasks in the experiments is reflected in the high asymptote parameter value confirmed for each fit (as > 95%). The second parameter of this fit captures the inflection or “indifference” point at which the pigeons respond equally to both Task 1 and 2. The fitted inflection points are near and slightly after the reversal midpoint of each session (expressed as a percentage of the session—top panel: 52.3%, middle: 55.6%, and bottom: 53.6%). The third parameter captures the rate of change in task responding across the function. Here the three experiments appear to differ. The conditional discrimination is more difficult for the birds to reverse than either of the two simple discriminations. It has a shallower slope reflected in a slower rate parameter (b = −13.10) than the other two discriminations (middle b = −5.02 & bottom b = −8.13), which show sharper transitions between the two halves of the session. Nonetheless, all three experiments reveal the same basic changes in accuracy and gradual transition from Task 1 to Task 2 across the session. This transition results in two distinct types of errors made by the pigeons within and across sessions.

The first type of error is that of anticipation. This is where responses appropriate to the second task occur before the reversal. These errors are interesting as their orderly nature and increasingly greater occurrence near the reversal reveals them to be task-related mistakes. Thus, there is some degree of competition or loss of stimulus control between the two tasks near the reversal. Revealingly, these anticipatory errors persist and reoccur, despite the fact that there is never a reward for switching to the second task early. In fact, as the reversal point approaches, the frequency of anticipatory errors increases. This reversal-related increase suggests that these errors are internally generated intrusions of the second task into performance of the first one.

The second type of error is making perseverative choices. These are errors where choice responses from the first task continue on after the reversal. These errors also persist despite the consistent feedback after the reversal that responses appropriate to the first task have not, and therefore likely will not, result in further reinforcement in that session. The information provided by these two types of errors is not equivalent. When errors are made in anticipation to the second task, the animal is informed that its choices on the current trial were incorrect, but it does not guarantee that the next Task 1 response will not be rewarded. When a perseverative error occurs, however, it should provide unambiguous information to the animal that Task 1 responses will no longer be reinforced in that session and, therefore, the remaining responses should only be to Task 2. The persistence of these perseverative errors throughout training, however, suggests that this latter type of feedback does not easily result in rapid switching to Task 2 in pigeons (see section below on control of switching behavior by reinforcement cues).

Despite the potential difference in feedback provided by these two errors, they seem to occur symmetrically around the reversal. This generalized pattern among the competing tasks and choices indicates that the behavior of the pigeon is strongly organized by some mechanism that is not solely based on reinforcement feedback across the course of a session. The next key question thus becomes what cue or cues modulate the changing choice behavior as the pigeons sequentially move from performing one action to a competing action across a session.

Control of Switching Behavior in Mid-Session Reversal

An MSR task can be considered to have two distinct types of cues. The first type is consistent with the traditional discriminative cues that directly receive choice behaviors and lead to rewarded outcomes as required by simple or conditional discriminations. These are the stimuli available to the animal during a trial to which it can respond (e.g., red and green key lights). The second type of cue can be thought of as the switching cue. The switching cue is a conditional, context-like cue that controls or sets the occasion for which discriminative cue the pigeon should select during each part of a session. Because the reversal most typically occurs at the midpoint of each session in MSR tasks, this predictability introduces multiple potential sources of switching cues that could be used to predict the reversal. Among the internal or endogenous switching cues potentially available are time elapsed within the session, counts of the number of trials that have occurred, or relative satiety. Among the external or exogenous cues would be the reinforcement outcomes from recent responses or other additional external cues that could identify each portion of a session (Rayburn-Reeves, Qadri, Brooks, Keller, & Cook, under review). Isolating the source of the switching cue controlling the sequential performance of the two discriminations has been a top priority in the initial analyses of MSR.

Endogenous sources of switching control were among the first type of cue to be examined. For instance, the degree of relative satiety was ruled out early on as a switching cue (Cook & Rosen, 2010). Using a pre-feeding manipulation prior to a session, no evidence was observed that the pigeons were using their degree of hunger as the basis for changing their choice behavior across a session. Instead the experimental evidence has consistently indicated that the pigeons are using elapsed time within a session as the primary means for resolving the competing choice behaviors of each task. This is important to establish because the vast majority of MSR research has typically used a simple count of the number of trials to easily program the computer to reverse the tasks within a session. This creates the possibility that pigeons could have used counts or estimations of the number of trials, or perceived amount of behavioral experience, from the start of a session to guide their transition in choice behavior. These do not appear to be the case. Time seems to be the essential switching cue.

Compelling evidence for use of elapsed time as the switching cue has come from direct manipulations of time within and across sessions. Cook and Rosen (2010) found that inserting an empty temporal gap of different durations into the middle of the first half of a session resulted in systematic shifts in the onset of subsequent oddity-based behaviors. This suggests that pigeons were timing through the gap and prematurely switching based on the total elapsed time since the session start, regardless of their experience. They also trained pigeons on 20, 40, and 80 min sessions where the switch occurred at the temporal midpoint of the session (i.e., 10, 20, or 40 min., respectively). Therefore, regardless of the number of trials initiated by the pigeons during these time periods, the reversal occurred on an exclusively time-based schedule. Non-differentially reinforced probe sessions after this training confirmed that the same highly regulated switching behavior occurred near the temporal midpoint of the sessions. As the internal perception of the passage of time was the only reliable source of information in these procedures (i.e., nothing externally physical is changing from trial to trial), the resulting switch from matching to oddity behavior would have had to be controlled by mechanisms related to this temporal cue.

Further support for the use of a timing cue in MSR was also provided by McMillan and Roberts (2012). They trained pigeons on a simple discrimination using red and green stimuli with a 6 s inter-trial interval (ITI) between each trial. Probe sessions were then conducted in which this ITI duration was either doubled to 12 s or halved to 3 s. With the longer ITIs the pigeons made significantly more anticipatory errors as the birds prematurely switched to the second behavior earlier than when the ITI was 6 s or 3.0 s. Correspondingly, they made significantly more perseverative errors with the 3 s ITI duration, by switching to the second behavior later than during sessions with 6 s or 12 s ITI lengths. Both of these systematic changes in errors as a function of ITI indicate the pigeons were using the elapsed time, as opposed to counting trials, to predict the location of the discrimination reversal. Given such findings, the observation of any anticipation errors prior to the reversal is likely a good indirect marker that some form of elapsed time is the switching cue controlling an animal’s choice behavior in MSR.

A recent study by McMillan et al. (2015) has added further insights into this general issue. They tested a go/no-go successive discrimination version of an MSR task (see bottom panel of Figure 1). While the summary results with this task look similar to the other discriminations in Figure 1, the cues controlling the switching behavior of the pigeons were further illuminated by the authors examining pecking behavior separately for the different reinforced go and non-reinforced no-go trials on each side of the reversal (see Figure 1a from McMillan et al., 2015). This breakdown revealed that the pigeons rarely failed to peck the correct stimulus of each task on either side of the reversal. This approach ensured maximizing reward on all reinforced go trials. It was only on the no-go stimulus of each task that birds made errors. The pattern of these errors suggests that the behavior with the two tasks was likely controlled by different mechanisms. As a session progressed, the pigeons increasingly failed to inhibit pecking to the first task’s no-go stimulus as the trials approached the reversal. These anticipation errors prior to the reversal were likely mediated by some criterion-based timing cue generated from between-session averaging of the time to the reversal. This ensured maximizing reinforcement because the pigeons were pecking the upcoming correct Task 2 stimulus on every presentation just prior to the reversal, as well as pecking the correct Task 1 stimulus at this point. After the reversal, preservative errors to the formerly correct Task 1 stimulus also persisted and required a number of non-reinforced responses to extinguish. These persistent errors after the reversal are, in contrast to timing, likely mediated by within-session excitation related to the pigeons’ recent experience with the first task (i.e., repeated reinforcement for Task 1 responses up to the reversal). Overall, this two-part approach by the pigeons ensured that all rewards were collected on both types of go trials with only one of the two discriminative stimuli (the correct stimulus of Task 2) being timed, and the other (the correct stimulus of Task 1) being controlled by within-session excitation. This single-stimulus timing account is likely a product of the successive nature of the go/no-go procedure where separate presentations of a single stimulus occur on each trial. It is hard to imagine how such a single-stimulus timing mechanism could account for MSR reversal behavior involving more complex stimulus arrangements where more rules and stimulus combinations are involved, such as in a matching-to-sample procedure.

Daniel, Cook, and Katz (2015), for instance, recently conducted an MSR experiment to examine whether pigeons could learn to conditionally switch behavior between two abstract concepts over a session. Pigeons were trained to switch from a matching-to-sample (MTS) to an oddity-from-sample (OFS) task within a session, similar to the procedure used by Cook and Rosen (2010). Of more importance, however, was the use of much larger stimulus sets to train each concept. This was done because large stimulus sets are known to promote concept learning in pigeons (Bodily, Katz, & Wright, 2008; Cook & Wasserman, 2006; Katz & Wright, 2006; Wasserman, Kiedinger, & Bhatt, 1988; Wright, Cook, Rivera, Sands, & Delius, 1988). Over a series of extensive training sessions and stages, the set size of the number of randomly combined stimuli involved in each of those two tasks was increased from three to six to 12 items. Most critically to the question of timing is that all birds showed highly similar, almost linear, switching functions that exhibited large degrees of anticipatory and perseverative behavior. This was true regardless of the number of stimuli involved within each part of the session. Because of the very large and changing number of stimulus pairs involved, it is hard to see how asymmetrical timing of a single stimulus could be involved. This is perhaps because every stimulus is simultaneously a correct and incorrect stimulus intermixed within each part of a session. The switching function indicates that a timing mechanism based on collective groups of matching and oddity relations seem to be involved. Interestingly, no evidence was found in transfer testing that the pigeons had learned to time the twin general concepts of matching or oddity as the means to switch behaviors on the MSR task. Instead the pigeons seemed to learn each portion of the MSR task by learning to time the different sample-specific arrangements. This need to memorize and track many arrangements simultaneously may be one reason why strongly linear switching functions were observed in that experiment in comparison to the typically sigmoidal function seen with simpler discriminations (see Figure 1).

A possibly related result can be seen in unpublished results of an MSR experiment done subsequently with the same birds as tested in Cook and Rosen (2010). In this case, each bird learned in succession a series of MSR discriminations involving MTS and OFS conditional discriminations with three different groupings of sample stimuli (first set: red & cyan; second set: yellow & violet; third set: blue & green). After training each of these stimulus groupings separately in succession, pigeons were given sessions where all three groupings of sample stimuli were randomly intermixed across trials, with the reversal from MTS to OFS remaining at the midpoint of each session. Figure 2 shows the averaged switching functions from the last 10 sessions where the three sample groups were randomly intermixed within each portion of the same session. Each stimulus group exhibits similar overlapping functions with comparable degrees of anticipatory and perseverative errors. These overlapping functions occurred despite the fact that the testing order of the specific samples was completely randomized within each session. These results rule out the possibility that the amount of experience or timing of specific stimuli were critically involved in mediating switching in this conditional discrimination MSR task. It instead suggests that the switching function is the result of increasing internal competition between the representations of Task 1 and Task 2 as a function of reversal proximity.

The general preference for using a time-based cue in the MSR paradigm parallels similar research on time-place learning (TPL) experiments with animals. Time-place learning experiments require the animal to shift to different spatial locations based specifically on an elapsed time during which reinforcement is available in each location (Wilkie, 1995). Thus, TPL tasks directly test the ability of animals to utilize time as a predictive cue for which location is currently providing reinforcement. For example, Wilkie and Willson (1992) trained two pigeons in an operant task using 90-s trials in which intermittent reinforcement was available for 30 s at each of three key locations across a trial. They found that the pigeons allocated the majority of their responses to the correct key during the time in which it provided reinforcement. They also found that the pigeons sometimes began responding on the to-be reinforced response key before that key provided food. They suggested this behavior was evidence that pigeons were anticipating the change in reinforcement across keys as a function of an interval timing cue. One major difference between TPL and MSR tasks, however, is that in the former, the passage of time is the best cue for where reinforcement will be located. This is because reinforcement occurs probabilistically due to the use of an intermittent interval schedule of reinforcement. This creates ambiguity as to when a particular key will stop providing reward. If the schedule of reinforcement was not probabilistic, it might be assumed that animals would use the reinforcement outcome as the primary feedback cue to switch key responses. The MSR task illuminates the fact that even providing unambiguous outcome information does not result in primary use of the reinforcement cue. Rather, both TPL and MSR tasks reveal that pigeons regularly rely on the passage of time as a cue for organizing sequences of behavior over a repeated series of events.

Control of Switching Behavior by Reinforcement Cues

The above results indicate that switching behavior in pigeons in an MSR task is predominantly controlled by time, at least when using visual stimuli as the relevant dimension for the discrimination. This reliance on time as the switching cue, however, is not the optimal solution. Short of counting each trial, one of the best cues would be to attend to the consequences of recent choices. For example, humans are excellent at MSR tasks (Cook & Rosen, 2010; Rayburn-Reeves et al., 2011). The reason for this is because humans learn to stay with the correct choice associated with the first task until an error occurs. At this point, humans immediately switch their choice behavior to the second task. This behavior is indicative of a win-stay/lose-shift strategy (Levine, 1975; Restle, 1962), where responses to one alternative persist until the first non-reinforced trial, where responses then shift to the other alternative. It is thought to be the optimal strategy in reversal tasks because it minimizes errors and results in rapid, flexible shifts in behavioral responses between the two tasks. Pigeons appear not to greatly attend to this valuable and highly useful exogenous cue in MSR tasks. This is a bit of mystery and has received considerable experimental attention. The next section reviews this material.

Variable Reversal Locations

To examine the relative contribution of reinforcement cues, much of this MSR research has focused on reducing the relevancy of time and increasing the saliency of reinforcement as the switching cue. Many studies have attempted to reduce the relevancy and predictability of the timing cue by randomly varying across sessions the trial at which the discrimination reversal occurs (McMillan, Kirk, & Roberts, 2014; Rayburn-Reeves, Laude, & Zentall, 2013; Rayburn-Reeves et al., 2011; Rayburn-Reeves & Zentall, 2013; A. P. Smith, Pattison, & Zentall, 2016).

In an initial study, Rayburn-Reeves et al. (2011) found that this variable reversal manipulation did increase the contribution of reinforcement cues to the control of pigeons’ switching behavior. They showed that when reversals occurred early in the session, pigeons produced more perseverative than anticipatory errors, but clearly responded to the reinforcement shift by switching to the second task earlier than when the reversal was presented later in a session. During sessions when the reversal occurred at these later points within the session, perseverative errors decreased and anticipatory errors increased. Furthermore, the fewest combined errors of both types were found when the reversal occurred at the midpoint of the session. The latter result suggests the birds were likely using temporal information from across a number of sessions to compute an aggregate expectation of when Task 1 or 2 would be in effect. Similar molar aggregations from across sessions seem to have occurred in other studies using variable reversal locations as well (McMillan et al., 2014; Rayburn-Reeves, Laude, et al., 2013). In fact, any anticipation of a reversal within a session, variable or not, represents this type of molar aggregation operation from across prior sessions.

Nonetheless, because of the asymmetry in the rate and types of errors made across the different reversal locations, it appears the pigeons can be sensitive to the changing reinforcement contingencies, at least as experienced over a number of trials. If pigeons were solely using the time within the session as a cue based on an aggregate of previous sessions’ reversal locations, the actual location of the reversal event during the current session should not have made a difference. Taken together, these results suggest that when time-based cues are made less useful, control by recent outcomes increases in MSR. More important, it indicates that both external and internal cues can be used to guide behavioral choice in MSR, although their relative strength may vary depending on the circumstances. One interesting possibility is that reinforcement acts as a molecular or local cue, adjusting levels of excitation and inhibition across trials within sessions, whereas the timing cue is generated over a number of sessions, acting on a molar level in regulating responses across a single session. It seems to be the interplay between these two sources of information that combine to control the animals’ momentary course of action.

Role of Spatial Cues

Another factor that apparently produces greater attention to reinforcement as a switching cue involves using a spatial dimension as the critical discriminative stimulus for the first and second tasks. Across a number of experiments, pigeons have been tested with the two portions of the session involving a switch of reinforcing responses from one side key (e.g., left) to the other (e.g., right) at the midpoint of the session. From these experiments, the pigeons are clearly more sensitive to reinforcement outcome as a switching cue and appear less controlled by temporal cues than when tested on visual discriminations.

McMillan and Roberts (2012), for example, trained pigeons on an MSR task using a combination of relevant visual and spatial information. Across three phases, pigeons first experienced discriminations in which both spatial and visual cues were relevant (Phase 1), then only visual cues were relevant (Phase 2), and finally back to both cues being relevant (Phase 3). During Phases 2 and 3, probe sessions were intermixed in which the ITI length was either doubled (12 s) or halved (3 s). As described above, this allows assessment of the relevance of time as a switching cue. They found that accuracy around the reversal location was improved and sharpened when both dimensions were relevant in comparison to the visual-only phase. Further, probe sessions with the ITI manipulation resulted in large and expected temporal differences in the visual-only condition, consistent with pigeons’ use of elapsed time in a session as being the primary switching cue. In the combined visual-spatial task, however, these same ITI manipulations had little effect as the birds exhibited the same switching function in each ITI condition. This indicates that elapsed time was not the primary cue causing the switch from one response to the other. McMillan et al. (2014) and McMillan et al. (2015) found similar results indicating that the use of a spatial discrimination consistently sharpens switching accuracy in MSR tasks. This sharper discriminative transition at the reversal and pigeons’ general insensitivity to ITI manipulations when spatial information is directly relevant to the MSR each suggest that the pigeons were increasingly relying on local reinforcement contingencies to guide their switching behavior. These findings and those of Rayburn-Reeves et al. (2011) indicate that pigeons are sensitive to recent response-reinforcement feedback, especially when time is difficult to use and when space is the relevant discriminative dimension.

One possibility brought up by McMillan and Roberts (2012) was that spatial information provides a form of prospective cuing allowing the animal to appropriately orient toward the correct stimulus. Visual tasks do not afford that type of information, as the positions of the visual cues randomly change across trials. Having a spatial cue may increase accuracy by assisting the animals to bridge the gap between trials and making it easier for them to recognize changes in the reinforcement contingencies.

Memory and Reinforcement Cues

In an attempt to better clarify the role of the memory for prior trials versus sustained postural or location orientation during the discriminations, Rayburn-Reeves, Laude, et al. (2013) manipulated the time between trials to see whether shorter or longer ITI durations would produce more efficient use of the previous trials’ outcomes. One hypothesis based on a theory of memory decay is that if the animal is given too much time between trials it may not be able to remember its most recent experiences because of the decay in memory that occurs over time. Likewise, if given too little time, proactive interference between trials may become too great. In either case, too much or too little time between trials would make it difficult for pigeons to use their memories of prior choices and outcomes to guide behavior.

To examine this issue, Rayburn-Reeves, Laude, et al. (2013) trained three groups of pigeons on a spatial MSR discrimination task. Each group was given a different ITI duration (1.5 s, 5.0 s, or 10.0 s) during initial training. The hypothesis was that, if the memory for the previous trial weakened as a function of time, then the pigeons trained on the shorter ITI should perform better than the other two groups. Indeed, they found that pigeons trained on the 5.0 s and 10.0 s ITI lengths showed the typical anticipation errors around the reversal location, suggesting use of the time-based switching cue. In contrast, pigeons trained with the 1.5 s ITI length showed almost no anticipation, producing a strong stepwise function that suggested use of reinforcement cues. In a follow-up experiment, half of the pigeons from the longer ITI groups were transferred to the 1.5 s ITI task, while the remaining half continued with their previous ITI durations. Once transferred, the pigeons retrained on the 1.5 s ITI task also began showing near optimal performance, similar to the pigeons initially trained on the shortest duration. Finally, all groups were given training on the variable reversal task created by Rayburn-Reeves et al. (2011). Pigeons retained on the 5.0 s and 10.0 s ITI durations showed large numbers of anticipatory and perseverative errors as found in Rayburn-Reeves et al.’s (2011) study, whereas pigeons trained with 1.5 s ITI durations appeared to base responding almost entirely on the reinforcement cue, with few anticipatory and perseverative errors across reversal locations.

Laude, Stagner, Rayburn-Reeves, and Zentall (2014) further manipulated independently the ITI duration (1.5 and 5.0 s) and the relevant stimulus dimension (visual vs. spatial) by training four groups of pigeons on each combination of these two variables. These groups were chosen to parcel out whether the reduction in ITI length, the relevant stimulus dimension, or some combination of both was contributing to the differences seen in the previous research. If it was simply that ITI durations of longer than 1.5 s resulted in rapid declines in memory for the most recent trial, then the relevant stimulus dimension should not have mattered. Likewise, if the spatial as opposed to the visual discrimination affords greater use of reinforcement contingencies, then ITI should not matter. Laude et al. (2014) found that only the group trained with a combination of the spatial discrimination and a 1.5 s ITI length showed significant reductions in anticipation prior to the reversal as compared with the other three groups. These results strongly suggest that both the reduction in ITI length and the use of a spatial discrimination are necessary for optimizing performance by pigeons in MSR. Either element alone does not appear to be sufficient to produce a stepwise function indicative of a possible win-stay/lose-shift strategy. Taken together, these results suggest that when outcome information is recent enough in spatial discriminations, it seems as though pigeons can better utilize this type of information to refine and optimize their reversal behavior. Rayburn-Reeves, Laude, et al. (2013) point out that evidence for the use of reinforcement-based cues under short ITI durations needs to be assessed within the possibility that animals are using their postural orientation or physical location in the chamber as an important basis for choice.

An important observation about the pigeons’ ITI behaviors observed in the video records from their experiment sheds light on what might have been controlling performance across the session. During the long ITI durations, the pigeons regularly moved significantly more around the chamber than with the 1.5 s ITI. With less than two seconds between the hopper offset and the onset of the next trial, a pigeon only had enough time to raise its head and orient to the previously pecked location in space, resulting in it often standing and remaining on that side of the chamber and reaching toward the hopper with its head to eat. The authors suggested that one reason why the short ITI group performed so well was because they were able to develop a form of “procedural” memory based on a repetitive spatial peck-eat pattern. It was this pattern that could be easily disrupted by the absence of reinforcement on the reversal. This in turn caused them to move to the other side of the chamber on the following trials. Thus, positional orientation may be an important part of why spatial discriminations in general support much better use of reinforcement cues than visual discriminations in guiding switching behavior. By allowing the competing tasks (first left+, then right+) to be both more distinct and memorable, it may allow the animals a vehicle for reducing competition between discriminative choice behaviors near the session’s transition point.

Species Differences

As described, the majority of MSR research has been conducted using pigeons. One interesting comparative question is whether other animals would show similar MSR findings, such as the general preference for using temporal information over recent reinforcement information as the switching cue. Humans are clearly quite tuned to reinforcement outcomes. Their behavior is the gold standard for the exclusive use of the win-stay/lose-shift strategy, indicative of highly flexible behavioral patterns needed to optimize behavior in dynamic and complex environments. Does this extend to other mammals? To date, there have only been three studies examining MSR performance in rats, for instance. Additionally, in a concession to the rats’ poor visual acuity (Slotnick, 1984), all three involved spatial discriminations. Nonetheless, there is a history of processing differences in how rats and pigeons may attempt to solve different types of discrimination problems (Bitterman, 1965; Cheng & Roberts, 1989; Mackintosh, 1975; Mackintosh & Cauty, 1971). The same thing may possibly be true of MSR as rats seem to show a greater sensitivity to reinforcement outcomes as a switching cue than do pigeons.

Rayburn-Reeves, Stagner, Kirk, and Zentall (2013) trained rats on a spatial discrimination using standard operant levers. Under these conditions, rats learned the two competing tasks (left+ then right+) separated by 5.0 s ITIs to a high degree of accuracy, showing no anticipation prior to the reversal and little perseveration afterward. In a follow-up experiment, rats were then transferred to the variable reversal task and finally given training on multiple reversals within a single session. Throughout the training phases, rats continued to show behavior indicative of win-stay/lose-shift responding based on reinforcement outcomes. The majority of errors occurred only on the first reversal trial. In a similar study, A. P. Smith et al. (2016) trained two groups of rats on an MSR spatial task with 5 s ITI durations using either lever presses or nose-pokes as the two choice response manipulanda. Both groups showed similar acquisition rates to each other and very few errors around the reversal, suggesting that the nature of the response is not critical to the rats.

McMillan et al. (2014) attempted to better understand why rats might be so good at MSR and why they appeared to show increased sensitivity to trial outcomes. They reasoned that in spatial operant tasks, rats may be able to remain more stationary in comparison to pigeons during the intertrial intervals (see earlier discussion of postural and location orientation). If their spatial orientation between trials was helping to mediate choice behavior with the levers, then testing rats with a procedure in which this orientation could not be maintained might increase attention to timing cues. They tested rats in an open T-maze apparatus in which responses to the left and right arms were reinforced for the first and second halves of a 24-trial session. After each trial, the rats were restarted from the same central start box. This central start location essentially eliminated the ability of rats to spatially orient to the correct response location between trials, which is more akin to procedures typically using a central warning signal to start visual discriminations for pigeons in operant tasks.

In contrast to the excellent performance by rats in an operant setting, McMillan et al. (2014) found that the rats tested in a T-maze showed large numbers of anticipatory and perseverative errors around the reversal. This suggests that time may have been the more important switching cue in the T-maze setting. Even when the point of reversal was varied across sessions in a follow-up experiment, T-maze switching behavior did not markedly improve. The results from McMillan et al. (2014) suggest that prior differences between rats and pigeons in MSR may not reflect qualitative differences across these two species; rather, they may be due to the ability of the animal to spatially orient to the previously correct alternative during the delay between trials. It is possible that pigeons are simply more active during ITIs than rats in general, thus requiring reduced ITI length to mitigate the pigeons’ tendency to move around between trials.

Together such results suggest that the benefits of testing most spatial discriminations may stem from allowing animals to use and maintain postural or location orientation cues during the ITI. This allows them to be more sensitive to reinforcement cues and reduce their reliance on using time as the main switching cue in MSR. That said, a reliable finding with T-maze procedures in rats is that of spontaneous alternation (Brushfield, Luu, Callahan, & Gilbert, 2008; Dudchenko, 2004). Rats tend not to repeat a previous response in spatially constructed apparatuses, such as the T-maze, Y-maze, and radial arm maze, even with lengthy delays between trials (Dudchenko, 2004; Evenden & Robbins, 1984). Being such a robust finding in rats, this spontaneous alternation, or tendency not to revisit recent locations, might be indicative of a predisposition for exploratory behavior that is likely advantageous in the rat’s natural environment. Therefore, it may be that a task that requires repeated visits to a single location competes with a natural tendency not to repeat behavior in this manner, thereby creating competition between tendencies to alternate and perseverate. Thus, the use of previous reinforcement as a cue in spatial apparatus may produce competing sources of information (repeat vs. don’t repeat) for subsequent behavior in MSR tasks, which might be enough to shift attention to the time-based cue to mediate behavior. Such differences in procedures between the operant chamber and T-maze complicate the interpretation of MSR in rats with reference to their use of memory for the previous response-reinforcement association.

At the moment, it is unclear whether there exists a qualitative comparative difference between how rats and pigeons solve MSR. While rats, like humans, seem to attend more to reinforcement outcomes than pigeons, it remains to be clarified if this has a methodological source or not. Future research will need to parcel out better whether rats, as well as other types of species, learn to mediate the transition between the two competing tasks in MSR in a way that is qualitatively different from pigeons. Regardless of the final resolution, appreciating how other animals solve MSR across different circumstances will contribute to our understanding of how animals solve such complex sequential discriminations.

Within-Session Modulation of Switching Cue Competition

In the MSR tasks considered thus far, the ability of animals to use switching cues, such as elapsed time or reinforcement, seems to depend partially on factors related to memory, session organization, physical orientation, and the relevant stimulus dimension. Presumably, the specific use of any particular cue depends on its relative utility in comparison to all the available cues within a session (e.g., Egger & Miller, 1962; Mackintosh, 1975; Rescorla & Wagner, 1972).

The usefulness of different environmental cues or physiological processes in the real world is often transitory, however. In complex environments where multiple sources of information can exist and serve to cue significant upcoming events, their relative usefulness may depend on each other or interact over time. As a result, it would be important for animals to be able to flexibly adjust to such changing circumstances across time depending on the relative utility of available cues signaling which actions will lead to the most positive outcomes.

Given these kinds of considerations, we have recently been investigating the relative contribution of simultaneously available switching cues to the control of MSR by pigeons (Rayburn-Reeves et al., under review). One means of doing so involved the addition of distinctive external visual cues during the ITI to assist in identifying each portion of a session. The idea behind the addition of these visual cues was to see whether they served to differentiate the two tasks within the session and reduce control by the timing cue during MSR.

In these experiments, pigeons were given training with alternating sessions where distinctive color cues during the ITI were either present or absent. Cue-absent sessions mirrored standard MSR tasks where no external visual cues were available to denote each portion of a session. During cue-present sessions, the front screen was briefly illuminated by a blue hue during the ITIs of the first half of each session and a yellow hue during the ITIs of the second half of each session. As would be expected, pigeons were much better at the task with the addition of these external switching cues, showing reductions in both anticipatory and perseverative errors as compared with the cue-absent MSR condition.

Next, we put the external visual and internal timing cues in conflict with one another to assess their relative strength within sessions. Using probe sessions with the cue-absent condition, we presented the second half yellow hues during selective trials in the first half of the session (otherwise blue-cued) and first half blue cues during trials in the second half of the session (otherwise yellow-cued). By varying where in the session these conflicting “miscues” appeared, we could assess their relative influence and contribution to performance across a session. We found that the impact of the conflicting external cue depended on the location within the session at which it was inserted. Figure 3 presents a subset of the miscuing data from Experiment 4 reported in Rayburn-Reeves et al. (under review). The figure illustrates the baseline performance of the cue-absent session type (gray triangles) as compared with the cue-absent miscue session type (black circles), in which 10 trials were assigned as miscue trials within this session type (indicated in green). As can be seen, when conflicting external cues from the second half were presented at the beginning of the session, pigeons based responding primarily on the time within the session. That is, they responded appropriately to Task 1 indicating their choice behavior was being strongly controlled by the temporal cue and was not being influenced by the conflicting cue that just appeared during the ITI. Likewise, conflict cues from the first half of the session presented at the end of the session also produced a similar and non-influential outcome. At this point, too, all of their choice behavior was appropriate to Task 2 even with the conflicting cue. As proximity to the reversal increased, however, the conflicting external cues increasingly impacted choice behavior. This can be seen in the middle of the figure by the increasing number of choice responses that were specific to the ITI color cue. Thus, in the middle of the session, the conflict cues had a much greater influence on choice behavior than when they occurred at the beginning or the end of the session. Finally, there appeared to be a greater influence of miscuing prior to the reversal as compared with after.

Such results indicate the pigeons were using both internal timing and external color cues to discriminate Task 1 from Task 2. More important, there was a trade-off between these cues’ influence depending on how close the pigeons were to the reversal. Thus, the relevance of particular switching cues appears to change over the course of a session. Pigeons appear to be dominated by the time at the session endpoints where time-based cues would be highly reliable. As the difficulty of using the timing switching cue increases near the temporally ambiguous reversal, the external ITI cues come to dominate as exhibited by their stronger influence on choice behavior. Thus, it appears pigeons are tracking multiple cues during the session and their attention to each of the cues changes depending on their relative utility. This is consistent with previous results where pigeons utilized the external cue provided by previous response-outcome associations to a greater degree when the timing cue was made less reliable (Rayburn-Reeves et al., 2011). That animals might be controlled by different cues depending on their utility is a well-established notion. The interesting and important new development from the above MSR experiments is that the influence of these different cues may change dynamically over the course of a session. Different cues may have various impacts at different times within a session. Dynamic cue use as a function of time is an interesting avenue of research that has received relatively little attention in the field of animal behavior and comparative cognition.

Analysis of the Switching Function in MSR

The published studies reviewed above identified key properties of MSR and advanced our understanding of the factors controlling behavioral change over a session. This section explores the possible theoretical underpinnings of these situations and their implications for the structure of animal discrimination learning and the organization of behavior. One good starting point is an important question centered on the sigmoidal behavioral pattern seen in these studies (see Figure 1). The exact contour of this function is modulated by several types of switching cues (i.e., time, external visual or spatial stimuli, and reinforcement) and these influence the function’s sharpness at the reversal. One essential question to address is the mechanism of control during the region of “poorer” accuracy covering the transition point between the two tasks. One distinct possibility is that the gradual transition between tasks at the reversal midpoint reflects the increased psychological competition and eventual resolution between the behaviors involved. If so, the MSR paradigm would be an ideal preparation for examining how such representational competition is involved in the sequential and temporal organization of behavior. Before accepting such an account, however, other possibilities need to be ruled out.

One alternative account of this gradual midsession transition considers the training history of the two competing tasks. The mixed nature of reinforcement at the temporally ambiguous midpoint may potentially result in increasingly less accurate choice behavior simply because the animals do not learn the tasks during this portion of a session. In this way, the transition through the 50% range, or inflection point (x0), would directly reflect an absence of knowledge based on a loss of stimulus control by the separate tasks. This confusion account appears to be unsupported. Evidence against this confusion account comes from experiments in which multiple choices or discriminations have been tested at the same time.

Cook and Rosen (2010) conducted an MSR task involving three different sample-choice pairings presented in different successive combinations across the two different portions of each session. By looking at the pattern of choice errors made to the different samples across a session, they could determine whether the pigeons were guessing at the transition point between the two tasks. If the transition reflected an absence of stimulus control, choice errors would be equally distributed among the incorrect alternatives regardless of the sample, the present task, or the stimulus assignments of the upcoming task. Alternatively, if the birds were controlled by the competing structure of each task, they would make choice errors that are specific to each sample organization at the time (i.e., each sample mapping onto the test stimuli) and possibly the upcoming organization of the sample-test mapping of the next segment. The evidence from the distribution of choice errors was unequivocal. Errors were far from equally distributed. For each sample during Task 1, the birds increasingly made errors only to the upcoming incorrect choice alternative linked to that sample in Task 2. Errors to the third “irrelevant” choice alternative never increased for a sample, despite that this same test alternative would be relevant at the same time for the other two samples. Thus, the anticipation errors prior to the reversal reflect specific competition caused by the increasing activation of the sample-test representations involved with the upcoming task, rather than any confusion about what to do. It appeared the birds were always engaged in one task or the other and not just choosing at random as predicted by a confusion account.

Further evidence against a confusion hypothesis comes from McMillan and Robert’s (2015) study. They tested pigeons with a variation of an MSR task in which three different discrimination tasks were programmed to occur successively during one-third of each session (i.e., Tasks 1, 2, and 3). Again, choice errors were not equally distributed across the alternatives as a function of time. In both visual (e.g., red, green, blue) and spatial (e.g., left, center, right) forms of the task, the anticipatory errors made prior to each task switch were directed toward the choices associated with the upcoming task or tasks in the sequence. The anticipatory and perseverative errors reflected competing control between the adjacent solutions to each task at the transition point in each session. Together, this type of choice evidence indicates that the sigmoidal pattern seen at the transition point mirrors the amount of competition between the adjacent tasks during the different portions of each session. Anticipatory errors thus represent the intrusion of the next task before a reversal, and perseverative errors reflect the continuing influence of the most recent task after the reversal.

The diagram in Figure 4 shows one way to conceptualize the representations involved in MSR. The physical inputs on each trial are the discriminative visual or spatial stimuli critical to reward on any simple or conditional trial. These stimuli are tied to internal representations of the tasks and their learned solutions for each part of a session. These are symbolized as the separate representations that mediate behavior in Task 1 (e.g., depending on the experiment; red+, left+ or matching-like rules) and mediate behavior in Task 2 (e.g., green+, right+, or oddity-like rules). Successful activation and resolution of this information for each task provides the impetus for a motor action to a potentially correct stimulus. There are other potential external contextual stimulus inputs that could act as switching cues as well. These include the reinforcement of recent choices or external visual or spatial switching cues that could help the animal determine which portion of a session it might be in. In addition, and importantly, there is an internal clock that serves to support timing as a switching cue. This presumably reflects some form of an accumulator that is able to track the elapsed time within a session. Accumulation of time is thought to enter into a short-term memory value that is regularly compared with a learned clock criterion value held in long-term memory (although see Bizo and White [1994] for an alternative model of timing via reinforcement accumulation). The clock criterion value of the switch point is thought to be based on an aggregate of recently experienced temporal durations of reversals from recent sessions. During each session, the timed interval begins at the start of a session and ends once the task reversal occurs. This most recent value is averaged into the values from previous sessions, which form the basis of the criterion clock value. The output of this timing mechanism serves as the endogenous switching cue between the two tasks. The changing amount of activation between these two competing tasks, as mediated by time, is reflected in the sigmoidal switching function of MSR. From this combination of these cues, the internal competition between the two task representations and its resolution are at the core of MSR and its implications for understanding animal behavior.

Can we better characterize the properties of this internal competition between the representational states that determine choice behavior? For instance, does the increased competition between the two tasks near the transition change how quickly animals respond across a session as measured by reaction time (RT)? RT could differ over the session if, for example, the increased competition near the reversal resulted in longer choice times. This lag could be due to a low level of discriminability or increased competition between the values of the two choice responses as compared with their stronger values at the session’s endpoints. Another possibility is that the simultaneous activation of both tasks near the reversal might result in faster responding on these trials. Here they might only need to encounter a single stimulus to make a choice as either stimulus might rapidly activate an independent representation, resulting in immediate responses to whichever stimulus was first encountered. Finally, RT may not be affected by the level of competition between the two tasks and therefore remain steady across the session.

Figure 5 shows choice RT data from two MSR experiments involving either a simple or a conditional discrimination. The simple discrimination results are derived from 40 post-acquisition sessions drawn from the experiments described in Rayburn-Reeves et al. (under review), and the conditional discrimination results come from the same sessions as reported in Experiment 1 in Cook and Rosen (2010). The overall pattern in both studies suggests that RT does not vary systematically in the way that behavioral choice changes across trials. Overall, the RT function across sessions from each study is generally flat, except for a slight “warm-up” effect at the beginning of a session. This effect was consistently seen across birds, suggesting that they needed a few trials to get into the more regular pattern of behavior seen for the remainder of the session. This effect may merit further research, as it may indicate a critical period where the processes controlling behavior from previous sessions are reactivated, or it could simply be that some pigeons would benefit from a period of darkness in the chamber prior to the start of the first trial in order to acclimate to the chamber. Further research would be needed to clarify this issue. In any case, the overall pattern of results support the idea that RT is not affected by the level of competition between the two tasks in the same way that behavioral choice is controlled.

However, there were differences in RT among individual birds that might merit more investigation. One conditional bird did seem to slow down in making its test choices just after the reversal. This suggests there might be increased competition at this point. That said, this effect was not observed in the other two birds. Furthermore, the slower RTs for the simple discrimination in comparison to the conditional discrimination task also reflect one bird that was much slower than the others (although the overall function was still flat). Although more research is needed to clarify such details, the processes resolving the competition in the middle of the session seem to not dramatically affect or interfere with choice time. As a general rule, and aside from the warm-up effect, the birds appear to take approximately the same time to respond on a trial independent of the level of competition present between tasks.

Another central question about the competing nature of the two discriminations pivots around the nature of the averaged switching function’s sigmoidal shape. The shape of such averaged functions can stem from two different sources. First, the shape might represent the averaging of a collection of sessions in which the animal makes a single switch from performing Task 1 to Task 2. Presumably, because variability possibly affects both the estimation of accumulated time within a session and also the averaged time to the criterion duration, the temporal location of this “all-or-none” shift varies from session to session. Averaging these variable, single switch points across a number of sessions may produce the gradual transition seen during the midpoint of the session. A second possibility is that the middle part of the function represents a period of ongoing competition between the two tasks. Thus, there is an intermediate transition period where the relative activations of the two task representations overlap enough to cause a large number of alternating responses across the choice stimuli. This results in multiple switches across the tasks in each session. In this possibility, the sigmoidal shape of the average function is a direct representation of the degree of this competition within a single session.

The best way to examine these alternatives is to look at the behavior of individual birds from single sessions. Is a typical session characterized by a single switch from one task to the other task, or is it comprised of a region of multiple switches? Since each session might only contain a single switch or data point at the transition of each session, a large number of sessions from different animals is needed. As mentioned above, we had 40 post-acquisition sessions drawn from different experiments described in Rayburn-Reeves et al. (under review) that tested birds on a spatial MSR task for which we could analyze a considerable amount of single-session data. In these sessions, time was the only switching cue available, although these time-only sessions were embedded within ongoing sessions that had other external switching cues available. We examined only the time-only sessions for each bird to explore the question at hand.

We found evidence indicative of both single-switch and multiple-switch representations. Figure 6 depicts four single-session MSR performances from two of the four pigeons tested in Rayburn-Reeves et al. (under review). We selected these two birds because they each best represent the range of patterns observed. The left panel shows four representative sessions from one bird, #2L, that most frequently and regularly exhibited a single switch from responding Task 1 to Task 2 within a session. In the top leftmost panel, this bird made 42 consecutive choice responses to Task 1, followed by 38 successive choice responses to Task 2. This single switch sometimes came before or after, but always near, the reversal. The other three sessions depicted show a similar behavioral pattern. This pattern was typical for bird #2L. Approximately 67% of the sessions examined contained only one or two switches, suggesting this bird typically made a single action to switch. As a result, this pigeon’s anticipatory and perseverative errors stemmed from either switching too early or too late based on variations in its estimation of elapsed time. This bird’s representations of the two tasks across a session were likely quite separate and generated minimal competitive interference, allowing it to maintain long strings of the first and second task behaviors. This is depicted in the schematic at the bottom of Figure 6, which shows an initial region of choice behavior dominated by choice of Task 1, a smaller intermediate region where there is increasing overlap and competition in the control of responding between the two tasks, and a large terminal region where responses are strongly controlled by the choice of the second task.

The second pigeon, #1B, exhibited a different profile. This is depicted on the right side of Figure 6. In contrast to pigeon #2L, this bird showed a much larger intermediate period where there was considerable competition between the two task representations for control of behavior. In the top rightmost panel, this pigeon responded initially to the correct choice from Task 1, but then suffered a number of separate intrusions from Task 2 as it neared the reversal. After the reversal, there was also an extended period of perseveration from Task 1 before a final and terminal switch to Task 2. As can be seen in the figure, the other example sessions for this pigeon show similar patterns of multiple switches from Task 1 to Task 2 around the reversal. The various switches between responding to the two tasks are clearly temporally related, as they cluster towards the middle of the session. Unlike the first bird, this pigeon made two or fewer switches on only 17% of his sessions. This bird’s capacity to keep the two tasks separated was much poorer than the first bird’s and resulted in considerably more competition, especially in the middle of the session. The schematic at the bottom of Figure 6 captures this increased intermediate phase where the two tasks competed for behavior and smaller sections of strongly controlled behaviors at the session endpoints. The two other birds not shown from this experiment landed somewhere in between the patterns of the first and second bird. One bird from the experiment looked more similar to pigeon #2L, but with a few more tightly grouped switches per session. The other bird had an extended region of intermediate competing choices more like the second pigeon’s pattern, with a clear region of competition that was smaller in range, as it typically began making errors later and ending them earlier in a session.

From these results, it appears that MSR has three broad phases. The first and third are relatively extended segments at the beginning and end of each session where the animal is strongly controlled by either Task 1 or Task 2, respectively. This is determined by the clarity of the switching cue’s value. Between those phases is an intermediate phase where ongoing competition between the two tasks is much higher. The duration of this middle region seems to vary among birds depending on their approach to the task and their ability to segregate the two portions of the session.

We tried to capture the size of this intermediate phase of competition by looking at the trial locations where the first and last errors occurred within a session. This is not a perfect measure. Pigeons make errors that likely do not have much to do with competition (e.g., warm-up, mistaken actions, etc.). Thus, the first and last errors do not precisely mark the onset and offset of competition. Nonetheless, this easily computed measure does provide boundaries on the period over which the representations of each task and the mechanisms controlling responding to Task 1 and 2 appear simultaneously active. For the “single-switch” bird (#2L), the average of the last five sessions of the first error (­average trial number = 32.2) and last error (average trial number = 40.8) occupied a small range (8.6 trials). The other bird (#1B), however, showed a wider range of 37.4 trials between the first error (average trial number = 17.0) and last error (average trial number = 54.4). These values were (59.0 − 33.4 = 25.6) and (50.4 − 22.8 = 27.6) for the two birds (#3M & #4G) that are not shown in Figure 6.

The above pattern indicates that animals in MSR typically engage in medium to long runs of Task 1 responses before beginning to suffer from anticipation interference from Task 2, followed by medium to long runs of Task 2 responses after an intermediate period of perseveration on Task 1. To capture the character of these two extended runs, we next analyzed the starting location within a session of where the longest run of correct responding to Task 1 and Task 2 occurred. We determined for each of the 40 sessions the trial where the longest sequence of correct responding started for both tasks. The resulting pattern is shown in Figure 7. This figure shows the relative frequency distribution of the starting trial position for the longest runs of Task 1 and for those of Task 2. The results show that the pigeons consistently begin responding correctly to Task 1 on Trial 1 or 2 and repeat this response for an extended period of the first half of the session. As can be seen in the figure, only infrequently did these runs start later than Trial 5 over this portion of the session (be observant of the break point on the y-axis). The results for Task 2 are more variable as might be expected given their later location in a session. The distribution of the longest Task 2 runs begins prior to and peaks just after the reversal location, reflecting the regular initiation of the longest Task 2 runs around the reversal. The greater variability in Task 2 run behavior may reflect greater cue competition than is present at the start of the session. At the reversal, animals have to deal with switching cue imprecision, memory for recent changes in outcomes, and competing memories for the long block of reinforced Task 1 choices. These challenges are never shared by the start requirements for Task 1 runs.

Finally, in addition to examining long runs of each choice type, we also examined the distribution of starting locations of short runs. For this purpose, we defined short as runs of three or fewer trials of the same response. These types of runs likely represent the places within a session where relative activation and competition between the two task representations is greatest. Shown in Figure 8 is the relative frequency distribution of the starting location of all short runs recorded from all four birds. Consistent with the earlier analysis of MSR into three phases, this distribution has the expected highest accumulation of short runs in the middle of the session. As a direct reflection of these data, the greatest competition for control of action occurs during the transition between the tasks. This transition produces a greater amount of alternation between competing responses, although the level of competition does seem to vary among animals. There is also an interesting asymmetry in the distribution of short runs on either side of the reversal, with more short runs after the reversal than before. We believe this might come from differences of within- and between-session influences on responding. Before the reversal, the major source of competition comes from temporal anticipation of Task 2. The origins of this competition must come from the birds’ previous experience of when the reversal occurred during past sessions and the imprecision of measuring where they might be, temporally, in the current session. After the reversal, however, the birds seem to have a much greater degree of conflict as exhibited by the increased frequency of short runs. Besides the difficulties of temporal imprecision and their memory of the last several choices, there may be greater conflict created by the extended period of recently reinforced choices of Task 1 within the session. This increase in behavioral variability just after the reversal has been recently documented under changing reinforcement probability conditions, where a change from high to low rates of reinforcement occurs between the two halves of the session (Stahlman & Leising, 2016). Thus, perseveration errors likely have contributions from both within- (i.e., short-term and intermediate-term memories) and between-session (i.e., long-term memories) sources of experience. Anticipatory errors, on the other hand, stem from predominantly between-session representations or long-term memories. With MSR tasks, subjective differences in reinforcement probability might also occur at the endpoints of the session when the birds are good at each task. As these probabilities subjectively converge in the more challenging middle part of the session, pigeons might also begin dynamically exploring or sampling the two alternatives to a greater degree (Dunlap & Stephens, 2012; Lea, McLaren, Dow, & Graft, 2012). In the future, it will be interesting to see if the configurations of other MSR tasks result in the same properties of short and long run locations. For example, with more complex tasks, like conditional discriminations, the contributions of within-session experiences might produce greater levels of response competition and more short runs because of the mixed nature of reinforcement for both stimuli inherent in such discriminations.

Summary

This paper integrates recent studies examining MSR and provides several new analyses of how internal and external sources of information compete for control of responding across a session. The MSR procedure is an excellent preparation for better understanding how animals organize time and order their sequential behavior, especially given the procedural simplicity of MSR. Animals in these tasks need to use and integrate two possibly different or independent sets of cues. One set is the traditional discriminative spatial and visual cues that have been regularly studied in discrimination learning settings for many years (Mackintosh, 1974; Shettleworth, 1998; Thorndike, 1898). The interesting twist in MSR is the introduction of the switching cue. This additional, critical cue allows the animals to emergently organize their behavior across a session to solve the competing demands of the reversed discrimination task. For pigeons, the results seem to indicate that temporal cues are the primary source of information for switching between the competing tasks, with use of this cue moderated by other factors like ITI length, stimulus dimension, and type of apparatus. Although several newer studies have illuminated how changes in task demands can modulate the degree of this temporal control (Daniel et al., 2015; McMillan et al., 2014; McMillan & Roberts, 2012; McMillan, Sturdy, Pisklak, Spetch, 2016; Rayburn-Reeves, Laude, et al., 2013; Rayburn-Reeves et al., under review), it is unclear why time has such a powerful influence on this species. Humans, and perhaps all mammals, appear to be far less influenced by time and more likely to attend to recent outcomes in guiding their switching behavior (Rayburn-Reeves et al., 2011; Rayburn-Reeves, Stagner, et al., 2013; A. P. Smith et al., 2016).

It is possible that this difference in performance on MSR reflects differences in the relative contribution of rule-governed and associative learning mechanisms between pigeons and humans and again, possibly other mammals (e.g., Ashby, Alfonso-Reese, Turken, & Waldron, 1998; Daw, Niv, & Dayan, 2005). In tests using Ashby et al.’s (1998) diagnostic procedures for isolating the contributions of these separate mechanisms, pigeons appear to rely on associative mechanisms in settings where humans easily use rule-based learning (J. D. Smith et al., 2011; J. D. Smith et al., 2012). If that is the case, the differences in the efficacy of different switching cues may reflect an extension of this division in MSR. Perhaps pigeons can only solve the relations between the two tasks using associative-based learning mechanisms that rely heavily on timing. Humans, on the other hand, can rapidly pick up on the rule-based organization of the task and use more executive functioning or rule-based mechanisms to solve this problem. It would be informative to test these different species with organizations of the task that would favor associative mechanisms over exclusive attention to rule-based information. One direction for future research will be to test various species to determine whether there are differences in how other species or classes of animals approach and solve this task. In this respect, triangulating research from behavioral ecology and neuroscience will likely provide a deeper understanding and basis for predicting various sources of information that regulate animal behavior over time. Understanding the differences in the natural ecology of various species, as well as assessing converging physiological changes in cognitive processing across species should provide a more complete picture of cue use in dynamic environments.

Another essential direction for future research stems from the different labels we have used to distinguish discriminative and switching cues. While useful for the purposes of presentation, analysis, and discussion, it is not clear that they are functionally different from the animal’s point of view. Consider time as a switching cue. Although time perception mediates how pigeons organize and partition their successful choice behaviors among the competing tasks, there are at least two broad classes of alternatives for thinking about its processing and contribution to the task. The first is that time is just another type of discriminative cue that is part of the entire complex of cues that determine each trial’s response. In this line of thinking, time acts as a discriminative cue to determine responding the same way color or location does. That is, the time-based switching cue is not fundamentally different in its role from other cues. Based on this theory, one could build a simple neural net model that could use time as an input along with the regular discriminative roles for spatial or visual inputs. The value and weighting of all these cues would then be calculated using the same associative rules. As a result, time would just be part of the vector of cues that determine momentary responding. Thus, the animal learns associatively to do the different behaviors at the right time depending on the input. Daniel Brooks, a post-doctoral fellow working in the Cook lab, has built such a model and it can readily produce the standard switching function seen in MSR (Brooks, personal communication).

The major alternative to the associative model of momentary choice is that the switching cue serves a hierarchical or modulatory function, acting more like a context cue than a discriminative cue (Cooper & Shallice, 2000; Monsell, 2003). In this case, the switching cue provides a context or occasion setter that helps the animal modulate use of one or the other task representations (Holland, 1992). Bouton (2007) has suggested that context-dependent shifts in behavior, similar to the ones considered here, are modulated by such additional inhibitory modifications. In the present case, the animals may well know how to independently perform both Task 1 and Task 2, but time or other switching cues serve to determine which specific behavior is expressed. At the moment, it is not possible to distinguish between these two broad classes of explanations. Whether switching cues in MSR are part of the associative complex that determines momentary responding across a session or instead they serve a modulatory, hierarchical, or contextual role in resolving different representations is another important research direction.

Even with these open questions, MSR remains an exciting new tool for studying how different learned behaviors, representations, or brain states compete to control behavior. In this vein, MSR has properties that are shared with the myriad of cue competition studies in Pavlovian settings, such as the effects of overshadowing and blocking. One advantage of MSR is that it allows repeated testing of the same competitive relation over many sessions rather than looking at the accumulation of sequential stages of training that regularly occurs in cue competition studies. It provides new and better opportunities for the dissection of the simultaneous activation of alternatively learned behaviors and, on a larger scale, competing brain states. How the brain resolves and organizes such competing states to produce a singular stream of actions is a fundamental question and one that is in need of more investigation (e.g., Daw et al., 2005; Dennett & Kinsbourne, 1992).

It is evident from the analyses considered above that there is an intermediate period of variable task competition in MSR that is regularly resolved by the animals. These resolution processes seem to cause no greater increase in processing time based on the amount of competition or level of task complexity, at least as measured by choice RT. Still further, these processes always seem to converge on selecting one of the two learned outcomes. The pigeons do not engage in other types of behaviors (i.e., guessing). Theoretically speaking, one of the most basic and valuable mechanisms that we use for the resolution among competing behaviors is associative strength (e.g., Pearce & Bouton, 2001; Rescorla & Wagner, 1972; Spence, 1936). The strongest value, association, or state at any one moment wins. This has and will continue to serve us well. It is almost certain that during the intermediate phase of MSR, the birds do not always “choose” the best alternative with the greatest overall strength (at least as computed across sessions). Instead, momentary within-session influences aside from associative strength must be causing the birds to select less “optimal” behaviors. MSR thus provides a new vehicle for dissecting such long-term and short-term influences as they can be regularly and repeatedly produced in this setting. As a result, we can analyze in detail why and how animals make the specific choices they do at a particular moment in time. This new capacity to regularly produce competition between different task activations makes MSR a powerful tool for helping us better understand how animals organize and select their ongoing behavior.

References

Andreasen, N. C., Paradiso, S., & O’Leary, D. S. (1998). “Cognitive dysmetria” as an integrative theory of schizophrenia: A dysfunction in cortical-subcortical-cerebellar circuitry? Schizophrenia Bulletin, 24, 203–218. doi:10.1093/oxfordjournals.schbul.a033321

Antle, M. C., & Silver, R. (2009). Neural basis of timing and anticipatory behaviors. The European Journal of Neuroscience, 30, 1643–1649. doi:10.1111/j.1460-9568.2009.06959.x

Ashby, F. G., Alfonso-Reese, L. A., Turken, A. U., & Waldron, E. M. (1998). A neuropsychological theory of multiple systems in category learning. Psychological Review, 105, 442–481. doi:10.1037/0033-295X.105.3.442

Barkley, R. A. (1997). Attention-deficit/hyperactivity disorder, self-regulation, and time: Toward a more comprehensive theory. Journal of Developmental & Behavioral Pediatrics, 18, 271–279. doi:10.1097/00004703-199708000-00009

Bitterman, M. E. (1965). Phyletic differences in learning. American Psychologist, 20, 396–410. doi:10.1037/h0022328

Bitterman, M. E. (1975). The comparative analysis of learning. Science, 188, 699–709. doi:10.1126/science.188.4189.699

Bizo, L. A., & White, K. G. (1994). The behavioral theory of timing: Reinforcer rate determines pacemaker rate. Journal of the Experimental Analysis of Behavior, 61, 19–33. doi:10.1901/jeab.1994.61-19

Bodily, K. D., Katz, J. S., & Wright, A. A. (2008). Matching-to-sample abstract-concept learning by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 34, 178-184. doi:10.1037/0097-7403.34.1.178

Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & Memory, 11, 485–494. doi:10.1101/lm.78804

Brushfield, A. M., Luu, T. T., Callahan, B. D., & Gilbert, P. E. (2008). A comparison of discrimination and reversal learning for olfactory and visual stimuli in aged rats. Behavioral Neuroscience, 122, 54–62. doi:10.1037/0735-7044.122.1.54

Buhusi, C. V., & Meck, W. H. (2005). What makes us tick? Functional and neural mechanisms of interval timing. Nature Reviews Neuroscience, 6, 755–765. doi:10.1038/nrn1764

Cheng, K., & Roberts, W. A. (1989). Timing multimodal events in pigeons. Journal of the Experimental Analysis of Behavior, 52, 363–376. doi:10.1901/jeab.1989.52-363

Cheng, K., & Roberts, W. A. (1991). Three psychophysical principles of timing in pigeons. Learning & Motivation, 22, 112–128. doi:10.1016/0023-9690(91)90019-5

Cook, R. G., & Rosen, H. A. (2010). Temporal control of internal states in pigeons. Psychonomic Bulletin & Review, 17, 915–922. doi:10.3758/PBR.17.6.915

Cook, R. G., & Wasserman, E. A. (2006). Relational discrimination learning in pigeons. In E. A. Wasserman & T. R. Zentall (Eds.), Comparative Cognition: Experimental Explorations of Animal Intelligence, 307–324. London: Oxford University Press.

Cooper, R., & Shallice, T. (2000). Contention scheduling and the control of routine activities. Cognitive Neuropsychology, 17, 297–338. doi:10.1080/026432900380427

Daniel, T. A., Cook, R. G., & Katz, J. S. (2015). Temporal dynamics of task switching and abstract-concept learning in pigeons. Frontiers in Psychology, 6: 1334. doi:10.3389/fpsyg.2015.01334

Daw, N. D., Niv, Y., & Dayan, P. (2005). Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control. Nature Neuroscience, 8, 1704–1711. doi:10.1038/nn1560

Dennett, D. C., & Kinsbourne, M. (1992). Time and the observer: The where and when of consciousness in the brain. Behavioral and Brain Sciences, 15, 183–247. doi:10.1017/S0140525X00068229

Dudchenko, P. A. (2004). An overview of the tasks used to test working memory in rodents. Neuroscience & Biobehavioral Reviews, 28, 699–709. doi:10.1016/j.neubiorev.2004.09.002

Dunlap, A. S., & Stephens, D. W. (2012). Tracking a changing environment: Optimal sampling, adaptive memory and overnight effects. Behavioural Processes, 89, 86–94. doi:10.1016/j.beproc.2011.10.005

Egger, M. D., & Miller, N. E. (1962). Secondary reinforcement in rats as a function of information value and reliability of the stimulus. Journal of Experimental Psychology, 64, 97–104. doi:10.1037/h0040364

Evenden, J. L., & Robbins, T. W. (1984). Win-stay behaviour in the rat. The Quarterly Journal of Experimental Psychology Section B: Comparative and Physiological Psychology, 36, 1–26. doi:10.1080/14640748408402190

Holland, P. C. (1992). Occasion setting in Pavlovian conditioning. The Psychology of Learning and Motivation, 28, 69–125. doi:10.1016/S0079-7421(08)60488-0

Katz, J. S., & Wright, A. A. (2006). Same/different abstract-concept learning by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 32, 80–86. doi:10.1037/0097-7403.32.1.80

Lashley, K. S. (1951). The problem of serial order in behavior. In L. A. Jeffress (Ed.), Cerebral Mechanisms in Behavior, 112–136. New York: John Wiley & Sons.

Laude, J. R., Stagner, J. P., Rayburn-Reeves, R. M., & Zentall, T. R. (2014). Midsession reversals with pigeons: Visual versus spatial discriminations and the intertrial interval. Learning & Behavior, 42, 40–46. doi:10.3758/s13420-013-0122-x

Lea, S. E. G., McLaren, I. P. L., Dow, S. M., & Graft, D. A. (2012). The cognitive mechanisms of optimal sampling. Behavioural Processes, 89, 77–85. doi:10.1016/j.beproc.2011.10.004

Levine, M. (1975). A Cognitive Theory of Learning: Research on Hypothesis Testing. Oxford, UK: Lawrence Erlbaum.

Lewis, P. A., & Miall, R. C. (2006). Remembering the time: A continuous clock. Trends in Cognitive Science, 10, 401–406. doi:10.1016/j.tics.2006.07.006

Mackintosh, N. J. (1974). The Psychology of Animal Learning. New York: Academic Press.

Mackintosh, N. J. (1975). A theory of attention: variations in the associability of stimuli with reinforcement. Psychological Review, 82, 276–298. doi:10.1037/h0076778

Mackintosh, N. J., & Cauty, A. (1971). Spatial reversal learning in rats, pigeons, and goldfish. Psychonomic Science, 22, 281–282. doi:10.3758/BF03335956

Martin, B., Giersch, A., Huron, C., & van Wassenhove, V. (2013). Temporal event structure and timing in schizophrenia: Preserved binding in a longer “now”. Neuropsychologia, 51, 358–371. doi:10.1016/j.neuropsychologia.2012.07.002

Matell, M. S., & Meck, W. H. (2000). Neuropsychological mechanisms of interval timing behavior. Bioessays, 22, 94–103. doi:10.1002/(SICI)1521-1878(200001)22:1<94::AID-BIES14>3.0.CO;2-E

McMillan, N., Kirk, C. R., & Roberts, W. A. (2014). Pigeon (Columba livia) and rat (Rattus norvegicus) performance in the midsession reversal procedure depends upon cue dimensionality. Journal of Comparative Psychology, 128, 357–366. doi:10.1037/a0036562

McMillan, N., & Roberts, W. A. (2012). Pigeons make errors as a result of interval timing in a visual, but not a visual-spatial, midsession reversal task. Journal of Experimental Psychology: Animal Behavior Processes, 38, 440. doi:10.1037/a0030192

McMillan, N., & Roberts, W. A. (2013). Interval timing under variations in the relative validity of temporal cues. Journal of Experimental Psychology: Animal Behavior Processes, 39, 334–341. doi:10.1037/a0032470

McMillan, N., & Roberts, W. A. (2015). A three-stimulus midsession reversal task in pigeons with visual and spatial discriminative stimuli. Animal Cognition, 18, 373–383. doi:10.1007/s10071-014-0808-2

McMillan, N., Sturdy, C. B., Pisklak, J. M., & Spetch, M. L. (2016). Pigeons perform poorly on a midsession reversal task without rigid temporal regularity. Animal Cognition Epub, 1–5.. doi:10.1007/s10071-016-0962-9

McMillan, N., Sturdy, C. B., & Spetch, M. L. (2015). When is a choice not a choice? Pigeons fail to inhibit incorrect responses on a go/no-go midsession reversal task. Journal of Experimental Psychology: Animal Learning and Cognition, 41, 255–265. doi:10.1037/xan0000058

Meck, W. H., & Church, R. M. (1983). A mode control model of counting and timing processes. Journal of Experimental Psychology: Animal Behavior Processes, 9, 320–334. doi:10.1037/0097-7403.9.3.320

Miller, R. R., & Barnet, R. C. (1993). The role of time in elementary associations. Current Directions in Psychological Science, 2, 106–111. doi:10.1111/1467-8721.ep10772577

Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7, 134–140. doi:10.1016/S1364-6613(03)00028-7

Pearce, J. M., & Bouton, M. E. (2001). Theories of associative learning in animals. Annual Review of Psychology, 52, 111–139. doi:10.1146/annurev.psych.52.1.111

Rayburn-Reeves, R. M., Laude, J. R., & Zentall, T. R. (2013). Pigeons show near-optimal win-stay/lose-shift performance on a simultaneous-discrimination, midsession reversal task with short intertrial intervals. Behavioural Processes, 92, 65–70. doi:10.1016/j.beproc.2012.10.011

Rayburn-Reeves, R. M., Molet, M., & Zentall, T. R. (2011). Simultaneous discrimination reversal learning in pigeons and humans: Anticipatory and perseverative errors. Learning & Behavior, 39, 125–137. doi:10.3758/s13420-010-0011-5

Rayburn-Reeves, R. M., Qadri, M. A. J., Brooks, D. I., Keller, A. M., & Cook, R. G. (under review). Dynamic cue use by pigeons in a midsession reversal task. Behavioural Processes.

Rayburn–Reeves, R. M., Stagner, J. P., Kirk, C. R., & Zentall, T. R. (2013). Reversal learning in rats (Rattus norvegicus) and pigeons (Columba livia): Qualitative differences in behavioral flexibility. Journal of Comparative Psychology, 127, 202–211. doi:10.1037/a0026311

Rayburn-Reeves, R. M., & Zentall, T. R. (2013). Pigeons use of cues in a repeated 5-trial sequence, single reversal task. Learning & Behavior, 41, 138–147. doi:10.3758/s13420-012-0091-5

Rescorla, R., & Wagner, A. R. (1972). A theory of Pavlovian conditioning. Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical Conditioning II: Current Research and Theory. New York: Appleton-Century-Crofts.

Restle, F. (1962). The selection of strategies in cue learning. Psychological Review, 69, 329–343. doi:10.1037/h0044672

Roberts, S. (1981). Isolation of an internal clock. Journal of Experimental Psychology: Animal Behavior Processes, 7, 242–268. doi:10.1037/0097-7403.7.3.242

Shettleworth, S. J. (1998). Cognition, Evolution, and Behavior. New York: Oxford University Press.

Slotnick, B. (1984). Olfactory stimulus control in the rat. Chemical Senses, 9, 157–165. doi:10.1093/chemse/9.2.157

Smith, A. P., Pattison, K. F., & Zentall, T. R. (2016). Rats’ midsession reversal performance: The nature of the response. Learning & Behavior, 44, 49–58. doi:10.3758/s13420-015-0189-7

Smith, J. D., Ashby, F. G., Berg, M. E., Murphy, M. S., Spiering, B., Cook, R. G., & Grace, R. C. (2011). Pigeons’ categorization may be exclusively nonanalytic. Psychonomic Bulletin & Review, 18, 414–421. doi:10.3758/s13423-010-0047-8

Smith, J. D., Berg, M. E., Cook, R. G., Murphy, M. S., Crossley, M. J., Boomer, J., . . .  Grace, R. C. (2012). Implicit and explicit categorization: A tale of four species. Neuroscience & Biobehavioral Reviews, 36, 2355–2369. doi:10.1016/j.neubiorev.2012.09.003

Spence, K. W. (1936). The nature of discrimination learning in animals. Psychological Review, 43, 427–449. doi:10.1037/h0056975

Staddon, J., & Higa, J. (1999). Time and memory: Towards a pacemaker-free theory of interval timing. Journal of the Experimental Analysis of Behavior, 71, 215–251. doi:10.1901/jeab.1999.71-215

Stahlman, W. D., & Leising, K. J. (2016). Midsession shifts in reward probability and the control of behavioral variability. Learning & Behavior, 1-13. doi:10.3758/s13420-016-0211-8

Thorndike, E. L. (1898). Animal intelligence: An experimental study of the associative processes in animals. Psychological Monographs: General and Applied. doi:10.1037/10780-000

Tinklepaugh, O. L. (1928). An experimental study of representative factors in monkeys. Journal of Comparative Psychology, 8, 197–236. doi:10.1037/h0075798

Vicario, C. M. (2013). Cognitively controlled timing and executive functions develop in parallel? A glimpse on childhood research. Frontiers in Behavioral Neuroscience, 7, 1–4. doi:10.3389/fnbeh.2013.00146

Wasserman, E. A., Kiedinger, R. E., & Bhatt, R. S. (1988). Conceptual behavior in pigeons: Categories, subcategories, and pseudocategories. Journal of Experimental Psychology: Animal Behavior Processes, 14, 235–246. doi:10.1037//0097-7403.14.3.235

Wilkie, D. M. (1995). Time-place learning. Current Directions in Psychological Science, 4, 85–89. doi:10.1111/1467-8721.ep10772332

Wilkie, D. M., & Willson, R. J. (1992). Time-place learning by pigeons, Columba livia. Journal of the Experimental Analysis of Behavior, 57, 145–158. doi:10.1901/jeab.1992.57-145

Wright, A. A., Cook, R. G., Rivera, J. J., Sands, S. F., & Delius, J. D. (1988). Concept learning by pigeons: Matching-to-sample with trial-unique video picture stimuli. Animal Learning & Behavior, 16, 436–444. doi:10.3758/Bf03209384

What Can Nest-Building Birds Teach Us?

Alexis J. Breen, Lauren M. Guillette, and Susan D. Healy
School of Biology, University of St Andrews, UK

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

For many years nest building in birds has been considered a remarkable behaviour. Perhaps just as remarkable is the public and scholarly consensus that bird nests are achieved by instinct alone. Here we take the opportunity to review nearly 150 years of observational and experimental data on avian nest building. As a result we find that instinct alone is insufficient to explain the data: birds use information they gather themselves and from other individuals to make nest-building decisions. Importantly, these data confirm that learning plays a significant role in a variety of nest-building decisions. We outline, then, the multiplicity of ways in which learning (e.g., imprinting, associative learning, social learning) might act to affect nest building and how these might help to explain the diversity both of nest-building behaviour and in the resulting structure. As a consequence, we contend that nest building is a much under-investigated behaviour that holds promise both for determining a variety of roles for learning in that behaviour as well as a new model system for examining brain-behaviour relationships.

Keywords: nest building, learning, cognition, comparative cognition, birds

Author Note: Correspondence concerning this article should be addressed to Alexis J. Breen at ab297@st-andrews.ac.uk, Lauren M. Guillette at lmg4@st-andrews.ac.uk, and ­Susan D. Healy at sdh11@st-andrews.ac.uk.

Acknowledgments: We thank the School of Biology at the University of St Andrews for funding (AJB) and the BBSRC (LMG: BB/M013944/1 and SDH: BB/I019634/1). We would also like to thank three anonymous reviewers for providing useful and insightful comments on an earlier version of the manuscript.

---

Introduction

The notion that learning might be involved in nest building was not lost on inquiring minds in the 19th century, including that of Alfred Russell Wallace (1823–1913). He may have been the first to argue that nest building in birds was not due entirely to instinct: “[t]his point [. . .] is always assumed without proof, and even against proof, for what facts there are, are opposed to it” (Wallace, 1867). Ironically, despite the passing of nearly 150 years, Wallace’s statement is as relevant today with regard to both the popular and scientific opinion as it was in his time. Indeed, at present nest building in birds is a behaviour considered to reflect nothing more than genes (Bluff, Weir, Rutz, Wimpenny, & Kacelnik, 2007; Hansell & Ruxton, 2008; Raby & Clayton, 2009; Seed & Byrne, 2010; Zentall, 2006). This view, however, continues to be based largely on untested assumptions, as there are very few data on how birds ‘know’ what type of nest to build.

Helpfully, there are other aspects of birds’ nest building that are quite well described. Indeed, several excellent bodies of work provide a broad overview and thorough discussion of this key component of avian reproductive biology (e.g., Collias & Collias, 1984; Deeming & Reynolds, 2015; Hansell, 2000). In brief, there are considerable data on the inter- and intraspecific variation in nest-site selection, composition, morphology, and building techniques. This wealth of data reveal an abundance of diversity in all these features of building: (a) birds build nests in an extraordinary range of different sites (Hansell, 2000); (b) where the individual builders of most species are known, nest building is not necessarily restricted to one of the sexes and contribution by one or both partners varies considerably from species to species (Collias & Collias, 1984; Hansell, 2000); and (c) nest material composition is highly variable, encompassing a broad range of both natural (e.g., grasses, leaves, twigs, sticks, mud, mosses, lichens, feathers, and/or arthropod silk) and man-made (e.g., cigarette butts, polypropylene string, and bits of fence wire) materials (Antczak et al., 2010; Hansell, 2000; Nicolakakis & Lefebvre, 2000; Suárez-Rodríguez, López-Rull, & Garcia, 2013).

The techniques with which birds build their nests have also been described in some detail and are also various: they range from the sculpting of burrows or cavities from substrate excavation, through the moulding of mud or salivary mucus by vibrating head and/or shaping breast and feet movements, the piling up of materials where subsequent bill manipulations, coupled with side-to-side shaking movements, may be made in order to entangle or intertwine nest components, to the weaving of hanging nest baskets using intricate tuck, looping, interlocking, winding, and knotting bill-made stitches to fasten and secure grassy materials (Collias & Collias, 1984; Hansell, 2000).

At the most ‘basic’ level, nest-building efforts may culminate in a seemingly haphazard arrangement of piled-up materials, such that a mound or plate (located on or above the ground, respectively) of sticks and twigs provides birds with a nesting substrate, whereas more intricate nests incorporate distinct egg-holding cavities and/or overhead, protective roofing (Hansell, 2000). Lastly, yet most pertinent, nest building is key to (most) birds’ reproductive success. Indeed, it has been argued that birds’ continued radiation following the end of the Cretaceous period may have been due to the provision of a protected area for egg incubation, enabling those species to buffer their offspring from a rapidly changing environment (Deeming & Ferguson, 1989).

The building process, then, begins with nest-site selection and is followed by the appropriate choice of available materials from the environment, which individuals then manipulate and/or modify into the structure we call a nest. The builders may subsequently continue to modify that structure even when it contains eggs, chicks, or an incubating parent. For all this variability in all aspects of nest building, a question that has been rarely asked is whether one or more parts of the nest-building process involve decision making. Given that birds have long been models for investigating learning and memory and for examining brain-behaviour relationships, it is curious that such a familiar avian behaviour has been so little explored. It seems timely, then, to review the available data, to examine the interpretations reached from those data, and to suggest directions for future research.

We begin with a historical backdrop to the current view that birds’ nest building is entirely innate (Table 1). We next present observational and experimental data collected from both the field and the laboratory to demonstrate that nest building in birds: (a) is not fixed; (b) is experience-­dependent; and (c) reflects inter- and intraspecific information use (Tables 2 and 3). In doing so, these data offer belated confirmation of Wallace’s belief that, indeed, learning (defined here as a change in an individual’s behaviour in response to previous experience with a given, or similar, stimulus; Domjan, 2014) plays a dominant role in birds’ nest building. Coupling this evidence with the enormous diversity in the structures produced, we suggest that nest building in birds might provide a useful comparative model behaviour system because its study allows for both the experimental examination of the processes of learning and memory as well as of the underlying neurobiology.

A Brief History

One form of learning that is well studied in birds is imprinting (Bateson, 1966, 2015). Imprinting is the mechanism whereby an animal acquires a preference for a particular stimulus due to exposure to that stimulus during a sensitive, typically juvenile, phase (Bateson, 1966). The possibility that young birds may learn what nest they are to build as adults through imprinting on some feature(s) of their natal nest experience was the impetus behind a series of early experiments (Table 1, Section 1) addressed at determining whether the nest a bird builds as an adult bears any relationship to that from which they fledged (Sargent, 1965). Sargent manipulated three distinct features of the young birds’ natal nest: (a) the colour of the nest material he provided (brown, green, or red); (b) the form of the structure in which the birds could build (an enclosed or open nest-cup); and (c) the location of the nestbox/cup (inside the cage or in a cage extension). Sargent reared zebra finches Taeniopygia guttata in an experimental design with these three components cross-factored and, when the males (they are the builder in this species) built their first nest, Sargent tested their preferences for material colour, structure, and location. Natal nest colour did not appear to influence adult colour preference as all the males strongly preferred to build with brown nest material, regardless of the colour of the nest in which they had been reared. Similarly, natal nest experience did not appear to affect preference for the kind of structure in which to build: all males chose to build their nests in open nest-cups. Some evidence of imprinting on the natal nest, however, came from the males’ preference for building their nest in the location that matched the one in which they were reared. Taken together, Sargent concluded that innate predispositions played the much larger role in his birds’ nest-building decision making. To our knowledge, this study is the only one in which the role of imprinting in birds’ nest building has been investigated to any significant extent.

We find a few more data on the effect of early-life experience from a handful of deprivation experiments (Table 1, Section 1) where the typical method has been to deny juvenile birds access to building materials until the birds are sexually mature. These inexperienced adults are then presented with appropriate material (and a mate if required) in order to determine (a) whether the bird(s) can build a nest at all and, if yes, (b) the form the nest takes.

Such deprivation experiments show that a lack of early experience with nest-building material appears to vary in its effect on subsequent nest-building efforts, depending on the species tested (Table 1, Section 1). Hand-reared American robins Turdus migratorius, for example, could not construct a robust nest, even after repeated attempts, although they could line the inside of a nest-cup successfully when provided with such a cup (Scott, 1902). Similarly, a pair of hand-raised rose-breasted grosbeaks Pheucticus ludovicianus failed to construct a nest but they could line an artificial container when one was provided (Scott, 1904). In contrast, domesticated canaries Serinus canaria ­domestica could construct a complete nest even when deprived entirely of appropriate nest materials when young (Hinde & Warren, 1959; Verlaine, 1934). It may be that these canary data are a major source for the opinion that nest building is essentially ‘hardwired’. Even with the experience-deprived canaries’ apparent success, however, behaviours observed during their nest building were described as atypical and the resultant nests appeared ‘poorer’ in quality (Verlaine, 1934).

That experience is important to first-time nest builders is a view supported by detailed observations on the ontogeny of building behaviour in aviary-housed village weaverbirds Ploceus cucullatus. These birds benefitted from practicing their motor skills through direct manipulation and/or handling of nest material (Collias & Collias, 1964; Table 1, Section 1): when provided with access to nest material, juvenile weaverbirds improved at collecting material from plants (e.g., where to tear the leaf, in what direction, and how to perch in order to do so), and they become more proficient at building (by increasing the number of pieces woven within a 3.5 h observation period by 26%). The nests of these first-time builders, however, were characterised as ‘crude’ in comparison to those built by experienced adults. It appears that motor learning then, at least for weaverbirds, is important to nest building.

Collectively, these early data on nest-building behaviour are, however, sparse and suggestive at most: they do not confirm that learning plays a dominant role in nest building by birds but nor do they confirm its irrelevance. The key problem is that there are so very few data. In the meantime, investigations of building behaviour have focussed largely on the invertebrates, work that has lead to the general conclusion that relatively simple rules can be sufficient to explain even apparently complex structures (Hansell, 1984). Building by invertebrates has, unsurprisingly, lead to the assumption that building by birds might have a similar mechanistic basis, and may explain why Wallace’s view has received so little attention. Those invertebrate data may also help to explain why there has been little attempt to determine how a bird ‘knows’ what nest to build: we thought we knew. Recent data from both the field and laboratory on a number of components of bird nest building, however, show that we don’t actually know what we think we do: learning does play a role in nest building.

Learning in Nest-Building Birds: Evidence from the Field

Measuring Heritability

Virtually all behaviour reflects an interaction between genetic endowment, ontogenetic processes, and later ­experience-dependence. One way to assess the extent to which genes and experience may or may not influence birds’ nest-building behaviours is to measure the repeatability of those behaviours over time (i.e., the upper limit of heritability; Boake, 1989; Lessells & Boag, 1987). As solitary weaverbirds build multiple nests over the course of a single breeding season, they provide an opportunity to examine repeatability in the gross morphology (length, width, height) of their nests. If male weaverbirds build nests using a genetic ‘template’ one would expect high inter-male repeatability (R) with regard to nest dimensions. Examination of 93 nests build by 20 individual males (e.g., southern masked Ploceus velatus and village weaverbirds P. cucullatus; ~4 nests/male) showed this was not the case (southern masked weavers: R = 0.21; village weavers: R = 0.07; Walsh, Hansell, Borello, & Healy, 2010; Table 2, Section 1). Moreover, the repeatability of building actions (such as carrying, inserting, and dropping grass) the birds performed during two key phases (initial attachment and ring phase; Figure 1) was variable: in the initial attachment phase none of the measured behaviours were repeatable, whereas in the ring phase, three out of the five behaviours were (Walsh, Hansell, Borello, & Healy, 2011).

That simple rule-based explanations did not necessarily account for the constituent behaviours used by southern masked weavers when building their nests was confirmed by examination of the movements the birds made when building sequential nests (Walsh, Hansell, Borello, & Healy, 2013; Table 2, Section 1). If the birds used either of two possible rules to build a nest: (a) stigmergy (i.e., using predetermined stimulus-response reactions) or (b) stereotypy (i.e., using a set sequence of repetitive actions), one would expect to see a sequential and nonoverlapping progression of discrete, discriminable building phases. Additionally, both of these rule-based behaviours should result in birds completing a nest before beginning the construction of a new one. But the birds did not do this. Rather there was no discernible pattern to the order in which building weavers visited their nests to tidy or to add new material.

A pattern in individual weaverbird nest building, however, has been detected using computer-aided image texture classification (Bailey et al., 2015; Table 2, Section 1): from examination of textural surface patterns, the identity of the builder of 96 village and southern masked weaverbird nests could be identified with varying (up to 81.82%) accuracy (~5 nests/male). As it appears that there is some compositional consistency or ‘signature’ to the final nest structure built by each weaverbird (such as can be assigned to pieces painted by individual artists), it is possible that weaverbirds develop their own ‘style’ of building, which then becomes fixed. If this is the case, then nest building might bear a resemblance to closed-end vocal learning in some songbirds with acquisition of variation in the behaviour limited to a sensitive period (Beecher & Brenowitz, 2005). Further work is required to examine this possibility.

Taken together, these observational data show that the nests built by the same male can vary. They do not, however, necessarily provide evidence for learning. Observational and experimental data from the field on nest-site selection, however, offer more compelling support that birds do use their own experience when making nest-building decisions.

Nest-Site Selection

One of the decisions a builder needs to make is where to build. A variety of evidence shows that this decision depends on a bird’s previous breeding experience (Table 2, Sections 2–7; Lima, 2009). For example, long-term mark and recapture surveys on populations of breeding birds indicate that birds that have an unsuccessful breeding attempt are more likely to build their next nest in a new location (e.g., Northern flickers Colaptes auratus; Doligez, Danchin, Clobert, & Gustafsson, 1999; collared flycatchers Ficedula albicollis; Dow & Fredga, 1983; goldeneyes Bucephala clangula; Fisher & Wiebe, 2006; Table 2, Section 2). Builders might also change the kind of location in which they build. Stonechats Saxicola rubicola, for instance, were more likely to nest in a different vegetation type after suffering nest predation (Greig-Smith, 1982; Table 2, Section 2), whereas Siberian jays Perisoreus infaustus chose to build their next nest in lower, denser vegetation after losing their nest to predation (Eggers, Griesser, Nystrand, & Ekman, 2006; Table 2, Section 3). Egg removal experiments confirm that breeding experience influences an individual’s subsequent nest-site selection: for example, ‘unsuccessful’ orange-breasted sunbirds Anthobaphes violacea dispersed twice as far between nesting attempts as those that fledged chicks (Grégoire & Cherry, 2007; Table 2, Section 2). Similarly, between-year breeding-site fidelity was stronger in American robins and brown thrashers Taxostoma rufum that had not experienced simulated nest predation (Haas, 1998; Table 2, Section 2).

In addition to avoiding a previously unsuccessful nest site, birds can also learn to return to a site in which they have raised young successfully (Table 2, Section 2). Mountain bluebirds Sialia currucoides, for example, chose to nest in one of two nestbox types (painted and unpainted) if they had previously fledged chicks from that box type (Herlugson, 1981). Similarly, cliff-nesting kittiwakes Rissa ­tridactyla will return to nest on cliffs on which they have bred successfully (Danchin, Boulinier, & Massot, 1998; Suryan & Irons, 2001) and spotted antbirds Hylophylax naevioides may reuse extant nests from which they have produced young (Styrsky, Brawn, & Robinson, 2005). These data show that birds can assess and modify their building strategies, in ecological time, in response to environmental variables and outcomes that are associated with nesting success.

There are also observational data to suggest that birds can learn to associate nest success and predation risk (Table 2, Section 3). Between-year breeding-territory fidelity of nesting red-backed shrikes Lanius collurio, for example, changed in relation to the spatial distribution of corvid nest predators (Roos & Pärt, 2004). Similarly, some pairs of ovenbirds Seiurus aurocapilla in northwestern Pennsylvania will nest in atypical breeding habitat (occupying the forest edges) to lower the risk of losing their eggs and chicks to chipmunks, a nest predator (Morton, 2005).

Whether or not birds modify their nest-building decisions in response to individual predator species has not been explicitly tested (but see Chen, Liu, Yan, & An, 2011; Table 2, Section 3). There are observational data to show, however, that at least some birds can discriminate between different sources of nest failure: pinyon jays Gymnorhinus cyanocephalus, for example, rebuild in cooler (less exposed) nest sites after nest depredation but will rebuild in warmer (more exposed) sites when a nest fails due to heavy snowfall (Marzluff, 1988; Table 2, Section 3). These data suggest that these decisions are amenable to experimental manipulations of predator presentations (e.g., Hakkarainen, Ilmonen, Koivunen, & Korpimäki, 2001; Peluc, Sillet, Rotenberry, & Ghalambor, 2008; Table 2, Section 3).

A bird might also respond to nest predation by modifying the structure of its nest (Table 2, Section 4). For example, house wrens Troglodytes aedon appear to change the structure of their nest in response to an increase in perceived threat exposure: when building in a nestbox with a large (versus a small) entrance hole, these birds construct a far taller stick wall between their nest cup and the entrance hole, which in turn makes access, at least by human hands, more difficult (Stanback et al., 2013). Rather than using sticks, cavity-nesting rock wrens Salpinctes obsoletus vary the number of stones they place around the cavity entrance depending on the size of the entrance (Merola, 1995; Smith, 1904; Warning & Benedict, 2015; Figure 2). This structural modification of rock wrens’ nesting environment may itself reduce predator access, but it also appears to amplify the sound of a simulated predator approach, potentially aiding in predator detection (Warning & Benedict, 2015). Ground nesting black larks Melanocorypha yeltoniensis also seem to build an anti-predator defense for their nest: artificial nests not surrounded by the typical dung ‘pavement’ of faecal deposits collected from nearby domestic livestock were more likely to be trampled by cows than were those surrounded by a dung pavement (Moiseev, 1980; Fijen et al., 2015). Whether or not birds facultatively change the structure of the nest itself in response to such environmental variables is not yet clear.

Social Learning and Nest-Site Selection

Although it has rarely been tested, birds may learn about nest building from watching the choices of other individuals, that is, social learning (Heyes, 1994). For example, observations of nest-site selection by yellow wagtails Motacilla flava (Schiermann, 1939) and redheads Aythya Americana (­Hochbaum, 1955) suggested, at least to these authors, that the appearance of new nesting ‘traditions’ (building in shrubs or on dry land, respectively) were the result of these birds learning from others’ nest-building decisions (Table 1, Section 2).

Indeed, kittiwakes Rissa tridactyla (Figure 3) appear to favour information about the relative success or failure of neighbours’ nesting attempts when choosing where to build their nest over their own experience: those birds that experienced simulated nest predation (egg removal) tended to return to the same nesting territory if their conspecific neighbours had raised young successfully (Boulinier, McCoy, Yoccoz, Gasparini, & Tveraa, 2008; Table 2, Section 5). Prothonotary warblers Protonotaria citrea that lose their nest to predation also have stronger site fidelity when their neighbours were successful nesters than when their neighbours were not (Hoover, 2003). Some species take information from their neighbours very seriously indeed: unsuccessful nesting piping plovers Charadrius melodus melodus with unsuccessful neighbours built their next nest more than 34 times further away than did those plovers with neighbours that raised multiple offspring (Rioux, Amirault-Langlais, & Shaffer, 2011).

Migratory pied Ficedula hypoleuca and collared flycatchers F. albicollis pay attention not only to the nest-box type used by heterospecific tit species, but also to the relative success of the tits, and were less likely to copy the tits’ apparent nestbox choice (by not moving into a nestbox assigned the same arbitrary, geometric symbol) if the pair of tits had laid a small clutch (Seppänen & Forsman, 2007; Seppänen, Forsman, Mönkkönen, Krams, & Salmi, 2011; Table 2, Section 6). It seems that the flycatchers can assess the nesting success of the tits based on some assessment of the number or size of eggs/chicks in the tit nestbox. It is not yet, however, clear by which mechanism the flycatchers ‘count’ the eggs/chicks they can see when they visit and peek into tit nestboxes (Forsman & Thomson, 2008). As an aside, the nesting tits appear to attempt to occlude the peeking flycatchers’ view of their young: these birds, which typically cover their eggs with nest material (Haftorn & Slagsvold, 1995), bring more material to cover experimentally exposed eggs in response to a flycatcher call than that of a noncompeting heterospecific (Loukola, Laaksonen, Seppänen, & Forsman, 2014; Table 2, Section 7).

Birds can, then, use different kinds of interspecific information to estimate nest-site quality when choosing where to locate their own nest. Whether or not birds use social information in other aspects of nest building is still to be addressed.

Social Learning and Building

There are a number of features of nest building that might be learned from conspecifics such as appropriate materials, effective handling techniques, and the structure to be achieved. And we might expect to see evidence of the influence of building behaviours (e.g., material choice) of more experienced and familiar individuals, particularly on the decisions of first-time builders. There is tantalizing evidence that at least some birds appear to “follow the fashion” of others’ material choice: over a 10-year period, Baltimore orioles Icterus galbula provided with both natural (plant) and unnatural (coloured yarn) nest materials increasingly selected pieces of coloured yarn with which to build their nests and the material preferences of all orioles converged to choosing only white yarn, apparently following the example set by a conspecific who would build with nothing else (Williams, 1934; Table 2, Section 8). It is not clear, however, whether first-time builders were more likely to incorporate the coloured yarn than were other birds or whether the birds that chose the white yarn were more likely to have hatched into a nest containing white yarn. The incorporation of coloured yarn into nests by neighbouring kingbirds Tyrannus spp., American robins, and cedar waxwings Bombycilla cedrorum beginning in the same year the orioles’ exclusively chose to use white material tells us little about why these birds chose this material but suggests that offering coloured yarn may prove a useful experimental manipulation with which to examine both interspecific as well as intraspecific transmission of material choices.

Differences in aromatic plant species composition in blue tit nests across two distinct, but environmentally similar, study plots also suggest a role for social transmission for material choice (Mennerat, Perret, & Lambrechts, 2009; Table 2, Section 8). Aside from these anecdotal reports, however, there is, as yet, no experimental evidence that birds learn how to build, which materials to use, or what structure they should build from observing others.

Nest Design

The question of how a bird knows what structure to build and whether asocial experience might shape that product remains equally untested, at least in the wild (but see below for recent laboratory data). But as there is evidence that variation in nest structure or materials can lead to variation in reproductive success, it would appear that there is potential for builders to learn how to change the structure of their next nest. Black lark chicks, for example, reared in nests where females had incorporated a greater amount of surrounding livestock dung had higher tarsus growth rates than did chicks reared in nests surrounded by less dung (Fijen et al., 2015). Similarly, removal or addition of feathers from nests of tree swallows Tachycineta bicolor either decreased or increased chick growth rate, respectively (Winkler, 1993; Dawson, O’Brien, & Mlynowski, 2011).

If material choice enhances a bird’s reproductive success, then that bird might change the construction of its next nest by, for example, altering the number of feathers (in the case of the swallows) or quantity of faecal matter (in the case of the larks) based on prior breeding success or failure. They might also learn to select or avoid material(s) based on their structural suitability: biomechanical analysis shows that female blackbirds Turdus merula consistently allocate stronger, thicker, and more rigid material elements differentially within the nest structure, although it is not clear that they have learned to do this (Biddle, Deeming, & Goodman, 2015).

Learning in Nest-Building Birds: Evidence from the Laboratory

Model Species

Experimental investigation into the cognition of nest building has been focussed on nest building in male zebra finches. Although more widely recognized for their pivotal role in investigations related to birdsong (e.g., neurological, developmental, functional), these birds are a useful laboratory model species for investigating a range of behaviours and cognition (Healy, Haggis, & Clayton, 2010) for several key reasons: zebra finches (a) readily breed and build nests under laboratory conditions with a variety of materials (Figure 4); (b) have short (90 day) generation times; and (c) immediately recommence nest building when their young have fledged.

Material Choice

Although it seems plausible that birds may benefit from choosing certain materials for building their nest, this assumes that birds ‘know’ what constitutes a ‘good’ material choice. But it does appear that zebra finches can, at least, assess the quality of building materials (whether by breeding experience and/or sensitivity to material properties) and then respond to them selectively (Table 3, Sections 1–2). For example, males that built a nest using material of a colour they did not prefer but from which they raised chicks, subsequently preferred that colour of material for building their next nest, while males that were unsuccessful in raising chicks in a nest built with material of the preferred colour did not switch preference for material colour (Muth & Healy, 2011; Table 3, Section 1). Why this latter group of birds did not switch their preference is not yet clear. But it does appear that there is complexity and subtlety to what birds learn from their own building experience. The cause of initial colour preferences requires further investigation, although birds do not appear to prefer the colour of nest material from which they fledged (Muth & Healy, 2012; Muth, Steele, & Healy, 2013; Sargent, 1965; Table 3, Section 2).

Table 3

Like animal tool users (e.g., Manrique, Sabbatini, Call, & Visalberghi, 2011; Sanz, Call, & Morgan, 2009; St Clair & Rutz, 2013), male zebra finches will also attend to the structural properties of nest materials. For example, finches given wire-mesh nestboxes with either a small or a large entrance selected material of the appropriate length for their nestbox (long material for a large entrance hole and short for a small one) when both of these materials were made available (Muth & Healy, 2014; Table 3, Section 2). Over the course of handling the material, however, the males building in the nestboxes with a small entrance hole modified the way in which they held the material such that they could build with both the short and long material. The birds’ initial choices would suggest that they had assessed which was the appropriate material for the size of their nestbox entrance, but that they could learn to use a different motor technique in order to make more effective use of the available material. This change in material choice as the birds built suggests at least two things: (a) that building decisions can be updated through the building process; and (b) that the points at which decisions change or their causes may not be apparent from examination of the resulting structure. Detailed observations and/or manipulations of building will be required to investigate decision making and any contribution made to building by cognition.

That male zebra finches learn about the structural properties of nest materials is also supported by evidence that, after building a nest with long (15 cm), flexible string, birds preferred to build a second nest with 15 cm stiff string (Bailey, Morgan, Bertin, Meddle, & Healy, 2014; Table 3, Section 2). This preference appears to be due to the birds’ having learned that the flexible string was costly: males that built a nest with flexible string required twice as many pieces as did males that built with the stiff string.

Neural Architecture

Although there is a considerable literature on the neural and hormonal underpinnings of reproduction in birds, the nest-building component of this process has received little attention (Hall, Bertin, Bailey, Meddle, & Healy, 2014; Kingsbury, Jan, Klatt, & Goodson, 2015; Klatt & Goodson, 2013). Given the interest in developing avian models for the study of cognitive neuroscience (e.g., Clayton & Emery, 2015), nest building might be such a system. For example, immediate early gene activity shows that a number of neural circuits, such as the anterior motor pathway (involved in motor learning and sequencing), the social behaviour network (involved in a suit of social and reproductive behaviour in vertebrates), and the mesotocinergic, vasotocinergic, and dopaminergic reward system (involved in the motivation, and production of, social and reproductive behaviour), are all active when birds build nests (Hall et al., 2014; Hall, Healy, & Meddle, 2015).

Examination of the primate brain suggest that at least some of the neural processes involved in nest building may extend to construction behaviours more generally. Indeed, functional brain imaging techniques (Obayashi et al., 2001) have revealed that tool use by Japanese macaques Macaca fuscata produces similar activation patterns in the anterior motor pathway to those observed in birds building nests (Hall et al., 2014). The nature of the apparent similarity in the neurobiological processes underlying tool use and nest building, however, is not yet clear.

Given the phenotypic similarity between nest building and tool use in birds, it seems plausible that the neural underpinning of the two behaviours might also share common features. This possibility is supported by evidence that (a) tool users have a more foliated cerebellum (a brain region involved in motor control) than do non-tool users (Barton, 2012; Iwaniuk, Lefebvre, & Wylie, 2009), and that (b) cerebellar foliation increases with nest structure complexity (no nest < platform nest < cup nest; Hall, Street, & Healy, 2013). These data are just the beginning for what promises to be a productive brain-behaviour model.

Nest Building as a Model in Comparative Cognition

A handful of models for comparative cognition have been chosen because the species in question has a purported adaptive specialization for one cognitive ability or another (e.g., spatial memory in food-storing species; Biegler, McGregor, Krebs, & Healy, 2001; transitive inference in species that differ in social structure; Bond, Kamil, & Balda, 2003; timing in nectarivorous animals, e.g., bees and hummingbirds; Skorupski & Chittka, 2006; Henderson, Hurly, Bateson, & Healy, 2006; respectively; categorisation in songbirds; Sturdy, 2007). With the increasing interest in understanding physical cognition, particularly in birds (Guillette & Healy, 2015; Shettleworth, 2009), and the development and neural basis of cognition (Clayton & Emery, 2015), nest building in birds provides a range of features that may make it a useful comparative ‘model’ system for investigating brain-behaviour relationships. These features include: (a) variability within and across species (in nest structure, materials used, building technique, identity of builder; Hansell, 2000); (b) evidence of the role that learning plays in multiple aspects of nest building (Tables 2 and 3); and (c) extensive, in-depth study of avian neuroanatomy (Ziegler & Marler, 2012). Importantly, nest building is also amenable to experimental manipulation, both in the laboratory and in the field.

Model system or not, for the comparative cognition enthusiast, it is the second point that offers the most promise for future work. For, although there is much known about some aspects of avian nest building (Healy, Morgan, & Bailey, 2015), there is still an awful lot to learn. For example, do imprinting and/or other kinds of early-life experience play a role, and if so, on what components of building (e.g., motor skills, material choices, what structure to build)? To what extent does social learning play a role? Do birds continuously update their nest-building skills and decisions? Does skill at nest building correlate with ability to solve so-called physical cognition tasks? Are there sex differences in any of the relevant abilities, and to what extent is there interspecific variation? There is much scope for comparison to be made.

References

Antczak, M., Hromada, M., Czechowski, P., Tabor, J., Zablocki, P., Grzybek, J., & Tryjanowski, P. (2010). A new material for old solutions—the case of plastic string used in Great Grey Shrike nests. Acta Ethologica, 13(2), 87–91. doi:10.1007/s10211-010-0077-2

Bailey, I. E., Backes, A., Walsh, P. T., Morgan, K. V., Meddle, S. L., & Healy, S. D. (2015). Image analysis of weaverbird nests reveals signature weave textures. Royal Society Open Science, 2(6), 150074. doi:10.1098/rsos.150074

Bailey, I. E., Morgan, K. V., Bertin, M., Meddle, S. L., & Healy, S. D. (2014). Physical cognition: Birds learn the structural efficacy of nest material. Proceedings of the Royal Society B, 281(1784), 20133225. doi:10.1098/rspb.2013.3225

Barton, R. A. (2012). Embodied cognitive evolution and the cerebellum. Philosophical Transactions of the Royal Society B, 367(1599), 2097–2107. doi:10.1098/rstb.2012.0112

Bateson, P. P. G. (1966). The characteristics and context of imprinting. Biological Reviews, 41, 177–220. doi:10.1111/j.1469-185X.1966.tb01489.x

Bateson, P. P. G. (2015). Thirty years of collaboration with Gabriel Horn. Neuroscience & Biobehavioral Reviews, 50, 4–11. doi:10.1016/j.neubiorev.2014.09.019

Beecher, M. D., & Brenowitz, E. A. (2005). Functional aspects of song learning in songbirds. Trends in Ecology & Evolution, 20(3), 143–149. doi:10.1016/j.tree.2005.01.004

Biddle, L. E., Deeming, D. C., & Goodman, A. M. (2015). Morphology and biomechanics of the nests of the Common Blackbird Turdus merula. Bird Study, 62(1), 87–95. doi:10.1080/00063657.2014.988119

Biegler, R., McGregor, A., Krebs, J. R., & Healy, S. D. (2001). A larger hippocampus is associated with longer-lasting spatial memory. Proceedings of the National Academy, USA, 98(12), 6941–6944. doi:10.1073/pnas.121034798

Bluff, L. A., Weir, A. A., Rutz, C., Wimpenny, J. H., & Kacelnik, A. (2007). Tool-related cognition in New Caledonian Crows. Comparative Cognition & Behavior Reviews, 1, 1–25. doi:10.3819/ccbr.2008.20001

Boake, C. R. B. (1989). Repeatability: Its role in evolutionary studies of mating behavior. Evolutionary Ecology, 3(2), 173–182. doi:10.1007/BF02270919

Bond, A. B., Kamil, A. C., & Balda, R. P. (2003). Social complexity and transitive inference in corvids. Animal Behaviour, 65(3), 479–487. doi:10.1006/anbe.2003.2101

Boulinier, T., McCoy, K. D., Yoccoz, N. G., Gasparini, J., & Tveraa, T. (2008). Public information affects breeding dispersal in a colonial bird: Kittiwakes cue on neighbours. Biology Letters, 4(5), 538–540. doi:10.1098/rsbl.2008.0291

Chen, J. N., Liu, N. F., Yan, C., & An, B. (2011). Plasticity in nest site selection of Black Redstart (Phoenicurus ochruros): A response to human disturbance. Journal of Ornithology, 152(3), 603–608. doi:10.1007/s10336-010-0622-9

Clayton, N. S., & Emery, N. J. (2015). Avian models for human cognitive neuroscience: A proposal. Neuron, 86(6), 1330–1342. doi:10.1016/j.neuron.2015.04.024

Collias, E. C., & Collias, N. E. (1964). The development of nest-building behavior in a weaverbird. The Auk, 81(1), 42–52.

Collias, N. E., & Collias, E. C. (1984). Nest Building and Bird Behavior. Princeton, NJ: Princeton University Press.

Danchin, E., Boulinier, T., & Massot, M. (1998). Conspecific reproductive success and breeding habitat selection: Implications for the study of coloniality. Ecology, 79(7), 2415–2428. doi:10.1890/0012-9658(1998)079[2415:CRSABH]2.0.CO;2

Dawson, R. D., O’Brien, E. L., & Mlynowski, T. J. (2011). The price of insulation: Costs and benefits of feather delivery to nests for male tree swallows Tachycineta bicolor. Journal of Avian Biology, 42(2), 93–102. doi:10.1111/j.1600-048X.2010.05208.x

Deeming, D. C., & Ferguson, M. (1989). In the heat of the nest. New Scientist, 121(1657), 33–38.

Deeming, D. C., & Reynolds, S. J. (2015). Nests, eggs and incubation: New ideas about avian reproduction. Oxford, UK: Oxford University Press.

Doligez, B., Danchin, E., Clobert, J., & Gustafsson, L. (1999). The use of conspecifc reproductive success for breeding habitat selection in a non-colonial, hole-nesting species, the collared flycatcher. Journal of Animal Ecology, 68(33), 1193–1206. doi:doi:10.1676/13-108.1

Domjan, M. (2014). The Principles of Learning and Behavior. Stamford, CT: Cengage Learning.

Dow, H., & Fredga, S. (1983). Breeding and natal dispersal of the goldeneye, Bucephala clangula. Journal of Animal Ecology, 52(3), 681–695.

Eggers, S., Griesser, M., Nystrand, M., & Ekman, J. (2006). Predation risk induces changes in nest-site selection and clutch size in the Siberian jay. Proceedings of the Royal Society B, 273(1587), 701–706. doi:10.1098/rspb.2005.3373

Fijen, T. P. M., Kamp, J., Lameris, T. K., Pulikova, G., Urazaliev, R., Kleijn, D., & Donald, P. F. (2015). Functions of extensive animal dung “pavements” around the nests of the Black Lark (Melanocorypha yeltoniensis). The Auk, 132(4), 878–892. doi:10.1642/AUK-15-38.1

Fisher, R. J., & Wiebe, K. L. (2006). Breeding dispersal of Northern Flickers Colaptes auratus in relation to natural nest predation and experimentally increased perception of predation risk. Ibis, 148(4), 772–781. doi:10.1111/j.1474-919X.2006.00582.x

Forsman, J. T., & Thomson, R. L. (2008). Evidence of information collection from heterospecifics in cavity-nesting birds. Ibis, 150(2), 409–412. doi:10.1111/j.1474-919X.2007.00773.x

Grégoire, A., & Cherry, M. I. (2007). Nesting success and within-season breeding dispersal in the Orange-breasted Sunbird Anthobaphes violacea. Ostrich, 78(3), 633–636. doi:10.2989/OSTRICH.2007.78.3.10.321

Greig-Smith, P. W. (1982). Dispersal between nest-sites by Stonechats Saxicola torquata in relation to previous breeding succes. Ornis Scandinavica, 13(3), 232–238. doi:10.2307/3676304

Guillette, L. M., & Healy, S. D. (2015). Nest building, the forgotten behaviour. Current Opinion in Behavioural Sciences, 6, 90–96. doi:10.1016/j.cobeha.2015.10.009

Haas, C. (1998). Effects of prior nesting success on site fidelity and breeding dispersal: An experimental approach. The Auk, 115(4), 929–936.

Haftorn, S., & Slagsvold, T. (1995). Egg covering in birds: Description of the behaviour in tits (Parus spp.) and a test of hypotheses of its function. Fauna Norvegica, Series C Cinclus, 18(2), 85–106.

Hakkarainen, H., Ilmonen, P., Koivunen, V., & Korpimäki, E. (2001). Experimental increase of predation risk induces breeding dispersal of Tengmalm’s owl. Oecologia, 126(3), 355–359. doi:10.1007/s004420000525

Hall, Z. J., Bertin, M., Bailey, I. E., Meddle, S. L., & Healy, S. D. (2014). Neural correlates of nesting behavior in zebra finches (Taeniopygia guttata). Behavioural Brain Research, 264, 26–33. doi:10.1016/j.bbr.2014.01.043

Hall, Z. J., Healy, S. D., & Meddle, S. L. (2015). A role for nonapeptides and dopamine in nest-building behaviour. Journal of Neuroendocrinology, 27(2), 158–165. doi:10.1111/jne.12250

Hall, Z. J., Street, S. E., & Healy, S. D. (2013). The evolution of cerebellum structure correlates with nest complexity. Biology Letters, 9(6), 20130687. doi:10.1098/rsbl.2013.0687

Hansell, M. (1984). Animal architecture and building behaviour. London: Longman.

Hansell, M. (2000). Bird nests and construction behaviour. Cambridge, UK: Cambridge University Press.

Hansell, M., & Ruxton, G. D. (2008). Setting tool use within the context of animal construction behaviour. Trends in Ecology and Evolution, 23(2), 73–78. doi:10.1016/j.tree.2007.10.006

Healy, S. D., Haggis, O., & Clayton, N. S. (2010). Zebra Finches and cognition. Emu, 110(3), 242–250. doi:10.1071/MU10004

Healy, S. D., Morgan, K. V., & Bailey, I. E. (2015). Nest construction behaviour. In C. D. Deeming & J. S. Reynolds (Eds.), Nests, eggs and incubation: New ideas about avian reproduction, 16–28. Oxford, UK: Oxford University Press.

Henderson, J., Hurly, A. T., Bateson, M., & Healy, S. D. (2006). Timing in free-living rufous hummingbirds, Selasphorus rufus. Current Biology, 16, 512–515. doi:10.1016/j.cub.2006.01.054

Herlugson, C. J. (1981). Nest site selection in Mountain Bluebirds. Condor, 83(3), 252–255.

Heyes, C. M. (1994). Social learning in animals: Categories and mechanisms. Biological Reviews of the Cambridge Philosophical Society, 69, 207–231. doi:10.1111/j.1469-185X.1994.tb01506.x

Hinde, R. A., & Warren, R. P. (1959). The effect of nest building on later reproductive behaviour in domesticated canaries. Animal Behaviour, 7(1–2), 35–41. doi:10.1016/0003-3472(59)90028-4

Hochbaum, H. A. (1955). Travels and traditions of waterfowl. Minneapolis, MN: Univ. of Minnesota Press.

Hoover, J. P. (2003). Decision rules for site fidelity in a migratory bird, the Prothonotary Warbler. Ecology, 84(2), 416–430. doi:10.1890/0012-9658(2003)084[0416:DRFSFI]2.0.CO;2

Iwaniuk, A. N., Lefebvre, L., & Wylie, D. R. (2009). The comparative approach and brain-behaviour relationships: A tool for understanding tool use. Canadian Journal of Experimental Psychology, 63(2), 150–159. doi:10.1037/a0015678

Kingsbury, M. A., Jan, N., Klatt, J. D., & Goodson, J. L. (2015). Nesting behavior is associated with VIP expression and VIP–Fos colocalization in a network-wide manner. Hormones and Behavior, 69, 68–81. doi:10.1016/j.yhbeh.2014.12.010

Klatt, J. D., & Goodson, J. L. (2013). Sex-specific activity and function of hypothalamic nonapeptide neurons during nest-building in zebra finches. Hormones and Behavior, 64(5), 818–824. doi:10.1016/j.yhbeh.2013.10.001

Lessells, C. M., & Boag, P. T. (1987). Unrepeatable repeatabilities: A common mistake. Auk, 104, 116–121. doi:10.2307/4087240

Lima, S. L. (2009). Predators and the breeding bird: Behavioral and reproductive flexibility under the risk of predation. Biological Reviews, 84(3), 485–513. doi:10.1111/j.1469-185X.2009.00085.x

Loukola, O. J., Laaksonen, T., Seppänen, J. T., & Forsman, J. T. (2014). Active hiding of social information from information-parasites. BMC Evolutionary Biology, 14(1), 32. doi:10.1186/1471-2148-14-32

Manrique, H. M., Sabbatini, G., Call, J., & Visalberghi, E. (2011). Tool choice on the basis of rigidity in capuchin monkeys. Animal Cognition, 14(6), 775–786. doi:10.1007/s10071-011-0410-9

Marzluff, J. M. (1988). Do pinyon jays alter nest placement based on prior experience? Animal Behaviour, 36(1), 1–10. doi:10.1016/S0003-3472(88)80244-6

Mennerat, A., Perret, P., & Lambrechts, M. M. (2009). Local individual preferences for nest materials in a passerine bird. PLoS ONE, 4(4), 2–7. doi:10.1371/journal.pone.0005104

Merola, M. (1995). Observations on the nesting and breeding behavior of the Rock Wren. Condor, 97(2), 585–587. doi:10.2307/1369046

Moiseev, A. (1980). The breeding biology of the Black Lark in Central and Northern Kazakhstan. In A. A. Kovshar (Ed.), Biology of the Birds of Naurzum State Protected Area, 1st ed., 120–126. Nauka, Alma-Ata, Kazakhstan.

Morton, E. S. (2005). Predation and variation in breeding habitat use in the Ovenbird, with special reference to breeding habitat selection in northwestern Pennsylvania. The Wilson Bulletin, 117(4), 327–335. doi:10.1676/04-135.1

Muth, F., & Healy, S. D. (2011). The role of adult experience in nest building in the zebra finch, Taeniopygia guttata. Animal Behaviour, 82(2), 185–189. doi:10.1016/j.anbehav.2011.04.021

Muth, F., & Healy, S. D. (2012). Zebra finches build nests that do not resemble their natal nest. Avian Biology Research, 5(4), 218–226. doi:10.3184/175815512X13531725598475

Muth, F., & Healy, S. D. (2014). Zebra finches select nest material appropriate for a building task. Animal Behaviour, 90, 237–244. doi:10.1016/j.anbehav.2014.02.008

Muth, F., Steele, M., & Healy, S. D. (2013). Colour preferences in nest-building zebra finches. Behavioural Processes, 99, 106–111. doi:10.1016/j.beproc.2013.07.002

Nicolakakis, N., & Lefebvre, L. (2000). Forebrain size and innovation rate in European birds: Feeding, nesting and confounding variables. Behaviour, 137(11), 1415–1429.

Obayashi, S., Suhara, T., Kawabe, K., Okauchi, T., Maeda, J., Akine, Y., Onoe, Hirotaka, & Iriki, A. (2001). Functional brain mapping of monkey tool use. NeuroImage, 14(4), 853–861. doi:10.1006/nimg.2001.0878

Peluc, S. I., Sillett, T. S., Rotenberry, J. T., & Ghalambor, C. K. (2008). Adaptive phenotypic plasticity in an island songbird exposed to a novel predation risk. Behavioral Ecology, 19(4), 830–835. doi:10.1093/beheco/arn033

Raby, C. R., & Clayton, N. S. (2009). Prospective cognition in animals. Behavioural Processes, 80(3), 314–24. doi:10.1016/j.beproc.2008.12.005

Rioux, S., Amirault-Langlais, D. L., & Shaffer, F. (2011). Piping Plovers make decisions regarding dispersal based on personal and public information in a variable coastal ecosystem. Journal of Field Ornithology, 82(1), 32–43. doi:10.1111/j.1557-9263.2010.00305.x

Roos, S., & Pärt, T. (2004). Nest predators affect spatial dynamics of breeding red-backed shrikes (Lanius collurio). Journal of Animal Ecology, 73(1), 117–127. doi:10.1111/j.1365-2656.2004.00786.x

Sanz, C., Call, J., & Morgan, D. (2009). Design complexity in termite-fishing tools of chimpanzees (Pan troglodytes). Biology Letters, 5(3), 293–296. doi:10.1098/rsbl.2008.0786

Sargent, T. D. (1965). The role of experience in the nest building of the Zebra Finch. The Auk, 48–61.

Schiermann, G. (1939). “Stammesgenossenschaften” bei Vögeln. Ornithologische Monatsbererichte, 47, 1–3.

Scott, W. E. D. (1902). Instinct in song birds, method of breeding in hand-reared robins (Merula migratoria). Science, 16(393), 70–71. doi:10.1126/science.16.393.70

Scott, W. E. D. (1904). The inheritance of song in passerine birds. Further observations on the development of song and nest-building in hand-reared Rose-breasted Grosbeaks, Zamelodia ludoviciana (Linnaeus). Science, 20(504), 282–283. doi:10.1126/science.20.504.282

Seed, A., & Byrne, R. (2010). Animal tool-use. Current Biology, 20(23), R1032–R1039. doi:10.1016/j.cub.2010.09.042

Seppänen, J. T., & Forsman, J. T. (2007). Interspecific social learning: Novel preference can be acquired from a competing species. Current Biology, 17(14), 1248–1252. doi:10.1016/j.cub.2007.06.034

Seppänen, J. T., Forsman, J. T., Mönkkönen, M., Krams, I., & Salmi, T. (2011). New behavioural trait adopted or rejected by observing heterospecific tutor fitness. Proceedings of the Royal Society B, 278(1712), 1736–1741. doi:10.1098/rspb.2010.1610

Shettleworth, S. J. (2009). The evolution of comparative cognition: Is the snark still a boojum? Behavioural Processes, 80(3), 210–217. doi:10.1016/j.beproc.2008.09.001

Skorupski, P., & Chittka, L. (2006). Animal cognition: An insect’s sense of time? Current Biology, 16(19), 853–855. doi:10.1016/j.cub.2006.08.069

Smith, P. (1904). Nesting habits of the Rock wren. Condor, 6, 109–110.

St Clair, J. J. H., & Rutz, C. (2013). New Caledonian crows attend to multiple functional properties of complex tools. Philosophical Transactions of the Royal Society B, 368(1630), 20120415. doi:10.1098/rstb.2012.0415

Stanback, M. T., Bartholomew, J. E., Bergner, L. M., Cline, E. L., Helms, P. I., McGovern, P. G., . . . Roth, J. E. (2013). House wrens alter nest architecture to compensate for cavity vulnerability. The Wilson Journal of Ornithology, 125, 174–178. doi:10.1242/jeb.089763

Sturdy, C. B. (2007). Auditory category perception as a natural cognitive activity in songbirds. Comparative Cognition & Behavior Reviews. doi:10.3819/ccbr.2008.20006

Styrsky, J. N., Brawn, J. D., & Robinson, S. K. (2005). Juvenile mortality increases with clutch size in a neotropical bird. Ecology, 86(12), 3238–3244. doi:10.1890/04-1613

Suárez-Rodríguez, M., López-Rull, I., & Garcia, C. M. (2013). Incorporation of cigarette butts into nests reduces nest ectoparasite load in urban birds: New ingredients for an old recipe? Biology Letters, 9(1), 20120931. doi:10.1098/rsbl.2012.0931

Suryan, R. M., & Irons, D. B. (2001). Colony and population dynamics of black-legged kittiwakes in a heterogeneous environment. The Auk, 118(3), 636–649.

Verlaine, L. (1934). L’instinct et l’intelligence chez les oiseaux. Recherches Philosoph, 3, 285–305.

Wallace, A. R. (1867). The philosophy of birds’ nests. Intellectual Observer, 11, 413–420.

Walsh, P. T., Hansell, M., Borello, W. D., & Healy, S. D. (2010). Repeatability of nest morphology in African weaver birds. Biology Letters, 6(2), 149–151. doi:10.1098/rsbl.2009.0664

Walsh, P. T., Hansell, M., Borello, W. D., & Healy, S. D. (2011). Individuality in nest building: Do Southern Masked weaver (Ploceus velatus) males vary in their nest-building behaviour? Behavioural Processes, 88(1), 1–6. doi:10.1016/j.beproc.2011.06.011

Walsh, P. T., Hansell, M., Borello, W. D., & Healy, S. D. (2013). Are elaborate bird nests built using simple rules? Avian Biology Research, 6(2), 157–162. doi:10.3184/175815513X13629302805186

Warning, N., & Benedict, L. (2015). Paving the way: Multifunctional nest architecture of the Rock Wren. The Auk, 132(1), 288–299. doi:10.1642/AUK-14-186.1

Williams, H. S. (1934). Nest building-new style. Natural History, 34, 431–446.

Winkler, D. W. (1993). Use and importance of feathers as nest lining in tree swallows (Tachycineta bicolor). The Auk, 110(1), 29–36. doi:10.2307/4088228

Zentall, T. R. (2006). Mental time travel in animals: A challenging question. Behavioural Processes, 72(2), 173–183. doi:10.1016/j.beproc.2006.01.009

Ziegler, H. P., & Marler, P. (2012). Neuroscience of Birdsong. New York, NY: Cambridge University Press.

Functional Magnetic Resonance Imaging of the Domestic Dog: Research, Methodology, and Conceptual Issues

Andie M. Thompkins
Dept. of Psychology, Auburn University, Auburn, AL, USA

Gopikrishna Deshpande
Dept. of Psychology, Auburn University, Auburn, AL, USA
AU MRI Research Center, Dept. of Electrical & Computer Engineering, Auburn University, Auburn, AL, USA
Alabama Advanced Imaging Consortium, Auburn University and University of Alabama Birmingham, AL, USA

Paul Waggoner
Canine Performance Sciences, College of Veterinary Medicine, Auburn University, Auburn, AL, USA

Jeffrey S. Katz
Dept. of Psychology, Auburn University, Auburn, AL, USA
AU MRI Research Center, Dept. of Electrical & Computer Engineering, Auburn University, Auburn, AL, USA
Alabama Advanced Imaging Consortium, Auburn University and University of Alabama Birmingham, AL, USA

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

Neuroimaging of the domestic dog is a rapidly expanding research topic in terms of the cognitive domains being investigated. Because dogs have shared both a physical and social world with humans for thousands of years, they provide a unique and socially relevant means of investigating a variety of shared human and canine psychological phenomena. Additionally, their trainability allows for neuroimaging to be carried out noninvasively in an awake and unrestrained state. In this review, a brief overview of functional magnetic resonance imaging (fMRI) is followed by an analysis of recent research with dogs using fMRI. Methodological and conceptual concerns found across multiple studies are raised, and solutions to these issues are suggested. With the research capabilities brought by canine functional imaging, findings may improve our understanding of canine cognitive processes, identify neural correlates of behavioral traits, and provide early-life selection measures for dogs in working roles.

Keywords: canine fMRI, dog cognition, dog neuroimaging

Author Note: Correspondence may be addressed to: Andie M. Thompkins, email: andie.thompkins@auburn.edu; Gopikrishna Deshpande, email: gopi@auburn.edu; Paul Waggoner, email: waggolp@auburn.edu; or Jeffrey S. Katz, email: katzjef@auburn.edu.

Acknowledgments: We wish to thank Jennifer Robinson, Lewis Barker, and Ana Franco-Watkins for their considerate feedback in the development of this review. Writing of this review was supported by Defense Advanced Research Projects Agency (government grant/contract number W911QX-13-C-0123). The views, opinions, and/or findings contained in this article are those of the authors and should not be interpreted as representing the official views or policies, either expressed or implied, of the Defense Advanced Research Projects Agency, U.S. Department of Defense, or the federal Government of the United States of America.

---

Introduction

The domestic dog (Canis familiaris) has become one of the primary subjects of recent comparative cognition research. The “rise of the dog” is unsurprising given its foundations. Bensky, Gosling, and Sinn (2013) discussed these foundations in a comprehensive review, highlighting the interspecific communication and relationship characteristic of dog-human cohabitation, dogs as models for human cognitive deficiencies, the trainability and availability of dog subjects, and a steadfast interest in dog cognition by the general public. The authors cite exponential growth in the number of dog cognition publications, covering an expanse of sensory modalities, research questions, and dog ages and populations. A more recent development in the rise of the dogs has been the comparative neuroimaging of awake dogs. In this review, we summarize the extant literature on functional magnetic resonance imaging (fMRI) and discuss its implications for comparative research and application.

The inclusion of dogs in comparative neuroimaging studies has come about due to an ideal combination of scientific relevance and training protocols. Due to the evolutionary history that humans and dogs have shared over tens of thousands of years, dog subjects and human participants enter research with similar environmental experiences and an extensive interspecific social repertoire (e.g., sharing daily environments and companions). Further, the social bond shared between humans and dogs and the corresponding receptivity of dogs to human cues helps alleviate the need for restraint and sedation in neuroimaging studies. Rather, dogs can be trained to lie motionless and awake for neuroimaging scans. Though the dog fMRI literature is in its infancy, lagging far behind the human fMRI literature in scope and number, the unique challenges of this field (e.g., training time, parameter selection) become surmountable when data acquisition and analysis techniques are sophisticated enough to compensate accordingly. Consequently, canine fMRI may require methodological advancements over and above the state-of-the-art in human fMRI for addressing these unique challenges.

The advent of fMRI technology in the 1990s presented the scientific and medical communities with a safe and noninvasive means of imaging brain activity with high spatial resolution. FMRI makes use of the activity-dependent flow of oxygenated blood in order to localize mental functions to specific structures in the brain. That is, when neurons in the brain are activated, an increased volume of oxygenated blood flows to the region in which they are located in order to meet the localized energy demand by the neurons. The volume of oxygen in this blood exceeds what is consumed by the neuronal activity, and therefore a surplus of oxygenated blood in that region gives rise to a localized decrease in magnetic susceptibility (as oxyhemoglobin is diamagnetic) and a concomitant increase in MR signal intensity. Thus, the signal used in fMRI is referred to as blood oxygenation level dependent (BOLD) signal (Ogawa, Lee, Kay, & Tank, 1990).

The increase in blood flow that occurs subsequent to a period of neuronal activity is called the hemodynamic response. In humans, this response temporally lags in comparison to the neuronal activity, reaching its maximum level approximately five seconds after neuronal activity, which may have occurred over the course of milliseconds. Subsequent to this hemodynamic response peak is an undershoot period, by which the signal does not return to baseline until 15 to 20 seconds after its peak. The course of the hemodynamic response (from rise, to peak, to fall, and return to baseline) to an external stimulus is referred to as the hemodynamic response function (HRF). The convolution of the external stimulus and the HRF represents the expected signal in brain regions activated by the external stimulus. By matching this expectation with the measured response using a linear mathematical model, brain regions subserving the processing of the external stimulus can be pinpointed (Ogawa et al., 1992; Poldrack, Mumford, & Nichols, 2011).

Comparatively, the HRF will differ across species due to differences in vasculature necessitated by varying brain size and shape. For example, in awake rodents, the latency of the HRF peak has been shown to be 2 seconds (Martin, Martindale, Berwick, & Mayhew, 2006), while the canonical HRF used in humans has a peak latency of 6s (Henson, 2004). The precise shape of the HRF has not been determined in canines, and this is an important limitation for future research to address to further validate canine fMRI. Until such advances have been made, it is reasonable to use time and dispersion derivatives in the general linear activation analysis so as to explicitly model and regress out the variability of the HRF (with respect to the canonical HRF) in experimental data (Jia, Hu, & Deshpande, 2014).

In order to measure the BOLD response, several magnetic resonance imaging components must come together. In MR technology, a strong static magnetic field serves to align the protons in the body. Emission of radio frequencies is used to intermittently disrupt this alignment of protons, after which the protons realign with the static magnetic field while necessarily emitting energy. This resonance energy is picked up by receiving coils and creates the signal by which fMRI data are obtained (Smith, 2010).

While our emphasis in this review is on fMRI, MRI also allows for detailed structural scans of the brain. The use of MRI for investigating anatomical structure of canine brains predates the use of fMRI for investing canine brain function. The main reason for this is that anatomical MRI could be performed on anesthetized dogs without any loss of information since anesthesia does not (at least immediately) alter structure. Hence, MRI has become an invaluable tool in the veterinary field and has been used primarily for clinical purposes (e.g., degenerative diseases, cognitive dysfunction, herniation). There have also been specific applications (e.g., aging, tractography) for domestic dogs based on the anatomical/white-matter structure and development of the brain (e.g., Anaya García, Hernández Anaya, Marrufo Meléndez, Velázquez Ramírez, & Palacios Aguiar, 2015; Baxi et al., submitted; Gross, Garcia-Tapia, Riedesel, Ellinwood, & Jens, 2010; Su et al., 2005; Jacqmot et al., 2013).

As interest has risen for assessing nonhuman cognition via functional MRI studies, a growing variety of species have been imaged in MR scanners. The current driving force of progress in the expansion of fMRI research stems from the possibility of keeping animals in a still, wakeful, and attentive state during scanning. Experimental and training techniques to allow for awake scanning have been developed for rats (King et al., 2005; Lahti, Ferris, Li, Sotak, & King, 1998), pigeons (De Groof et al., 2013), monkeys (e.g., Chen, Wang, & Dillenburger, 2012) and dogs (e.g., Jia, Pustovyy, et al., 2014). In rats, Lahti et al. (1998) used fMRI to localize somatosensory cortex activation upon shock, and King et al. (2005) furthered methodological development by investigating the effects of experiment acclimation on stress levels. De Groof et al. (2013) used traditional and resting state fMRI to investigate visual system connectivity in awake pigeons. In monkeys, fMRI has been used to explore visual area activation as well as protocol and parameter adjustments for improving image quality. As we will review in detail, Jia et al. (2015) used awake dog subjects to uncover olfaction-driven activations. The key difference with dogs, though, is that, unlike other animals, they do not have to be restrained and can be trained to hold their head still as humans do, making the experiment more valid for comparisons to humans.

Although various training techniques have not been systematically explored, an overarching goal of any training methodology is to reduce training time while maintaining success in behavior. When training dogs to lie still for fMRI, researchers have used a variety of techniques including chaining (e.g., Berns, Brooks, & Spivak, 2013), target stick (e.g., Jia, Pustovyy, et al., 2014), and model-rival (e.g., Andics, Gácsi, Faragó, Kis, & Miklósi, 2014) methods. In most cases, training builds incrementally from basic contingencies outside the scanner room (e.g., head on chin rest in mock coil, touching nose to target) to inside the scanner room (e.g., prone position on scanner bed) to inside the scanner bore (e.g., head still in coil for scan). Regardless of training method used, comparisons cannot be made across the current body of literature because the availability of the dogs and handlers to participate in training has varied significantly. Pragmatically, developing techniques that promote rapid acclimation to the scanner environment with minimal stress to the dogs is ideal.

History of fMRI in the Dog

The use of fMRI provides an exciting and fairly unchartered area of comparative cognition and neuroimaging research with domestic dogs. Explorations in dog MRI and fMRI began with the use of sedation to answer questions about anatomy and physiology, primarily for the purposes of veterinary education and research. Such studies have provided knowledge of canine neural responsiveness, cognitive effects of aging, neuroimaging efficacy, and health viability, and thus we first discuss the work done with anesthetized dogs in this paragraph. Bach et al. (2013) used fMRI to successfully identify neural regions associated with processing of auditory stimuli, as well as establish the efficacy of fMRI with anesthetized dogs in regard to auditory stimulus presentation. Su et al. (2005) used longitudinal structural MRI to investigate the time course of neural correlates of canine cognitive decline (e.g., ventricular enlargement, lesions), strengthening the potentiality of the dog as a model of human aging. The efficacy of using high-field MRI to image dog brain structure was explored by Martín-Vaquero et al. (2011), in which it was found that the 3T MRI provided more consistent and reliable anatomical imaging data than did 7T MRI, contrary to what one might expect given generally superior field strength and image quality at 7T. In regard to health concerns surrounding MRI with dog subjects, Venn, McBrearty, McKeegan, and Penderis (2014) published findings of post-scan hearing loss, emphasizing the need for hearing protection when imaging dogs in MRI environments.

Though prior research on cognitive process in dogs has been conducted with anesthetized dogs, the cognitive processes of their natural, attentive state are of great comparative interest. The use of anesthesia necessarily impedes attentiveness and alters the state of consciousness, as well as reduces rates of blood flow and respiration. The amalgamation of these reduced biomarkers of stimulus processing leaves much to be desired in the data set, as brain regions or activation patterns involved in cognitive processing may be minimized or lost altogether (Jia, Pustovyy, et al., 2014). In search of valid and viable findings, neuroimaging research with dogs has begun a transition to functional imaging using highly trained dogs that do not require anesthesia for image acquisition. In the first published instance of MR images obtained through awake dog imaging, Tóth, Gácsi, Miklósi, Bogner, and Repa (2009) established successful data acquisition with dogs that were trained in a stepwise fashion to remain still and ignore scanner noise. This study consisted of only structural scans, but it was not long after that functional scans were achieved in awake dogs. The movement in canine functional imaging has been pioneered by laboratories at Auburn University, Emory University, and Eotvos Lorand University. Figure 1 documents the timeline of canine fMRI research in awake dogs from these laboratories. Table 1 summarizes the existent literature presenting the number of subjects, tasks, stimuli, and the brain areas activated. Next, we review in a chronological fashion the methods and findings listed in Table 1.

Berns, Brooks, and Spivak (2012) first published research on fMRI data acquisition simple discrimination task in the awake and unrestrained domestic dog. The authors addressed three major challenges in using fMRI technology with dogs: subject motion, which distorts acquired data; use of anesthesia, which eliminates the viability of a cognitive assessment; and immobilization. To target these challenges, the authors developed a set of behavioral and technical methodologies for imaging dogs while they remained motionless, awake, and attentive to a cognitive task. Further, this methodological set was used to assess the reward-prediction error theory of dopamine release in dogs via use of reward signals and attention to activation changes in the ventral striatum. Specifics of this study are presented next.

Proof of Concept

Two dogs were used as subjects in Berns et al. (2012), one of which had been previously trained in agility. Each dog was incrementally trained, using positive reinforcement, in a mock MRI scanner consisting of a replica of the head coil, scanner bore, and patient table. Additionally, the dogs were exposed to presentations of the scanner noises and sound levels that they would experience in the scanner. The discrimination task was trained by assigning reward conditions to each of two hand signals given by a handler: a hand held straight up signaled forthcoming presentation of food reward, and two hands held horizontally facing one another signaled no reward.

Once the dogs performed to criteria in the mock scanners, they were moved to true fMRI scanning in a Siemens 3T Trio over a period of six weeks. Initial scanning provided both an assimilation period and an assessment of image acquisition feasibility, followed by a subsequent session to optimize scanning parameters, and finally followed by image acquisition during the randomized instrumental reward task. In this final session, the handler (positioned at the end of the bore) presented 10-second durations of the reward/no-reward task as previously trained.

Analysis of the obtained functional data focused on the head of the caudate in order to target the ventral striatum. The ventral striatum served as the predicted area for activation according to reward-prediction error learning, which anticipates dopamine release and corresponding neural activation of the ventral caudate upon expectation of reward. Reward and no-reward conditions served as the contrast of interest, revealing significant activation differences in the right caudate, though the meaning of the lateralized activation is unclear. These activation differences highlighted a distinct hemodynamic response for reward signal presentations as compared to no-reward signals, thus providing support for the notion that dopamine is released in response to unexpected events that signal future reward and, here specifically, a representation of positive reward prediction in the domestic dog.

Replication of the Reward/No-Reward Task

Berns, Brooks, and Spivak (2013) followed their initial 2012 study of fMRI with dogs with an assessment of the replicability of their methodology. Further, the authors sought to potentially reduce signal variability of caudate responses to the instrumental reward task with additional experimental improvements. In this replication, 13 dogs of various training background (e.g., service, agility, basic obedience) completed positive-reinforcement training on the mock scanners (this time, with a mock knee coil instead of head coil) and the reward/no-reward task.

In this expanded subject set, 62% of dogs showed significant differential positive activation in the caudate for reward signals. These findings were consistent with Berns et al. (2012); however, substantial signal variability was found across subjects for overall caudate activation. Berns et al. (2013) discuss several potential reasons for this variability between subjects, including greater human attachment in service and therapy dogs, the inherent noise of imaging data, the difficult balance between imaging repetition and efficacy of the task, mislocation of regions of interest, and individual motivational differences. Interestingly, the authors note that when the dog fMRI data collected from the instrumental reward task is compared to that of humans, it may indeed be less variable than human caudate activity. Overall, this replication of awake, unrestrained fMRI with dogs supported the efficacy of reliable training in demonstrating differential activations in the dog brain. Further, the results of this study and Berns et al. (2012) provide support for the possibility of dog models of human cognitive function.

Temperament and Stimulus Source

To further expand on their developments in dog fMRI, Cook, Spivak, and Berns (2014) modified their reward/no-reward task to assess activation differences driven both by subject temperament and stimulus source. The same dogs as used in Berns et al. (2013) were employed in this study, and all were evaluated for 14 factors of temperament (e.g., attachment, trainability, Hsu & Serpell, 2003) using the owner self-report Canine Behavioral Assessment and Research Questionnaire (C-BARQ). Stimulus sources were divided equally among reward/no-reward hand signal presentations given by a familiar person or an unfamiliar person, as well as digitized hand signal displays presented on a projection screen. Analyses revealed that across the subject set, the caudate was differentially active by condition, indicating further support for the implication of the ventral striatum in reward anticipation. Further, activations revealed that the dogs could generalize the meaning of the hand signals across stimulus sources. When C-BARQ temperament factors, particularly aggressivity, were taken into account, activation differences were found according to a stimulus source of familiar human versus unfamiliar human or projection by computer. That is, dogs with lower aggressivity levels showed greater activation for reward signals given by the familiar person than by the unfamiliar person or computer. Alternatively, dogs with higher aggressivity levels showed greater activation for reward signals given by an unfamiliar person or computer. Cook et al. (2014) note that, because the striatal response is dependent upon arousal and stimulus salience, higher aggressiveness correlates to higher salience for the novel situations of unfamiliar person and computer, and that lower aggressiveness correlates to lower anxiety and higher salience with a familiar person. In their conclusion, the authors stress the possibility of differences across dogs in their reactions to various contexts, and emphasize the need for consideration of this possibility when making claims from dog studies without temperament testing.

In all, these initial studies of functional imaging with dogs provided strong support for the opportunities presented by the merger of canine cognition with fMRI technology. The establishment of successful training and imaging techniques allows for the expansion of this research to involve more specific regions of interest along with a greater range of subjects and ontogenic histories. Notably, success with visually based experiments provided an interesting opportunity to investigate processing in other sensory modalities.

Audition

Andics et al. (2014) have also used positive-reinforcement training to conduct fMRI studies with awake and unrestrained dogs. Here, the authors conducted comparative research into the function and location of voice-sensitive brain regions in dogs and humans. Because humans and dogs have long shared a natural environment, Andics et al. (2014) questioned how voice-sensitive regions in both populations would respond to conspecifics and heterospecifics, and whether they would show similar processing of emotional cues in these signals. Eleven dogs and 22 humans participated in scans during which an identical set of auditory stimuli was presented. This stimulus set consisted of human (e.g., laugh, cough, yawn) and dog vocalizations (e.g., growl, pant, bark) ranging in emotional valence along with environmental sounds and silence. The silence condition was used to functionally assign the auditory region of interest by contrasting the fMRI response to silence against the response during sound presentation.

Cortical sound sensitivities were revealed in the perisylvian regions for the dogs and the superior temporal sulcus and inferior frontal cortex for humans. Both species showed sensitivity in the medial geniculate body. In the dog brain, subregions were identified that activated maximally for dog vocalizations as well as to human vocalizations and environmental sounds. On the contrary, nearly all human auditory regions of activation were maximal for human vocalizations, although the medial geniculate body showed a maximal activation for dog vocalizations.

Olfaction

Jia, Pustovyy, et al. (2014) utilized positive-reinforcement training and fMRI with awake and unrestrained dogs to investigate olfactory processing and the effects of anesthesia on the quality of neural data. The authors note that there is a large body of literature pertaining to both the cellular and behavioral correlates of olfaction in dogs, but little research has been done on the cognitive processes that underlie olfaction. Thus, their study aimed to serve as a comparison of the neural response in the brain to varying odor concentrations in awake versus anesthetized dogs. Six dogs served as the subjects for this study, and a specialized computer-controlled odorant delivery system was designed with MR safety restrictions and parameters in mind (e.g., elimination of ferromagnetic objects in the scanner room, motion control). This delivery system was used to precisely present 10-second periods of odorant (ethyl butyrate, eugenol, & carvone mixture) to dogs across five randomized blocks. Further, for the first time in canine fMRI research, these authors used a single external infrared camera to track dog head motion and retrospectively correct for motion-related artifacts in the data, especially faster and jerky movements which cannot be captured by the poor temporal resolution available to image-based rigid body registration methods which are commonly used in fMRI analysis.

Both awake and anesthetized dogs demonstrated strong activation in the olfactory bulb and bilateral piriform lobes upon presentation of both high and low concentrations of odor. However, the intensity of activations, as well as their spatial extent, was mediated by concentration, with larger activations for higher odor concentrations. Awake dogs exhibited activations in areas including the medial, superior, and orbital frontal cortices and the cerebellum, all of which are tied to cognitive processes, whereas anesthetized dogs did not. Given the findings, the authors concluded that anesthesia degrades processing of odors and that the use of fMRI can and will provide a useful investigation into the neural substrates of the olfactory system.

Recent findings by Jia et al. (2015) expanded on this work and revealed olfactory enhancement with the addition of zinc nanoparticles to odorant presentations. Using conditions of pure odorants, odorants plus zinc nanoparticles, and gold nanoparticles, as well as zinc nanoparticles alone in water vapor, Jia et al. (2015) hypothesized that zinc nanoparticles, previously implicated in enhancement of odor response in vitro, would lead to greater activity in the brain regions for olfactory processing that were revealed in Jia, Pustovyy, et al. (2014). Indeed, activations in the olfactory bulb and hippocampus were greater in awake dogs exposed to odorants with zinc nanoparticles compared to pure odorants, pure zinc nanoparticles, and odorants with gold nanoparticles. Acknowledging the need for confirmation of increased odor sensitivity via behavioral tests, Jia et al. (2015) highlight the possible utility of zinc for enhancing the abilities of working odor detection dogs.

Berns, Brooks, and Spivak (2015) sought to investigate the canine perceptual experience of socially related stimuli via the processing odors of familiar and unfamiliar people and dogs. In order to investigate the driving social relationship between a human and dog, the authors again utilized the dopamine theory of reward-error prediction, hypothesizing that if the relationship between a dog and its most familiar person includes reward expectancy, then caudate activation will be greater when the scent of that person is being processed, as opposed to another person or a dog. The same dogs that were used in their prior research (Berns et al., 2012; Berns et al., 2013; Cook et al., 2014) were enlisted for this study. Additional training was needed to acclimatize the dogs to smelling odors on a cotton swab while withholding approach. For presentation of swabs during scanning sessions, odors for the familiar and unfamiliar human were obtained from the armpits, and odors for the familiar dog, unfamiliar dog, and the dog’s own self were obtained from the perineal-genital areas. In order to maintain compliance and motivation, the dogs were presented with interspersed reward trials during odor-presentation runs.

Analyses of the obtained imaging data focused on two regions of interest: the olfactory bulb and the caudate nucleus. The olfactory bulb was generally significantly activated by the task and this activation was non-differential across all five odor types. However, the caudate nucleus showed differential activation according to odor type. For all dogs, the caudate was maximally activated for the odor of a familiar person, suggesting that a positive reward association is in place for the scent of a familiar human, even in their physical absence. Interestingly, service dogs once again stood out with greater overall caudate activation as compared to dogs with other histories.

Collectively, the studies conducted by Jia, Pustovyy, et al. (2014) and Jia et al. (2015) and Berns et al. (2015) provide evidence for the efficacy of olfactory neuroimaging with dogs. The olfactory bulb has been consistently implicated in the processing of odors, and the use of anesthesia and the intensity of odors have been directly tied to olfactory processing. Additionally, reward-based processing of odorants was supported by activations in the caudate nucleus. Given the findings of these studies, there is clear evidence that fMRI can be utilized for future research to systematically explore olfactory processing in dogs.

Face Processing

In the first published fMRI investigation of face processing in awake dogs, Dilks et al. (2015) presented eight fMRI-experienced canine subjects (Cook et al., 2014) with movie clips and static images. The dogs viewed movie clips of human faces, scenes, objects, and scrambled objects, each for three seconds. In the static images condition, the dogs were presented with black and white images of human faces, dog faces, objects, scenes, and scrambled faces, each for 600 milliseconds. Imaging data was analyzed for six of the dogs, and movie clip contrasts localized dog and human face processing to the inferior temporal cortex in the right hemisphere. The data also revealed significant category effect for static images when face images were compared with objects and scenes. Because the response profile did not map onto the dog visual cortex, low-level feature processing is unlikely to account for the activation patterns seen in the temporal lobe. Rather, Dilks et al. (2015) conclude that the activations represent the first evidence of a face-processing region in dogs.

Resting State

The methodology of resting state fMRI has gained traction in the past decade because of distinct advantages it offers in terms of experimental design. Foremost, it does not require the subjects to perform any task, and hence is less stressful to subjects in human patient populations. Next, task-based activation studies have to be carefully designed so that any differences in responses may not be attributable to differences in task performance metrics (such as accuracy and reaction time). These measures may not be always possible to achieve. No such requirements are placed in resting state studies. Finally, analysis of task-based activation studies are primarily model driven (although data-driven methods also exist, it is difficult to interpret all time-locked evoked responses obtained from them), and this poses a challenge because one would have to explicitly model all sources of variance in measured data. However, in resting state studies, one could simply correlate experimentally measured time series from different brain regions (or perform an independent component analysis) to uncover underlying brain networks which are coevolving in time. These advantages of resting state studies in the human context are even more applicable in the context of awake dog imaging, as it is harder to make dogs perform a task (active or passive) and assure compliance and uniform performance.

Kyathanahally et al. (2015) used resting state fMRI to identify whether the default mode network (DMN), found reliably in humans (Buckner, Andrews-Hanna, & Schacter, 2008) and monkeys (Mantini et al., 2011) but much less frequently in rodents (Becerra, Pendse, Chang, Bishop, & Borsook, 2011; Upadhyay et al., 2011), exists in the domestic dog. Resting state fMRI is conducted with subjects that do not perform any cognitive tasks, but rather lie still with eyes open and relax. In humans, the core part of the DMN is active during rest and consists of two connected subnetworks—the posterior part consisting of the posterior cingulate cortex (PCC) and inferior parietal cortical areas, as well as an anterior part consisting of medial frontal structures. (Note that we are referring to the core part of the DMN and not the extended DMN, which also consists of lateral and medial temporal cortices. Please see Buckner et al., 2008, for details.) This network has been implicated in cognition and self-referential processing, and it has been found reliably in human resting state fMRI investigations. Additionally, this network’s activity is depressed when a patient is under anesthesia (Greicius et al., 2008). Most importantly, though, the long-range connectivity between the anterior and posterior parts of the DMN is lacking in very young children and seems to develop with age in humans (Fair et al., 2009). The establishment of long-range connectivity between anterior and posterior parts of the DMN is thought to facilitate large-scale information integration required for higher cognitive processes. Further, in humans, connectivity magnitude and associated network structure for various resting state networks, specifically the DMN, have been shown to be more informative in predicting behavior as well as traits compared to activation alone (Cole, Yarkoni, Repovš, Anticevic, & Braver, 2012; Jia, Hu, & Deshpande, 2014). Therefore, investigation of resting state networks in awake dogs is a promising area of research.

To assess the presence of a DMN in dogs and to understand the effects of anesthesia on its activation, Kyathanahally et al. (2015) scanned six dog subjects in both awake and anesthetized states. Seed-based and independent component analyses (ICA) were used and identified dissociation between the anterior and posterior regions of the DMN. Further, while this dissociation was seen for both awake and anesthetized dogs, the degree of dissociation was higher in anesthetized dogs in keeping with prior human results that anesthesia modulated the structure of resting state networks such as the DMN (Deshpande, Kerssens, Sebel, & Hu, 2010). In all, this investigation into resting state fMRI with dogs revealed comparative differences in the traits of the DMN between humans/monkeys and dogs, namely localized anterior and posterior subnetworks in dogs and a connected DMN in humans. The findings suggest differences in cognitive processing that are perhaps due to evolutionary differences.

In summary, the research and findings discussed herein are representative of the current excitement and expansion of canine cognitive research into functional imaging. As interest and conceptual foundations in this area continue to grow, the cognitive processes and behavior of the domestic dog may be better linked to develop a comprehensive understanding of man’s best friend. Further, such linking of cognition and behavior will allow for more informed comparisons to be made across species, as well as allow for greater understanding of the environment effects of domestication into the human social world. Though this area of research offers much promise, there are many challenges left to be addressed, both in respect to training and imaging methodologies and conceptual issues of cognitive investigation. In the following sections, we review those challenges most pertinent to future canine neuroimaging studies.

Methodological Issues and Solutions

The ability to obtain fMRI data on awake and unrestrained dogs provides expansive opportunity for research, but the advancement of this body of research does not come without significant methodological requirements and considerations. Here, we outline a variety of requisite considerations for safe scanning of dogs and provide solutions for each (Figure 2). We explore safety concerns, potential stressors for dog subjects, suitable scan parameters, desired experimental rigor, and the generalizability of imaging results to the greater dog population.

Safety

An important methodological concern of fMRI with dogs is the well-being and safety of individual canine subjects. Due to the extremely high sound pressure levels (SPLs) found in the MR environment, noise-induced discomfort and hearing loss are a concern for MR experiments. Sound levels in an MR suite range between 65 and 95 dB, with peaks from 120 to 131 dB, and these levels have been shown to result in significant short-term hearing loss in dogs (Venn et al., 2014). Physiological effects of sound stress include elevations in heart rate and blood pressure, as well as changes in metabolism. Additionally, the noise experienced in MRI may cause inner ear pain, distress, and inhibited communication abilities in dogs (Lauer, El-Sharkawy, Kraitchman, & Edelstein, 2012). Though long-term hearing loss due to scanner exposure has not been investigated in dogs (for long-term effects of cochlear damage in mice, see Kujawa & Liberman, 2009), such long-term effects could prove catastrophic to the safety and trainability of dogs, especially those working in search and rescue, bomb detection, and police work, as they are often physically separated from their handlers and require attention to vocal cues at a distance.

When designing experiments for canine fMRI, the most desirable solution to the problem of scanner noise is the use of sound-attenuating earmuffs in combination with careful selection of scan parameters. Earmuffs such as “Mutt Muffs” can provide upwards of 28dB of sound reduction when fitted properly to each dog (www.safeandsoundpets.com) and have been successfully used in past experiments (Cook et al., 2014). For further noise reduction where experimental demands allow, the use of a “whisper” mode and/or alterations in scanning parameters can attenuate sound pressure levels during scanning (Baker, 2013).

In addition to noise levels, safety concerns arise regarding the specific absorption rates (SAR) of radio frequencies for dogs. When RF energy in the scanner is absorbed by the body, tissues may rise in temperature. Tissue heating is almost always negligible; however, in order to eliminate the risk of thermal injury, SAR levels should be measured throughout MR scans (Smith, 2010). A variety of scan parameters influence SAR levels, including frequency, TR, coil selection, and the orientation of the body. Further, RF absorption by the body is determined by exposure duration, the thermoregulatory system, and health conditions (Shellock & Crues, 2004). In the case of human patients, the Food and Drug Administration (FDA) has established guidelines for the maximum allowable radio frequency absorption by tissue, defined as 4 watts per kilogram for a period of up to 15 minutes, or 3 watts per kilogram over the head for a period up to 10 minutes. Unfortunately, no such guidelines exist for nonhuman subjects, and this makes it the responsibility of researchers using dogs in MRI experiments to assess the SAR levels experienced by subjects. Until such research has been conducted, it seems best to adhere to the FDA standards for humans when working with dogs inside the scanner (Berns et al., 2013). Additionally, Berns et al. (2013) note that a reduction in flip angle may prevent rises in SAR levels, as do shorter scans. Given that SAR values generally increase with body weight, adhering to human SAR levels may be enough to protect the dogs, assuming that humans in general weigh more than domestic dogs. Nevertheless, it is a good idea to weigh the dogs and enter their body weight while running MR sequences so that the scanner can get a realistic estimate of SAR levels for individual dogs.

A large portion of the domestic dog population that may be ideal for study in the MRI environment is without recorded life history. That is, in many cases, there is some period of a dog’s life history that cannot be accounted for in terms of potential safety hazards (e.g., metallic objects in the body) or medical ailments and/or procedures. Because of this gap in life history, safety precautions must be taken before placing a dog in the MR environment. Of key concern is the potential for ferromagnetic objects that have been implanted in or ingested by a dog (Smith, 2010). The presence of conductive materials within the body may lead to excessive heating and third-degree burns (Shellock & Crues, 2004). Due to this potential hazard, it is imperative that each dog be screened for the presence of metallic objects in the body before being enlisted in an experiment and entering the scanner environment. Shellock and Crues (2004) note that while an object may be demonstrated as safe under a given set of MRI conditions, the same object may be unsafe in other conditions, especially those using stronger fields, greater radio frequencies, and different RF coils. This must be taken into consideration when assessing individual dogs for new scan conditions or replications in new scanner environments. Sensitive hand-held metal detectors custom-built for MR prescreening must be employed prior to the dog’s entering the scanner room in order to make sure that the dog does not have any ferromagnetic material inside its body.

Stress

Alongside the importance of safety within the scanner is the need to eliminate undue stress to the animal being investigated. Undue stress may prevent generalizability of the data. Stress may be defined as something that challenges the homeostasis within an individual or that places demands on the individual for which they do not have adequate resources. Such a stressor leads to physiological and behavioral responses that engage and mobilize the animal for action. Short-term stressors (such as the scanner environment) lead to increases in vigilance, alarm, and orientation, as well as physiologic responses such as tachycardia, metabolic changes, and increased respiration (Morgan & Tromborg, 2007). King et al. (2005) note that increased respiration and heart rate, as well as behavioral changes such as head motion, may alter the BOLD signal in such a way that activation changes in the brain may partially be attributed to noise rather than just the manipulation of an experimental variable.

In experiments aimed at acquiring imaging data from dogs, sources of stress may include noise, environmental novelty, enclosed spaces, restriction of movement, separation from the owner/handler, and long durations of assessment. Currently, the different techniques employed by dog neuroimaging laboratories inherently address stress reduction in training, as they lead to willful compliance. Methodologies using progression from mock to true scanning (e.g., Berns et al., 2012; Jia, Pustovyy, et al., 2014) gradually introduce dogs to the space constraints, noise levels, and time requirements of the scanning environment. A similar note can be made of the gradual introduction for dogs participating as observers in the model-rival method (Andics et al., 2014). Gradual exposure not only allows for reduced risk of stress in an overwhelming environment, but also gives trainers an opportunity to identify and eliminate stress signals emitted by a dog at any point in training. Acclimating dogs to the aforementioned sources of stress during training is essential to successful scanning, and improvements in training techniques aimed at further reducing sources of stress will be important for future research. In addition, stressors not accounted for or eliminated by training methodologies may still be reduced by thoughtful selection of experimental parameters, and such parameter manipulations should be explored.

Auditory experiments in fMRI pose a unique challenge due to the high sound pressure levels within the scanner room. In order to optimize the amount of auditory stimuli that can be heard and processed, a sparse temporal sampling (STS) procedure may be used, as in Bach et al. (2013) and Andics et al. (2014). This sort of imaging paradigm allows for periods of scanner silence during which the auditory stimuli of interest may be presented without interference. Because the BOLD response lags in time behind the neural response, acquiring imaging data shortly after, but not during, auditory stimulus presentation allows capture of the stimulus-evoked hemodynamic response function. Bach and colleagues (2013) implemented this procedure and compared it with a standard scanning procedure without intermittent silent periods. Though the dogs were anesthetized and some degree of signal attenuation could be expected, the authors found reliable activation of auditory areas for all of the dogs. Of interest to the use of STS, conditions during which auditory stimuli were presented during silent period resulted in significantly higher activation levels than those presented during continuous scanning, supporting past evidence that STS procedures may increase signal strength by 21%. When using awake dogs for auditory fMRI experiments, the use of STS procedures is an ideal way to ensure optimal BOLD signal and may be easily introduced into positive-reinforcement behavioral training.

Scanning Parameters

As mentioned in relation to noise and stress reduction, careful selection of the imaging sequence and scanning parameters is an essential component of canine fMRI methodology. While we do not attempt to explain all technological aspects here, interested readers may find the cited articles useful. Goals of parameter selection are aimed at increasing efficiency in data collection and in subsequent processing of obtained data. Parameters of consideration include the choice of (a) field strength, (b) imaging sequence and its parameters, (c) RF coil, (d) signal contrast, and (e) body position. First, in some cases, the experimenter may have the option of scanning in a 3T or 7T scanner. Martín-Vaquero et al. (2011) sought to parse out differences in image quality between the two field strengths using anesthetized dogs. Contrary to what might be expected, only eight of 32 anatomical structures had better image quality in the 7T scanner as compared to the 3T. Most structures (19/32) were of comparable image quality for both scanners, and five were better at 3T field strength. Specifically, Martín-Vaquero and colleagues found that when performing high-resolution scans, the noise due to magnetic susceptibility and chemical shift were much more apparent in the 7T scanner, and thus suggest using a 3T scanner for these sequences. However, this contradicts a large body of evidence that suggests that increasing field strengths offers substantial benefits in terms of SNR (Duyn, 2012). Unlike the 3T, getting fMRI data of good quality from 7T requires the choice of proper sequence parameters (which change with field strength) and use of higher order shimming. Therefore, if done right, higher field strengths such as the 7T could potentially offer increased SNR and smaller voxels, which are crucial in canine imaging since brain structures in dogs are relatively smaller than in humans. Such advantages of ultra-high field have already been demonstrated for imaging relatively smaller structures in the human brain (Denison, Vu, Yacoub, Feinberg, & Silver, 2014; Satpute et al., 2013; Suthana et al., 2015) as well as other smaller mammals such as rodents (Schafer, Kida, Xu, Rothman, & Hyder, 2006).

Second, in regard to choice of sequence, Chen et al. (2012) found that in MR imaging of awake monkeys, the signal-to-noise ratio (SNR) was reduced by using segmented echo-planar imaging (EPI) in exchange for single-shot EPI, and also by optimizing echo time. Recent innovations such as the multiband EPI (Feinberg et al., 2010) will allow us to choose a shorter TR for the same voxel size compared to regular EPI. Also, the use of parallel imaging in sequences may not only reduce scan time and acoustic noise, but also be beneficial for mitigating artifacts due to off-resonance effects (Golay, de Zwart, Ho, & Sitoh, 2004). Further, zoomed resolution approaches may be employed for obtaining higher spatial resolution from specific structures (Yacoub, Harel, & Uğurbil, 2008). For performing dog fMRI in 3T scanners, which are widely available, the choice of scan parameters depends on the scientific question being investigated. The available SNR could be traded for either spatial or temporal resolution. Therefore, if the scientific investigation surrounds smaller structures such as the nucleus accumbens or other nuclei in the brain stem, then it may be a good idea to have smaller voxel size and a longer TR. However, if one is interested in temporal properties of the signal in relatively larger regions of the cortex, employing a shorter TR and relatively larger voxel size may be preferable. In general, since the size of the dog brain is smaller than that of the human brain (Roth & Dicke, 2005), scanning sequences must strive for achieving voxel sizes that are smaller than that typically used in human scans. Briefly returning to stress and safety concerns, the experimenter must also make note of the effects of timing and slice selection on noise and scan duration and select accordingly in order to create optimal experimental conditions.

Third, as noted before, different types of coils (e.g., head, neck, human knee, flex) have been used by different investigators. It is difficult to compare the SNRs obtained from these coils since corresponding raw data is not available in the public domain. Nevertheless, an important factor to bear in mind irrespective of the coil used is that one needs to ensure that the coil is close to the dog’s brain. Standard quality control procedures must be employed to assess whether the signal obtained from a coil is of acceptable quality. However, given the wide range of options available (such as surface coils, linear volume coils, birdcages, phased arrays), further studies are required to determine what type of coil is necessary and sufficient for performing routine fMRI in awake dogs. Specifically, efforts must be made for developing size- and shape-matched head coils for imaging the canine brain for achieving both high sensitivity and specificity. If successful, the higher performance obtained from a custom-built coil may aid in obtaining much higher temporal and spatial resolution, as well as much larger functional contrast-to-noise ratio. Shorter measurement times could help to avoid image deterioration by motion artifacts, for instance physiological motion artifacts due to respiration and heart rate within the skull. The performance of such a custom-built coil could then be compared with human coils that have been adapted for canine imaging so far in order to determine whether the investments required for building custom coils in terms of capital and technical expertise are indeed justified.

Fourth, various options are available for imaging function in the brain using MRI. These include the widely used BOLD signal contrast, perfusion-based methods such as arterial spin labeling (ASL; Petcharunpaisan, Ramalho, & Castillo, 2010), and Vascular space occupancy (VASO; Lu & van Zijl., 2012). However, the use of signal contrast mechanisms other than BOLD in animal research has been primarily driven by pharmacological fMRI in preclinical studies (Nasrallah, Lee, & Chuang, 2012). Unless a study is using a dog model of human illness and testing pharmacological effects of different drugs, the BOLD contrast provides the right ingredients in terms of sensitivity required for probing normal cognition in the canine brain. Finally, all canine fMRI studies published so far have used dogs in the prone position, which seems like a natural choice. Further, the coil geometry also lends itself to this position very well. We do not see a reason to image dogs in other positions such as supine, unless specifically required for an application.

Experimental Rigor

Given that canine fMRI is a nascent field, initial reports were more interested in demonstrating the proof of concept rather than follow experimental rigor that is customary in human fMRI studies. We can observe the following shortcomings: (a) lack of controlled delivery of stimuli such that they are devoid of subjectivity and timing error (e.g., Berns et al., 2012), which can be detrimental in general linear modeling of the BOLD signal, (b) temporal discontinuity between trials (e.g.. Berns et al., 2013), which can invalidate assumptions made during activation analyses, (c) small sample sizes resulting in smaller effect sizes, (d) lack of compliance measurements, and (e) inadequate attention toward motion artifacts. Given that the field has moved beyond the proof-of-concept stage, we suggest that future studies make every attempt to employ best practices used in human fMRI studies. For example: (a) given that controlled delivery of olfactory stimuli by a custom-built device was demonstrated by Jia, Pustovyy, et al. (2014) and likewise with auditory stimuli by Andics et al. (2014), similar approaches could be used while presenting stimuli of other modalities as is routinely done in human fMRI research; (b) training dogs to remain in the scanner for a longer period of time (>1–2 minutes, as in Jia, Pustovyy, et al., 2014, and Andics et al., 2014). would naturally obviate the necessity to have the dog pull out of the coil after every trial (or two) or move between trials, which introduces temporal discontinuities in the signal and necessitates that data be discarded; (c) recruiting larger numbers of dogs and/or obtaining more runs from available dogs coupled with spatial normalization techniques (discussed next) would allow group inferences with robust statistics rather than qualitative inferences in individual dogs; (d) if dogs are presented visual stimuli, manual checking of compliance by other humans or automatic compliance measurement via eye-tracking is desirable.

In regard to motion control, there is increasing awareness in the neuroimaging community about the detrimental effects of head motion on fMRI data quality (Power et al., 2014). Canine fMRI studies have taken comfort in the fact that the motion parameters obtained from rigid body registration (i.e., three translations and three rotations) can be inspected to choose only data not corrupted by motion (e.g., Berns et al., 2013). However, this approach does not take into account the facts that: (a) spin history effects and through-plane motion are not modeled in rigid-body registration or by censoring only affected TRs, (b) the frame-wise displacement of different voxels in the brain are different from each other, and (c) rapid motion that occurs between TRs can affect data quality in ways that cannot be restored by rigid-body registration or even censoring. Some of these issues were partially addressed by Jia, Pustovyy, et al. (2014) and Kyathanahally et al. (2015) by employing a single external infrared camera to record dogs’ head motion with high temporal resolution (order of milliseconds) and spatial precision (order of micrometers) and then correcting for those effects post hoc. However, in an ideal scenario, we suggest employing prospective motion correction by either employing an external camera (Todd, Josephs, Callaghan, Lutti, & Weiskopf, 2015; Maclaren et al., 2012) or using image-based tracking (as in 3D PACE; Thesen, Heid, Mueller, & Schad, 2000).

Generalizability

Across breeds, the domestic dog demonstrates significant variability in brain morphology (Roberts, McGreevy, & Valenzuela, 2010) and therefore presents a challenge when attempting to spatially normalize images into a stereotactic space and generalize findings from fMRI experiments with dogs. In order to encourage spatial normalization across experiments for imaging data obtained from canine fMRI, Datta et al. (2012) used 15 mesocephalic dogs to create a 3T template. However, they note the influence of the mesocephalic characteristics of the subjects and suggest that differences in encephalization may lead to differences in cortical folding that preclude simple transformations, rendering the normalized template suitable only for dogs similar in skull and brain structure to those used in development of the template. Such concern is evidenced by brain structure analyses conducted by Roberts et al. (2010), in which it was found that brachycephalic dogs possess skulls with comparatively rotated cerebral hemispheres, pitched brain angle at the anterior pole, and repositioning of the olfactory lobe. To combat the exclusivity of normalized images due to such differences, Bach et al. (2013) suggest a wider scope of canine structural imaging to include a variety of skull shapes. Along those lines, they suggest the development of a database similar to what is seen in BrainMap for human data (Laird, Lancaster, & Fox, 2005), where imaging data from a multitude of canine fMRI studies may be accumulated, analyzed, and normalized. In the meantime, thought should be given to the viability of using a normalized template with a given population of dogs. For those not matching optimally to the template, it will be best to use within-group spatial normalization as seen in Jia, Pustovyy, et al. (2014).

Conceptual Issues and Solutions

Once the canine researcher has addressed the methodological challenges described above, another set of challenges related to conceptual foundations must be addressed. Of utmost importance is the knowledge of what is already known about the structure and function of the domestic dog brain. This knowledge provides the framework with which to build viable hypotheses and draw conclusions from sometimes convoluted activation data. When developing these hypotheses and conclusions, it is also important to use a theoretical background and logical arguments in order to parse out the many possibilities for structure-function relationships. Steps to develop this conceptual framework will now be discussed.

Of great benefit to both fMRI methodology and comparative cognition research is the shared natural environment of dogs and humans. Miklósi and Topál (2013) discuss the exploration of “human-like” and “infant-like” functional traits in dogs and emphasize the importance of careful control in experiments aimed at identifying cognitive mechanisms. When it comes to domestic dogs, criteria of task-demand and environmental similarity have already been met and are easily accounted for when designing a task. This is because the natural environment of dogs is the human environment, rather than the environment of their distant relatives and ancestors. Because dogs are encultured in human society, there are far fewer methodological concerns regarding environmental generalizability when it comes to interaction with experimenters, presentation of commands and/or rewards by humans, stimuli found in the human environment, and ambient aspects such as sound and lighting. Here, these aspects of generalizability can be assumed so long as the experimenter takes strides to equate exposure and task difficulty for humans and dogs.

The crux of in-scanner validity may be the mode of stimulus presentation. As with most laboratory experiments, the ability to replicate real-life scenarios in the scanner is limited, as visual stimuli are typically presented via projector screen, auditory stimuli via speaker or earphone, and olfactory stimuli via localized sampling. Because of the constrained nature of such presentations, skeptics may rightly question whether activations are representative of what would occur in a natural situation. Evidence from prior canine fMRI studies would suggest that imaging data from experimental analogs are at least indicative, if not wholly representative, of real-world scenarios (e.g., auditory samples, Andics et al. 2014; olfactory samples, Berns et al. 2015). However, findings from other research keep certainty of generalizability at bay. For example, Snow et al. (2011) found that in humans, patterns of activation differed between presentation of 3D objects and 2D images, highlighting that 3D objects offer more information about a stimulus than a 2D image of the same stimulus, and may even provide motivation for attention due to the possibility of interaction. Given the possibility of generalizability risks and the lack of literature investigation regarding this risk in dogs, researchers should attempt to eliminate these concerns by providing naturalistic stimuli whenever possible. As noted, this issue of stimulus generalizability between stimuli in the testing environment and the real world (or 2D vs. 3D stimuli) is not unique to the MR scanner environment (e.g., Spetch, 2010). However, steps should be taken to minimize the discrepancy between stimuli and their real-life counterparts as well as acknowledge this remaining discrepancy when analyzing and discussing imaging data.

By its nature, fMRI requires multiple instances of neural activations and corresponding BOLD responses to create a clear localization of function in the brain. Due to this repetitive nature, attention and habituation become concerns as number of presentations and time in the scanner increase. These concerns are exacerbated in dogs, as there is not a good means of communicating the need for attention, nor is there a desirable method for tracking decreasing attention span or habituation while behavioral responses are being withheld. Snow et al. (2011) found that when presenting human participants with repeating instances of a single 2D image, functional data showed robust repetition effects and degradation of signal. It is desirable to be cautious of such effects in canine data, as similar repetition effects have been found in non-human primates (Miller, Li, & Desimone, 1991). Further, continued attention to any stimuli is a concern when requiring dogs to remain still for extended periods of time. Cook et al. (2014) anticipated attentional changes with increased scan duration when presenting dogs with familiar human, unfamiliar human, and computerized image presentations of reward signals. To combat deficits in attention, the researchers arranged the experimental conditions such that the signals presented later in scanning would be more stimulating and motivational (in this case, presentations of a familiar human). The disadvantage of such an approach would be introduction of order effects.

In any fMRI experiment, there exists the possibility of overzealous or inaccurate connections being made between structure and function. This becomes especially true when attempting to separate active and passive processing in nonhuman animals, as we cannot be sure what sort of cognitive process they are engaging in without concurrent behavioral measures. Whereas human participants may exhibit a given cognitive process behaviorally during a scan (e.g., via a response box) or report on strategies after scanning, we do not have the luxury of obtaining such information from dog subjects. The importance of distinguishing between active and passive processing is great, as human research has shown differential activations between the two cognitive processes (e.g., O’Craven, Rosen, Kwong, Treisman, & Savoy, 1997). As a first measure to preventing inaccurate designation of cognitive process to activation, researchers can ensure that dogs have had extensive out-of-scanner training on a behavioral measure for the cognitive task in question. For example, in the reward signal research previously discussed (e.g., Cook et al., 2014), dogs were extensively exposed to the experimental conditions before entering the scanner. Such exposure ensures the highest probability possible that the dogs will engage in the same cognitive functioning during scans presenting the same conditions.

Post hoc attributions of function to structure also pose a risk to the quality and validity of canine fMRI studies. Extrapolating a cognitive process from an area of activation (usually one that was unexpected) is known as reverse inference (Poldrack, 2006) and is better left for creating hypotheses for future research rather than making definitive conclusions. Poldrack (2006) suggests that reverse inference may hold ground if a margin of confidence and probability is used, such that areas that are activated by a large number of cognitive processes are given low levels of confidence when engaging in reverse inference, but areas that are activated by few processes are given a higher level of confidence. Henson (2006) explains that while reverse inference may not be the ideal way to look at data, it does lend itself to the identification of successful experimental replication and the ability to connect relating cognitive processes.

To further address the potential complications of reverse inference, researchers may make use of forward inference by seeking qualitatively different brain activations when comparing competing cognitive theories (Henson, 2006). That is, one can design an experiment with conditions that engage different cognitive processes according to one theory, but that do not in another theory. With this framework, the resulting activity patterns will be evidence for one theory. In a more general sense, contrasting a working hypothesis with a null hypothesis will allow more concise conclusions to be made from data obtained in functional imaging studies.

Methodological adjustments may also be suitable for improving the conceptual issue of inference. By considering the potential utility of pure insertion, researchers may enhance conclusions that differences in imaging data are due to the differences in experimental conditions, without the interference of confounding variables and extraneous sources of variance (Henson, 2006). This is especially important in studies with dogs, again due to the lack of feedback and explanation of strategies that exists in human research. Henson (2006) explains that when using pure insertion, if no variables other than those of experimental interest vary, then there is no reason to expect an underlying qualitative difference in brain activity. Thus, if the independent variable is precisely and singularly manipulated in studies of canine cognition using fMRI, then researchers may be better suited to make conclusions about canine brain structure-function relationships.

Applications and Future Directions

One of the most pressing applications of fMRI research with domestic dogs is the investigation of bio-behavioral markers of successful working dogs. Cobb, Branson, McGreevy, Lill, and Bennett (2015) define a working dog as one which is “operational in a private industry, government, assistant, or sporting context,” while noting that these dogs may also simultaneously serve as human companions. Typically, these dogs fulfill roles in emotional support (service dogs) or in threat prevention (odor detection dogs). The working dog industry has been and is continuing to grow at a rapid rate, with dogs being trained in increasingly complex duties and the breeding programs producing greater numbers of puppies. Unfortunately, upwards of 50% of these dogs fail at some point in their training (e.g., Dalibard, 2009; Maejima et al., 2007; Slabbert & Odendaal, 1999; Wilsson & Sinn, 2012), resulting in large-scale concerns of wasted funding, lost revenue, and a deficit in ethical considerations for individual dogs as transitions between working and pet dog roles can lead to stress and adoption difficulties.

While training programs and working purposes vary among working dog organizations, fMRI methodologies can be developed to identify common activation areas and patterns, as well as behavioral correlates, among dogs that pass rigorous training and succeed in the workforce. Discovering such bio-behavioral markers using cross-sectional and longitudinal designs over specific training histories may lead to better standards of identification, training, and treatment of dogs intended for working roles. An endophenotype for a specific working dog role may be developed in a stepwise fashion by using behavioral assessments to identify the most viable behavioral tendencies to fulfill the role and then correlating scores on such identifiers with brain activation data, such as neural responsiveness to target odors, auditory cues, or visual markers.

The use of fMRI, while in its infancy, may also bolster the reliability and validity of prior cognitive research with dogs. By adapting behavioral tasks for use in the scanner and/or correlating behavioral measures with in-scanner techniques, underlying cognitive processes may be better examined and evidenced. Such adaptation and correlation in human studies have preceded dog research, as the scientific community has successfully translated questions of human cognition historically targeted by self-report and behavioral measures into tasks to be completed in MRI. Such imaging capabilities have allowed simultaneous behavioral measurement of cognitive processes (i.e., the direct responding from the participant) and brain activation patterns that capture the once-illusive covert neural processing of cognitive tasks. Further, the convergence of multiple measures allows for thorough assessment of the strengths and weaknesses of individual cognitive theories. Building on the rise of cognitive neuroscience in human cognition, the merger of behavioral and neural responses by dog subjects will provide researchers with comprehensive and expansive data sets, and questions that were previously left to speculation may be explained in terms of neural structure and activation. Finally, the gamut of advanced analysis methods in human fMRI research, such as connectivity models (Jia, Hu, et al., 2014) and multivariate pattern analysis and learning models (Deshpande, Libero, Sreenivasan, Deshpande, & Kana, 2013) can be employed on dog fMRI data, potentially alleviating some of the issues with traditional activation models and giving us new insights into underlying neural mechanisms.

Comparative Mechanisms

Comparisons between human and dog cognitive processes may be directly analyzed with the use of comparative fMRI methods (e.g., Andics et al., 2014). By presenting the same task to both humans and dogs, activation areas and patterns may be directly discussed. For example, questions of domestication and development of heterospecific social processing may target analogous neural structures and networks in dogs and humans over the individual lifespan. If one seeks to understand the similarities between emotion recognition in humans and dogs, then they may present both human and dog participants with the same stimulus set (e.g., humans smiling versus frowning) in the scanner and investigate neural activation patterns. This form of identical measurement eliminates common points of ambiguity in results due to methodology, such as differences in stimulus presentation and environment. With increased clarity of comparative measures between humans and dogs, investigations into the evolutionary development of neural structure and processes across species will be broader, more robust, and easier to implement. There is great potential to better understand which processes are due to convergent evolution or homology and the interaction between phenotype and ontology.

Conclusions

In summary, canine fMRI is a new and exciting frontier in comparative cognition research. The trainability of dogs, as well as their close social connection to humans, makes them a prime species for study in the MR scanner in an awake and unrestrained state. With continued refinement of methodology and conceptual ideas highlighting the utility of fMRI with dogs, we can expect to see an increase in the information we know about the function and structure of the canine mind. When embarking on a study of domestic dogs in fMRI, researchers must carefully consider the experimental design parameters, from subject to coil selection and from stimulus modality to presentation order. Conceptual issues should be addressed during experimental design, rather than post hoc, in order to ensure the viability of imaging data and the conclusions that are reached. With careful attention to each interlocking aspect of fMRI design, not only should the comparative cognition literature advance, but the neuroimaging literature as a whole should advance as well. To be sure, the recent combination of the “rise of the dogs” with neuroimaging has formed the foundation for the cognitive neuroscience of canine cognition.

References

Anaya García, M. S., Hernández Anaya, J. S., Marrufo Meléndez, O., Velázquez Ramírez, J. L., & Palacios Aguiar, R. (2015). In Vivo study of cerebral white matter in the dog using diffusion tensor tractography. Veterinary Radiology & Ultrasound, 56(2), 188-195. doi:10.1111/vru.12211

Andics, A., Gácsi, M., Faragó, T., Kis, A., & Miklósi, Á. (2014). Voice-sensitive regions in the dog and human brain are revealed by comparative fMRI. Current Biology, 24(5), 574–578. doi:10.1016/j.cub.2014.01.058

Bach, J.-P., Lüpke, M., Dziallas, P., Wefstaedt, P., Uppenkamp, S., Seifert, H., & Nolte, I. (2013). Functional magnetic resonance imaging of the ascending stages of the auditory system in dogs. BMC Veterinary Research, 9, 210. doi:10.1186/1746-6148-9-210

Baker, M. A. (2013). Reduction of MRI acoustic noise achieved by manipulation of scan parameters: A study using veterinary MR sequences. Radiography, 19(1), 11–16. doi:10.1016/j.radi.2012.09.004

Baxi, M., Robinson, J. L., Katz, J., Waggoner, W., Beyers, R., Morrison, E., . . . Deshpande, G. (Under review). Characterization of structural connectivity of the default mode network in dogs using diffusion tensor imaging. Scientific Reports, under review.

Becerra, L., Pendse, G., Chang, P. C., Bishop, J., & Borsook, D. (2011). Robust reproducible resting state networks in the awake rodent brain. PLoS One, 6(10), e25701. doi:10.1371/journal.pone.0025701

Bensky, M. K., Gosling, S. D., & Sinn, D. L. (2013). The world from a dog’s point of view: A review and synthesis of dog cognition research. Advances in the Study of Animal Behavior, 45, 209–406. doi:10.1016/B978-0-12-407186-5.00005-7

Berns, G. S., Brooks, A. M., & Spivak, M. (2012). Functional MRI in awake unrestrained dogs. PLoS ONE, 7(5), e38027. doi:10.1371/journal.pone.0038027

Berns, G. S., Brooks, A., & Spivak, M. (2013). Replicability and heterogeneity of awake unrestrained canine fMRI responses. PLoS ONE, 8(12), e81698. doi:10.1371/journal.pone.0081698

Berns, G. S., Brooks, A. M., & Spivak, M. (2015). Scent of the familiar: An fMRI study of canine brain responses to familiar and unfamiliar human and dog odors. Behavioural Processes, 110, 37–46. doi:10.1016/j.beproc.2014.02.011

Buckner, R. L., Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain’s default network: Anatomy, function, and relevance to disease. Annals of the New York Academy of Sciences, 1124, 1–38. doi:10.1196/annals.1440.011

Chen, G., Wang, F., & Dillenburger, B. (2012). Functional magnetic resonance imaging of awake monkeys: Some approaches for improving imaging quality. Magnetic Resonance Imaging, 30(1), 36–47. doi:10.1016/j.mri.2011.09.010

Cobb, M., Branson, N., McGreevy, P., Lill, A., & Bennett, P. (2015). The advent of canine performance science: Offering a sustainable future for working dogs. Behavioural Processes, 110, 96–104. doi:10.1016/j.beproc.2014.10.012

Cole, M. W., Yarkoni, T., Repovš, G., Anticevic, A., & Braver, T. S. (2012). Global connectivity of prefrontal cortex predicts cognitive control and intelligence. The Journal of Neuroscience, 32(26), 8988–8999. doi:10.1523/JNEUROSCI.0536-12.2012

Cook, P. F., Spivak, M., & Berns, G. S. (2014). One pair of hands is not like another: Caudate BOLD response in dogs depends on signal source and canine temperament. PeerJ, e596. doi:10.7717/peerj.596

Dalibard, G. H. (2009). Parameters influencing service dogs’ quality of response to commands: Retrospective study of 71 dogs. Journal of Veterinary Behavior: Clinical Applications and Research, 4(1), 19–24. doi:10.1016/j.jveb.2008.08.008

Datta, R., Lee, J., Duda, J., Avants, B. B., Vite, C. H., Tseng, B., . . . Aguirre, G. K. (2012). A digital atlas of the dog brain. PLoS ONE, 7(12), e52140. doi:10.1371/journal.pone.0052140

De Groof, G., Jonckers, E., Güntürkün, O., Denolf, P., Van Auderkerke, J., & Van der Linden, A. (2013). Functional MRI and functional connectivity of the visual system of awake pigeons. Behavioural Brain Research, 239(1), 43–50. doi:10.1016/j.bbr.2012.10.044

Denison, R. N., Vu, A. T., Yacoub, E., Feinberg, D. A., & Silver, M. A. (2014). Functional mapping of the magnocellular and parvocellular subdivisions of human LGN. NeuroImage, 102, 358–369. doi:10.1016/j.neuroimage.2014.07.019

Deshpande, G., Kerssens, C., Sebel, P. S., & Hu, X. (2010). Altered local coherence in the default mode network due to sevoflurane anesthesia. Brain Research, 1318, 110–121. doi:10.1016/j.brainres.2009.12.075

Deshpande, G., Libero, L. E., Sreenivasan, K. R., Deshpande, H. D., & Kana, R. K. (2013). Identification of neural connectivity signatures of autism using machine learning. Frontiers in Human Neuroscience, 7, 670. doi:10.3389/fnhum.2013.00670

Dilks, D. D., Cook, P., Weiller, S. K., Berns, H. P., Spivak, M. H., & Berns, G. (2015). Awake fMRI reveals a specialized region in dog temporal cortex for face processing. PeerJ, e1115. doi:10.7717/peerj.1115

Duyn, J. H. (2012). The future of ultra-high field MRI and fMRI for study of the human brain. Neuroimage, 62(2), 1241–1248. doi:10.1016/j.neuroimage.2011.10.065

Fair, D. A., Cohen, A. L., Power, J. D., Dosenbach, N. U., Church, J. A., Miezin, F. M., . . . Petersen, S. E. (2009). Functional brain networks develop from a “local to distributed” organization. PLoS Computational Biology, 5(5), e1000381. doi:10.1371/journal.pcbi.1000381

Feinberg, D. A., Moeller, S., Smith, S. M., Auerbach, E., Ramanna, S., Gunther, M., . . . Yacoub, E. (2010). Multiplexed echo planar imaging for sub-second whole brain fMRI and fast diffusion imaging. PloS ONE, 5(12), e15710. doi:10.1371/journal.pone.0015710

Golay, X., de Zwart, J. A., Ho, Y. C. L., & Sitoh, Y. Y. (2004). Parallel imaging techniques in functional MRI. Topics in Magnetic Resonance Imaging, 15(4), 255–265. doi:10.1097/01.rmr.0000142829.79609.d4

Greicius, M. D., Kiviniemi, V., Tervonen, O., Vainionpää, V., Alahuhta, S., Reiss, A. L., & Menon, V. (2008). Persistent default–mode network connectivity during light sedation. Human Brain Mapping, 29(7), 839–847. doi:10.1002/hbm.20537

Gross, B., Garcia-Tapia, D., Riedesel, E., Ellinwood, N. M., & Jens, K. (2010). Normal canine brain maturation at magnetic resonance imaging. Veterinary Radiology and Ultrasound, 51(4), 361–373. doi:10.1111/j.1740-8261.2010.01681.x

Henson, R. N. (2004). Analysis of fMRI timeseries: Linear Time-Invariant models, event-related fMRI and optimal experimental design. In Frackowiak, Friston, Frith, Dolan, & Price (Eds.), Human Brain Function, 2nd ed. (pp. 793-822). London: Elsevier.

Henson, R. (2006). Forward inference using functional neuroimaging: Dissociations versus associations. Trends in Cognitive Sciences, 10(2), 64–69. doi:10.1016/j.tics.2005.12.005

Hsu, Y., & Serpell, J. A. (2003). Development and validation of a questionnaire for measuring behavior and temperament traits in pet dogs. Journal of the American Veterinary Medical Association, 223(9), 1293–1300. doi:10.2460/javma.2003.223.1293

Jacqmot, O., Van Thielen, B., Fierens, Y., Hammond, M., Willekens, I., Schuerbeek, P. V. (2013). Diffusion tensor imaging of white matter tracts in the dog brain. The Anatomical Record, 296, 340–349. doi:10.1002/ar.22638

Jia, H., Hu, X., & Deshpande, G. (2014). Behavioral relevance of the dynamics of the functional brain connectome. Brain Connectivity, 4(9), 741–759. doi:10.1089/brain.2014.0300

Jia, H., Pustovyy, O. M., Waggoner, P., Beyers, R. J., Schumacher, J., Wildey, C., . . . Deshpande, G. (2014). Functional MRI of the olfactory system in conscious dogs. PLoS ONE, 9(1), e86362. doi:10.1371/journal.pone.0086362

Jia, H., Pustovyy, O., Wang, Y., Waggoner, P., Beyers, R., Schumacher, J., . . . Desphande, G. (2015). Enhancement of odor-induced activity in the canine brain by zinc nanoparticles: A functional MRI study in fully unrestrained conscious dogs. Chemical Senses. doi:10.1093/chemse/bjv054

King, J. A., Garelick, T. S., Brevard, M. E., Chen, W., Messenger, T. L., Duong, T. Q., & Ferris, C. F. (2005). Procedure for minimizing stress for fMRI studies in conscious rats. Journal of Neuroscience Methods, 148(2), 154–160. doi:10.1016/j.jneumeth.2005.04.011

Kujawa, S. G., & Liberman, M. C. (2009). Adding insult to injury: Cochlear nerve degeneration after “temporary” noise-induced hearing loss. The Journal of Neuroscience, 29(45), 14077–14085. doi:10.1523/JNEUROSCI.2845-09.2009

Kyathanahally, S. P., Jia, H., Pustovyy, O. M., Waggoner, P., Beyers, R., Schumacher, J., . . . Deshpande, G. (2015). Anterior–posterior dissociation of the default mode network in dogs. Brain Structure and Function, 220(2), 1063–1076. doi:10.1007/s00429-013-0700-x

Lahti, K. M., Ferris, C. F., Li, F., Sotak, C. H., & King, J. A. (1998). Imaging brain activity in conscious animals using functional MRI. Journal of Neuroscience Methods, 82(1), 75–83. doi:10.1016/S0165-0270(98)00037-5

Laird, A. R., Lancaster, J. L., & Fox, P. T. (2005). BrainMap: The social evolution of a functional neuroimaging database. Neuroinformatics, 3, 65–78. doi:10.1385/NI:3:1:065

Lauer, A. M., El-Sharkawy, A. M. M., Kraitchman, D. L., & Edelstein, W. A. (2012). MRI acoustic noise can harm experimental and companion animals. Journal of Magnetic Resonance Imaging, 36(3), 743–747. doi:10.1002/jmri.23653

Lu, H., & van Zijl, P. C. (2012). A review of the development of Vascular-Space-Occupancy (VASO) fMRI. Neuroimage, 62(2), 736–742. doi:10.1016/j.neuroimage.2012.01.013

Maclaren, J., Armstrong, B. S., Barrows, R. T., Danishad, K. A., Ernst, T., Foster, C. L., . . . Zaitsev, M. (2012). Measurement and correction of microscopic head motion during magnetic resonance imaging of the brain. PLoS ONE 7(11): e48088 doi:10.1371/journal.pone.0048088

Maejima, M., Inoue-Murayama, M., Tonosaki, K., Matsuura, N., Kato, S., Saito, Y., . . . Ito, S. (2007). Traits and genotypes may predict the successful training of drug detection dogs. Applied Animal Behaviour Science, 107, 287–298. doi:10.1016/j.applanim.2006.10.005

Mantini, D., Gerits, A., Nelissen, K., Durand, J. B., Joly, O., Simone, L., . . . Vanduffel, W. (2011). Default mode of brain function in monkeys. The Journal of Neuroscience, 31(36), 12954-12962. doi:10.1523/JNEUROSCI.2318-11.2011

Martin, C., Martindale, J., Berwick, J., & Mayhew, J. (2006). Investigating neural–hemodynamic coupling and the hemodynamic response function in the awake rat. Neuroimage, 32(1), 33–48. doi:10.1016/j.neuroimage.2006.02.021

Martín-Vaquero, P., da Costa, R. C., Echandi, R. L., Tosti, C. L., Knopp, M. V., & Sammet, S. (2011). Magnetic resonance imaging of the canine brain at 3 and 7T. Veterinary Radiology & Ultrasound, 52(1), 25–32. doi:10.1111/j.1740-8261.2010.01747.x

Miklósi, Á., & Topál, J. (2013). What does it take to become ‘best friends’? Evolutionary changes in canine social competence. Trends in Cognitive Sciences, 17, 287–294. doi:10.1016/j.tics.2013.04.005

Miller, E. K., Li, L., & Desimone, R. (1991). A neural mechanism for working and recognition memory in inferior temporal cortex. Science, 254(5036), 1377–1379. doi:10.1126/science.1962197

Morgan, K. N., & Tromborg, C. T. (2007). Sources of stress in captivity. Applied Animal Behaviour Science, 102(3–4), 262–302. doi:10.1016/j.applanim.2006.05.032

Nasrallah, F. A., Lee, E. L., & Chuang, K. H. (2012). Optimization of flow–sensitive alternating inversion recovery (FAIR) for perfusion functional MRI of rodent brain. NMR in Biomedicine, 25(11), 1209–1216. doi:10.1002/nbm.2790

O’Craven, K. M., Rosen, B. R., Kwong, K. K., Treisman, A., & Savoy, R. L. (1997). Voluntary attention modulates fMRI activity in human MT–MST. Neuron, 18(4), 591–598. doi:10.1016/S0896-6273(00)80300-1

Ogawa, S., Lee, T. M., Kay, A. R., & Tank, D. W. (1990). Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proceedings of the National Academy of Sciences, 87(24), 9868–9872. doi:10.1073/pnas.87.24.9868

Ogawa, S., Tank, D. W., Menon, R., Ellermann, J. M., Kim, S. G., Merkle, H., & Ugurbil, K. (1992). Intrinsic signal changes accompanying sensory stimulation: Functional brain mapping with magnetic resonance imaging. Proceedings of the National Academy of Sciences, 89(13), 5951–5955. doi:10.1073/pnas.89.13.5951

Petcharunpaisan, S., Ramalho, J., & Castillo, M. (2010). Arterial spin labeling in neuroimaging. World Journal of Radiology, 2(10), 384. doi:10.4329/wjr.v2.i10.384

Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences, 10(2), 59–63. doi:10.1016/j.tics.2005.12.004

Poldrack, R. A., Mumford, J. A., & Nichols, T. E. (2011). Handbook of functional MRI data analysis. New York, NY: Cambridge University Press. doi:10.1017/cbo9780511895029

Power, J. D., Mitra, A., Laumann, T. O., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2014). Methods to detect, characterize, and remove motion artifact in resting state fMRI. Neuroimage, 84, 320–341. doi:10.1016/j.neuroimage.2013.08.048

Roberts, T., McGreevy, P., & Valenzuela, M. (2010). Human induced rotation and reorganization of the brain of domestic dogs. PLoS ONE, 5(7), e11946. doi:10.1371/journal.pone.0011946

Roth, G., & Dicke, U. (2005). Evolution of the brain and intelligence. Trends in Cognitive Sciences, 9(5), 250–257. doi:10.1016/j.tics.2005.03.005

Satpute, A. B., Wager, T. D., Cohen-Adad, J., Bianciardi, M., Choi, J. K., Buhle, J. T., . . . Barrett, L. F. (2013). Identification of discrete functional subregions of the human periaqueductal gray. Proceedings of the National Academy of Sciences, 110(42), 17101–17106. doi:10.1073/pnas.1306095110

Schafer, J. R., Kida, I., Xu, F., Rothman, D. L., & Hyder, F. (2006). Reproducibility of odor maps by fMRI in rodents. Neuroimage, 31(3), 1238–1246. doi:10.1016/j.neuroimage.2005.12.060

Shellock, F. G., & Crues, J. V. (2004). MR procedures: Biologic effects, safety, and patient care. Radiology, 232(3), 635–652. doi:10.1148/radiol.2323030830

Slabbert, J. M., & Odendaal, J. S. J. (1999). Early prediction of adult police dog efficiency: A longitudinal study. Applied Animal Behaviour Science, 64, 269–288. doi:10.1016/S0168-1591(99)00038-6

Smith, J. A. (2010). Hazards, safety, and anesthetic considerations for magnetic resonance imaging. Topics in Companion Animal Medicine, 25(2), 98–106. doi:10.1053/j.tcam.2010.01.003

Snow, J. C., Pettypiece, C. E., McAdam, T. D., McLean, A. D., Stroman, P. W., Goodale, M. A., & Culham, J. C. (2011). Bringing the real world into the fMRI scanner: Repetition effects for pictures versus real objects. Scientific Reports, 1, 130. doi:10.1038/srep00130

Spetch, M. L. (2010). Understanding how pictures are seen is important for comparative visual cognition. Comparative Cognition & Behavior Reviews, 5, 163–166. doi:10.3819/ccbr.2010.50013

Su, M.-Y., Tapp, P. D., Vu, L., Chen, Y.-F., Chu, Y., Muggenburg, B., . . . Head, E. (2005). A longitudinal study of brain morphometrics using serial magnetic resonance imaging analysis in a canine model of aging. Progress in Neuro-Psychopharmacology & Biological Psychiatry, 29(3), 389–97. doi:10.1016/j.pnpbp.2004.12.005

Suthana, N. A., Donix, M., Wozny, D. R., Bazih, A., Jones, M., Heidemann, R. M., . . . Bookheimer, S. Y. (2015). High-resolution 7T fMRI of human hippocampal subfields during associative learning. Journal of Cognitive Neuroscience, 27(6), 1194–206. doi:10.1162/jocn_a_00772

Thesen, S., Heid, O., Mueller, E., & Schad, L. R. (2000). Prospective acquisition correction for head motion with image–based tracking for real–time fMRI. Magnetic Resonance in Medicine, 44(3), 457–465. doi:10.1002/1522-2594(200009)44:3<457::AID-MRM17>3.0.CO;2-R

Todd, N., Josephs, O., Callaghan, M. F., Lutti, A., & Weiskopf, N. (2015). Prospective motion correction of 3D echo-planar imaging data for functional MRI using optical tracking. NeuroImage, 113, 1–12. doi:10.1016/j.neuroimage.2015.03.013

Tóth, L., Gácsi, M., Miklósi, Á., Bogner, P., & Repa, I. (2009). Awake dog brain magnetic resonance imaging. Journal of Veterinary Behavior Clinical Applications and Research, 4, 50. doi:10.1016/j.jveb.2008.09.021

Upadhyay, J., Baker, S. J., Chandran, P., Miller, L., Lee, Y., Marek, G. J., . . . Day, M. (2011). Default-mode-like network activation in awake rodents. PloS ONE, 6(11), e27839. doi:10.1371/journal.pone.0027839

Venn, R. E., McBrearty, A. R., McKeegan, D., & Penderis, J. (2014). The effect of magnetic resonance imaging noise on cochlear function in dogs. Veterinary Journal, 202(1). doi:10.1016/j.tvjl.2014.07.006

Wilsson, E., & Sinn, D. L. (2012). Are there differences between behavioral measurement methods? A comparison of the predictive validity of two ratings methods in a working dog program. Applied Animal Behaviour Science, 141(3–4), 158–172. doi:10.1016/j.applanim.2012.08.012

Yacoub, E., Harel, N., & Uğurbil, K. (2008). High-field fMRI unveils orientation columns in humans. Proceedings of the National Academy of Sciences, 105(30), 10607–10612.

Comparative Cognition Outside the Laboratory

Suzanne E. MacDonald and Sarah Ritvo
Department of Psychology, York University, Toronto, ON, Canada

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

With its roots firmly planted in behaviorist and animal learning traditions, lab-based research is an enduring and pervasive characteristic of comparative cognition. In this review, we discuss progress in comparative cognition research in other experimental settings such as zoos, captive animal parks, and wild settings. Zoos provide access to a large array of species housed in seminatural environments that allow a reasonable degree of experimental control. Thanks to the advent of computer technology, a wide range of complex cognitive processes is increasingly being successfully studied in zoo environments. Further, cognitive research provides enrichment for captive animal participants, reducing anxiety and promoting psychological well-being. The results of cognitive research also benefit the welfare of captive animals through preference assessment, species-specific exhibit design, and behavioral management. Field settings also offer unique advantages and have allowed researchers to systematically study such diverse topics as spatial cognition, cultural transmission, problem solving, and preference. Not only does field research expand our understanding of the evolutionary and ecological drivers of animal cognition, but it also can directly inform conservation efforts. Although venturing out of the lab presents tangible challenges, including the restriction of testable hypotheses and conclusions that can be inferred from results, the benefits to be gained outweigh the costs.

Author Note: Dr. Suzanne E. MacDonald, Department of Psychology, York University, Toronto, ON M3J 1P3.

Correspondence concerning this article should be addressed to Dr. Suzanne E. MacDonald at suzmac@yorku.ca.

Keywords: comparative, cognition, application, primate, enrichment,
behavioral management; wildlife; field studies; preference<

---

Comparative cognition is a new field, the birth of which is often traced back to the publication of two influential books just over 30 years ago (Hulse, Fowler, & Honig, 1978; Roitblat, Bever, & Terrace, 1984). Of course, the roots of the field are much older, dating back to Thorndike, Tolman, Skinner, and the animal learning theorists of the middle of the last century. With the onset of the cognitive revolution in the 1970s, the focus of traditional animal learning expanded to include such daring (for the time) topics as short-term memory (e.g., Roberts & Grant, 1976), spatial memory (Olton & Samuelson, 1976), and abstract concept formation in animals (e.g., Zentall & Hogan, 1974). The types of questions being asked about animal brains became much more similar to those being asked about human brains.

What did not change was the way the questions were addressed. Early animal cognition research was done in the same laboratories that had previously studied animal learning, with the addition of larger and more complex environments, like the water maze (Morris, 1984) and variants on the radial arm maze (e.g., Suzuki, Augerinos, & Black, 1980). The ecological niche of the species studied also began to be considered, with the animals given more freedom to explore and forage naturally (e.g., Spetch & Honig, 1988). The idea that evolution might have resulted in species-specific cognitive abilities, with potentially diverse underlying mechanisms (Shettleworth, 1972), was a driving feature for the new field of comparative cognition. The discovery that food-storing birds demonstrated astounding spatial memory abilities (e.g., Shettleworth, 1990; Shettleworth & Krebs, 1982) resulted in a dramatic increase in the number of taxa and individual species studied. What remained constant for most comparative cognition researchers, however, was the focus on bringing these species-specific behaviors into a laboratory setting to tease apart the variables influencing the behaviors. However, some researchers gradually began to venture out of the laboratory into other settings, such as zoos, wild animal parks, and wild settings.

Studying Comparative Cognition in Zoos

Zoos have been around for many hundreds of years, and have evolved along with our views on animals (Baratay & Hardouin-Fugier, 2003; Kisling, 2000). Although Egyptian rulers kept animals in captivity dating back to 1500 BC, the first publicly accessible “menageries” date back to Europe in the 1700s. The first Zoological Garden, the precursor of modern zoos, opened in London in 1828, with a wide variety of exotic animals displayed in cages to the wonderment of visitors. For the past 40 years, modern zoos have strived to create environments that mimic the natural setting of the particular species, with more or less success. In most zoos, however, the legacy of the zoological garden lives on, with different species physically separated from each other. To a comparative psychologist, many zoos resemble a very large laboratory (without per diem charges!), with a range of species that Darwin would envy. Testing closely related species with different ecological niches to explore possible adaptive specializations of cognitive processes (e.g., Brodbeck & Shettleworth, 1995; Krebs, Healy, & Shettleworth, 1990; Pravosudov & Clayton, 2002; Sherry, Jacobs, & Gaulin, 1992) in a zoo setting becomes much easier. Another advantage found in most zoos is the relatively large degree of experimental control over the environment. Zoo animals are housed with similar routines to those found in laboratory settings, moving into their public exhibit areas in the morning and back into their private holding spaces in the evening. This allows researchers the ability to enter exhibit spaces without the animals present, to alter the environmental cues, or to position experimental stimuli. Sample size is by necessity much smaller than that possible in lab settings, and individual life histories can vary wildly, depending on whether the animals are captive or wild born. But some questions can be addressed effectively (Saudargas & Drummer, 1996), and some species, like nonhuman primates, are particularly well suited as subjects for zoo research, as many primate species are neophilic (e.g., Day, Coe, Kendal, & Laland, 2003; Joubert & Vauclair, 1986) and readily explore new objects in their environment (e.g., MacDonald & Pinel, 1991).

Nonhuman primate cognition has always been a topic of interest to comparative cognition researchers. Not only are nonhuman primates our closest living relatives, but they also offer a wide array of closely related species with different ecological and social niches. However, maintaining primates in lab settings is costly, and can be ethically questionable for many species, especially the great apes. Primates that live in zoos offer an opportunity to collect much-needed basic cognitive data on species that would otherwise not be available for study. Because of the inherent difficulty in obtaining control over experimental variables like life history or individual experience, much of the early research in zoo settings focused on whether a species could do a particular task. For example, MacDonald and Wilkie (1990) used a free-foraging variant of the radial arm maze with an Old World monkey species, and found that both monkeys tested were highly accurate at remembering the location of hidden food, even after a delay of up to 24 hours. Further, both animals used a “least distance” or “traveling salesman” strategy to minimize the distance they traveled to retrieve food. This paradigm has been successfully used with other zoo-housed primate species, like New World monkeys (e.g., MacDonald, Pang, & Gibeault, 1994) and great apes (gorillas: Gibeault & MacDonald, 2000; MacDonald, 1994; orangutans: MacDonald & Agnes, 1999) to compare and contrast the cognitive abilities of different species. The same basic task has also been used in field settings (Bicca-Marques & Garber, 2004; Garber & Pacuilli, 1997; Janson, 1998) and has provided a more complete picture of spatial cognition in the Primate order.

In addition to asking whether an animal has a particular ability, one can also ask how a species solves a cognitive problem. The search for the mechanisms underlying primate cognitive processes has moved out of the lab and into zoo settings, thanks to the advent of computer technology. Using computer touch screens in a zoo is challenging, primarily because the enclosures are not usually designed to accommodate the equipment. Data are often collected under less-than-ideal circumstances, in small holding areas, and with limited access to the animals. Training protocols that might take a few weeks in a lab setting can often take a year or more in zoos, simply because the daily zoo routine does not allow for regular data collection. Isolating individual animals and eliminating the distractions present in a zoo setting are often difficult, and depend on the close cooperation of zoo staff. However, some zoos are now building exhibits centered around cognition research, which give researchers easier access and which allow members of the public to see firsthand how comparative cognition research is done. The “Think Tank” at the National Zoo in Washington, DC, is an excellent example of this type of exhibit, although many others now exist, notably at Zoo Atlanta; Wolf Park, Indiana; the Wolf Science Centre near Vienna; The Seas at Epcot in Florida; and at the Leipzig Zoo in Germany. All of these facilities serve not only to advance comparative cognition as a science, but also to educate and inform the public about the importance of our field.

In a typical experiment, a computer touch screen or test apparatus is made available to free-ranging animals, either individually or in a group setting, and the animal(s) can interact with the screen and researcher if they so choose. Reinforcement is typically given manually, although some purpose-built exhibits have automated the process so the animals can participate throughout the day. For example, Marsh, Spetch, and MacDonald (2011) investigated how orangutans (Pongo abelli) used landmarks in a spatial task presented on a computer touch screen. The animals participated individually, and they interacted with the touch screen by touching it with a wooden dowel. On each trial, a square array of two-dimensional “landmarks” were presented at a random location on the touch screen. Orangutans were trained to locate the goal hidden in the center of the array, and then were given an expansion task, in which the distance between the landmarks was increased, while maintaining the same geometric relationship between them. Unlike human adults, who continue to search in the “middle” of the array on expansion tasks (MacDonald, Spetch, Kelly, & Cheng, 2004), the orangutans focused their searching along absolute directional vectors from the individual landmarks. An unexpected advantage to doing this type of research in a zoo setting is that the opportunity to collect comparable data from human children—zoo visitors—often exists when the task is one that can be completed in a few minutes, while children are visiting the animals’ exhibit with their parents. In this way, Marsh et al. (2011) and Marsh, Adams, Floyd, and MacDonald (2013) were able to collect directly comparable data from children across a wide age range, without the usual delay and time commitments necessary to obtain data from children in school settings.

Complex cognitive processes, including imitation (e.g., Stoinski, Wrate, Ure & Whiten, 2001), numerosity judgments (e.g., Anderson et al., 2005), categorization and concept formation (e.g., Marsh & MacDonald, 2008; Vonk & MacDonald, 2002, 2004), and metacognition (e.g., Marsh & MacDonald, 2011, 2012) have all been studied in zoo-housed primates, and the list of institutions establishing cognitive research programs for their animals is growing every day. Complex cognition is now being investigated in other taxa as well, often in species that cannot be easily studied in either lab or field settings. For example, Asian elephants have demonstrated sophisticated cooperative behavior (Foerder, Galloway, Barthel, Moore, & Reiss, 2011; Plotnik, Lair, Suphachoksahakun, & de Waal, 2011) as well as tool use (Hart, Hart, McCoy, & Sarath, 2001; Whiten, Horner & de Waal, 2005) and self-recognition (Plotnik, de Waal, & Reiss, 2006). Similarly, dolphins at Disney’s Epcot Center have been participating in cognitive research for over 25 years (e.g., Harley, Fellner, & Stamper, 2010). Dolphins use their astonishing echolocation and communicative abilities to allow researchers to study echoic object and shape recognition (e.g., Harley & DeLong, 2008), as well as more cognitively complex behaviors such as pointing (Xitco, Gory, & Kuczaj, 2001, 2004). The ability to carefully control environmental cues and experimental stimuli in an aquarium setting allows for insights into dolphin cognition that would be next to impossible to gain in the wild.

An overarching benefit of all cognition research in zoos is the cognitive stimulation that the participants receive while doing various experimental tasks. Participating in cognitive research is increasingly seen as an important form of enrichment for captive animals, as a means of keeping individuals mentally stimulated (e.g., Mason, Clubb, Latham, & Vickery, 2007). Modern zoos are moving away from the traditional “environmental” enrichment paradigm to one of behavioral management (Tresz, 2006; Weed & O’Neill-Wagner, 2015), although this has been a slow process. Forthman & Ogden (1992) were among the first to call for a new focus on the cognitive and social requirements of animals when designing zoo exhibits and daily routines. Since then, the tools of comparative cognition and applied behavior analysis have been adopted by many zoos, and have been used effectively in a wide range of contexts, from determining appropriate breeding partners (Watters & Powell, 2012) to reducing aggression in chimpanzees (Bloomsmith, Laule, Alford, & Thurston, 1994), to improving exhibit design for primates (Hosey, 2005; Hosey & Druck, 1987), to reducing anxiety-related behavior in polar bears (Kelly, Harrison, Size, & MacDonald, 2015; Renner & Kelly, 2006).

Reducing anxiety and promoting psychological well-being in captive animals is an important goal for zoos, both from an animal welfare standpoint and to ensure that highly endangered species breed successfully in captivity. Providing stimulation to relieve boredom is common; however, until recently, assessing how stimuli are perceived by the animals has not been a priority in many zoos. Comparative cognition methods can be used effectively to evaluate these interventions, and to suggest new stimuli that may be effective, based on the cognitive abilities of target species. For example, music is consistently used as environmental enrichment in primate facilities around the world, under the assumption that music is as engaging for animals as it is for humans (Hinds, Raimond, & Purcell, 2007; Lutz & Novak, 2005). For the most part, music selection is based on the preferences of human facilitators despite the fact that there is little to no indication that human and nonhuman music preferences correspond (Lutz & Novak, 2005).

A complication of this area of study from both a theoretical and a methodological standpoint relates to preference assessment in subjects generally incapable of directly communicating internal sentiments. Solutions to this challenge have been sourced from human infant preference assessment literature. Review of established methodologies indicate three main approaches: (a) behavioral observation during stimulus exposure, (b) the least-aversive or most-desired choice paradigm, and (c) participant-controlled procedures (Ritvo & Allison, 2014). Participant-controlled procedures uniquely allow subjects to spontaneously and autonomously choose the type and duration of stimulus exposure. This makes participant-controlled procedures the most precise and appropriate approach for inferring that subjects “like” one stimulus more than another, as opposed to “dislike” one stimulus less than another and is, consequently, a recommended methodology for assessing nonhuman primate preferences (Lamont, 2005; Ritvo & Allison, 2014; Ritvo & MacDonald, unpublished manuscript).

Accordingly, Ritvo and MacDonald (unpublished manuscript) employed a participant-controlled dichotomous-choice design in their investigation of music preference and discrimination in Sumatran orangutans (Pongo abelli). Three orangutans at the Toronto Zoo, two females (ages 21 and 28) and one male (age: 6), were trained to indicate preference via touch screen choices. Six music genres were tested based on conventional and popular North American genres that subjects would be familiar with (i.e., via radio music employed as auditory enrichment at the Toronto Zoo). A seventh genre, Tuva throat singing, was also selected because both the music and the way it is physically produced resemble orangutan long calls. Specific genre exemplars were selected based on human preference indicated by the greatest number of purchases on iTunes (Apple Inc., 2013).

In Study 1, preference for music vs. silence was explored. Following exposure to a sample of one of the seven music genres, subjects chose to continue to listen to the music sample previously played by touching one side of the screen, or to listen to the equivalent duration of silence instead by touching the other side of the screen. In Study 2, orangutans’ ability to discriminate music from scrambled music was assessed using a standard delayed matching-to-sample (DMTS) task. In one condition, orangutans were rewarded for correct classification via touch screen of auditory stimuli as “music” or “scrambled music.” In the second, control condition, subjects were rewarded for correct classification via touch screen of auditory stimuli as a female zookeeper’s voice or a male zookeeper’s voice. Contrary to expectation, results indicated that (a) subjects preferred silence to music (or were indifferent), (b) they did not display a preference for any specific musical genre employed, (c) they did not discriminate “music” from “scrambled music,” and (d) only a single female discriminated between a male zookeeper’s and a female zookeeper’s voices.

Given that the orangutans tended to choose silence or to stop participating entirely during times of alarm or commotion, their preference for silence over music may relate to the potential for music to mask valuable information that other auditory stimuli provide, (e.g., food is being prepared or conspecifics are distressed). However, the results of Study 2 imply a more profound explanation; orangutans do not perceive music analogously to humans. In particular, results suggest that human-defined music exemplars and scrambled versions of the same music were not qualitatively discerned by orangutans. This finding could explain why the orangutans did not appear to find the music employed in Study 1 appealing. Whereas humans perceive music as a united, stable, rhythmic and harmonious stimulus, orangutans may not perceive music as qualitatively different from other fluctuating auditory stimuli, or they may perceive music as more akin to indiscriminate noise. Antagonistic behavior observed in Study 1 supports this explanation, suggesting that our participants found human-defined music to be mild-to-moderately aversive.

In either case, our results suggest that the music employed in these investigations was not rewarding for the orangutans. Consequently, the common practice of using Western music as an enrichment tool in primate care facilities appears unfounded and could in fact result in negative behavioral or psychological effects. Whether other species have similar reactions to music enrichment remains to be empirically tested. What is key to this type of research, though, is the idea of preference or choice on the part of the animal.

This issue has recently received a lot of public attention with the ongoing court battle in Argentina to grant Sandra, a 29-year-old female Sumatran orangutan housed in the Buenos Aires Zoo, human rights on the basis of her cognitive capacities (Jacobs, 2015). Zoos around the world are following this court case with interest, and regardless of the final ruling, it has brought to the forefront the issue of an animal’s right to control aspects of its captive home, such as when and what to eat, where to sleep, and with whom to socialize. Comparative cognition researchers can assist zoos in designing exhibits that make the most of animals’ species-specific cognitive abilities, thus improving animal welfare and building the new scientific field of behavioral management.

Studying Comparative Cognition in the Wild

Behavioral ecologists have been studying cognitive processes in the field for decades. Comparative cognition researchers coming out of the animal learning tradition are now realizing the exciting potential of expanding their research to include data from animals living in the wild. Of course, field settings are much more challenging places in which to work from a practical standpoint. Dealing with the costs and hassle of long-distance travel, lodging, wildlife permits, ethics approval, not to mention the ever-present contingent of biting bugs, can be a daunting prospect. More important, obtaining sufficient control over extraneous variables is often impossible. Life history information for individuals is usually absent, and so studying large, long-lived species is difficult. Just as in zoo settings, the lack of control over many environmental variables constrains the types of questions that can be asked. And, just as in the zoo setting, spatial cognition is a fertile area for study in the field. Space limitations are eliminated, and there are a large number of environmental cues that can be experimentally manipulated. Systematic, long-term, and creative field research on wild birds (e.g., Healy & Hurly, 2004) and invertebrates like ants (e.g., Graham & Cheng, 2009; Narendra, Sulikowski, & Cheng, 2007; Wystrach, Beugnon, & Cheng, 2012) and bees (e.g., Menzel et al., 2005) has broadened our understanding of the range of species exhibiting complex spatial abilities and, even more important, elucidated the underlying cognitive mechanisms involved.

Although working with wild animals does have some challenges, there are also some exciting opportunities to branch out and investigate behaviors that are best expressed in a complex environmental context. Cognition and culture comprise an area that is difficult to explore in lab settings, but one in which researchers who study freely behaving wild animals have been able to glean new insights. The social transmission of information has been studied extensively in chimpanzees and other primates, (e.g., Boesch & Boesch, 1990; Huffman & Quiatt, 1986; Inoue-Nakamura & Matsuzawa, 1997; Kendal et al., 2010; Perry & Manson, 2003). More recent research has shown cultural transmission and even “teaching” behavior in a range of species, including cetaceans (e.g., Greggor, 2012), meerkats (e.g., Thornton & Raihani, 2010), and great tits (Aplin et al., 2015). Aplin et al. (2105) experimentally introduced a novel foraging behavior into a population of great tits (Parus major) and studied the dissemination and persistence of the behavior over two generations. They found that individual birds adopted social information preferentially over personal information; essentially, they preferred to learn from other birds, rather than learn by experience. This demonstration of complex cultural transmission in a non-primate species suggests that there remains much more to be learned from studying comparative cognition in wild populations.

Problem solving is another area that is a fruitful avenue for field research. Presenting individuals with a novel stimulus and observing their behavior reduces the possibility of prior learning effects. This can be done with either captive or wild animals. Extensive research on problem solving has been done with birds, including New Caledonian crows (e.g., von Bayern, Heathcote, Rutz, & Kacelnik, 2009), keas (e.g., Auersperg, von Bayern, Gajdon, Huber, & Kacelnik, 2011), passerines (Webster & Lefebvre, 2001), and ravens (Heinrich & Bugnyar, 2005). Problem solving in wild mammals is also being explored. For example, Benson-Amram & Holekamp (2012) studied problem solving in wild spotted hyenas (Crocuta crocuta) by presenting individuals with a large puzzle box containing meat. To obtain the food, a hyena had to perform two distinct behaviors in sequence, first sliding a latch and then swinging a door. There were considerable individual differences between hyenas, and only 15% of the participants were able to solve the task, although those that tried more solutions were more successful. Similar to results from human subjects, reduced neophobia and increased exploratory behaviors were both important predictors of problem-solving success.

Undertaking field work with free-ranging species does present additional challenges. For example, MacDonald (2015) investigated problem solving and exploratory behavior in wild raccoons, comparing raccoons from an urban population with those from a rural environment. The first step in this multiyear project was to trap wild urban raccoons and fit them with GPS-collars to determine home range size and individual movement (Dupuis-Desormeaux & MacDonald, 2011). Because individual identification in wild animals is often difficult, determining the size and location of home ranges is critically important so individual animals are not sampled more than once. We found that urban home range sizes were much smaller than expected–about three square blocks–and that animal movement was constrained by busy city streets. In comparison, estimated rural home ranges for raccoons vary between 60 and 90 ha (Beasley, Devault, & Rhodes, 2007). MacDonald (2015) then sampled raccoons from nonoverlapping home ranges throughout the greater Toronto area, and from across rural southern Ontario. Two different container types, each baited with highly preferred food, were placed in the raccoons’ home ranges. One container was familiar to both urban and rural animals: a standard 40-liter garbage can, fitted with a “bungee” cord across the lid to hold it in place. One container was novel: a hanging bucket suspended 30 cm above the ground from a rope anchored to nearby trees. The baited locations were equipped with motion-capture infrared tracking cameras, which recorded video in complete darkness. More than 120 tracking nights over a two-year period resulted in 800 hours of video. After eliminating data from a wide range of other species (domestic cats, coyotes, and black bears were the most common), and only using data from an individual raccoon’s first encounter with the objects, a total of 22 rural and 22 urban samples remained. Contrary to predictions, the novel object was explored and depleted quickly by all the urban raccoons and by their rural counterparts. Although the hanging object did move and spin in an unpredictable manner, locating the hidden food contained inside was a simple one-stage process. However, the familiar object—the garbage can—proved to be more of a challenge than expected for the rural animals. While 17 of the 22 urban animals successfully depleted the food from the can, none of the 22 rural raccoons were successful, despite many attempts. The urban animals were much more persistent in manipulating the object, and they also employed additional strategies not observed in the rural animals. This was true for both male and female animals, as well as young raccoons, who were at the time of testing only five or six months old and navigating the environment for the first time. These data support the tantalizing possibility that the anthropogenic selection is at work, with our cities—and human behavior—selecting for particular cognitive abilities in raccoons. Persistence, neophilia, and high levels of exploratory behavior may result in increased survival and reproduction in the urban setting, and thus we may be observing cognitive evolution in action in this species. Comparing the exploratory abilities and behavioral flexibility of infant raccoons from rural and urban populations is the next step to determine whether differential experience explains the gap between urban and rural raccoons, or whether these traits are heritable and stable across generations, as has been found in other wild species (e.g., great tits: Dingemanse, Both, Drent, van Oers, & van Noordwijk, 2002; Cole, Morand-Ferron, Hinks, & Quinn, 2012; cane toads: Candler & Bernal, 2014; Herborn et al., 2010; freshwater fish: Smith, Philips, & Reichard, 2015).

Just as preference and choice are becoming a topic of interest in zoo research, studying preference in wild populations can be a valuable contribution of comparative cognition to conservation efforts for endangered species. One of the most pressing current issues in the field is the effect of human–animal conflicts, which often result in the culling of “problem animals.” Reducing these conflicts is vital to ensure the livelihood of human farmers and fishers, and also to ensure the continued survival of highly endangered species. The African elephant is an excellent example of this complex problem. Given the endangered status of this species, and the increasingly limited habitat available for them, reducing human conflict is essential. In many African countries, including Kenya, Zimbabwe, and South Africa, elephant migratory and ranging patterns traverse lands owned by subsistence farmers (e.g., Hoare, 1999; Loarie, van Aarde, & Pimm, 2009). Crop raiding by elephants can destroy an entire year’s harvest in a single night (Sitati & Walpole, 2006), so tensions are understandably high between farmers and elephant herds. To date, the focus has been on training elephants to avoid human habitation and fences, using a variety of stimuli as punishers, including the sound of angry bees (King, Douglas-Hamilton, & Vollrath, 2007), electrified fences, beating drums, throwing rocks, firecrackers, chili peppers, and cowbells (e.g., Osborn & Parker, 2002, 2003; Sitati, Walpole, & Leader-Williams, 2005). Currently, Zitzer & MacDonald (2015) are exploring the “flip side” of avoidance, looking at elephant food and olfactory preferences, with the goal of using positive reinforcement to encourage elephants to choose alternative routes, away from human settlements. We are using motion-capture tracking cameras in the field in South Africa to measure which natural vegetation items elephants choose in an experimental preference test (akin to a large buffet for elephants). We are combining these preference data with vegetation surveys to look at the damage done by elephants in a range of habitats. The preference and habitat data will provide concrete information to local landowners about the real effect that elephants are having on the landscape, and will also provide positive solutions to mitigate conflicts in the future. Of course, this is just one example. Many more opportunities exist for fruitful collaborations between comparative cognition researchers and conservationists, as nicely outlined by Greggor, Clayton, Phalan, & Thornton (2014).

Conclusions

Venturing out of the lab is not easy. Reduced experimental control can restrict the types of hypotheses that can be tested and may limit the conclusions that can be drawn from results. Establishing cognitive mechanisms can be difficult with so many potential confounding variables. There may be significant financial costs, as well as time and effort involved in conducting long-term studies that may be dependent on weather and limited by unreliable access. However, the benefits to be gained far outweigh the costs. Extending the range of species with whom we work will result in truly comparative research, and lead to a better understanding of the diverse ways that evolution has shaped the brains of animals, both human and nonhuman. Results from zoo and field settings can inform and inspire lab-based research, and vice versa. Placing a species in its ecological context can lead to new empirical questions and exciting new directions for research. In these days of uncertain, limited funding and public distrust of lab-based animal research, it is more important than ever to share our comparative approach and methods and collaborate with local institutions and communities.

References

Anderson, U. S., Stoinski, T. S., Bloomsmith, M. A., Marr, M. J., Smith, A. D., & Maple, T. L. (2005). Relative numerousness judgment and summation in young and old Western lowland gorillas. Journal of Comparative Psychology, 119(3), 285–295. doi:10.1037/0735-7036.119.3.285

Aplin, L. M., Farine, D. R., Morand-Ferron, J., Cockburn, A., Thornton, A., & Sheldon, B. C. (2015). Experimentally induced innovations lead to persistent culture via conformity in wild birds. Nature, 518(7540), 538–541. doi:10.1038/nature13998

Apple, Inc. (2013). iTunes application software. Retrieved from http://itunes.apple.com.

Auersperg, A. M. I., von Bayern, A. M. P., Gajdon, G. K., Huber, L., & Kacelnik, A. (2011). Flexibility in problem solving and tool use of kea and New Caledonian crows in a multi access box paradigm. PLOS ONE, 6(6), e20231. doi:10.1371/journal.pone.0020231

Baratay, E., & Hardouin-Fugier, E. (2003). Zoo: A history of zoological gardens in the West. London, England: Reaktion Books.

Beasley, J. C., Devault, T. L., & Rhodes, O. E. (2007). Home‐range attributes of raccoons in a fragmented agricultural region of northern Indiana. The Journal of Wildlife Management, 71(3), 844–850. doi:10.2193/2006-022

Benson-Amram, S., & Holekamp, K. E. (2012). Innovative problem solving by wild spotted hyenas. Proceedings of the Royal Society of London B: Biological Sciences, rspb20121450. doi:10.1098/rspb.2012.1450

Bergman, T. J., & Kitchen, D. M. (2008). Comparing responses to novel objects in wild baboons (Papio ursinus) and geladas (Theropithecus gelada). Animal Cognition, 12(1), 63–73. doi:10.1007/s10071-008-0171-2

Bicca-Marques, J. C., & Garber, P. A. (2004). Use of spatial, visual, and olfactory information during foraging in wild nocturnal and diurnal anthropoids: A field experiment comparing Aotus, Callicebus, and Saguinus. American Journal of Primatology, 62(3), 171–187. doi:10.1002/ajp.20014

Bloomsmith, M. A., Laule, G. E., Alford, P. L., & Thurston, R. H. (1994). Using training to moderate chimpanzee aggression during feeding. Zoo Biology, 13(6), 557–566. doi:10.1002/zoo.1430130605

Boesch, C., & Boesch, H. (1990). Tool use and tool making in wild chimpanzees. Folia Primatologica, 54(1–2), 86–99. doi:10.1159/000156428

Brodbeck, D. R., & Shettleworth, S. J. (1995). Matching location and color of a compound stimulus: Comparison of a food-storing and a nonstoring bird species. Journal of Experimental Psychology: Animal Behavior Processes, 21(1), 64. doi:10.1037/0097-7403.21.1.64

Candler, S., & Bernal, X. E. (2014). Differences in neophobia between cane toads from introduced and native populations. Behavioral Ecology, aru162. doi:10.1093/beheco/aru162

Cole, E. F., Morand-Ferron, J., Hinks, A. E., & Quinn, J. L. (2012). Cognitive ability influences reproductive life history variation in the wild. Current Biology, 22(19), 1808–1812. doi:10.1016/j.cub.2012.07.051

Day, R. L., Coe, R. L., Kendal, J. R., & Laland, K. N. (2003). Neophilia, innovation and social learning: A study of intergeneric differences in callitrichid monkeys. Animal Behaviour, 65(3), 559–571. doi:10.1006/anbe.2003.2074

Dingemanse, N. J., Both, C., Drent, P. J., van Oers, K., & van Noordwijk, A. J. (2002). Repeatability and heritability of exploratory behaviour in great tits from the wild. Animal Behaviour, 64(6), 929–938. doi:10.1006/anbe.2002.2006

Dupuis-Desormeaux, M., & MacDonald, S. E. (2011, August). Behavior and movement patterns of urban raccoons (Procyon lotor). Paper presented at symposium Exploring the self-worlds of nonhuman animals via advances in biotelemetry technology at the annual meeting of the American Psychological Association, Washington, DC.

Foerder, P., Galloway, M., Barthel, T., Moore, D. E., & Reiss, D. (2011). Insightful problem solving in an Asian elephant. PLOS ONE, 6(8), e23251. doi:10.1371/journal.pone.0023251.

Forthman, D. L., & Ogden, J. J. (1992). The role of applied behavior analysis in zoo management: Today and tomorrow. Journal of Applied Behavior Analysis, 25(3), 647–652. doi:10.1901/jaba.1992.25-647

Garber, P. A., & Paciulli, L. M. (1997). Experimental field study of spatial memory and learning in wild capuchin monkeys (Cebus capucinus). Folia Primatologica, 68(3–5), 236–253. doi:10.1159/000157250

Gibeault, S., & MacDonald, S. E. (2000). Spatial memory and foraging competition in captive western lowland gorillas (Gorilla gorilla gorilla). Primates, 41(2), 147–160. doi:10.1007/BF02557796

Graham, P., & Cheng, K. (2009). Ants use the panoramic skyline as a visual cue during navigation. Current Biology, 19(20), R935–R937. doi:10.1016/j.cub.2009.08.015

Greggor, A. L. (2012). A functional paradigm for evaluating culture: An example with cetaceans. Current Zoology, 58(2), 271–286.

Greggor, A. L., Clayton, N. S., Phalan, B., & Thornton, A. (2014). Comparative cognition for conservationists. Trends in Ecology & Evolution, 29(9), 489–495. doi:10.1016/j.tree.2014.06.004

Harley, H. E., & DeLong, C. M. (2008). Echoic object recognition by the bottlenose dolphin. Comparative Cognition and Behavior Reviews, 3, 46–65. doi:10.3819/ccbr.2008.30003

Harley, H. E., Fellner, W., & Stamper, M. A. (2010). Cognitive research with dolphins (Tursiops truncates) at Disney’s The Seas: A program for enrichment, science, education and conservation. International Journal of Comparative Psychology, 23, 331–343.

Hart, B. L., Hart, L. A., McCoy, M. and Sarath, C. R. (2001). Cognitive behaviour in Asian elephants: Use and modification of branches for fly switching. Animal Behaviour, 62(5), 839–47. doi:10.1006/anbe.2001.1815.

Healy, S. D., & Hurly, T. A. (2004). Spatial learning and memory in birds. Brain, Behavior and Evolution, 63(4), 211–220. doi:10.1159/000076782

Heinrich, B., & Bugnyar, T. (2005). Testing problem solving in ravens: String‐pulling to reach food. Ethology, 111(10), 962–976. doi:10.1111/j.1439-0310.2005.01133.x

Herborn, K. A., Macleod, R., Miles, W. T. S., Schofield, A. N. B., Alexander, L., & Arnold, K. E. (2010). Personality in captivity reflects personality in the wild. Animal Behaviour, 79(4), 835–843. doi:10.1016/j.anbehav.2009.12.026

Hinds, S. B., Raimond, S., & Purcell, B. K. (2007). The effect of harp music on heart rate, mean blood pressure, respiratory rate, and body temperature in the African green monkey. Journal of Medical Primatology, 36(2), 95–100. doi:10.1111/j.1600-0684.2006.00157.x

Hoare, R. E. (1999). Determinants of human–elephant conflict in a land‐use mosaic. Journal of Applied Ecology, 36(5), 689–700. doi:10.1046/j.1365-2664.1999.00437.x

Hosey, G. R. (2005). How does the zoo environment affect the behaviour of captive primates? Applied Animal Behaviour Science, 90(2), 107–129. doi:10.1016/j.applanim.2004.08.015

Hosey, G. R., & Druck, P. L. (1987). The influence of zoo visitors on the behaviour of captive primates. Applied Animal Behaviour Science, 18(1), 19–29. doi:10.1016/0168-1591(87)90251-6

Huffman, M. A., & Quiatt, D. (1986). Stone handling by Japanese macaques (Macaca fuscata): Implications for tool use of stone. Primates, 27(4), 413–423. doi:10.1007/BF02381887

Hulse, S. H., Fowler, H., & Honig, W. K. (Eds.). (1978). Cognitive processes in animal behavior. Hillsdale, NJ: Lawrence Erlbaum. Retrieved from http://psycnet.apa.org/psycinfo/1980-51406-000

Inoue-Nakamura, N., & Matsuzawa, T. (1997). Development of stone tool use by wild chimpanzees (Pan troglodytes). Journal of Comparative Psychology, 111(2), 159. doi:10.1037/0735-7036.111.2.159

Jacobs, S. (2015). Orangutan named Sandra fights for freedom in Argentina court. Retrieved May 21, 2015, from http://www.smh.com.au/environment/animals/orangutan-named-sandra-fights-for-freedom-in-argentina-court-20150521-gh6f3e.html

Janson, C. H. (1998). Experimental evidence for spatial memory in foraging wild capuchin monkeys, Cebus apella. Animal Behaviour, 55(5), 1229–1243. doi:10.1006/anbe.1997.0688

Joubert, A., & Vauclair, J. (1986). Reaction to novel objects in a troop of Guinea baboons: Approach and manipulation. Behaviour, 96(1), 92–104. doi:10.1163/156853986X00234

Kelly, K. R., Harrison, M. L., Size, D. D., & MacDonald, S. E. (2015). Individual effects of seasonal changes, visitor density, and concurrent bear behavior on stereotypical behaviors in captive polar bears (Ursus maritimus). Journal of Applied Animal Welfare Science, 18(1), 17–31. doi:10.1080/10888705.2014.924832

Kendal, R. L., Custance, D. M., Kendal, J. R., Vale, G., Stoinski, T. S., Rakotomalala, N. L., & Rasamimanana, H. (2010). Evidence for social learning in wild lemurs (Lemur catta). Learning & Behavior, 38(3), 220–234. doi:10.3758/LB.38.3.220

King, L. E., Douglas-Hamilton, I., & Vollrath, F. (2007). African elephants run from the sound of disturbed bees. Current Biology, 17(19), R832–R833. doi:10.1016/j.cub.2007.07.038

Kisling, V. N. (Ed.). (2000). Zoo and aquarium history: Ancient animal collections to zoological gardens. Boca Raton, Florida: CRC Press. doi:10.1201/9781420039245

Krebs, J. R., Healy, S. D., & Shettleworth, S. J. (1990). Spatial memory of Paridae: Comparison of a storing and a non-storing species, the coal tit, Parus ater, and the great tit, P. major. Animal Behaviour, 39(6), 1127–1137. doi:10.1016/S0003-3472(05)80785-7

Lamont, A. M. (2005). What do monkeys’ music choices mean? Trends in Cognitive Sciences, 9(8), 359–361. doi:10.1016/j.tics.2005.06.014

Loarie, S. R., van Aarde, R. J., & Pimm, S. L. (2009). Fences and artificial water affect African savannah elephant movement patterns. Biological Conservation, 142(12), 3086–3098. doi:10.1016/j.biocon.2009.08.008

Lutz, C. K., & Novak, M. A. (2005). Environmental enrichment for nonhuman primates: Theory and application. Institute for Laboratory Animal Research Journal, 46(2), 178–191. doi:10.1093/ilar.46.2.178

MacDonald, S. E. (1994). Gorillas’ (Gorilla gorilla gorilla) spatial memory in a foraging task. Journal of Comparative Psychology, 108(2), 107. doi:10.1037/0735-7036.108.2.107

MacDonald, S. E. (2015). Problem solving and exploratory behavior in wild urban and rural raccoons: Cognitive evolution in action? Manuscript submitted for publication.

MacDonald, S. E., & Agnes, M. M. (1999). Orangutan (Pongo pygmaeus abelii) spatial memory and behavior in a foraging task. Journal of Comparative Psychology, 113(2), 213. doi:10.1037/0735-7036.113.2.213

MacDonald, S. E., Pang, J. C., & Gibeault, S. (1994). Marmoset (Callithrix jacchus jacchus) spatial memory in a foraging task: Win-stay versus win-shift strategies. Journal of Comparative Psychology, 108(4), 328. doi:10.1037/0735-7036.108.4.328

MacDonald, S. E., & Pinel, J. P. (1991). Information gathering: A component of the defensive behavior of rats and Old-World monkeys. The Psychological Record 41, 207–215. Retrieved from http://psycnet.apa.org/psycinfo/1991-26603-001

MacDonald, S. E., Spetch, M. L., Kelly, D. M., & Cheng, K. (2004). Strategies in landmark use by children, adults and marmoset monkeys. Learning and Motivation, 35, 322–347. doi:10.1016/j.lmot.2004.03.002

MacDonald, S. E., & Wilkie, D. M. (1990). Yellow-nosed monkeys’ (Cercopithecus ascanius whitesidei) spatial memory in a simulated foraging environment. Journal of Comparative Psychology, 104(4), 382. doi:10.1037/0735-7036.104.4.382

Marsh, H. L., Adams, L., Floyd, C., & MacDonald, S. E. (2013). Feature versus spatial strategies by orangutans (Pongo abelii) and human children (Homo sapiens) in a cross-dimensional task. Journal of Comparative Psychology, 127(2), 128. doi:10.1037/a0030591

Marsh, H. L., & MacDonald, S. E. (2008). The use of perceptual features in categorization by orangutans (Pongo abelli). Animal Cognition, 11(4), 569–585. doi:10.1007/s10071-008-0148-1

Marsh, H. L., & MacDonald, S. E. (2011). Information seeking by orangutans: A generalized search strategy? Animal Cognition, 15(3), 293–304. doi:10.1007/s10071-011-0453-y

Marsh, H. L., & MacDonald, S. E. (2012). Orangutans (Pongo abelii) “play the odds”: Information-seeking strategies in relation to cost, risk, and benefit. Journal of Comparative Psychology, 126(3), 263. doi:10.1037/a0025906

Marsh, H. L., Spetch, M. L., & MacDonald, S. E. (2011). Strategies in landmark use by orangutans and human children. Animal Cognition, 14(4), 487–502. doi:10.1007/s10071-011-0382-9

Mason, G., Clubb, R., Latham, N., & Vickery, S. (2007). Why and how should we use environmental enrichment to tackle stereotypic behaviour? Applied Animal Behaviour Science, 102(3–4), 163–188. doi:10.1016/j.applanim.2006.05.041

Menzel, R., Greggers, U., Smith, A., Berger, S., Brandt, R., Brunke, S., . . . Watzl, S. (2005). Honey bees navigate according to a map-like spatial memory. Proceedings of the National Academy of Sciences of the United States of America, 102(8), 3040–3045. doi:10.1073/pnas.0408550102

Morris, R. (1984). Developments of a water-maze procedure for studying spatial learning in the rat. Journal of Neuroscience Methods, 11(1), 47–60. doi:10.1016/0165-0270(84)90007-4

Narendra, A., Si, A., Sulikowski, D., & Cheng, K. (2007). Learning, retention and coding of nest-associated visual cues by the Australian desert ant, Melophorus bagoti. Behavioral Ecology and Sociobiology, 61(10), 1543–1553. doi:10.1007/s00265-007-0386-2

Olton, D. S., & Samuelson, R. J. (1976). Remembrance of places passed: Spatial memory in rats. Journal of Experimental Psychology: Animal Behavior Processes, 2(2), 97. doi:10.1037/0097-7403.2.2.97

Osborn, F. V., & Parker, G. E. (2002). Community-based methods to reduce crop loss to elephants: Experiments in the communal lands of Zimbabwe. Pachyderm, 33, 32–38.

Osborn, F. V., & Parker, G. E. (2003). Towards an integrated approach for reducing the conflict between elephants and people: A review of current research. Oryx, 37(1), 80–84. doi:10.1017/S0030605303000152

Perry, S., & Manson, J. H. (2003). Traditions in monkeys. Evolutionary Anthropology: Issues, News, and Reviews, 12(2), 1–81. doi:10.1002/evan.10105

Plotnik, J. M., Lair, R., Suphachoksahakun, W., & de Waal, F. B. M. (2011). Elephants know when they need a helping trunk in a cooperative task. Proceedings of the National Academy of Sciences, 108(12), 5116–5121. doi:10.1073/pnas.1101765108

Plotnik, J. M., de Waal, F. B. M., & Reiss, D. (2006). Self-recognition in an Asian elephant. Proceedings of the National Academy of Sciences, 103(45), 17053–17057. doi:10.1073/pnas.0608062103

Pravosudov, V. V., & Clayton, N. S. (2002). A test of the adaptive specialization hypothesis: Population differences in caching, memory, and the hippocampus in black-capped chickadees (Poecile atricapilla). Behavioral Neuroscience, 116(4), 515. doi:10.1037/0735-7044.116.4.515

Renner, M. J., & Kelly, A. L. (2006). Behavioral decisions for managing social distance and aggression in captive polar bears (Ursus maritimus). Journal of Applied Animal Welfare Science 9(3), 333–339. doi:10.1207/s15327604jaws0903_5

Ritvo, S. E., Allison, R. S., (2014). Challenges related to nonhuman animal-computer interaction: Usability and “liking.” In Adjunct Proceedings of the International Conference on Advances in Computer Entertainment Technology—Workshop on Animal Human Computer Interaction (ACE ’14). Funchal, Madeira. doi:10.1145/2693787.2693795

Ritvo, S. E., & MacDonald, S. E. (unpublished manuscript). Music preference and discrimination in three Sumatran orangutans (Pongo abelii). York University (Canada), Toronto, Canada.

Roberts, W. A., & Grant, D. S. (1976). Studies of short-term memory in the pigeon using the delayed matching to sample procedure. In D. L. Medin, W. A. Roberts, & R. T. Davis (Eds.), Processes of animal memory, 79–112. Hillsdale, NJ: Erlbaum.

Roitblat, H.L., Terrace, H.S., & Bever, T.G. (Eds.). (1984). Animal Cognition. Hillsdale, NJ: Lawrence Erlbaum Associates.

Saudargas, R. A., & Drummer, L. C. (1996). Single subject (small N) research designs and zoo research. Zoo Biology, 15(2), 173–181. http://dx.doi.org/10.1002/(SICI)1098-2361(1996)15:2<173::AID-ZOO7>3.0.CO;2-8

Sherry, D. F., Jacobs, L. F., & Gaulin, S. J. C. (1992). Spatial memory and adaptive specialization of the hippocampus. Trends in Neurosciences, 15(8), 298–303. doi:10.1016/0166-2236(92)90080-R

Shettleworth, S. J. (1972). Constraints on learning. Journal of Experimental Psychology: Animal Behavior Processes, 8(4), 354–375. doi:10.1016/S0065-3454(08)60006-6

Shettleworth, S. J. (1990). Spatial memory in food-storing birds. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 329(1253), 143–151. doi:10.1098/rstb.1990.0159

Shettleworth, S. J., & Krebs, J. R. (1982). How marsh tits find their hoards: The roles of site preference and spatial memory. Journal of Experimental Psychology: Animal Behavior Processes, 8(4), 354. doi:10.1037/0097-7403.8.4.354

Sitati, N. W., & Walpole, M. J. (2006). Assessing farm-based measures for mitigating human-elephant conflict in Transmara district, Kenya. Oryx, 40(3), 279–286. doi:10.1017/S0030605306000834

Sitati, N. W., Walpole, M. J., & Leader‐Williams, N. (2005). Factors affecting susceptibility of farms to crop raiding by African elephants: Using a predictive model to mitigate conflict. Journal of Applied Ecology, 42(6), 1175–1182. doi:10.1111/j.1365-2664.2005.01091.x

Smith, C., Philips, A., & Reichard, M. (2015). Cognitive ability is heritable and predicts the success of an alternative mating tactic. Proceedings of the Royal Society B, 282(1809), 20151046. doi:10.1098/rspb.2015.1046

Spetch, M. L., & Honig, W. K. (1988). Characteristics of pigeons’ spatial working memory in an open-field task. Animal Learning & Behavior, 16(2), 123–131. doi:10.3758/BF03209054

Stoinski, T. S., Wrate, J. L., Ure, N., & Whiten, A. (2001). Imitative learning by captive western lowland gorillas (Gorilla gorilla gorilla) in a simulated food-processing task. Journal of Comparative Psychology, 115(3), 272. doi:10.1037/0735-7036.115.3.272

Suzuki, S., Augerinos, G., & Black, A. H. (1980). Stimulus control of spatial behavior on the eight-arm maze in rats. Learning and Motivation, 11(1), 1–18. doi:10.1016/0023-9690(80)90018-1

Thornton, A., & Raihani, N. J. (2010). Identifying teaching in wild animals. Learning & Behavior, 38(3), 297–309. doi:10.3758/LB.38.3.297

Tresz, H. (2006). Behavioral management at the Phoenix Zoo: New strategies and perspectives. Journal of Applied Animal Welfare Science, 9(1): 65–70. doi:10.1207/s15327604jaws0901_6

von Bayern, A. M. P., Heathcote, R. J. P., Rutz, C., & Kacelnik, A. (2009). The role of experience in problem solving and innovative tool use in crows. Current Biology, 19(22), 1965–1968. doi:10.1016/j.cub.2009.10.037

Vonk, J., & MacDonald, S. E. (2002). Natural concepts in a juvenile gorilla (Gorilla gorilla gorilla) at three levels of abstraction. Journal of the Experimental Analysis of Behavior, 78(3), 315–332. doi:10.1901/jeab.2002.78-315

Vonk, J., & MacDonald, S. E. (2004). Levels of abstraction in orangutan (Pongo abelii) categorization. Journal of Comparative Psychology, 118(1), 3. doi:10.1037/0735-7036.118.1.3

Watters, J. V., & Powell, D. M. (2012). Measuring animal personality for use in population management in zoos: Suggested methods and rationale. Zoo Biology, 31(1), 1–12. doi:10.1002/zoo.20379

Webster, S. J., & Lefebvre, L. (2001). Problem solving and neophobia in a columbiform–passeriform assemblage in Barbados. Animal Behaviour, 62(1), 23–32. doi:10.1006/anbe.2000.1725

Weed, J. L., & O’Neill-Wagner, P. L. (2015). Animal behaviour research findings facilitate comprehensive captive animal care: The birth of behavioral management. Retrieved August 31, 2015, from http://awic.nal.usda.gov/environmental-enrichment-nonhuman-primates-resource-guide-animal-behavior-research-findings.

Whiten, A., Horner, V., & de Waal, F. B. M. (2005). Conformity to cultural norms of tool use in chimpanzees. Nature, 437(7059), 737–740. doi:10.1038/nature04047

Wystrach, A., Beugnon, G., & Cheng, K. (2012). Landmarks or panoramas: What do navigating ants attend to for guidance? Frontiers in Zoology, 8(1), 21–21. doi:10.1186/1742-9994-8-21

Xitco, M. J., Gory, J. D., & Kuczaj, S. A. (2001). Spontaneous pointing by bottlenose dolphins (Tursiops truncates). Animal Cognition, 4(2), 115–123. doi: 10.1007/s100710100107

Xitco, M. J., Gory, J. D., & Kuczaj, S. A. (2004). Dolphin pointing is linked to the attentional behavior of a receiver. Animal Cognition, 7(4), 231–238. doi: 10.1007/s10071-004-0217-z

Zentall, T., & Hogan, D. (1974). Abstract concept learning in the pigeon. Journal of Experimental Psychology, 102(3), 393. doi: 10.1037/h0035970

Zitzer, H., & MacDonald, S. E. (2015). The role of food preferences in habitat damage by wild African elephants (Loxodonta africana). Manuscript in preparation.

From the Pigeon Lab to the Courtroom

John T. Wixted
University of California, San Diego

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

The task of detecting the presence or absence of a stimulus based on a diagnostic evidence variable is a pervasive one. It arises in basic experimental circumstances, such as a pigeon making a decision about whether or not a stimulus was presented 10 seconds ago, as well as in applied circumstances, such as a witness making a decision about whether or not a suspect is the guilty perpetrator. Understanding how to properly conceptualize and analyze performance on a signal-detection task like that is nontrivial, and advances in this area have come mainly from experimental psychologists studying performance on basic memory and perception tasks. One illustrative example from the pigeon memory literature is considered here in some detail. Unfortunately, lessons learned by basic experimental psychologists (e.g., the value of using signal-detection theory to guide thinking, appreciating the distinction between discriminability and response bias, understanding the utility of receiver operating characteristic analysis, etc.), while having a major impact on applied fields such as diagnostic medicine, have not always been fully appreciated by applied psychologists working on issues pertaining to eyewitness misidentification. In this regard, signal-detection-based analyses can greatly enhance our understanding of important applied issues such as (a) the diagnostic accuracy of different police lineup procedures and (b) the relationship between eyewitness confidence and accuracy. The application of signal-detection theory to issues like these can reverse what many believe to be true about eyewitness identifications made from police lineups.

Keywords: Keywords: pigeon memory; ROC analysis; signal-detection theory; eyewitness memory, confidence and accuracy

Author Note: John T. Wixted, Department of Psychology, University of California, San Diego.

Correspondence concerning this article should be addressed to John T. Wixted at jwixted@ucsd.edu.

---

A ubiquitous task, both in the laboratory and in everyday life, involves making a decision about whether a stimulus occurred (Outcome A) or not (Outcome B). A pigeon, for example, might have to decide whether a keylight was briefly presented 5 s ago (Outcome A) or not (Outcome B) by pecking a red choice key (decision: the keylight was presented) or a green choice key (decision: the keylight was not presented). Similarly, a human might have to decide whether a test word appeared on a previously presented list (Outcome A) or not (Outcome B) by saying “old” or “new.” Or an eyewitness to a crime might have to decide whether a person shown to them by the police is the one who committed the crime (Outcome A) or not (Outcome B) by making a positive identification or not. These examples are all from the domain of memory, but similar detection tasks come up in many other domains. In diagnostic medicine, for example, a patient has an illness (Outcome A) or not (Outcome B) and a medical test is used to make a decision about which condition applies to this patient. And in a jury trial, the defendant is guilty (Outcome A) or not (Outcome B), and the jury makes a decision to convict or not. In all of these cases, a binary (dichotomous) decision has to be made based on what is often assumed to be a continuous evidence variable. The question facing the decision maker in each of these cases is whether or not there is sufficient evidence along some continuous scale to warrant making the decision to classify the item into Outcome A or not.

The continuous evidence variable upon which the decision is based changes as a function of the task, but the decision-making logic is the same in each case. On recognition memory tasks, the continuous evidence variable is, theoretically, the internal strength of the memory signal (which ranges from low to high), and the question is whether or not the memory signal is strong enough to say (for example) that the suspect in the photo is the person who committed the crime. In the medical context, the continuous variable might be the blood count of some protein, and the question is whether or not the blood count is high enough to warrant the diagnosis. And in a jury trial, the evidence variable is literally the sum total of incriminating evidence against a defendant. Is there enough evidence to conclude that the defendant is guilty beyond a reasonable doubt or not?

Signal-detection theory offers an illuminating framework for understanding how decisions like these are made (Green & Swets, 1966; Macmillan & Creelman, 2005). It not only provides a way to conceptualize why one decision is made instead of the other; it also suggests a measurement strategy that one would not likely hit upon in its absence. That measurement strategy is called receiver operating characteristic (ROC) analysis. ROC analysis has two broad purposes: (a) to distinguish between competing theories of decision making (e.g., between two nonidentical signal-detection models, or between a signal-detection model and a non-detection model), and (b) to measure discriminability (i.e., the ability to distinguish between the two relevant states of the world) in theory-free fashion. The second purpose may be the more important of the two because it is how ROC analysis is used in applied settings (such as diagnostic medicine). It might seem counterintuitive that a method like ROC analysis, which is so closely tied to signal-detection theory, can be described as “theory free,” but it is. Signal-detection theory brings you to and helps you conceptualize ROC analysis—indeed, it is hard to imagine conceiving of that approach in the absence of signal-detection theory—but once it does, for applied questions, the theory is no longer needed to interpret the data.

Using ROC Analysis to Test Theory

My own introduction to signal-detection theory (which eventually brought me to ROC analysis) began with a pigeon memory task. In a typical delayed matching-to-sample task, a trial begins with the presentation of (for example) a red or green light and then, after a delay, red and green choice keys are presented simultaneously. A response to the matching color is rewarded with food, whereas a response to the non-matching color ends the trial. A variant of this basic task involves the use of initial sample stimuli that are asymmetric in salience. For example, some investigations have involved sample stimuli consisting of presentations of food versus no food. Typically, the presentation of one of these samples is followed, after some delay, by a choice between two comparison stimuli (e.g., red and green). A response to one comparison is reinforced following samples of food, and a response to the other comparison is reinforced following samples of no food. A consistent finding in these studies is that performance following samples of food declines as the retention interval increases, whereas performance following samples of no food does not (Colwill, 1984; Colwill & Dickinson, 1980; Grant, 1991; Wilson & Boakes, 1985).

The same asymmetrical decay functions were observed by Grant (1991) when samples consisted of the presence versus absence of a variety of stimuli (including colors, shapes, and food). In each case, performance following the presence of an event declined as the retention interval increased, but performance following the absence of an event did not. In still other cases, the sample stimuli consisted of a short-duration houselight (e.g., 2 sec) versus a long-duration houselight (e.g., 10 sec), or a white keylight that required 40 keypecks to extinguish it versus a white keylight that required only 10 keypecks (Colwill, 1984; Fetterman & MacEwen, 1989; Sherburne & Zentall, 1993; Spetch & Wilkie, 1983). In these cases, too, the forgetting functions following the two samples are usually asymmetric. Figure 1 shows an example of asymmetric forgetting functions from a sample/no-sample task (Wixted, 1993).

Typically in these experiments, performance following the less salient sample (e.g., no sample, no food, a short sample, or a sample requiring relatively few keypecks) begins at a high level and remains accurate as the delay interval increases. Performance following the more salient sample (e.g., a sample in a sample/no-sample procedure, food in a food/no-food procedure, a long sample, or a sample requiring many keypecks) decreases rapidly as the retention interval increases and eventually falls to well below 50% correct (if the retention interval is long enough). What explains that pattern? Below, I present two competing theoretical accounts in some detail and describe an empirical investigation designed to differentiate between them. The following discussion may seem far removed from important social questions such as how to minimize eyewitness misidentifications, but my contention is that any such impression is far from the truth.

The Default Response (High-Threshold) Hypothesis

Various theories have been offered to account for asymmetric forgetting functions, but one theory is of particular interest because of its close connection to a theory of human recognition memory that prevailed in the years prior to the introduction of signal-detection theory. Colwill (1984), Wilson and Boakes (1985), and Grant (1991) all argued that the absence of a retention interval effect on no-sample trials suggests that memory plays no role on these trials. Instead, in the absence of memory, pigeons theoretically adopt a default response strategy of choosing the comparison stimulus associated with the absence of a sample. The default strategy is overridden on trials involving a sample so long as the memory trace has not completely faded. This explanation accounts for the flat retention function on no-sample trials because, whether the retention interval is short or long, no memory trace is ever present to override the default response. The same account explains why performance on sample trials is often significantly below chance at longer retention intervals: When the memory trace fades completely, subjects revert to their default strategy and reliably choose the wrong comparison stimulus.¹

A theory along these lines makes perfect sense, but one important and nonobvious feature of the theory is its implicit assumption that memory for the sample exists in one of only two discrete states, present vs. absent (i.e., memory strength is construed as an all-or-none variable, not as a continuous variable). According to this theory, when memory for the sample is present, that memory guides the response, leading to a correct choice. When memory for the sample is not present, the default response is implemented instead. If memory operated in that fashion, then a simple, algebraic model could be applied to the data to estimate a variable of interest, namely, the proportion of sample trials in which the sample was remembered (p). Imagine that, for a given task, the true value of p happened to be .80 (i.e., on 80% of the sample trials, the sample is remembered and guides choice performance). What might the pigeon do on the 20% of sample trials (and on 100% of the no-sample trials) in which memory for the sample is absent? The assumption is that on these trials, the pigeon implements its default strategy of choosing the no-sample alternative. The default strategy might not be to always choose the no-sample alternative under no-memory conditions, so the probability of choosing the no-sample alternative in the absence of memory can be represented by d, where d falls between 0 and 1. The value of d might be 1.0 (the pure default-response model), or it might instead be .90 or .80 without changing the basic pattern of results that this model predicts.

Generally speaking, the probability of choosing the sample alternative on a sample trial, p(“S” | S), is p (the probability that the sample is remembered) + (1 – p) times (1 – d), where 1 – p is the probability that the sample is not remembered and 1 – d is the probability that the sample alternative is selected on no-memory trials. If d = 1.0, then determining p is simple and straightforward: one need only measure the proportion of sample trials in which the sample alternative is chosen because p(“S” | S) = p + (1 – p) × (1 – d) = p + (1 – p) × 0 = p. Across conditions, one might find that the value of p is .80 in conditions involving a short retention interval and .20 in conditions involving a long retention interval (i.e., the probability of remembering the sample on sample trials decreases as the retention interval increases).

If the value of d is not equal to 1.0, then determining the value of p is slightly more complicated but still easy to do. The value of d is first determined by measuring the proportion of no-sample trials in which the no-sample alternative is correctly chosen. Imagine that on no-sample trials, pigeons choose the no-sample alternative 90% of the time (d = .90). This result would mean that when no memory for the sample is present (as must be true on no-sample trials), the bird’s default response is to choose the no-sample alterative 90% of the time and to choose the sample alternative 10% of the time. If d = .90, then performance on sample trials no longer provides a direct readout of p because, as indicated above, sample trial performance is theoretically equal to p + (1 – p) × (1 – d). Note that we can replace 1 – d with g, where g is the probability of “guessing” that the sample was presented despite no memory for the sample. Because g = 1 – d, we can write the equation as p(“S” | S) = p + (1 – p) × g.

If the observed probability of correctly choosing the sample alternative on sample trials, p(“S” | S), is called the hit rate (HR) and the observed probability of incorrectly choosing the sample alternative on no-sample trials, p(“S” | NS), is called the false alarm rate (FAR), then this simple model yields two equations:

HR = p + (1 – p) × g(1)

and

FAR = g(2)

This model is diagrammed in Figure 2. Note that the equation for performance on no-sample trials (Equation 2) places the focus on the non-default response (i.e., the probability of choosing the sample alternative by default in the absence of memory), but it is just another way of indicating that on no-sample trials, the probability of correctly choosing the no-sample alternative is equal to 1 – g, which is to say that it equals d.

Whereas the false alarm rate on no-sample trials provides a direct estimate of g, the hit rate on sample trials must be corrected to estimate p. This can easily be done by substituting FAR from Equation 2 for g in Equation 1:

HR = p + (1 – p) × FAR(3)

With a little algebraic rearrangement, we can solve for p to yield:

p = (HR – FAR) ∕ (1 – FAR)(4)

Using this equation, actual memory-based performance on sample trials can be directly computed from the data.

A concrete example will show how these equations work. Imagine that the size of the retention interval is manipulated within session, and performance on sample trials is 80% correct on short retention interval trials (HR_short = .80) and 40% correct on long retention interval trials (HR_long = .40). Further imagine that on no-sample trials, the no-sample alternative is correctly chosen by default 80% of the time. In other words, for both short and long retention intervals, d = .80 and g = .20. Using the equations above, we can estimate p for short and long retention interval trials:

p_short = (.80 – .20) ∙ (1 – .20) = .60 ∙ .80 = .75

p_long = (.40 – .20) ∙ (1 – .20) = .20 ∙ .80 = .25

In other words, when the retention interval is short, the bird remembers the sample on 75% of the trials, but when the retention interval is long, the bird remembers the sample on only 25% of the trials.

This algebraic approach to conceptualizing memory performance corresponds exactly to how recognition memory theorists once conceptualized human recognition memory performance on list-learning tasks. The theory was called the high-threshold theory of recognition memory (Green & Swets, 1966). In fact, although rarely used today, Equation 4 is the standard correction for guessing formula that was often used to measure recognition memory performance in list-learning experiments with humans (Macmillan & Creelman, 2005).

Perhaps the most important point to appreciate here is that, according to this model, various intuitively appealing measures of performance fully conflate two distinct properties of the decision-making process that ought to be separately estimated. For example, consider the most obvious choice of a dependent measure on a sample/no-sample task, overall proportion correct. Expressed in terms of the hit rate and false alarm rate, proportion correct is equal to [HR × N_Sample + (1 – FAR) × N_No-sample ] ∙ N, where N_Sample = the number of sample trials, N_No-sample = the number of no-sample trials, and N = N_Sample + N_No-sample (i.e., N = the total number of trials). If N_Sample = N_No-sample, as would typically be true, then proportion correct reduces to:

Proportion correct = [HR + (1 – FAR) ] ∙ 2(5)

Note that 1 – FAR is simply the proportion correct on no-sample trials, so this expression is the average of proportion correct on sample and no-sample trials.

Consider next what this proportion correct measure theoretically captures using the high-threshold model as a guide. We know from Equations 1 and 2 that HR = p + (1 – p) × g and FAR = g. Substituting these expressions for HR and FAR in Equation 5 yields:

Proportion correct = [p + (1 – p) g + (1 – g)] ∙ 2

which reduces to:

Proportion correct = [p (1 – g) + 1] ∙ 2

Thus, according to this model, if the ability to remember the sample remains constant across conditions but the likelihood of guessing changes across conditions, proportion correct will change. This change in the performance might lead the experimenter to conclude that memory in one condition is better than memory in the other, but that conclusion would be a mistake. The use of Equation 4 would reveal that memory is actually the same across both conditions (assuming that the high-threshold theory is correct).

From the perspective of high-threshold theory, p is the key measure. It is, for example, the measure that would be expected to be impaired in a group of amnesic patients (for humans tested using list memory), and it is the measure that would be expected to decrease as the retention interval increased (on a sample/no-sample task with birds or a list-memory task with humans). In addition, and critically, p would also be expected to remain constant (but the hit and false alarm rates would change) if the only aspect of performance that changed across conditions was g. According to Equations 1 and 2, both the hit rate and false alarm rate would be expected to increase as g increased, and both would decrease as g decreased. But only p is a measure of the ability to discriminate between the two states of the world. If p = 0, then performance on sample and no-sample trials would be the same (HR = FAR). In that case, the bird would show no evidence of being able to discriminate sample from no-sample trials. If p = 1, then performance on sample and no-sample trials would always be perfect if g = 0 (though it could be as low as 50% correct if g = 1).

The value of g is an experimentally manipulable variable. For example, a high rate of guessing (i.e., a high value of g) can be induced by arranging a differentially high payoff for correct sample choices (for pigeons) or correct “old” decisions (for humans on a list-learning task). A low rate of guessing (i.e., a low value of g) can be induced by arranging a differentially high payoff for correct no-sample choices (for pigeons) or correct “new” decisions (for humans on a list-learning task). This means that a set of hit and false alarm rate pairs, each of which is theoretically associated with a single level of memory-based performance (i.e., a single level of p), can be obtained across conditions by varying g. With those data in hand, one can plot hit rate vs. false alarm rate, and the resulting plot is known as the ROC.

Critically, Equation 3 provides the predicted shape of the ROC. That equation is in the form of the familiar equation for a straight line, y = m × x + b. In other words, according to Equation 3, the ROC, which is a plot of HR vs. FAR with memory (p) held constant, should be linear. The problem is that empirical ROCs are almost invariably curvilinear, which is why this simple model has been rejected, both in studies of human memory and (sometimes) pigeon memory. Signal-detection theory, which is considered next, offers a more viable interpretation of ROC data. For the moment, the most important take-home message is that a theoretical analysis of performance on a detection task draws a distinction between two distinct aspects of memory performance: response bias vs. the ability to differentiate between the two relevant states of the world. It is easy to lose sight of the importance of this distinction, which is what happened when researchers investigated recognition memory in the real world (an issue addressed later in this article).

Signal-Detection Theory

Figure 3 illustrates the signal-detection interpretation of performance on the sample/no-sample task (Wixted, 1993). The x-axis in this case represents the subjective strength of memory that a sample was presented earlier in the trial. Whereas the threshold model assumes the complete absence of a memory signal on these trials (which is intuitively sensible given that no sample was presented), the signal-detection model instead assumes that the act of retrospection always produces at least some false sense of sample occurrence. This is its most theoretically interesting departure from the high-threshold model. The strength of that false memory signal will vary from trial to trial due to noise in the neural system, but its mean value will be relatively low. Everything is the same on sample trials except that the average strength of the memory signal will be higher. Because the two distributions of memory strength signals partially overlap, there is no specific memory strength value that perfectly distinguishes between sample trials and no-sample trials. This is why the theory assumes that a criterion memory strength value, c, is set. On any trial in which the memory strength value exceeds the criterion, the sample alternative is chosen. This includes some no-sample trials in which the memory strength signal happens to be arbitrarily high. Thus, a false alarm in this model is based on a strong enough false memory signal, not on a random guess that occurs despite the complete absence of a memory signal (as in the threshold model). The proportion of the no-sample distribution that exceeds the criterion represents the false alarm rate. The proportion of the sample distribution that exceeds the criterion represents the hit rate. In this example, the hit rate ≈ .93 and the false alarm rate ≈ .07.

A longer retention interval will have no effect on the mean of the noise distribution (i.e., on the no-sample distribution) because it does not matter how long ago nothing occurred. In other words, on these trials, no memory trace is created that fades away. By contrast, the mean of the sample distribution will decrease with increasing retention interval as memory for the sample weakens. Thus, the longer the retention interval, the smaller the proportion of the sample distribution that exceeds the criterion and the lower the hit rate will be (assuming the criterion remains fixed as the retention interval increases). Eventually, the sample distribution will coincide with the no-sample distribution, and at that point, the hit rate will equal the false alarm rate. This is the empirical pattern that is observed on sample/no-sample tasks. That is, the hit rate decreases but the false alarm rate (or 1 – the false alarm rate, which is the no-sample measure plotted in Figure 1) remains constant.

Which interpretation is more consistent with the available evidence? The high-threshold account or the signal-detection account? As noted above, one way to answer that question is to empirically examine the shape of the ROC. Variations in payoffs described earlier, which were assumed to affect g (the probability of guessing “sample” or “old” despite the absence of memory) are now assumed to affect the location of the decision criterion, c. Payoffs that encourage choosing the sample alternative move the criterion to the left, such that more of the sample distribution and more of the no-sample distribution exceed it (corresponding to higher hit and false alarm rates). Payoffs that encourage choosing the no-sample alternative move the criterion to the right, such that less of the sample distribution and less of the no-sample distribution exceed it (corresponding to lower hit and false alarm rates).

In this model, the location of c on the memory axis is conceptually related to the magnitude of g in the high-threshold model. For example, as c moves to the left, or as g increases, the hit and false alarm rates both increase, an effect that would be referred to as a liberal response bias. What is the measure of memory in the signal-detection account that corresponds to the value of p in the high-threshold model? Critically, the relevant measure of memory is not a probability because there is no discrete event that corresponds to the probabilistic occurrence of memory for the prior presentation of the sample. Instead, there are only degrees of memory strength. Overall memory performance is high to the extent that the average strength of memory on sample trials is high compared to the average strength of memory on no-sample trials. That is, memory ability is theoretically captured by the distance between the means of the sample and no-sample distributions (scaled in standard deviation units), which is a measure known as d′. Theoretically, d′ indicates how well the organism’s brain separates the population of memory signals associated with sample trials vs. the population of memory signals associated with no-sample trials. In Figure 3, d′ = 3 (i.e., the means of the two distributions are three standard deviations apart).

One virtue of the signal-detection approach is that it naturally predicts a curvilinear ROC. Figure 4 shows two ROC plots, one that corresponds to a high d′ (e.g., as might occur if a short retention interval is used) and another that corresponds to a low d′ (e.g., as might occur if a long retention interval is used). In fact, these are actual data from a sample/no-sample experiment reported by Wixted (1993). Note the curvilinearity of the data in each case, which is more consistent with the signal-detection view than the pure threshold view (the dashed lines show the linear trend predicted by the threshold model). Again, the interesting theoretical implication of this result is that there are no “no-memory trials.” Instead, on every trial, the bird theoretically queries memory for evidence that that a sample was presented on that trial, and on every trial, a signal is returned by the brain. Sometimes (e.g., on no-sample trials), the signal that is returned is just noise in the nervous system. When a signal is returned, the bird then determines whether that signal is strong enough to decide that the sample was in fact presented (i.e., if the strength of that signal exceeds c).

The ROC data in this case were obtained by experimentally manipulating the birds’ decision criterion. In the neutral condition (the middle ROC point in each condition), the payoff for a correct “no-sample” decision was the same as the payoff for a correct “sample” decision. In both cases, the probability of food reinforcement for a correct response was .60. In the liberal condition (rightmost ROC point in each condition), the probability of food reinforcement was asymmetrical such that a correct sample choice was reinforced with probability 1.0, whereas a correct no-sample choice was reinforced with probability 0.20. These contingencies induced the birds to choose the sample alternative on a higher percentage of both sample and no-sample trials, thereby increasing both the hit rate and the false alarm rate relative to the neutral condition. To put this another way (and to make the results more relatable to the later discussion of eyewitness memory), the birds were more inclined to choose the sample alternative even when they were not especially confident that the sample had been presented. In the conservative condition (leftmost ROC point in each condition), the probability of food reinforcement was asymmetrical in the other direction such that a correct sample choice was reinforced with probability .20, whereas a correct no-sample choice was reinforced with probability 1.0. These contingencies induced the birds to choose the no-sample alternative on a higher percentage of both sample and no-sample trials, thereby decreasing both the hit rate and the false alarm rate relative to the neutral condition. In other words, the birds would only choose the sample alternative (with a payoff probability of only .20) when they were highly confident that the sample had in fact been presented on that trial. This would only occur if the memory strength on that trial were strong enough to exceed the high setting of the decision criterion.

The empirical ROC data are obviously closer to what the signal-detection model predicts than what the high-threshold model predicts. The same result is almost always observed on human memory tasks as well. As a result, signal-detection theory is generally regarded as the dominant account of human recognition memory, and d′ has become the standard dependent measure. This discriminability measure is essentially the same as log d in the Davison-Tustin (1978) model. Note that variations of the discrete-state high-threshold model can be found that will accommodate curvilinear ROC data, but my only purpose thus far has been to illustrate basic conceptual distinctions that separate the signal-detection view of memory from alternative theoretical views and to illustrate how the effort to test competing theoretical views brings one to ROC analysis.

To the Courtroom

The battle between high-threshold and signal-detection accounts of recognition memory has also played out (and in one form or another continues to play out) in the basic human memory literature. Although it sometimes seems like an abstract debate of interest only to math modelers, an argument could be made that a detailed inquiry into the underlying theoretics of recognition memory serves to underscore critical distinctions that are easily overlooked when the focus shifts to recognition memory in the real world (such as eyewitness memory). A critical distinction in the analyses considered above, and as noted earlier, is the well-known distinction between discriminability and response bias. In the high-threshold model, these two properties are captured by p and g, respectively, and in the signal-detection model, they are captured by d′ and c, respectively. Although the details of both models cannot simultaneously be true, the distinction they both draw between discriminability and response bias is similar and is far more important than it might seem to be at first glance. To see why, I turn next to the issue of the reliability of eyewitness identification and to the lab-based recognition memory tasks that are most commonly used to investigate it. For decades, this research has been mostly carried out without regard for the distinction between discriminability and response bias (for some notable exceptions, see Ebbesen & Flowe, 2002; Horry, Palmer, & Brewer, 2012; Meissner, Tredoux, Parker, & MacLin, 2005; Palmer & Brewer, 2012), and the reported results have had a profound effect on practices in the legal system. Without the guidance of basic theories of recognition memory (theories that protect one from compelling but often faulty intuitions), the argument can be made that eyewitness identification researchers got it wrong in several ways (Gronlund, Mickes, Wixted, & Clark, 2015).

From my perspective, this is a story about how basic psychological science and applied psychological science have drifted much too far apart from one another in recent years. As a result, mistakes have been made. My point is certainly not that eyewitness ID researchers got everything wrong, or that all eyewitness ID researchers made the influential mistakes I review next. The issues that the field got right (e.g., that memory is malleable and that eyewitness identification tests should not be biased against a suspect) are not controversial and are also not uniquely informed by signal-detection theory and ROC analysis. By contrast, the ones that the field got wrong are uniquely informed by signal-detection theory and ROC analysis, and those are the issues I focus on here.

Eyewitness Misidentification in the Real World

Many people, including, I would guess, most readers of this article, believe that eyewitness memory is inherently unreliable. And why not? Of the 334 wrongful convictions that have been overturned to date by DNA evidence since 1989, more than 70% were attributable, at least in part, to eyewitness misidentification (Innocence Project, 2015). A statistic like that is hardly a testimony to the impressive reliability of eyewitness identification. Instead, it seems like an obvious testimony to the catastrophic unreliability of eyewitness identification.

How can such tragic errors be reduced? For more than 30 years, applied psychological science has been brought to bear on this issue by using mock-crime laboratory studies. In a typical mock-crime study, participants (e.g., undergraduates) witness a mock crime (e.g., by watching a video of someone committing a crime, such as snatching a purse) and are later shown a photo lineup in which the perpetrator (the target) is either present or absent. A target-present lineup includes the perpetrator along with (usually five) similar fillers; a target-absent lineup is the same except that the perpetrator is replaced by another similar filler, as illustrated in Figure 5. That replacement filler can be designated as the innocent suspect. Note that not all studies pre-designate an innocent suspect in target-absent lineups, which adds complexity to the analysis of the data without changing anything of substance. Thus, in what follows, I shall assume that both target-present and target-absent lineups always have one suspect and five fillers, as real-world lineups typically do. Just as in a real-world investigation, a witness presented with a photo lineup in a mock-crime study can (a) identify a suspect (a suspect ID of an innocent or a guilty individual), (b) identify a filler (a filler ID), or (c) reject the lineup (no ID).

A suspect ID is the most consequential outcome of a lineup procedure because, as a general rule, only suspects who are identified from a lineup are placed at risk of prosecution. A suspect ID from a target-present lineup rightfully imperils the guilty perpetrator, but a suspect ID from a target-absent lineup wrongfully imperils an innocent suspect. A mistaken filler ID does not imperil anyone because the fillers are known to be innocent (e.g., the fillers might be database photos of people imprisoned in another state). The two key dependent measures in a mock-crime study are the correct ID rate (proportion of target-present lineups from which the guilty suspect is identified) and the false ID rate (proportion of target-absent lineups from which the innocent suspect is identified). In other words, this is a recognition task in which the hit rate and the false alarm rate are measured. Therefore, one’s thoughts should already be turning to signal-detection theory and ROC analysis, but many years went by (and extensive reforms were made to the legal system) before the first eyewitness ROC analysis was ever performed.

Simultaneous vs. Sequential Lineups in the Lab.

In light of the DNA exoneration cases, a major goal of scientific research (understandably) has been to find ways to reduce the false ID rate without appreciably reducing the correct ID rate. One simple change in the way that photo lineups are administered has long been thought to help protect innocent suspects from being mistakenly identified without much cost in terms of correctly identifying guilty suspects. Specifically, instead of presenting all six photos simultaneously (the traditional approach, as illustrated in Figure 5), the lineup photos are presented sequentially (one at a time) for individual yes/no decisions (Lindsay & Wells, 1985). The test effectively stops when someone is identified as the perpetrator. If the sequential test continues beyond that point, only the first identification typically counts (second laps are usually not allowed in lab studies, though they tend to be allowed in real-world sequential lineups).

Mock-crime studies have often found that sequential lineups result in a lower false ID rate. These same studies have often found that sequential lineups also lower the correct ID rate but to a lesser extent. In a review of the literature, Steblay, Dysart, and Wells (2011) reported that the average HR and FAR for the simultaneous lineup procedure equal 0.52 and 0.28, respectively, whereas the corresponding values for the sequential lineup procedure equal 0.44 and 0.15, respectively. Thus, on average, the sequential procedure yields both a lower HR and a lower FAR—an ambiguous outcome in terms of identifying the better procedure. Still, the drop in the FAR exceeds the drop in the HR. To the untrained eye, that seems to suggest a sequential superiority effect.

In an effort to quantify the diagnostic accuracy of the competing lineup procedures in terms of a single measure, eyewitness identification researchers have long relied on a statistic known as the diagnosticity ratio (correct ID rate ∕ false ID rate). Although the issue is contested (e.g., Clark, 2012; Gronlund, Carlson, Dailey, & Goodsell, 2009), some meta-analytic reviews of the mock-crime literature have concluded that the diagnosticity ratio is generally higher for sequential lineups (Steblay, Dysart, Fulero, & Lindsay, 2001; Steblay et al., 2011). For example, using the numbers reported by Steblay et al. (2011), the diagnosticity ratio for the sequential lineup procedure (0.44 ∕ 0.15 = 2.93) is higher than that of the simultaneous lineup procedure (0.52 ∕ 0.28 = 1.86), which led them to conclude that the sequential procedure is superior. The diagnosticity ratio increases because, when switching to the sequential procedure, the proportional drop in the FAR exceeds the proportional drop in the HR. That in itself seems like a positive outcome, thereby favoring the sequential procedure. The case in favor of the sequential procedure seems even more secure when one considers what the diagnosticity ratio actually measures. If half the lineups are target-present lineups and half are target-absent lineups (which is true of most of the relevant studies), then the diagnosticity ratio is a direct measure of the posterior odds of guilt. If the sequential procedure yields a higher diagnosticity ratio, then not only is the FAR rate lower, the posterior odds that an identified suspect is actually guilty are higher (i.e., the ID is more trustworthy) compared to a suspect identified from a simultaneous lineup. On the surface, the case in favor of the sequential procedure seems very strong indeed. Based on this interpretation of the empirical literature, approximately 30% of law enforcement agencies in the United States that use photo lineups have now adopted the sequential procedure (Police Executive Research Forum, 2013). Not many areas of psychological research can rival the real-world impact that eyewitness identification research has had.

Note that when using a lineup procedure, the essence of the task is to discriminate between innocent and guilty suspects. It is a detection task in much the same way that a sample/no-sample task with a pigeon is. It seems trickier because of the presence of fillers (what should one do with a filler ID?), but fillers have not stood in the way of computing the hit and false alarm rates that have convinced many that sequential lineups are diagnostically superior to simultaneous lineups. If there are 100 target-present lineups, and witnesses (a) identify the suspect in 52 of the lineups, (b) identify a filler in 16 of the lineups, and (c) reject the remaining 32 lineups, the hit rate is 52 ∕ 100 = .52. Similarly, if there are 100 target-absent lineups, and witnesses (a) identify the suspect in 24 of the lineups, (b) identify a filler in 32 of the lineups, and (c) reject the remaining 44 lineups, the false alarm rate is 24 ∕ 100 = .24. Thus, filler IDs are not typically counted when computing hit and false alarm rates (nor should they be). Ideally, the goal is to get the FAR as close to 0 as possible and the HR as close to 1.0 as possible. In other words, the goal is to maximize discriminability between innocent and guilty suspects. The sequential lineup procedure seems to do a good job of reducing the FAR without compromising the HR too much (though it would be better if it actually increased rather than slightly decreasing the HR). It may seem as if the data suggest that sequential lineups achieve the goal of increasing discriminability, but consider for a moment the fact that, so far, a compelling story in favor of the sequential procedure has been told with no mention of a measure of discriminability (and with no look at the ROC).

As discussed earlier in connection with high-threshold theory vs. signal-detection theory, a singular pair of hit and false alarm rates does not characterize the discriminability of a procedure. Instead, the whole ROC does. To say that the goal is to maximize discriminability is to say that the goal is to achieve the highest possible ROC. The ROC depicts the family of achievable hit and false alarm rates associated with a particular condition. If Condition A yields a higher ROC than Condition B, it means that both states of the world can be more accurately categorized in Condition A compared to Condition B. That is, if it yields a higher ROC, Condition A is capable of achieving both a higher HR and a lower FAR than Condition B.

Instead of performing ROC analysis, researchers in the field of eyewitness identification computed the diagnosticity ratio for each condition in an effort to determine which lineup format is superior. The problem is that this intuition-based approach cannot reveal the diagnostically superior condition when the HR and FAR both change in the same direction (which is the case here: the HR and the FAR are both lower for the sequential procedure). Why not? The reason why a higher diagnosticity ratio does not identify the superior procedure is most easily appreciated by examining a basic property of an ROC curve. Keep in mind that an ROC shows the full range of hit and false alarm rates that are achievable as response bias ranges from liberal to conservative (while holding discriminability constant). An important consideration that has only recently come to be understood in the field of eyewitness identification is that a natural consequence of more conservative responding (in addition to the fact that the correct and false ID rates decrease) is that the diagnosticity ratio increases (Gronlund, Wixted, & Mickes 2014; Wixted & Mickes, 2012, 2014). Critically, this occurs whether more conservative responding is induced for the simultaneous procedure (e.g., using instructions that encourage eyewitnesses not to make an ID unless they are confident of being correct) or more conservative responding is induced by switching to the sequential procedure. The diagnosticity ratio continues to increase as responding becomes ever more conservative, all the way to the point where both the correct and false ID rates approach 0, in which case administering a lineup would be practically useless even though the diagnosticity ratio would be very high (Wixted & Mickes, 2014). Thus, achieving the highest possible diagnosticity ratio by inducing ever more conservative responding is not a logical goal to pursue.

As noted by a recent National Academy committee report on eyewitness identification, “ROC analysis represents an improvement over a single diagnosticity ratio” (National Research Council, 2014, p. 80). To be sure, the committee did not judge ROC analysis to be such a flawless methodology that the field can now stop worrying about the best way to compare lineup procedures and use ROC analysis forevermore. Instead, the committee also expressed reservations about confidence-based ROC analysis because different eyewitnesses might exhibit differences in the inclination to express a certain level of confidence, such as high confidence. That is, the memory strength that warrants high confidence for one eyewitness might warrant only medium or low confidence for another. Thus, although the committee agreed that ROC analysis represents an advance over the diagnosticity ratio, it also called for new research to identify even better diagnostic methodologies. For the time being, however, there are only two choices: the diagnosticity ratio and ROC analysis. Given that choice, ROC analysis is clearly the better option. This is an important point to consider because the two approaches (namely, the diagnosticity ratio vs. ROC analysis) can yield opposite answers to the question of which lineup procedure is diagnostically superior.

To appreciate the advantage of ROC analysis, consider the two ROC curves illustrated in Figure 6. The ROC is a plot of the family of hit and false alarm rates (i.e., correct and false ID rates) associated with each procedure, and values shown next to each data point indicate the diagnosticity ratio (i.e., correct ID rate ∕ false ID rate) for that point. In this example, Procedure A is diagnostically superior to Procedure B because for any given false ID rate, Procedure A can achieve a higher correct ID rate. If only a single ROC point is computed for each procedure and those two points are then compared using the diagnosticity ratio (as was done in the vast majority of mock-crime lab studies comparing simultaneous and sequential lineups), the diagnostically inferior lineup procedure could be misconstrued as being the superior procedure (e.g., imagine computing only the rightmost ROC point for each procedure and comparing them using the diagnosticity ratio). The only way to determine the diagnostically superior procedure is to trace out the ROC (i.e., trace out the obtainable hit and false alarm rates) for each lineup procedure.

The easiest and by far the most common way to construct an ROC in experiments with humans is to collect confidence ratings. The overall hit and false alarm rates (i.e., the values that are usually reported as the correct and false ID rates) are computed using all correct suspect IDs from target-present lineups and all incorrect suspect IDs from target-absent lineups. In the example given above, there were 52 correct suspect IDs from 100 target-present lineups (made with varying degrees of confidence) and 24 incorrect suspect IDs from 100 target-absent lineups (again made with varying degrees of confidence). This yielded overall hit and false alarm rates of .52 and .24 respectively. ROC analysis essentially gives you permission to disregard suspect IDs that are made with low confidence (as the legal system might do). If you disregard low-confidence suspect IDs by treating them as effective non-IDs, then (a) you have adopted a more conservative standard for counting suspect IDs, and (b) you will have fewer correct and false IDs than you did before, so the correct and false ID rates will now both be lower. Imagine that two correct suspect IDs were made with low confidence and 10 incorrect suspect IDs were made with low confidence. Excluding these IDs leaves 52 – 2 = 50 correct suspect IDs from 100 target-present lineups and 24 – 10 = 14 incorrect suspect IDs from 100 target-absent lineups. Thus, the new hit and false alarm rates are .50 and .14, respectively. Now there are two points to plot on the ROC. When all IDs are counted regardless of confidence, the resulting correct and false ID rates correspond to the rightmost ROC point. Disregarding low-confidence IDs yields the next ROC point down and to the left.

Once you realize that you are not obligated to count IDs made with low confidence, it immediately follows that you are also not obligated to count IDs made with medium confidence. Excluding IDs made with low or medium confidence by treating them as effective non-IDs yields yet another pair of correct and false ID rates (i.e., another ROC point, again down and to the left). Critically, as noted above, the diagnosticity ratio increases monotonically as an ever-higher confidence standard is applied. Although it is easy to imagine that the diagnosticity ratio might not increase as responding becomes more conservative, it invariably occurs and is naturally predicted by signal-detection theory (see Appendix of Wixted & Mickes, 2014).

The point is that one must perform ROC analysis, not compute the diagnosticity ratio from a singular pair of hit and false alarm rates, to identify the diagnostically superior lineup procedure. The first ROC study of eyewitness identification procedures only appeared in late 2012 and it is reproduced here in Figure 7 (Mickes, Flowe, & Wixted, 2012). The results came as a surprise because they unexpectedly revealed a simultaneous superiority effect. Before that study was performed, there had not been a single suggestion that simultaneous lineups might be superior to sequential lineups. Instead, over the years, the debate had been whether there was a sequential superiority effect (because it tended to yield a higher diagnosticity ratio) or whether the two procedures were diagnostically equivalent. Now, multiple ROC studies of simultaneous vs. sequential lineups have been published, and they all show evidence of a simultaneous superiority effect, though the effect is not always significant (Carlson & Carlson, 2014; Dobolyi & Dodson, 2013; Gronlund et al., 2012; Mickes et al., 2012). To date, no ROC study has shown the slightest hint of a sequential superiority effect.

Simultaneous vs. Sequential Lineups in the Real World

Do the ROC results from lab studies generalize to the real world? In two recent police department field studies comparing the two lineup formats (one in Austin, Texas, and the other in Houston, Texas), evidence of a simultaneous superiority effect was observed. Using expert ratings of incriminating evidence against identified suspects, Amendola and Wixted (2015) found that, in Austin, the results significantly favored the simultaneous procedure. In other words, the results suggested that guilty suspects were more likely to be identified—and innocent suspects were less likely to be misidentified—using simultaneous lineups compared to sequential lineups. Similarly, in the Houston field study, Wixted, Mickes, Clark, Dunn, & Wells (in press) again found evidence of a simultaneous superiority effect based on police officer ratings of incriminating evidence against identified suspects. The effect was not always significant, depending on how the data were analyzed, but the trend was always in a direction that favored the simultaneous procedure. In addition, a separate signal-detection-based analysis of eyewitness confidence ratings in the Houston field study also favored the simultaneous procedure.

ROC analysis cannot be performed on data collected from real eyewitnesses because one does not know whether suspect IDs are correct or incorrect (information that is needed to compute the correct and false ID rates that make up the ROC). Instead, as noted above, the lineup performance measure used in these two police department field studies was “independent evidence of guilt,” which is a proxy for odds of guilt. This measure is conceptually identical to the diagnosticity ratio that has been used in lab studies for years. That is, the diagnosticity ratio—correct ID rate ∕ false ID rate—is also an odds-of-guilt measure. That fact raises an obvious question: Why is it acceptable to use an odds-of-guilt measure for real eyewitnesses when it is not acceptable to use it for lab studies?

An odds-of-guilt measure is problematic only when responding is more conservative for one lineup procedure than the other. In lab studies, sequential lineups often induce more conservative responding. Under those conditions, an odds-of-guilt measure like the diagnosticity ratio would be expected to favor the more conservative procedure whether or not it is the diagnostically superior procedure because that measure increases as responding becomes more conservative. However, in both police department field studies (the one conducted in Austin and the one conducted in Houston), responding happened to be similarly biased for simultaneous and sequential lineups in the sense that IDs were made with approximately equal frequency for both lineup types. Under those conditions only, an odds-of-guilt measure correctly identifies the diagnostically superior lineup procedure. In both police department field studies, the simultaneous procedure was favored according to the odds-of-guilt measure, just as would be predicted from recent lab-based ROC analyses.

Pushback from Proponents of the Sequential Procedure

Perhaps understandably, longstanding advocates of the sequential procedure have a different take on the data. For example, Wells, Steblay, & Dysart (2015a) argued that the results of the Austin Police Department field study, when combined with police department field data from three other study sites (San Diego, California; Tucson, Arizona; and Charlotte-Mecklenburg, North Carolina), actually favored the sequential procedure. Their argument was based not on independent incriminating evidence against identified suspects (as our analyses were) but was instead based on the fact that the filler ID rate for sequential lineups was lower than the filler ID rate for simultaneous lineups. Because fillers are known to be innocent, the interpretation was that sequential lineups better protect innocent suspects from being misidentified.

In addition, Wells, Steblay, and Dysart (2015b) and Steblay, Dysart, and Wells (2015) argued that the sample of identified suspects studied by Amendola and Wixted (2015) was, for unidentified reasons, biased against the sequential procedure. The basis of their concern about a possibly biased sample was that the ultimate case outcomes (i.e., proportion of suspects ultimately found guilty by jury or plea bargain) differed noticeably for the suspects identified in Austin (where our expert ratings study was conducted) compared to suspects identified in all four study sites aggregated together (Austin, Charlotte-Mecklenburg, San Diego, and Tucson). In Austin, the results showed that a higher proportion of suspects identified from simultaneous lineups was found guilty compared to sequential lineups; in the full data set, by contrast, the case outcomes for simultaneous and sequential lineups were more evenly balanced. Their interpretation of this pattern of data was that, for some reason, the Austin sample included an unusually high number of guilty suspects in the simultaneous condition. If so, it would not be surprising that independent expert ratings of guilt would also be higher for the simultaneous sample than for the sequential sample.

A subsequent analysis by Amendola and Wixted (2015) showed that the case made by Wells and colleagues, which is dependent on aggregating data across study sites, overlooks statistically significant evidence of site variance that effectively disallows aggregating data across sites. For example, based on an analysis of data aggregated across study sites, Wells et al. (2015a) argued that the lower filler ID rate observed for sequential lineups suggests a sequential lineup advantage. However, by examining the data separately by study site, Amendola and Wixted (2015) showed that the observed filler ID rate difference (like the higher diagnosticity ratio often associated with sequential lineups in lab studies) is entirely attributable to a conservative response bias that was evident in the three non-Austin study sites—a response bias that was absent in the Austin study site. As with lab studies, the conservative response bias sometimes induced by sequential lineups does not indicate a sequential superiority effect. Moreover, this previously unappreciated evidence of site variance also accounts for why Wells et al. (2015b) and Steblay et al. (2015) came to believe that the Amendola and Wixted (2015) sample was biased against the sequential procedure. As noted above, the basis of their concern about a possibly biased sample was that the case outcomes (i.e., proportion ultimately found guilty) for the suspects identified from lineups in Austin differed noticeably from the case outcomes for the suspects identified from lineups aggregated across all four study sites. However, given evidence of site variance, there is no reason why the Austin sample (where no conservative response bias was observed for sequential lineups relative to simultaneous lineups) should be representative of the data collapsed across study sites. Moreover, because response bias was similar for simultaneous and sequential lineups in the Austin sample only, any comparison of lineup performance based on incriminating evidence of guilt has to be limited to data from that site alone. For the Austin data, filler ID rates show no hint of a sequential superiority effect, and the Amendola and Wixted (2015) expert ratings data show clear evidence of a simultaneous superiority effect.

Theoretical Basis of the Simultaneous Superiority Effect.

In retrospect, the diagnostic advantage of simultaneous lineups should not have come as a surprise. The reason is that no theoretical explanation as to why sequential lineups might yield higher discriminability has ever been advanced, and the longstanding absence of a theoretical explanation to that effect probably should have been a cause for concern. To be sure, there is a prominent theory about why different patterns of responding are maintained by simultaneous and sequential lineups, but it is not a theory of discriminability. This well-known theory draws a distinction between absolute and relative decision strategies (Lindsay & Wells, 1985; Wells, 1984). According to this account, simultaneous lineups encourage a witness to identify the lineup member who most resembles the eyewitness’s memory of the perpetrator (a relative decision strategy). By contrast, sequential lineups encourage a witness to choose a lineup member only if the familiarity signal exceeds an absolute decision criterion. Wixted and Mickes (2014) argued that this is a theory of response bias (i.e., simultaneous lineups engender a more liberal response bias than sequential lineups), not a theory of discriminability. In agreement with this view, Wells (1984) wrote, “It is possible to construe of the relative judgments process as one that yields a response bias, specifically a bias to choose someone from the lineup” (p. 94).

But what about discriminability—that is, the ability to tell the difference between innocent and guilty suspects? Until recently, no theory of discriminability for the lineup task had ever been proposed. A case could be made that a theory of response bias in terms of absolute and relative responding is much less important than a theory of discriminability because manipulating response bias is easy to do using either kind of lineup procedure (i.e., one need switch lineup procedures to influence response bias). By contrast, in the absence of theoretical guidance, improving discriminability is hard. Fortunately, simple theoretical principles from the perceptual learning literature very naturally explain why simultaneous lineups should be diagnostically superior to sequential lineups in terms of discriminability. The basic idea, as argued by Wixted and Mickes (2014), is that simultaneous lineups immediately teach the witness that certain facial features are nondiagnostic and therefore should not be relied upon to try to decide whether or not the guilty suspect is in the lineup. The nondiagnostic features are the features that are shared by every member of the lineup (the fillers and suspect alike, and whether the suspect is innocent or guilty). These are the features that were used to select individuals for inclusion in the lineup—that is, features that match the physical description of the perpetrator. Because everyone in the lineup shares those features, relying on them to determine whether or not the guilty suspect is in the lineup can only harm discriminative performance. Simultaneous lineups immediately teach the witness what the nondiagnostic features are (namely, the features that are obviously shared by all six members of the lineup, such as the fact that they are all young white males), thereby allowing those features to be given less weight and, as a result, enhancing discriminative performance.

Again not surprisingly, longstanding advocates of the sequential procedure have taken issue not just with our interpretation of police department field study data but also with recent lab-based ROC analyses that confirmed the simultaneous superiority effect in terms of discriminability (e.g., Wells, Smalarz, & Smith, in press; Wells, Smith, & Smalarz, in press). In fact, they are taking the position that ROC analysis is not informative when it comes to comparing lineup procedures, declaring not only that I and other researchers are wrong about that but that so is the National Academy of Sciences committee that recently weighed in on the issue (National Research Council, 2014). In their view, the work that my colleagues and I performed on this issue misled the esteemed National Academy committee, thereby explaining why the committee made the mistake of endorsing ROC analysis over the diagnosticity ratio. As they put it: “Yes, the National Research Council (NRC) report got it wrong by interpreting ROC analyses on lineups as measures of underlying discriminability. But that is how the NRC eyewitness committee read and interpreted Wixted and Mickes’ work” (Wells, Smith, & Smalarz, in press). As part of an invited debate about these issues, Mickes and I have elaborated on the case in favor of ROC analysis (Wixted & Mickes, in press a; Wixted & Mickes, in press b). Lampinen (in press) recently joined the debate by arguing against the utility of ROC analysis.

It seems clear that the issue will continue to be debated in the years to come, but it is hard for me to imagine that the ultimate judgment will be that ROC analysis has nothing useful to offer. The mere fact that, in the past, researchers based their argument in favor of sequential lineups on the ratio of the correct ID rate to the false ID rate means that they already computed one point on the ROC. If there is a reasonable case to be made as to why it is mandatory to compute one point on the ROC to measure lineup performance yet is utterly inappropriate to examine any of the other points on the ROC, then that case should be made. Thus far, the anti-ROC arguments have avoided this basic consideration. Anyone interested in this topic would do well to read the various articles in this ongoing debate and then make their own judgment as to who has the stronger argument. The larger community of scientists will, of course, be the ultimate judge. To some extent, that is already happening (National Research Council, 2014; Rotello, Heit, & Dubé, 2015), and this is how it should be. For too long, some of the most influential applied research has been conducted by psychologists who are (in my view) too far removed from basic psychological science. Indeed, lineup format is not the only consequential issue that applied psychologists got wrong over the years. The other important issue where key mistakes have been made has to do with the very notion of eyewitness unreliability itself.

Eyewitnesses are nowhere near as unreliable as they have long been thought to be. As described in more detail below, eyewitness ID researchers have corrected this mistake with a compelling series of empirical calibration studies (most of which come from Neil Brewer and his colleagues), but the information seems almost exclusively confined to that small field. In my experience, the larger community of experimental psychologists typically reacts with shock in response to the claim that, under typical laboratory conditions (e.g., fair lineups, no administrator influence, etc.), eyewitness confidence is a strong indicator of reliability. Moreover, the earlier work suggesting that eyewitness identification is inherently unreliable even under pristine laboratory conditions has had a profound influence on the U.S. legal system, and that influence is growing, not shrinking. Courts across the land are increasingly inclined to disregard expressions of confidence made by eyewitnesses. A case can be made—and we do make the case—that this practice unnecessarily places innocent suspects at risk (exactly the opposite of what was intended).

Confidence and Accuracy

To many, the suggestion that eyewitnesses are not inherently unreliable may sound as implausible as the idea that ESP is real. However, my colleagues and I recently made the case that the blanket indictment of the reliability of eyewitness identification from a lineup is incorrect and serves only to place innocent suspects at greater risk of being wrongfully convicted (Wixted, Mickes, Clark, Gronlund, & Roediger, 2015). To appreciate why, it is essential to first draw a distinction between the initial eyewitness ID from a lineup and the much later ID that occurs at trial. There is nearly unanimous agreement that initial confidence can become artificially inflated for a variety of reasons such that by the time a trial occurs, an original ID made with low confidence (for example) can morph into an ID made with high confidence (Wells & Bradfield, 1998, 1999). The DNA exoneration cases that were associated with eyewitness misidentification involved high-confidence IDs of an innocent defendant made in front of a jury during a trial (and the jury interpreted the ID as compelling evidence of guilt). Errors like these make it clear that eyewitness identification can be unreliable under some conditions, such as at trial. Indeed, through decades of work, Loftus and her colleagues have established beyond any reasonable doubt that memory is malleable (Loftus, 2005; Loftus & Pickrell, 1995; Loftus & Palmer, 1974; Loftus, Miller, & Burns, 1978).

But is eyewitness memory always unreliable? The idea that eyewitness memory is generally unreliable (not just at trial) was set in stone by the fact that mock-crime studies once seemed to convincingly show that, even at the time of the initial identification from a lineup (long before a trial occurs and before memory contamination has much of an opportunity to take place), eyewitnesses who make a high-confidence identification are only somewhat more accurate than the presumably error-prone eyewitnesses who make a low-confidence identification (Devenport, Penrod, & Cutler, 1997). In other words, this research seemed to indicate that the relationship between confidence and accuracy is weak across the board.

The relationship between confidence and accuracy was originally measured by computing the standard Pearson r correlation coefficient between the accuracy of a response (e.g., coded as 0 or 1) and the corresponding confidence rating (e.g., measured using a five-point scale from “just guessing” to “very sure that is the person”). A correct response consists of (a) a suspect ID from a target-present lineup or (b) the rejection of a target-absent lineup, whereas an incorrect response consists of (a) a suspect ID from a target-absent lineup, (b) a filler ID from either type of lineup, or (c) the rejection of a target-present lineup. Because accuracy is coded as a dichotomous variable, the Pearson r in this case is known as a point-biserial correlation coefficient.

In an early review of the literature, Wells and Murray (1984) reported that the average point-biserial correlation between confidence and accuracy in studies of eyewitness identification was only .07 (declaring on that basis that confidence was “functionally useless” in forensic settings), but in a later meta-analysis, Sporer, Penrod, Read, and Cutler (1995) found that the relationship is noticeably stronger—about .41—when the analysis was limited to only those who make an ID from a lineup (i.e., when the analysis was limited to “choosers” who ID a suspect or a filler). Limiting the analysis to choosers is reasonable because only witnesses who choose someone would end up testifying in court against the suspect they identified. Still, even this higher correlation is generally viewed in a negative light. For example, Wilson, Hugenberg, and Bernstein (2013) recently stated that “. . . one surprising lesson that psychologists have learned about memory is that the confidence of an eyewitness is only weakly related to their recognition accuracy (see Sporer et al., 1995, for a review).” Thus, many still view the relationship between eyewitness confidence and accuracy as being of limited utility. In a well-known survey that is often cited in U.S. courts, Kassin, Tubb, Hosch, and Memon (2001) found that 90% of the respondents agreed with the following statement: “An eyewitness’s confidence is not a good predictor of his or her identification accuracy.” In addition, a recent amicus brief filed by the Innocence Project in Michigan states that “A witness’ confidence bears, at best, a weak relationship to accuracy.”

The problem with this conclusion is that it is based on a statistic that does not adequately characterize the relationship between confidence and accuracy (in much the same way that the diagnosticity ratio does not adequately characterize the diagnostic performance of a lineup procedure). Juslin, Olsson, and Winman (1996) definitively showed that, counterintuitively, a low correlation coefficient does not necessarily imply a weak relationship between confidence and accuracy. They argued that a better way to examine the relationship—the way that is more compatible with predictions made by signal-detection theory—would be to simply plot accuracy as a function of confidence.

Signal-Detection Predictions Concerning the Confidence–Accuracy Relationship

A key assumption of signal-detection theory is that a decision criterion is placed somewhere on the memory strength axis, such that an ID is made if the memory strength of a face (target or lure) exceeds it. The correct ID rate is represented by the proportion of the target distribution that falls to the right of the decision criterion, and the false ID rate is represented by the proportion of the lure distribution that falls to the right of the decision criterion. These theoretical considerations apply directly to eyewitness decisions made using a showup (i.e., where a single suspect is presented to the eyewitness), but they also apply to decisions made from a lineup once an appropriate decision rule is specified (Clark, Erickson & Breneman, 2011; Fife, Perry, & Gronlund, 2014; Wixted & Mickes, 2014). One simple lineup decision rule holds that eyewitnesses first determine the lineup member who most closely resembles their memory for the perpetrator and then identify that lineup member if subjective memory strength for that individual exceeds a decision criterion (see Clark et al. 2011 for a discussion of a variety of possible lineup decision rules).

Figure 8 shows how SDT conceptualizes confidence ratings associated with IDs that are made using a three-point scale (1 = low confidence, 2 = medium confidence, and 3 = high confidence). Theoretically, the decision to identify a target or a lure with low confidence is made when memory strength is high enough to support a confidence rating of 1 but is not high enough to support a confidence rating of 2 (i.e., when memory strength falls between the first and second decision criteria). Similarly, a decision to identify a target or a lure with the next highest level of confidence is made when memory strength is sufficient to support a confidence rating of at least 2 (but not 3). A high-confidence rating of 3 is made when memory strength is strong enough to exceed the rightmost criterion.

In this illustrative example, 37% of targets would be associated with memory strengths that exceed the highest confidence criterion (Figure 9, top left panel). By contrast, only about 2% of lures would be associated with memory strengths that exceed the highest confidence criterion (Figure 9, top right panel). If innocent and guilty suspects appeared equally often (i.e., if we assume equal base rates), this would mean that 37 out of 39 high-confidence IDs would be correct. Thus, the proportion correct for high-confidence IDs would be 37 ∕ (37 + 2) = .95.

Next, consider IDs made with medium confidence (a rating of 2 on the 1-to-3 confidence scale). Only 13% of targets in this example would be associated with memory strengths that fall above the criterion required to receive a confidence rating of 2 but below the criterion required to receive a high-confidence rating of 3 (Figure 9, middle left panel), whereas about 4% of lures would be associated with memory strengths that fall in that same range (Figure 9, middle right panel). Thus, the proportion correct for medium-confidence decisions is 13 ∕ (13 + 4) = .76.

Finally, 13% of targets in this example would be associated with memory strengths that fall above the criterion required to receive a low-confidence rating of 1 but below the criterion required to receive a medium-confidence rating of 2 (Figure 9, bottom left panel), whereas about 9% of lures would be associated with memory strengths that fall in that same range (Figure 9, bottom right panel). Thus, the proportion correct for low-confidence decisions drops even further to 13 ∕ (13 + 9) = .59.

Figure 10 is a graph of proportion correct versus confidence for the hypothetical example illustrated above. Obviously, SDT predicts a strong relationship between confidence and accuracy. The details of what the theory predicts will vary from case to case, but so long as the target distribution is shifted to the right of the lure distribution (i.e., so long as a diagnostic memory signal exists) and so long as the confidence criteria are monotonically arranged on the memory strength axis, the theory predicts that confidence and accuracy will be positively related, and this is true even if the correlation coefficient, as typically computed, is low.

Empirical Plots of the Confidence–Accuracy Relationship

A considerable body of research conducted in the time since the Sporer et al. (1995) review appeared has used a calibration approach (in which accuracy is plotted as a function of confidence measured using a 100-point scale), which is closer to how a signal-detection approach suggests that the data should be plotted (Figure 10). This body of research has almost always reported visually obvious evidence of a strong relationship between confidence and accuracy (Brewer & Wells, 2006; Brewer, Keast, & Rishworth, 2002; Palmer, Brewer, Weber, & Nagesh, 2013; Sauer, Brewer, Zweck, & Weber, 2010; Sauerland & Sporer, 2009; Brewer & Palmer, 2010; Weber & Brewer, 2004, 2006). This work explains why many eyewitness ID researchers no longer believe that confidence is, at best, a weak indicator of accuracy. Still, many do, and their views increasingly win the day in the U.S. legal system.

The dependent variable in a calibration analysis is an accuracy score of the general form correct IDs ∕ (correct IDs + incorrect IDs). If there were no incorrect IDs, then this measure would equal 1.0. If there were as many incorrect IDs as correct IDs, then it would equal .50. However, there is more than one way to compute a calibration accuracy score, depending on whether or not filler IDs are counted as incorrect IDs. Should filler IDs and innocent suspect IDs alike be counted as errors, or should only innocent suspect IDs be counted as errors?

Wixted et al. (2015) argued that for the information to be maximally informative to the legal system, filler IDs should not be included in the calculations (just as they are often not included when computing correct and false ID rates for diagnosticity ratio or ROC analyses). The eyewitness ID cases that end up before judges and juries are limited to identified suspects, and the question asked by the court is this: what does confidence tell us about the reliability of the suspect ID? Note that this is a question about the cases that go forward to prosecution using eyewitness identification as direct evidence of the suspect’s guilt, not about the full set of cases involving eyewitness who choose fillers or reject lineups. The accuracy of witnesses who identify suspects are of special interest, so the accuracy score of interest is guilty suspect IDs ∕ (guilty suspect IDs + innocent suspect IDs).

Although calibration studies typically count filler IDs as errors in their accuracy score (thereby lowering the accuracy score from what it would otherwise be), it is worth examining data from a representative study to see what suspect ID accuracy typically looks like as a function of confidence. A representative study by Dobolyi and Dodson (2013) can be used for this purpose. This was a face memory study in which the participating eyewitnesses were later tested using a six-person simultaneous or sequential photo lineup, and confidence ratings were taken using a 0-to-100 scale. The relationship between suspect ID accuracy and confidence—estimated from their data and collapsed across conditions—is shown in Figure 11. The results indicate that even low-confidence suspect IDs are fairly accurate (about 70% correct), though the 30% error rate would obviously be too high to justify a conviction based on a low-confidence ID alone. Remarkably, high-confidence suspect ID accuracy is almost perfect. These results are not atypical, though it is not uncommon to find high-confidence suspect ID accuracy to be closer to .95 than 1.0 (Wixted et al., 2015).

What data like these suggest is that the impression created by an earlier era of research that relied on the point-biserial correlation coefficient was misleading (suggesting, as it did, a weak confidence–accuracy relationship) and that the actual confidence–accuracy relationship is much more in line with what one would expect using signal detection as a guide. Unfortunately, the legal system increasingly accepts the idea that there is a weak correlation between confidence and accuracy, and jurors are increasingly encouraged to ignore expressions of eyewitness confidence (including initial confidence). If the relationship between eyewitness confidence and accuracy is initially strong, then an argument could be made that encouraging juries to disregard confidence places innocent suspects at increased risk of wrongful conviction. In his 2011 book Convicting the Innocent: Where Criminal Prosecutions Go Wrong, Brandon Garrett (2011a) analyzed trial materials for 161 DNA exonerees who had been misidentified by one or more eyewitnesses in a court of law. A key finding was that “. . . in 57% of these trial transcripts (92 of 161 cases), the witnesses reported that they had not been certain at the time of their earlier identifications” (p. 49, emphasis in original). Information about the initial confidence for the remaining 43% of cases was not available.

Figure 11 suggests that an expression of low confidence is how eyewitnesses communicate the fact that the ID they are making carries a high risk of being wrong. In fact, the DNA exoneration cases suggest that this is as true of the real world as it is of the laboratory. An initial ID made with low confidence is a red flag that the risk of eyewitness misidentification is high. Therefore, teaching jurors to ignore confidence is teaching them to ignore a critical clue that the identified suspect may be innocent. No fewer than 57% of DNA exonerees were convicted based on low-confidence initial IDs that later morphed into high-confidence IDs due to the malleability of memory. If it were understood that initial confidence (and only initial confidence) is clearly diagnostic of guilt, then many of these individuals might never have been convicted in the first place. Note that this is even true of what is perhaps the most famous case of eyewitness misidentification, the one that is usually used to illustrate how unreliable eyewitness identification can be. During a trial that was held in 1985, Jennifer Thompson confidently identified Ronald Cotton as the man who had raped her. Cotton was convicted largely on the basis of her testimony, but he was later exonerated by DNA evidence after spending more than 10 years in prison. Long before the trial, however, Thompson’s initial identification of Cotton from a photo lineup was characterized by a prolonged period of hesitation and indecision that lasted for nearly five minutes and that ended with a low-confidence verbal identification consisting of the words “I think this is the guy” (p. 33, Thompson-Cannino, Cotton, & Torneo, 2009; Garrett, 2011b). However, after confirmatory feedback from the police, Thompson quickly became confident that Cotton was the rapist. Her initial lack of confidence spoke volumes, but no one paid attention to it.

The same pattern continues to be seen in DNA exoneration cases. A new DNA exoneration case was in the news as this paper was being prepared for submission. The article says: “DNA testing methods were not as sensitive at the time of the trial and the convictions hinged on positive identifications by the three victims.” On the surface, this appears to be yet another testimony to the already well-established unreliability of eyewitness memory. But now consider something else mentioned in the article: “The judge noted that their initial identifications, however, were tentative and inconsistent in describing their assailant.” Once again, the initial red flag that the IDs were made with low confidence was disregarded. That mistake will likely be repeated with increasing frequency now that courts across the land are taking confidence off the table as a factor juries should use to assess the reliability of an eyewitness ID (e.g., New Jersey Courts, 2012; New Jersey Model Criminal Jury Charges, 2012).

Discussion

The most appropriate way to analyze recognition memory data turns out not to be intuitively obvious. That fact came to be appreciated by experimental psychologists working on problems that have no apparent applied relevance (e.g., list-memory studies with humans, sample/no-sample tasks with pigeons, etc.). Researchers conducting curiosity-driven research on basic issues like these hit upon a critical distinction between response bias and discriminability, and they developed theories to help conceptualize that distinction (e.g., signal-detection theory) and methods to help study it (e.g., ROC analysis). Applied psychologists instead focused their attention on issues of obvious social importance, such as the wrongful conviction of innocent defendants due to eyewitness misidentification, but they did so with limited theoretical guidance. The intuition-based approaches they used to investigate eyewitness misidentification led to the conclusion that sequential lineups are diagnostically superior to simultaneous lineups and to the further conclusion that eyewitness confidence is, at best, weakly related to accuracy. Both conclusions have had a profound impact on the legal system, but both are called into question when the data are conceptualized in terms of signal-detection theory and analyzed using ROC analysis and related methods.

Research on simultaneous and sequential lineups (apparently favoring the sequential procedure) and on the confidence–accuracy relationship in eyewitness identification (apparently indicating that the relationship is inherently weak) began more than 30 years ago and only recently came into contact with signal-detection-based concepts. When that contact was finally made, the conclusions changed rather dramatically. In fact, conclusions about the confidence–accuracy relationship were already changing in the positive direction due largely to the work of Neil Brewer (e.g., Brewer & Palmer, 2010) even though, thus far, that work has had limited impact on the legal system in the United States.

How could it happen that basic and applied psychologists became so insulated from each other? The answer is not clear, but what is clear is that the separation between the two disciplines is an unhealthy state of affairs. My own interpretation is that applied psychologists do not place much value on basic, curiosity-driven research, so they tend to largely ignore basic science. Indeed, the recent push toward “translational” research may be a larger manifestation of the same issue (namely, devaluing basic research in favor of direct application). Recent developments in the domain of eyewitness identification should perhaps be regarded as a case study of what can go wrong when that approach is taken too far. The push toward translational research is an attempt to favor studies with applied relevance over those without obvious applied relevance. The problem with that approach is that it is nearly impossible to tell in advance how important the results of a particular basic science experiment will turn out to be. In a very real way, trying to understand the asymmetrical pattern that pigeons exhibit on a sample/no-sample task (Figure 1) is what led me to realize that the effectiveness of different lineup procedures and the information value of eyewitness confidence were being investigated in ways that could (and, as it turns out, actually did) lead to the wrong answer. That seems like an important lesson in an era that seems hyper-focused on translational research, usually at the expense of the basic research that does not have an obvious applied connection.

Footnote

¹ In Figure 1, performance on sample trials is still above 50% correct at the longest retention interval, but when the retention interval is long enough to yield chance performance on asymmetric memory tasks, accuracy for the more salient sample typically falls well below 50%.

References

Amendola, K. L., & Wixted, J. T. (2015). Comparing the diagnostic accuracy of suspect identifications made by actual eyewitnesses from simultaneous and sequential lineups in a randomized field trial. Journal of Experimental Criminology, 11, 263–284. doi:10.1007/s11292-014-9219-2

Brewer, N., Keast, A., & Rishworth, A. (2002). The confidence-accuracy relationship in eyewitness identification: The effects of reflection and disconfirmation on correlation and calibration. Journal of Experimental Psychology: Applied, 8, 44–56. doi:10.1037/1076-898x.8.1.44

Brewer, N., & Palmer, M. A. (2010). Eyewitness identification tests. Legal and Criminological Psychology, 15, 77–96. doi:10.1348/135532509X414765

Brewer, N., & Wells, G. L. (2006). The confidence-accuracy relation in eyewitness identification: Effects of lineup instructions, foil similarity, and target-absent base rates. Journal of Experimental Psychology: Applied, 12, 11–30. doi:10.1037/1076-898X.12.1.11

Carlson, C. A. & Carlson, M. A. (2014). An Evaluation of Perpetrator Distinctiveness, Weapon Presence, and Lineup Presentation using ROC Analysis. Journal of Applied Research in Memory and Cognition, 3, 45–53. doi:10.1016/j.jarmac.2014.03.004

Clark, S. E. (2003). A memory and decision model for eyewitness identification. Applied Cognitive Psychology, 17, 629–654. doi:10.1002/acp.891

Clark, S. E., Erickson, M. A., & Breneman, J. (2011). Probative value of absolute and relative judgments in eyewitness identification. Law and Human Behavior, 35, 364–380. doi:10.1007/s10979-010-9245-1

Colwill, R. M. (1984). Disruption of short-term memory for reinforcement by ambient illumination. Quarterly Journal of Experimental Psychology, 36B, 235–258. doi:10.1080/14640748408402205

Colwill, R. M., & Dickinson, A. (1980). Short-term retention of “surprising” events by pigeons. Quarterly Journal of Experimental Psychology, 32, 539–556. doi:10.1080/14640748008401844

Davison, M. C., & Tustin, D. (1978). The relation between the generalized matching law and signal-detection theory. Journal of the Experimental Analysis of Behavior, 29, 331–336. doi:10.1901/jeab.1978.29-331

Devenport, J. L., Penrod, S. D., & Cutler, B. L. (1997). Eyewitness identification evidence: Evaluating commonsense evaluations. Psychology, Public Policy, and Law, 3, 338–361. doi:10.1037/1076-8971.3.2-3.338

Dobolyi, D. G., & Dodson, C. S. (2013). Eyewitness confidence in simultaneous and sequential lineups: A criterion shift account for sequential mistaken identification overconfidence. Journal of Experimental Psychology: Applied, 19, 345–357. doi:10.1037/a0034596

Ebbesen, E. B., & Flowe, H. D. (2002). Simultaneous v. sequential lineups: What do we really know? Retrieved December 4, 2015, from http://www2.le.ac.uk/departments/npb/people/hf49/manuscripts/SimSeq%20Submit.pdf

Fetterman, J. G., & MacEwen, D. (1989). Short-term memory for responses: The “choose-small” effect. Journal of the Experimental Analysis of Behavior, 52, 311–324. doi:10.1901/jeab.1989.52-311

Fife, D., Perry, C., & Gronlund, S. D. (2014). Revisiting absolute and relative judgments in the WITNESS model. Psychonomic Bulletin & Review, 21, 479–487. doi:10.3758/s13423-013-0493-1

Garrett, B. (2011a). Convicting the Innocent: Where Criminal Prosecutions Go Wrong. Cambridge, MA: Harvard University Press. doi:10.4159/harvard.9780674060982

Garrett, B. (2011b). Getting it wrong: Convicting the innocent. Slate. Retrieved March 29, 2016, from http://www.slate.com.

Grant, D. S. (1991). Symmetrical and asymmetrical coding of food and no-food samples in delayed matching in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 17, 186–193. doi:10.1037/0097-7403.17.2.186

Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. New York, NY: Wiley.

Gronlund, S. D., Carlson, C. A., Dailey, S. B., & Goodsell, C. A. (2009). Robustness of the sequential lineup advantage. Journal of Experimental Psychology: Applied, 15, 140–152. doi:10.1037/a0015082

Gronlund, S.D., Carlson, C.A., Neuschatz, J.S, Goodsell, C.A., Wetmore, S.A., Wooten, A., & Graham, M. (2012). Showups versus lineups: An evaluation using ROC analysis. Journal of Applied Research in Memory and Cognition, 1, 221–228. doi:10.1016/j.jarmac.2012.09.003

Gronlund, S. D., Mickes, L., Wixted, J. T., & Clark, S. E. (2015). Conducting an eyewitness lineup: How the research got it wrong. In B. H. Ross (Ed.), The psychology of learning and motivation, 63, 1–43. doi:10.1016/bs.plm.2015.03.003

Gronlund, S. D., Wixted, J. T., & Mickes, L. (2014). Evaluating eyewitness identification procedures using ROC analysis. Current Directions in Psychological Science, 23, 3–10. doi:10.1177/0963721413498891

Horry, R., Palmer, M. A., & Brewer, N. (2012). Backloading in the sequential lineup prevents within-lineup criterion shifts that undermine eyewitness identification performance. Journal of Experimental Psychology: Applied, 18, 346–360. doi:10.1037/a0029779

Innocence Project. (2015). Understand the causes: The causes of wrongful conviction. New York, NY: Innocence Project. Retrieved March 16, 2015, from http://www.innocenceproject.org/causes-wrongful-conviction

Juslin, P., Olsson, N., & Winman, A. (1996). Calibration and diagnosticity of confidence in eyewitness identification: Comments on what can be inferred from the low confidence-accuracy correlation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 1304–1316. doi:10.1037/0278-7393.22.5.1304

Kassin, S. M., Tubb, V. A., Hosch, H. M., & Memon, A. (2001). On the “general acceptance” of eyewitness testimony research: A new survey of the experts. American Psychologist, 56, 405–416. doi:10.1037/0003-066X.56.5.405

Lampinen, J. M. (in press). ROC analysis in eyewitness identification research. Journal of Applied Research in Memory and Cognition.

Lindsay, R. C. L., & Wells, G. L. (1985). Improving eyewitness identifications from lineups: Simultaneous versus sequential lineup presentation. Journal of Applied Psychology, 70, 556–564. doi:10.1037/0021-9010.70.3.556

Loftus, E. F. (2005). Planting misinformation in the human mind: A 30-year investigation of the malleability of memory. Learning & Memory, 12, 361–366. doi:10.1101/lm.94705

Loftus, E. F., Miller, D. G., & Burns, H. J. (1978). Semantic integration of verbal information into a visual memory. Journal of Experimental Psychology: Human Learning and Memory, 4, 19–31. doi:10.1037/0278-7393.4.1.19

Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and Verbal Behavior, 13, 585–589. doi:10.1016/S0022-5371(74)80011-3

Loftus, E. F., & Pickrell, J. E. (1995). The formation of false memories. Psychiatric Annals, 25, 720–725. doi:10.3928/0048-5713-19951201-07

Macmillan, N. A., & Creelman, C. D. (2005). Detection theory: A user’s guide (2nd ed.). Mahwah, NJ: Erlbaum.

Meissner, C. A., Tredoux, C. G., Parker, J. F., & MacLin, O. H. (2005). Eyewitness decisions in simultaneous and sequential lineups: A dual-process signal detection theory analysis. Memory & Cognition, 33, 783–792. doi:10.3758/BF03193074

Mickes, L., Flowe, H. D., & Wixted, J. T. (2012). Receiver operating characteristic analysis of eyewitness memory: Comparing the diagnostic accuracy of simultaneous and sequential lineups. Journal of Experimental Psychology: Applied, 18, 361–376. doi:10.1037/a0030609

National Research Council. (2014). Identifying the culprit: Assessing eyewitness identification. Washington, DC: The National Academies Press.

New Jersey Courts. (2012, July 19). Supreme Court Releases Eyewitness Identification Criteria for Criminal Cases. Retrieved March 29, 2016, from http://www.judiciary.state.nj.us/pressrel/2012/pr120719a.htm

New Jersey Model Criminal Jury Charges. (2012). Retrieved March 29, 2016, from http://www.judiciary.state.nj.us/pressrel/2012/jury_instruction.pdf

Palmer, M. A., & Brewer, N. (2012). Sequential lineup presentation promotes less biased criterion setting but does not improve discriminability. Law and Human Behavior, 36, 247–255. doi:10.1037/h0093923

Palmer, M., Brewer, N., Weber, N., & Nagesh, A. (2013). The confidence-accuracy relationship for eyewitness identification decisions: Effects of exposure duration, retention interval, and divided attention. Journal of Experimental Psychology: Applied, 19, 55–71. doi:10.1037/a0031602

Police Executive Research Forum. (2013). A National Survey of Eyewitness Identification Procedures in Law Enforcement Agencies. Retrieved March 29, 2016, from http://www.policeforum.org/

Rotello, C. M., Heit, E., & Dubé, C. (2015). When more data steer us wrong: Replications with the wrong dependent measure perpetuate erroneous conclusions. Psychonomic Bulletin & Review, 22, 944–954. doi:10.3758/s13423-014-0759-2

Sauer, J., Brewer, N., Zweck, T., & Weber, N. (2010). The effect of retention interval on the confidence-accuracy relationship for eyewitness identification. Law and Human Behavior, 34, 337–347. doi:10.1007/s10979-009-9192-x

Sauerland, M., & Sporer, S. L. (2009). Fast and confident: Postdicting eyewitness identification accuracy. Journal of Experimental Psychology: Applied, 15, 46–62. doi:10.1037/a0014560

Sherburne, L. M., & Zentall, T. R. (1993). Coding of feature and no-feature events by pigeons performing a delayed conditional discrimination. Animal Learning & Behavior, 21, 92–100. doi:10.3758/BF03213387

Spetch, M. L., & Wilkie, D. M. (1983). Subjective shortening: A model of pigeons’ memory for event duration. Journal of Experimental Psychology: Animal Behavior Processes, 9, 14–30. doi:10.1037/0097-7403.9.1.14

Sporer, S. L., Penrod, S., Read, D., & Cutler, B. (1995). Choosing, confidence, and accuracy: A meta-analysis of the confidence–accuracy relation in eyewitness identification studies. Psychological Bulletin, 118, 315–327. doi:10.1037/0033-2909.118.3.315

Steblay, N. K., Dysart, J., Fulero, S., & Lindsay, R. C. L. (2001). Eyewitness accuracy rates in sequential and simultaneous lineup presentations: A meta-analytic comparison. Law and Human Behavior, 25, 459–473. doi:10.1023/A:1012888715007

Steblay, N. K., Dysart, J. E., & Wells, G. L. (2011). Seventy-two tests of the sequential lineup superiority effect: A meta-analysis and policy discussion. Psychology, Public Policy, and Law, 17, 99–139. doi:10.1037/a0021650

Steblay, N. K., Dysart, J. E., & Wells, G. L. (2015). An unrepresentative sample is unrepresentative regardless of the reason: A rejoinder to Amendola and Wixted. Journal of Experimental Criminology, 11, 295–298. doi:10.1007/s11292-015-9233-z

Thompson-Cannino, J., Cotton, R., & Torneo, E. (2009). Picking cotton: Our memoir of injustice and redemption. New York, NY: St. Martin’s Press.

Weber, N., & Brewer, N. (2004). Confidence-accuracy calibration in absolute and relative face recognition judgments. Journal of Experimental Psychology: Applied, 10, 156–172. doi:10.1037/1076-898x.10.3.156

Weber, N., & Brewer, N. (2006). Positive versus negative face recognition decisions: Confidence, accuracy, and response latency. Applied Cognitive Psychology, 20, 17–31. doi:10.1002/acp.1166

Wells, G. L. (1984). The psychology of lineup identifications. Journal of Applied Social Psychology, 14, 89–103. doi:10.1111/j.1559-1816.1984.tb02223.x

Wells, G. L., & Bradfield, A. L. (1998). “Good, you identified the suspect”: Feedback to eyewitnesses distorts their reports of the witnessing experience. Journal of Applied Psychology, 83, 360–376. doi:10.1037/0021-9010.83.3.360

Wells, G. L., & Bradfield, A. L. (1999). Distortions in eyewitnesses’ recollections: Can the postidentification-feedback effect be moderated? Psychological Science, 10, 138–144. doi:10.1111/1467-9280.00121

Wells, G. L., & Murray, D. M. (1984). Eyewitness confidence. In G. L. Wells & E. F. Loftus (Eds.), Eyewitness testimony: Psychological perspectives (pp. 155–170). New York, NY: Cambridge University Press.

Wells, G. L., Smalarz, L., and Smith, A. M. (in press). ROC analysis of lineups does not measure underlying discriminability and has limited value. Journal of Applied Research in Memory and Cognition. doi:10.1016/j.jarmac.2015.08.008

Wells, G. L., Smith, A. M., & Smalarz, L. (in press). ROC analysis of lineups obscures information that is critical for both theoretical understanding and applied purposes. Journal of Applied Research in Memory and Cognition. doi:10.1016/j.jarmac.2015.08.010

Wells, G. L., Steblay, N. K., & Dysart, J. E. (2015a). Double-blind photo-lineups using actual eyewitnesses: An experimental test of a sequential versus simultaneous lineup procedure. Law and Human Behavior, 39, 1–14. doi:10.1037/lhb0000096

Wells, G. L., Steblay, N. K., & Dysart, J. E. (2015b). The flaw in Amendola and Wixted’s conclusion on simultaneous versus sequential lineups. Journal of Experimental Criminology, 11, 285–289. doi:10.1007/s11292-014-9225-4

Wilson, B., & Boakes, R. A. (1985). A comparison of the short-term memory performance of pigeons and jackdaws. Animal Learning & Behavior, 13, 285–290. doi:10.3758/BF03200022

Wilson, J. P., Hugenberg, K., & Bernstein, M. J. (2013). The cross-race effect and eyewitness identification: How to improve recognition and reduce decision errors in eyewitness situations. Social Issues and Policy Review, 7, 83–113. doi:10.1111/j.1751-2409.2012.01044.x

Wixted, J. T. (1993). A signal detection analysis of memory for nonoccurrence in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 19, 400–411. doi:10.1037/0097-7403.19.4.400

Wixted, J. T., & Mickes, L. (2012). The field of eyewitness memory should abandon “probative value” and embrace Receiver Operating Characteristic analysis. Perspectives on Psychological Science, 7, 275–278. doi:10.1177/1745691612442906

Wixted, J. T., & Mickes, L. (2014). A signal-detection-based diagnostic feature-detection model of eyewitness identification. Psychological Review, 121, 262–276. doi:10.1037/a0035940

Wixted, J. T., & Mickes, L. (in press a). Evaluating eyewitness identification procedures: ROC analysis and its misconceptions. Journal of Applied Research in Memory & Cognition.

Wixted, J. T., & Mickes, L. (in press b). ROC analysis measures objective discriminability for any eyewitness identification procedure. Journal of Applied Research in Memory & Cognition.

Wixted, J. T., Mickes, L., Clark, S. E., Dunn, J. C., & Wells, W. (in press). The reliability of eyewitness identifications from police lineups. Proceedings of the National Academy of Sciences. doi:10.1073/pnas.1516814112

Wixted, J. T., Mickes, L., Clark, S. E., Gronlund, S. D., & Roediger, H. L. (2015). Initial eyewitness confidence reliably predicts eyewitness identification accuracy. American Psychologist, 70, 515–526. doi:10.1037/a0039510

When Humans and Other Animals Behave Irrationally

Thomas R. Zentall
University of Kentucky

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

The field of comparative cognition has been largely concerned with the degree to which animals have analogs of the cognitive capacities of humans (e.g., imitation, categorization), but recently attention has been directed to behavior that is judged to be biased or suboptimal. We and some of our colleagues have studied several of these and have found that pigeons too show similar paradoxical behaviors. In the present review I will discuss three of these behaviors: sunk cost, justification of effort, and unskilled gambling. Sunk cost is the tendency to decide to spend more on a losing project because of the amount already invested. Pigeons show similar effects even when there is no ambiguity about the results of continuing versus changing alternatives. Justification of effort is the added value one often gives to a reward based on the effort exerted to obtain it. Pigeons too prefer stimuli that signal outcomes that they have had to work harder to obtain. Humans engage in unskilled gambling, like lotteries and slot machines, in which the return is typically less than the investment. And pigeons show a similar tendency to choose a low-probability, high-payoff alternative (gamble) over a more optimal, high-probability, low-payoff alternative. The fact that animals such as pigeons show behavior thought to be unique to humans suggests that the basis for such behaviors is not likely to result from culture or social mechanisms and may have basic behavioral origins.

Keywords: Sunk cost, justification of effort, gambling, contrast, pigeons

Author Note: Thomas R. Zentall, Department of Psychology, University of Kentucky, Lexington, Kentucky 40506-0044. Correspondence concerning this article should be addressed to Thomas R. Zentall at zentall@uky.edu.

I thank Jessica Stagner, Jennifer Laude, Tricia Clement, Rebecca Singer, Karen Roper, Cassandra Gipson, Andria Friedrich, Holly Miller, Kristina Pattison, Jerome Alessandri, Emily Klein, and Aaron Smith for their contribution to the research presented in this review.

---

The field of comparative cognition has devoted considerable attention to studying cognitive processes that have traditionally been viewed as uniquely within the human domain (see e.g., Wasserman & Zentall, 2006; Zentall & Wasserman, 2012). The goal has been to ask if other animals have cognitive abilities similar to those of humans. For example, will they imitate (Neilsen, Subiaul, Galef, Zentall, & Whiten, 2012), can they demonstrate transitive inference (McGonigle & Chalmers, 1977), can they categorize pictures (Wasserman, Brooks, & McMurray, 2015), do they have episodic memory (Zhou, Hohmann, & Crystal, 2012)? In each case putative evidence for such ability exists, and such results suggest either that other animals have similar cognitive abilities or possibly that the basis for these abilities has been attributed to higher cognitive abilities than is necessary. If higher cognitive abilities are not needed, the comparative research has implications for human behavior because it suggests that higher cognitive abilities may not be needed in humans either.

Comparative cognition research has devoted less attention to anomalies and contradictions in human behavior, such as the logically inconsistent behavior studied by Kahneman and Tversky (1979), Kahneman (2013), and Ariely (2010). When humans make decisions that might be called irrational or are inconsistent with optimal choice (choice that would maximize reinforcement), it may be even more important to understand the behavioral mechanisms involved because they have implications for the possibility of bringing them under behavioral control (Staw & Ross, 1978).

In the present paper we explore three examples of bias or suboptimal choice that many would be surprised to find in humans and other animals but which appear to be generally present. The first of these examples is sunk cost, in which the degree of prior investment affects the decision whether to continue “throwing good money after bad.” The second is cognitive dissonance or justification of effort, in which the effort expended in obtaining a reinforcer (or signal for reinforcement) affects the value of the reinforcer or signal. The third is the tendency to risk choosing a low probability of a high payoff over a more optimal high probability of a low payoff, an analog of human gambling behavior. In the first two cases, subjects give value to the effort given prior to the choice, but logically that effort should play no role in the decision. In the last case, subjects fail to give appropriate value to predictable losses. In all three cases, subjects act suboptimally.

The Sunk Cost Fallacy

One example of suboptimal behavior is commonly referred to as the sunk cost fallacy. A sunk cost is a cost that has already been incurred and cannot be recovered but is allowed to affect one’s future behavior. An example might be the following scenario: one goes to a movie and after half an hour decides it is quite dreadful, but one has paid good money to see it so one sits it out to the end. Not only does one not enjoy the rest of the movie, but one could have done something else more enjoyable with the time. The reason often given for this behavior is that leaving the theater would be a waste of the money spent on the ticket and we have been taught not to waste.

Another example of a sunk cost effect was in an experiment reported by Arkes and Blumer (1985). In this scenario one buys a ticket for a weekend ski trip to Michigan for $100. Later one buys a ticket for a weekend ski trip to Wisconsin for $50. One thinks that one will enjoy the ski trip to Wisconsin more than the Michigan ski trip. After purchasing the second ticket, one notices that the two ski trips are for the same weekend. Neither ticket is refundable, and it is too late to sell either. One must use a single ticket and not the other. Which ski trip shall one take? Although the rational decision would be to choose the Wisconsin ski trip, Arkes and Blumer found that only 46% of the subjects said they would choose that one.

At a more consequential level, some people continue to invest in a failing business because they have already invested so much in the business and the invested amount would be wasted if they gave up on it. Similarly, people may argue that they remain in a failing romantic relationship because they have invested too much in the relationship to leave.

Perhaps the most famous example of a response to sunk cost is the case of the development of the well-known Concorde plane. The supersonic aircraft was a joint project of the French and British governments. Long after it became clear that the project would generate little return on the investment, the project was continued because too much had been invested in it to quit (of course economic return may not be the only reason for continuing; see Arkes & Ayton, 1999).

Interestingly, people sometimes count on sunk cost to force themselves to continue behavior that they may find onerous. For example, people may buy a one-year gym membership because if they feel like quitting after a few visits to the gym, the investment already made may convince them to keep going.

Sunk cost also has been studied experimentally. For example, Arkes and Blumer (1985) found, paradoxically, that subjects who purchased season tickets to the theater at full price attended more plays than those who were willing to purchase tickets at full price but were offered the season tickets at half price. Apparently, the loss of the value of a full-price ticket was more aversive than the loss of the value of a half-price ticket, but of course in either case the cost of the ticket was already lost whether they attended or not. When people demonstrate the sunk cost fallacy (sometimes referred to as an escalation of commitment, McAfee, Mialon, & Mialon, 2010), they often increase their future investment in proportion to the amount already invested (Arkes & Blumer, 1985; Khan, Salter, & Sharp, 2000; Staw, 1976, 1981).

Although the sunk cost fallacy is so named because it is thought to result in suboptimal outcomes, some have argued that it may not always be evidence of irrational choices (McAfee et al., 2010). Sometimes, the greater the investment, the more likely the chance of success. In that case, the greater the investment already made, the closer one should be coming to success. Thus, it may be rational to take into account the size of the sunk investment when deciding whether to invest further because the effect of sunk costs on the willingness to make continued investments is ambiguous and whether one continues to invest will hinge on the function that predicts the rate of investment that will lead to success (sometimes referred to as the derivative of the hazard rate). McAfee et al. suggest that whether to invest more in a project may depend not only on the rate of progress made so far but also on whether the hazard function is increasing or decreasing. If progress is improving, added investment may be warranted, whereas if progress is declining, added investment may not be warranted.

Another important factor in deciding whether to continue is consideration of the alternative. Consider the following: A company is in decline, but it decides to embark on a high-risk strategy. Halfway through the new project there is evidence that the new strategy is not working. Should the company persist or go back to its former strategy? One could argue that it might be better to continue with the high-risk strategy because the alternative might be even worse. On the other hand, although the alternative of returning to the previous strategy might be a better investment, the fact that significant resources have been expended on the current strategy might mean there are insufficient funds to return to the earlier one.

McAfee et al. (2010) note another factor that may contribute to consideration of sunk cost: reputation. In the example given earlier about the Concorde, certainly the reputation of the French and British governments played an important role in their decision not to terminate the project. Termination would have had important consequences, not only of a financial nature but also concerning whether these two governments could be counted on to follow through on their commitments. Consider the impact that termination might have had on the decision to form a European trade agreement that eventually led to the European Economic Community and later to the Eurozone.

Finally, because generally there is some uncertainty about the future, giving up may be followed by a sense of regret. Would persistence have led to success? There may be some satisfaction in continuing until there is clear resolution, even if that resolution is failure. As consideration of sunk cost appears to be based on the human characteristic of regret, an overgeneralization of a “don’t waste” rule, and maintaining one’s reputation, it would be instructive to ask whether animals too consider sunk costs in deciding to persist with an initial activity. If one can find evidence for a sunk cost effect in animals, it would suggest that complex factors may reinforce the behavior in humans but may not be responsible for it.

Sunk Cost Research With Animals

Arkes and Ayton (1999) have argued that sunk cost effects are a uniquely human phenomenon because what has been offered as evidence for such effects in other animals can be attributed to simpler explanations. For example, animals engaged in a fixed action pattern may not be sensitive to the fact that they are not successful at arriving at a goal. They will continue to expend resources toward that goal irrespective of continued failure (Dawkins & Brockmann, 1980).

More recent research suggests, however, that sunk cost effects can be found in pigeons (Navarro & Fantino, 2005). In a cleverly designed experiment, Navarro and Fantino trained pigeons to peck at a light to receive rewards. For 50% of the trials, 10 pecks (FR10) were required for reinforcement, for 25% of the trials, 40 pecks were required (FR40) and for 12.5% of the trials, either 80 pecks (FR80) or 160 pecks (FR160) were required. However, at any time the pigeons could peck a second key to start a new trial after a 1-s delay. With these contingencies in mind, the ideal strategy would be to peck the reward key 10 times, and if reinforcement was not forthcoming, to start a new trial. When a change in key light signaled that one of the longer schedules was in effect, the pigeons started a new trial efficiently. However, when no change in key light was provided, three of the four pigeons completed the high FR trials and did not peck the other key to start the next trial (see also Macaskill & Hackenberg, 2012; Magalhães & White, 2014).

Although these experiments suggest that pigeons tend to persist with the current schedule longer than they should, there are several factors that should be considered before concluding that their decision is irrational. First, the decision to start a new trial after 10 pecks would require that the pigeons are able to count to 10, but there is no evidence that pigeons can do this. Instead, the pigeons must determine that 10 or more pecks have already been made. Given that the consequences of making fewer than 10 pecks and starting the next trial prematurely would be much worse than making a few extra pecks, one would expect that pigeons would tend to err in the direction of making too many pecks (i.e., being certain that no fewer than 10 pecks had been made). Although it is difficult to estimate how many pecks that would be, it would certainly be greater than 10. Furthermore, once the pigeon was certain that it had surpassed 10 pecks, it would also be difficult for the pigeon to estimate the number of remaining pecks that would be required to reach 40. Of course, starting a new trial would mean making only 10 more pecks and that should be discriminable from the number of pecks required to reach 40, but three other factors should be taken into account. First, starting a new trial incurs a delay that requires moving away from the reinforcement key to the new-trial key, a key that never provides reinforcement. Second, pecking the new-trial key results in a 1-s blackout that is likely to be more aversive than the 1-s delay of reinforcement. Finally, the delay of reinforcement associated with staying with the reinforcement key or starting a new trial was probabilistic. Starting a new trial would only provide reinforcement sooner 50% of the time. Half of the time, the delay would be even longer.

In the real world in which human sunk cost effects are found, outcomes are likely to be probabilistic, as well. Furthermore, those probabilities may be difficult to calculate. A financial advisor might say that the probability of success of a new business is small but that that is a best guess without taking into consideration several unknowns. In the pigeon experiments, the fixed probability of each of the outcomes makes the contingencies more certain, but it does not eliminate the probabilistic nature of the consequences of the decision to stay with the current uncertain fixed ratio or start a new trial in which the fixed ratio is also uncertain. That uncertainty may bias the animal to stay with the current trial rather than advance to the next trial.

Although the uncertainty of staying versus starting the next trial comes closer to the choices that humans make in the real world, we were interested in whether pigeons would show a sunk cost effect when there was little uncertainly in the consequences of staying versus switching. The question we asked was whether a pigeon would show a sunk cost effect by continuing to complete a known fixed ratio of responding rather than switch to a smaller fixed ratio. That is, would a pigeon consider the current investment (number of pecks already made) in making a decision to either continue to complete the number of pecks required or switch to a different alternative and make fewer required pecks to reinforcement.

In our first experiment (Pattison, Zentall, & Watanabe, 2012), we trained pigeons first to peck a red stimulus 30 times for reinforcement and a green stimulus 15 times for reinforcement. We then trained the pigeons to start pecking the red stimulus on the center key for a number of pecks (5, 10, 15, 20, or 25) and then turned off the center key and turned on one of side keys. The side key was either red, in which case it had to complete the number of pecks to red (25, 20, 15, 10 or 5, respectively) for reinforcement, or green, in which case it had to peck the green key always 15 times, independently of how many pecks it had already made to the red center key. Thus, on some trials they had to complete the pecks to red. On other trials they had to peck green 15 times. After several sessions of training the pigeons with forced trials to experience the consequences of continuing to peck red or switching to green, we gave the pigeons choice trials and found that they had a significant bias to complete the number of pecks to the red key.

To test the hypothesis that the pigeons preferred the variable pecks to the red stimulus over the fixed pecks to the green stimulus, independent of the number of pecks invested, in Experiment 2, we tested the pigeons with the same choice but without an initial investment. Had they learned the rule that a total of 30 pecks would be required to red, trials with no initial investment would have implied that 30 pecks would be required to the red side key, whereas only 15 pecks would be required to the green key. On the other hand, if red was preferred merely because of the variable number of pecks required to the red key following the initial investment, the red key should still be preferred. On these test trials we found that the pigeons had a clear preference for the green key. Thus, it was not just an attraction to the variable number of pecks required to the red key that determined the pigeons’ preference in Experiment 1.

In our third experiment (Pattison et al., 2012), we modified the procedure such that the initial stimulus appeared on one of the side keys. After the initial investment (5 to 25 pecks) the initial stimulus went off and the center key was turned on (white). A single peck to the white center key turned on the two side keys. Returning to the original colored side key required the pigeon to complete remaining number of pecks (25 to 5, respectively), whereas if the pigeon switched to the low fixed ratio color it required only 10 pecks for reinforcement. In this case we found an even stronger sunk cost effect (see Figure 1). In all cases of prior investment, we found a preference to complete the 30 pecks required for reinforcement rather than switch to the fixed ratio 10. We attributed the stronger effect in Experiment 3 to the fact that the initial investment and choice to complete the 30 pecks occurred at the same spatial location and there is evidence that there is considerable generalization decrement when pigeons match the same color on two different keys (Lionello-DeNolf & Urcuioli, 2000).

Figure 1. After training pigeons to peck green 30 times or red 10 times for food, they were trained to peck green a variable number of times and could choose between completing the remaining pecks to green or pecking red 10 times. The data are presented for each of the four pigeons.

Gibbon and Church (1981) found a similar phenomenon using time rather than number of pecks as the initial versus constant requirement. Although that was not the purpose of their experiments, they found a bias of almost 20% to continue with the “time left” alternative when the standard time came available. Thus, when the time left equaled the standard time, pigeons preferred the time left about 70% of the time.

The importance of these experiments is that the tendency to stay with the stimulus of the original investment does not appear to depend on the uncertainty of the requirements for reinforcement. In all of the earlier research and in most human examples of the sunk cost effect, there is the possibility that continuation will result in a better outcome than termination. The closest example of a human sunk cost effect that does not involve probabilistic outcomes is the ski trip example described earlier. That choice involved two initial investments, one of $100 and the other of $50, the sunk cost, and the judgment that the $50 ski trip would be more enjoyable would be the value of the less expensive ski trip. However, more than half of the subjects said that they would go on the more expensive ski trip, a clear example of a sunk cost effect in which the outcomes would not be considered probabilistic.

Once a decision has been made to consider the sunk cost of one’s prior effort, humans may attempt to justify that decision by modifying their prior belief. In the above example, one might reason that although skiing in Wisconsin might be more enjoyable, the Michigan ski trip has certain advantages that one had not considered. Perhaps one has not taken into account that the Michigan ski area is easier to drive to or the amenities at the lodge might be better. Such modification of prior belief is often described as a response to cognitive dissonance, the dissonance that may result from the discrepancy between one’s behavior (going on the Michigan ski trip) and one’s belief (the Wisconsin ski trip would have been more enjoyable). This brings us to the second line of research that asks if suboptimal behavior typically thought of as unique to humans can also be found in other animals.

Cognitive Dissonance

Cognitive dissonance can be defined as the discomfort that results from the occurrence of a discrepancy between one’s beliefs and one’s behavior. To resolve that discrepancy, it is assumed that one must find a way to modify one’s beliefs. Research with humans on cognitive dissonance can be traced back to the classic experiment by Festinger and Carlsmith (1959) in which humans were asked to take part in a boring task involving turning pegs in a peg board for an hour. The subjects were then paid either $1 or $20 to tell a waiting subject that the task was interesting. But before they did, they were asked to evaluate the experiment. Results indicated that the subjects who were paid only $1 rated the dull task as more enjoyable than those who were paid $20. The results are counterintuitive. One might think that the larger reward would be associated with judging that the task was more interesting. The authors explain their findings as follows: Being paid only $1 is not sufficient incentive for lying, so those who were paid $1 experienced dissonance. They could overcome that dissonance by coming to believe that the task was more interesting. Being paid $20, however, provided sufficient reason for lying, so there was no dissonance.

Bem (1967) argued that one does not have to hypothesize that dissonance is involved in the cognitive dissonance effect reported by Festinger and Carlsmith (1959). According to Bem, people look for reasons for their behavior. Those who were paid $20 used that as the reason for being willing to lie. Those who were paid $1 reasoned that it must not be a lie. Whether animals experience dissonance of this kind or look for reasons to explain their behavior is questionable, and if one could find a similar result in other animals it would suggest that a simpler mechanism might be responsible for this curious human behavior.

Although the experimental design used by Festinger and Carlsmith (1959) is clearly too complex to translate into a task that could be used with other animals, Aronson and Mills (1959) examined a version of cognitive dissonance called justification of effort that could more easily be adapted for use with other animals. In their experiment, a group of university students who volunteered to join a discussion group on the topic of the psychology of sex were asked to read a short passage to the experimenter. Subjects in the mild-embarrassment condition were asked to read aloud a list of sex-related words, whereas those in the severe-embarrassment condition were asked to read aloud a list of highly sexual words. Control subjects were not required to read aloud. All subjects then listened to a recording of a discussion about sexual behavior in animals that was reasonably dull and unappealing. When later asked to rate the group and its members, control and mild-embarrassment groups did not differ, but the severe-embarrassment group’s ratings were significantly higher. This group, whose initiation process was more difficult, was presumed to have increased the subjective value of the discussion group, presumably to resolve the dissonance (in this case between wanting to be part of the discussion but finding it embarrassing to read the material aloud). The advantage of such a design is that it focuses on the value of an outcome as influenced by the effort (in this case embarrassment) that preceded it. Presumably, the embarrassment and the desire to join the group created dissonance that was resolved by attributing greater value to the group.

Justification of Effort With Animals

The question we asked was whether an animal would attribute greater value to a reward if greater effort was required to obtain the reward (Clement, Feltus, Kaiser, & Zentall, 2000). Although we could have looked for rewards of different but similar value and manipulated the effort required to obtain each to see if the value of the rewards had changed, instead we held the actual rewards constant and used colored lights as surrogates or conditioned reinforcers to signal the occurrence of the rewards (see design in Figure 2). More specifically, we asked pigeons to peck at a white light. On some trials, a single peck was sufficient to turn off the white light and replace it with a red light. And pecking the red light led to reinforcement. On other trials, 20 pecks were required to turn off the white light and replace it with a green light. And pecking the green light led to reinforcement. Thus, red and green lights were equally followed by the same reinforcement. To be sure that the pigeons were attending to the red and green colors, we presented them in the context of a simultaneous discrimination. Thus, when the red stimulus was presented, it was accompanied by a yellow stimulus and pecks to the red stimulus were reinforced but pecks to the yellow stimulus were not. Similarly, when the green stimulus was presented, it was accompanied by a blue stimulus and pecks to the green stimulus were reinforced but pecks to the blue stimulus were not. Following over 20 sessions of training with this procedure, when we gave the pigeons a choice between the red and green colors, the two colors associated with reinforcement, they showed a significant preference for the color that followed 20 pecks. That is, the color that had required greater effort to obtain was preferred over the color that required less effort to obtain.

A similar result was reported by Kacelnik and Marsh (2002) with starlings that on some trials had to make 16 flights between two perches to obtain one colored light followed by reinforcement and on other trials had to make only four flights between the two perches to obtain a different colored light followed by the same reinforcement. When given a choice between the two colored lights, similar to the results of Clement et al. (2000), the starlings preferred the color that they had had to work harder to obtain.

To interpret this result as evidence of cognitive dissonance or justification of effort would have seemed unparsimonious. Although animals may have beliefs, any discrepancy between those beliefs and their behavior would not likely result in dissonance. Similarly, there would be no need for a pigeon to have to justify the effort that it had to expend to obtain a green light by believing that the food signaled by the green light was better than the food signaled by the red light. Instead, we proposed that contrast between the state of the pigeon immediately preceding the appearance of the colors and the appearance of the colors might account for the value given to the color (Clement et al., 2000). Specifically, we proposed that the pigeon was likely to be in a more negative hedonic state as it came close to meeting the 20 peck requirement than when it met the single peck requirement, and the contrast produced by the appearance of the green light gave the green light its additional value (see Figure 3). After examining various contrast effects that had been studied, we called this effect within-trial contrast.

The advantage of this within-trial contrast account is that it is not specific to the differential peck requirement used by Clement et al. (2000). Because the theory is based on the relative hedonic state of the organism prior to the appearance of reinforcement or the conditioned reinforcer that signals it, the theory predicts that any relatively aversive event that precedes the colored stimuli should produce a similar preference for the stimuli that follows. To test this prediction, we asked whether differential delay would produce a similar effect (DiGian, Friedrich, & Zentall, 2004). In this experiment, on some trials, pecks to a white stimulus resulted immediately in a choice between a red stimulus and a yellow stimulus, and choice of the red stimulus was reinforced. On other trials, pecks to a white stimulus resulted in a delay of 6 s followed by a choice between a green stimulus and a blue stimulus, and choice of the green stimulus was reinforced. On test trials, when given a choice between the red and green stimuli, the pigeons showed only a weak preference for the stimulus that followed the delay. However, in a follow-up experiment we found that if we signaled to the pigeon that the delay was coming (with a vertical line on the white key) or was not coming (with a horizontal line on the white key) the preference for the color that followed the delay was much stronger (see design in Figure 4). That is, the ability of the pigeon to anticipate the delay appeared to magnify the contrast effect.

As a further test of the within-trial contrast hypothesis, we repeated the DiGian et al. (2004) experiments using food and the absence of food as the differential hedonic events (Friedrich, Clement, & Zentall, 2005). We reasoned that in the context of food on some trials the absence of food on other trials would be a relatively aversive event that might increase the attractiveness of the conditioned reinforcer that followed. Thus, one of the colors was preceded by food and the other was preceded by the absence of food (see design in Figure 5). Once again, on test trials when the pigeons were given a choice between the two conditioned reinforcers, the color that was preceded by the absence of food was preferred over the color that was preceded by food. And once again, we found that signaling that food was or was not coming with line-orientation stimuli amplified the magnitude of the effect.

When these results were presented to colleagues, behavioral ecologists who study animals in their natural environment, they were curious to know why animals might develop such preferences. That led us to speculate that if food that was found far from an animal’s home base was more valued than food found close to home, although it might not necessarily encourage the animal to travel greater distances to obtain food, it might somewhat compensate for the greater effort required to obtain that food and encourage the animal to keep looking. To test this hypothesis, one could give an animal a choice between food found close to home and food found far from home, but we chose to use an analog of that procedure in an operant box by making the pigeon work hard (30 pecks) for food at one location (e.g., the left feeder) and not work very hard (one peck) for food at a different location (e.g., the right feeder) and then determine which feeder the pigeon preferred (Friedrich & Zentall, 2004). In this experiment, we assessed each pigeon’s initial feeder preference and monitored the change in feeder preference as a function of training. We found that as training progressed, there was a significant shift toward the feeder that they had to work harder to obtain but not for a control group for which the peck requirement was uncorrelated with feeder location. Furthermore, we found that the preference developed rather slowly (over the course of 72 sessions of training; see Figure 6).

In an interesting variation on the manipulated aversiveness of a prior event, Marsh, Schuck-Paim, and Kacelnik (2004) trained European starlings to peck a lit response key that was one color (e.g., red) on trials when they were prefed (presumably minimally hungry) and another color (e.g., green) on trials when they were not prefed (presumably more hungry). On test trials, when the starlings were given a choice between the red and green response keys (on test trials they were sometimes prefed, sometimes not prefed), they showed a significant preference for the color that in training was associated with the absence of prefeeding, and that preference was unaffected by whether they were prefed or not at the time of testing (see also Pompilio, Kacelnik, & Behmer, 2006; Vasconcelos & Urcuioli, 2008).

Results of Similar Experiments With Humans

The hypothesis that the procedure used with animals is analogous to the justification of effort phenomenon studied in humans would be strengthened if it could be shown that similar results can be found with humans using this procedure. Just such findings were reported in an experiment with children by Alessandri, Darcheville, and Zentall (2008). After training the children on some trials to obtain a simultaneous shape discrimination by clicking a computer mouse on an initial stimulus once or on other trials to obtain a different shape discrimination by clicking the computer mouse 20 times, when the children were given a choice between the two positive shapes, they showed a significant preference for the shape that required 20 clicks to obtain. Similarly, when university students were given a comparable task, they showed a very similar preference (Klein, Bhatt, & Zentall, 2005). Furthermore, when the students were asked about their rationale for choosing the positive shape that followed the 20 clicks, most of them said they just guessed and few of them even noticed the relation between the number of mouse clicks and discrimination that followed. Thus, it appears that the process underlying this contrast effect may be automatic and does not require awareness. Alessandri, Darcheville, Delevoye-Turrell, & Zentall (2008) repeated the study with university students but used differential pressure on a transducer as the prior aversive event and found similar results.

The Delay Reduction Hypothesis

An alternative account of the stimulus preference that follows the more aversive prior event is the delay reduction hypothesis proposed by Fantino and Abarca (1985). According to delay reduction theory, the value of a conditioned reinforcer depends on the degree to which it predicts reinforcement, relative to its absence. Thus, although the delay to reinforcement was the same in the presence of both conditioned reinforcers, according to delay reduction theory, it is the duration of the conditioned reinforcer relative to the total duration of the trial that determines its value. In the case of 20 pecks or a 6-s delay, the conditioned reinforcer that follows would occur relatively closer to reinforcement during the trial than a single peck or no delay; thus it should be a stronger conditioned reinforcer. Similarly, in the case of the absence of reinforcement as the relatively aversive event, the time since the prior reinforcement would have been longer; thus the conditioned reinforcement would occur relatively closer to reinforcement than when reinforcement preceded the conditioned reinforcer. The only result that appears to be inconsistent with the delay reduction account would be the result reported by Alessandri, Darcheville, et al. (2008) in which the prior event was the pressure applied to a transducer by human adults.

To distinguish between the within-trial contrast and delay reduction accounts of conditioned reinforcer preference, Singer and Zentall (2011) held the duration of the prior event constant and manipulated the schedule required to obtain the simultaneous discrimination. To obtain one simultaneous discrimination, the pigeons had to complete a modified fixed interval schedule (FI20; the first peck after 20 s produced the discrimination, modified as noted below). To obtain the other simultaneous discrimination, the pigeons had to complete a differential reinforcement of other behavior schedule (DRO20; the absence of pecking for 20 s produced the discrimination). The two schedules were alternated and were signaled by a vertical or horizontal line orientation on the key, and the duration of the FI schedule was modified to match the duration of the DRO schedule, trial by trial. To test within-trial contrast, it is important to first determine which of the two schedules is less preferred, and although it might seem obvious to some that schedules that do not require pecking would be preferred over those that require pecking, there is reason to believe that once time has been controlled for, pecking plays little role in schedule preference (Fantino & Abarca, 1985).

Interestingly, there were large individual differences in preference between the two schedules. Some pigeons preferred the FI schedule, whereas others preferred the DRO schedule, and most of the pigeons were relatively indifferent between the two schedules. Nevertheless, whatever preference the pigeons showed for the schedule, accurately predicted their preference for the conditioned reinforcer that followed, and the correlation was negative. That is, whichever schedule they less preferred, predicted their preference for the conditioned reinforcer that followed. The results of this experiment provided strong support for the within-trial contrast account of conditioned reinforcer preference found in these studies.

Failures to Replicate

Several experiments have been reported, however, that have failed to replicate the results reported by Clement et al. (2000). Although it is difficult to interpret the results of studies that fail to replicate an effect, the procedures used in these studies may identify limiting conditions under which the phenomenon can be reliably found. Vasconcelos, Urcuioli, and Lionello-DeNolf (2007) conducted several experiments under conditions that closely approximated those of Clement et al. In those experiments, pigeons were trained to criterion on the simultaneous discriminations and were then given 20 sessions of overtraining. As mentioned earlier, when Friedrich and Zentall (2004) showed that pigeons shifted their feeder preference when greater effort was required to obtain reinforcement from it and assessed feeder preference regularly throughout training, they found that 72 sessions of training were required to demonstrate a significant shift in preference relative to controls. Similarly, when Singer and Zentall (2011) manipulated the schedule (FI or DRO) prior to presentation of the simultaneous discriminations, they found that preference for the conditioned reinforcer that followed the less preferred schedule developed slowly and required 60 sessions of training before the preference was statistically significant. Thus, conditioned reinforcer preference may develop gradually with training, and the only way to know how it is developing is to monitor its progress during training.

Two experiments have been conducted, however, in which extended training was provided. The first experiment was by Vasconcelos and Urcuioli (2009) and it provided 60 sessions of overtraining. Although they found the expected preference for the conditioned reinforcer that followed the high peck requirement, the results failed to reach statistical significance (likely because of the lack of statistical power). In the second experiment, Arantes and Grace (2008) also gave extended training and failed to find a preference for the conditioned reinforcer that followed the high fixed ratio schedule. In this case, some of the pigeons previously had been used in an experiment involving relatively lean free-operant concurrent-chains schedules. As for the rest, something in their varied past histories might have led to the different results. It may be that transfer to the fixed ratio 20 schedule was not aversive enough to result in sufficient within-trial contrast to produce a significant conditioned reinforcer preference.

Another reason that a consistent preference for the conditioned reinforcer that follows the more aversive initial link has not always been found may be a function of the nature of choice given on test trials. When the pigeons are given a choice between two conditioned reinforcers, both of which have been associated with continuous reinforcement, it may be that the response strength to both may be strong enough to obscure reliable differences between them. In this regard, it is noteworthy that when pigeons have been tested for their preference between the two stimuli associated with the absence of reinforcement, the stimulus that was preceded by the higher fixed ratio was preferred and that preference was greater than the preference for its normally accompanied conditioned reinforcer (Clement et al., 2000). That is, in the case of the choice between conditioned reinforcers, the pigeons may respond impulsively to the first stimulus they see, whereas in the case of choice between the two stimuli not associated with reinforcement, choice latency is generally longer and thus the choice may be less impulsive.

Research involving the sunk cost effect and the justification of effort involve somewhat different procedures. Sunk cost involves the continuation of a response to a prior investment, whereas justification of effort involves giving added value to the stimulus that follows greater effort. However, it is possible to view justification of effort as resulting from a similar underlying mechanism. In the case of the justification of effort, one can view the greater effort (or prior relatively aversive event) as a greater investment, and for the sunk cost effect one can view the prior investment as giving greater value to the continuation of effort to complete the trial. It would be of interest to pursue the relation between these two effects in future research.

Suboptimal Choice an Analog of Human Gambling

When humans engage in commercial, unskilled gambling, they are responding suboptimally because they are choosing a low-probability, high-payoff alternative over a more optimal high-probability, low-payoff alternative (not gambling), such that the net expected return is less than what was wagered (e.g., slot machines and lotteries). Such choices can be thought of as impulsive in the sense that the gambler’s behavior suggests a failure to consider the long-term consequences of the decision. In fact, research has shown that patterns of decision making in pathological gamblers are marked by a preference for immediate gratification or relief from states of deprivation relevant to their addiction, despite negative long-term consequences (Yechiam, Busemeyer, Stout, & Bechara, 2005).

Recent research suggests that decision making depends on two different sources of input, primary processes governed by relatively simple associative learning that typically occurs impulsively, often without awareness, and secondary processes comprised of what we normally think of as thought processes, the conscious effort to consider possibilities, and an attempt to resolve dilemmas (Dijksterhuis, 2004; Evans, 2003; Kahneman, 2013; Klaczynski, 2005). It is widely held that nonhuman animals are thought to rely on primary decision processes associated with more primitive areas of the brain. Interestingly, pathological gamblers are also thought to arrive at decisions through the use of more primitive areas of the brain (Potenza, 2008). Thus, research with humans suggests that problem gambling, involving games of chance (rather than skill), involves automatic processes that also may apply to other species.

Our research into suboptimal choice began with a simpler question involving stimulus bias. The question we asked was whether pigeons would prefer an alternative that provided them with discriminative stimuli over nondiscriminative stimuli, if the probability of reinforcement was equated (Roper & Zentall, 1999). If they chose the discriminate stimulus alternative, they would receive, for example, a green stimulus 50% of the time, which was always followed by reinforcement, or a red stimulus 50% of the time, which was followed by the absence of reinforcement. If they chose the nondiscriminative stimulus alternative, they would receive, for example, a yellow or blue stimulus, each followed by reinforcement 50% of the time (the design of this experiment appears in Figure 7). Thus, the two alternatives each provided 50% reinforcement but in a sense we were asking whether pigeons would prefer information (the red or green stimulus) over the absence of information (the yellow or blue stimulus), and the answer was clearly that they showed a strong preference for the discriminative stimuli. Similar results have been reported by Fantino (1977) with pigeons, by Prokasy (1956) with rats, and by Bromberg-Martin and Hikosaka (2009) with monkeys.

As part of our experiment (Roper & Zentall, 1999, Experiment 1), we tested an important prediction of information theory: that the degree of preference should depend on the amount of information provided by the discriminative stimuli. Thus, information provided by the discriminative stimulus should be maximal when the probability of reinforcement was 50%. If at the time of choice, the probability of reinforcement was either greater than 50% or less than 50%, the discriminative stimuli would provide less information and the preference for that alternative should be reduced. This should be true because in both cases the discrepancy between what was expected at the time of choice and what occurred following choice (upon the appearance of the discriminative stimuli) would be less as the overall probability of reinforcement deviated from 50%.

In keeping with information theory, when the overall probability of reinforcement was increased to 87.5% (i.e., the green stimulus that predicted reinforcement occurred most of the time) and both the yellow and blue stimuli were associated with 87.5% reinforcement, the pigeons showed less of a preference for the discriminative stimulus alternative. However, when the overall probability of reinforcement was decreased to 12.5% (i.e., the green stimulus that predicted reinforcement occurred very seldom) and both the yellow and blue stimuli were associated with 12.5% reinforcement, preference for the discriminative stimulus alternative actually increased (see Figure 8). Thus, the symmetry predicted by information theory was not found. This finding led us to ask how much the pigeons would be willing to work for the discriminative stimuli if a single peck was required to obtain the nondiscriminative stimuli (Roper & Zentall, 1999, Experiment 2). What we found was that pigeons for which the overall probability of reinforcement was 50% became indifferent between the two alternatives when the ratio of responses (discriminative stimulus vs. nondiscriminative stimulus) was 8:1. When the overall probability of reinforcement was 87.5%, the pigeons were indifferent between the two alternatives when the ratio of responses was about 4.5:1; and when the overall probability of reinforcement was 12.5%, the pigeons still preferred the discriminative stimulus alternative when the ratio of responses was 16:1.

The results of these experiments suggested that the discrepancy between what was expected and the signal for reinforcement (“good news”) was more important than the discrepancy between what was expected and the signal for non-reinforcement (“bad news”). These results were consistent with what had been reported earlier by McDevitt, Spetch & Dunn (1997) who found that inserting a gap between the choice response and the terminal link S– had little effect on choice, whereas inserting a similar gap between choice and the suboptimal terminal link S+ resulted in a strong preference for the optimal alternative. Similarly, Spetch et al. (1994) found that increasing the duration of the suboptimal terminal link S+ resulted in a strong preference for the optimal alternative, whereas increasing the duration of the suboptimal terminal link S– had little effect on choice of the suboptimal alternative.

These results led us to ask whether pigeons would be willing to forgo food to obtain discriminative stimuli. There was already some evidence in the literature that they would, but the results of those experiments was mixed (Belke & Spetch, 1994; Fantino, Dunn, & Meck, 1979; Mazur, 1996; Spetch, Belke, Barnet, Dunn, & Pierce, 1990; Spetch, Mondloch, Belke, & Dunn, 1994). In those experiments, pigeons were given a choice between an alternative that provided a stimulus associated with 100% reinforcement (the optimal choice) and an alternative that 50% of the time provided a stimulus associated with 100% reinforcement and 50% of the time provided a stimulus associated with the absence of reinforcement (the suboptimal choice). In each experiment, several of the pigeons preferred the suboptimal alternative, however, others preferred the optimal alternative.

Originally, we thought that the difference between 50% reinforcement and 100% reinforcement may have been too great to observe consistent preferences, and we asked if we could get a more consistent preference for the suboptimal alternative by decreasing the value of the optimal alternative from 100% to 75% reinforcement (Gipson, Alessandri, Miller, & Zentall, 2009; see design in Figure 9). Although there were still individual differences, as a group the pigeons showed a significant preference for the suboptimal alternative.

The results of the Gipson et al. (2009) study, together with the results of our earlier study in which the overall probability of reinforcement was reduced to 12.5% (Roper & Zentall, 1999), suggested that we might be able to get a more pronounced suboptimal choice effect if we reduced the probability of reinforcement associated with the suboptimal choice alternative below 50%. Stagner and Zentall (2010) tested this hypothesis with a design in which in 20% of the trials, choice of the suboptimal alternative led to a predictor of reinforcement and in 80% of the trials, choice of the suboptimal alternative led to a predictor of non-reinforcement, whereas choice of the optimal alternative led to a predictor of reinforcement 50% of the time. The design is similar to that presented in Figure 9 except the probability of reinforcement associated with the suboptimal alternative was reduced to 20% and the probability of reinforcement associated with the optimal alternative was reduced to 50%. Thus, choice of the optimal alternative provided 2.5 times as much reinforcement as choice of the suboptimal alternative. In spite of this difference in the probability of reinforcement, the pigeons showed a consistently strong (better than 90%) preference for the suboptimal alternative. Furthermore, the suboptimal preference depended entirely on the discriminative stimuli that followed its choice because when both stimuli that followed the suboptimal choice predicted 20% reinforcement, the pigeons showed a strong (almost 90%) preference for the optimal alternative.

Stagner and Zentall (2010) demonstrated that a strong consistent preference for the suboptimal alternative could be found, but it is possible that the preference was at least partially determined by the unpredictability of reinforcement for choices of the optimal alternative. To test this hypothesis, we conducted an experiment in which we manipulated the magnitude of reinforcement rather than the probability of reinforcement (Zentall & Stagner, 2011a). In this experiment in 20% of the trials, choice of the suboptimal alternative led to a stimulus that predicted 10 pellets of food, whereas in 80% of the trials, choice of the suboptimal alternative led to a stimulus that predicted zero pellets. Choice of the optimal alternative always led to a stimulus that predicted three pellets of food (see design in Figure 10). Thus, choice of the optimal alternative resulted in three pellets, whereas choice of the suboptimal alternative led to an average of two pellets. Once again, a strong preference for the suboptimal alternative was found. Monkeys, too, show a similar effect when they prefer discriminative stimuli over nondiscriminative stimuli even when the discriminative stimuli predict less reinforcement on average than the nondiscriminative stimuli (Blanchard, Hayden, & Bromberg-Martin, 2015).

An alternative account of the Zentall and Stagner (2011a) experiment is that the pigeons chose suboptimally because they preferred the variable outcome (10 pellets vs. no pellets) more than the fixed outcome. To test this hypothesis, Zentall and Stagner made the probability of reinforcement associated with the discriminable stimuli the same (i.e., 20% of the time both stimuli associated with the suboptimal alternative were followed by 10 pellets), and now the pigeons consistently chose the optimal alternative. Thus, the variability of reinforcement given choice of the suboptimal alternative was not responsible for the suboptimal choice.

It appears that the predictive value of the conditioned reinforcer is what is responsible for the preference for the suboptimal alternative. In fact, it occurred to us that there may be something special about the certainty of the predictive value of the suboptimal alternative’s conditioned reinforcer. This phenomenon has come to be known as the Allais paradox (Allais, 1953). The Allais paradox can be illustrated by the following example: If humans are given a choice between a 100% chance of being given $3 and an 80% chance of being given $4, most subjects will prefer the suboptimal certainty of the $3. However, if they are given a proportionally similar choice between a 25% chance of being given $3 and a 20% chance of being given $4, most subjects will prefer the more optimal $4 choice. Although in humans this effect can be reduced with practice, it can be maintained when the discrimination is made more difficult (Shafir, Reich, Tsur, Erev, & Lotem, A., 2008).

To determine whether the Allais paradox might apply to pigeons, we asked whether pigeons would cease to prefer the suboptimal alternative if the conditioned reinforcer that followed that choice was followed by reinforcement only 80% of the time that it occurred (Zentall & Stagner, 2011b). Although the preference for the suboptimal alternative was somewhat reduced, the pigeons still showed better than an 80% preference for the suboptimal alternative. Thus, the suboptimal preference does not depend on the certainty of the predictive value of the conditioned reinforcer.

Choice of the suboptimal alternative does not appear to be rational. On the one hand, the lower the probability of the occurrence of the conditioned reinforcer, the stronger the preference is for that alternative (Roper & Zentall, 1999; Zentall & Stagner, 2011a). Although this preference would seem to be increasingly suboptimal, it is consistent with a prediction of the delay reduction hypothesis (Fantino & Abarca, 1985) because the less frequent the occurrence of the conditioned reinforcer is, the better it predicts reinforcement when it does occur, compared to when it is absent. But the less frequent the conditioned reinforcer is, the more frequent is the stimulus associated with the absence of reinforcement (the presumed conditioned inhibitor); and the more frequent the stimulus that predicts the absence of reinforcement is, the less the preference should be for the suboptimal alternative. Apparently, the stimulus that predicts the absence of reinforcement does not function as a conditioned inhibitor (see Spetch et al., 1994).

A more direct test of the conditioned inhibitory value of the stimulus that predicts the absence of reinforcement was provided by Laude, Stagner, and Zentall (2014). They used the design of Zentall and Stagner (2011a; see Figure 10) but substituted a black vertical line on a white background for the red stimulus that was associated with the absence of reinforcement. To test for inhibition, on probe trials they used a combined cue test (superimposing a presumed conditioned inhibitor on a known conditioned reinforcer, see Hearst, Besley, & Farthing, 1970) and tested the pigeons early in training, before a preference for the suboptimal alternative was observed, and later in training, after the pigeons were showing a preference for the suboptimal alternative. Early in training, Laude et al. found strong evidence for conditioned inhibition; however, later in training the evidence for inhibition was substantially reduced. These results support the hypothesis that the negative value of the stimulus that predicts the absence of reinforcement loses its effect with training.

The results of this line of research lead to a surprising conclusion. It is the value of the conditioned reinforcer that determines preference for each of the alternatives rather than their frequency, and the negative value of the stimulus that predicts the absence of reinforcement comes to play little role in that preference. Thus, the pigeons choose the alternative with the conditioned reinforcer that has the greatest value, independent of its frequency. To see how this predicts the results of the suboptimal choice experiments, we need to reconsider their designs. In the Gipson et al. (2009) experiment, the pigeons chose suboptimally because they preferred the signal for 100% reinforcement that follows, over the optimal alternative that provides a signal for 75% reinforcement. In the Zentall and Stagner (2011a) experiment, the pigeons showed a stronger suboptimal choice because they preferred the signal for 100% reinforcement that followed the suboptimal choice, over the signal for 50% reinforcement that followed the optimal choice. In the Zentall and Stagner (2011a) experiment the pigeons choose suboptimally because they preferred the signal for 10 pellets that followed the suboptimal choice, over the signal for three pellets that followed the optimal choice. We call this the conditioned reinforcer value hypothesis.

That choice appears to be governed by the value of the conditioned reinforcers rather than their frequency, and that stimuli associated with the absence of reinforcement play little role in that choice, encouraged us to reconsider the results of earlier research (Belke & Spetch, 1994; Fantino, Dunn, & Meck, 1979; Mazur, 1996; Spetch et al., 1990; Spetch et al., 1994) in which large individual differences in pigeons’ choice of the 100% reinforcement and 50% reinforcement were found. In all of those experiments, both alternatives had conditioned reinforcers that predicted reinforcement 100% of the time. According to the conditioned reinforcer value hypothesis, the pigeons should have been indifferent between the two alternatives, but in general they were not indifferent. Most of the pigeons showed a strong preference for one of the two alternatives. It should be noted, however, that in all of those experiments, the two alternatives were defined by their spatial location. That is, any natural spatial preference would be confounded with a preference for one alternative or the other. One possibility is that the pigeons were in fact indifferent between the two alternatives but they reverted to a spatial preference that appeared as a preference for one alternative or the other. However, that spatial preference tended to be idiosyncratic. Such a possibility was actually suggested by Spetch et al. (1994) and by Mazur (1996).

To test this hypothesis, we repeated the basic design of those experiments giving pigeons a choice between an alternative that resulted in discriminative stimuli, one presented in half of those trials that predicted reinforcement 100% of the time and the other presented in the remaining trials that predicted the absence of reinforcement (50% reinforcement overall) and an alternative that resulted in a stimulus that predicted reinforcement 100% of the time (Smith & Zentall, in press). But instead of defining the alternatives by their spatial location, we used shapes that varied in location from left to right to define the alternatives (see Figure 11). If the pigeons were indifferent between the two alternatives and they defaulted to a spatial preference, it would appear as indifference because the shapes would appear randomly on the left and right. In fact, the pigeons were indifferent between the two alternatives. To verify that the pigeons were able to discriminate between the shapes, we equated the reinforcement associated with the two stimuli that followed choice of the suboptimal alternative (both stimuli were followed by reinforcement 50% of the time) while maintaining the overall difference in reinforcement between choice of the suboptimal (50% reinforcement) and optimal (100% reinforcement) alternatives. Now the pigeons began to choose optimally. The results of this experiment were consistent with the hypothesis that the pigeons judge the value of the conditioned reinforcer that follows choice of that alternative and choose that alternative relatively independent of the frequency of the conditioned reinforcer (see Figure 12). Furthermore, it appears that the presumed conditioned inhibitor does not function to reduce the preference for that alternative.

It should be noted that the conditioned reinforcer value hypothesis may not account for all of the findings that have been reported. According to this hypothesis, with the procedure used by Spetch et al. (1990) and others, the frequency of optimal and suboptimal preferences should be equal, but several experiments have reported that more pigeons preferred the suboptimal alternative (Belke & Spetch, 1994; Dunn & Spetch, 1990). Furthermore, when the spatial locations of the optimal and suboptimal alternatives were reversed, several of the pigeons reversed their preference. Had the preference reflected indifference for the two alternatives, the pigeons would not have reversed their preference. Finally, as already noted, when the appearance of the conditioned reinforcer associated with the suboptimal alternative was delayed by 5 s, it resulted in a strong preference for the optimal alternative, whereas when the conditioned reinforcer associated with the optimal alternative was delayed by 5 s, it had much less effect on initial link choice. Thus, when pigeons perform this task with equal-valued conditioned reinforcers, they may not always be completely indifferent between the two alternatives. When the suboptimal link is chosen, the change in value between the initial link and the terminal link may result in delay reduction (Spetch et al., 1994) or contrast (Stagner & Zentall, 2010), which would not be present when the optimal link is chosen.

The finding that the value of the conditioned reinforcers in the terminal links can account in large part for the initial link preferences should in no way detract from the suboptimality of the initial link choice. Even when the pigeons are indifferent between the 50% and 100% reinforcement initial link alternatives (Smith & Zentall, in press; Stagner, Laude, & Zentall, 2012), they are choosing the suboptimal alternative 50% of the time, whereas they always should be choosing the optimal alternative.

A Good Analog of Human Gambling Behavior?

To what extent are the results of these analog gambling experiments with pigeons consistent with what we know about human gambling behavior? Consistent with the conclusion that with the pigeon procedure there is minimal inhibition associated with stimuli associated with the absence of reinforcement, it has been suggested that humans who gamble overstate their wins and underestimate their losses (Blanco, Ibañez, Sáiz-Ruiz, Blanco-Jerez, & Nunes, 2000), and Ladouceur, Mayrand, and Tourigny (1987) suggest that prolonged exposure to gambling may be associated with cognitive distortions and irrationality about the outcomes of their bets.

This may also explain why, for many gamblers, losing has little effect on the future probability of gambling (until one runs out of money). For example, there is generally a large increase in the number of lottery tickets sold when the value of the winning ticket increases, whereas it is not clear that variability in the probability of winning plays an important role in the number of tickets sold. Although one could argue that the concept of gambling odds is difficult for most of us to fully understand, problem gamblers who should have direct experience with the relation between odds and losing do not appear to be greatly affected by that experience. There is also evidence that problem gamblers show reduced sensitivity to aversive conditioning (Brunborg et al., 2010) and aversive conditioning should serve to inhibit the behavior that produced it.

The Effect of Deprivation on Suboptimal Choice

A paradoxical demographic of human gambling behavior is that people with higher needs (those of lower socioeconomic status) tend to gamble proportionally more than those of higher socioeconomic status (Lyk-Jensen, 2010; Worthington, 2001). If our pigeon model of suboptimal choice is a good analog of human gambling behavior, the level of pigeons’ food motivation should predict their degree of suboptimal choice. Laude, Pattison, and Zentall (2012) tested this hypothesis and found support for the relationship. They found that pigeons that were minimally food restricted (just motivated enough to participate in the experiment) had a strong preference for the optimal alternative, whereas those that were normally food restricted showed the typical suboptimal choice (see Figure 13).

This finding would appear to be inconsistent with optimal foraging theory (Stephens & Krebs, 1986) because animals should have evolved in such a way as to maximize their net energy input, while expending the least amount of energy in doing so. To account for these findings, one might appeal to models of risk-sensitive foraging (Stephens, 1981). Specifically, if an animal is on a negative energy budget such that the rate of energy gain is not sufficiently high for the animal to survive with the smaller but more frequent option (e.g., the reliable three pellet option) the animal’s only option may be for it to be risk prone and gamble on the variable option (10 pellets, 20% of the time). For example, Caraco (1981) found that juncos that were on a negative energy budget were risk prone (preferred variable rewards), whereas those that were on a positive energy budget preferred the constant reward. But as Kacelnik and Bateson (1996) suggested, for pigeons trained under the present high-restriction conditions, in which generally the majority of their daily ration is received in the experiment, the rate of food intake experienced is likely to be sufficient to result in a positive energy budget. Furthermore, food restrictions resulting in a negative energy budget would apply primarily to small birds with a high rate of metabolism that would likely die overnight unless they were risk prone, rather than to larger birds such as pigeons, which can go several days without food.

The mechanism responsible for the difference in suboptimal choice as a function of food restriction is likely to be impulsivity. Impulsivity has been proposed to be associated with human suboptimal choice involved in gambling (Michalczuk, Bowden-Jones, Verdejo-Garcia, & Clark, 2011; Nower & Blaszczynski, 2006). Impulsivity has been defined as the inability to delay reinforcement, and it has been assessed by way of delay discounting tasks in which an organism is given a choice between a small immediate reinforcer and a larger delayed reinforcer. The delay at which the organism is indifferent between the two alternatives defines the slope of the discounting function and the degree of impulsivity. Thus, impulsive individuals require that the delay to the larger reinforcer be relatively short before they will prefer it, and thus for them the slope of the discounting function would be relatively steep. We have recently found that the slope of the delay discounting function for pigeons is a good predictor of the degree to which they prefer the suboptimal choice in the suboptimal choice task (Laude, Beckmann, Daniels, & Zentall, 2014; see Figure 14).

The Effect of Environmental Enrichment on Suboptimal Choice

There is some suggestion from research with rats that various extra-experimental environmental factors such as social and physical enrichment can affect a rat’s propensity to self-administer drugs of addiction (Stairs & Bardo, 2009). Rats that are housed in an enriched group environment (a large cage with other rats and objects that are changed regularly) show a significantly reduced tendency to self-administer drugs than rats that are normally (individually) housed. The mechanism responsible for the reduced self-administration of drugs by environmental enrichment appears to be a reduction in impulsive behavior (Perry & Carroll, 2008) as well as the reduced effectiveness of conditioned reinforcers (Jones, Marsden, & Robbins, 1990). Impulsivity has also been implicated in human gambling behavior (Steel & Blaszczynski, 1998), and, as already noted, conditioned reinforcement has been proposed to account for suboptimal choice by animals (Dinsmoor, 1983). Furthermore, there is evidence that similar physiological mechanisms underlie compulsive gambling and drug addiction (Potenza, 2008).

In an attempt to determine the effect of housing on suboptimal choice, we gave one group of pigeons experience in an enriched environment (a large cage with four other pigeons for 4 hours a day), while the control pigeons remained in their normal one-to-a-cage housing (Pattison, Laude, & Zentall, 2013). When we exposed the pigeons from both groups to the suboptimal choice task, we found that the normally housed pigeons showed the typical suboptimal choice, whereas the enriched pigeons chose optimally for several sessions and then were much slower to begin to choose suboptimally (see Figure 15). Thus, enriched housing appears to have an effect on suboptimal choice, although that effect may be only temporary. These findings may have implications for the treatment of problem gambling behavior by humans. If these results can be generalized, they imply that exposing human gamblers to an environment that is socially and physically enriched may reduce their attraction to gambling.

Analysis of Gambling and Suboptimal Choice

To explain why pigeons prefer discriminative stimuli over nondiscriminative stimuli, Dinsmoor (1983) proposed that animals are attracted to conditioned reinforcers, but the stimuli associated with 0% reinforcement should result in conditioned inhibition, and in the case of 20% reinforcement those non-reinforced trials occur four times as often. We now know that although Dinsmoor ignored conditioned inhibition, he was probably correct to do so because the presumed conditioned inhibitors are relatively ineffective, even when they occur four times as often as the conditioned reinforcer. Furthermore, we now know that the frequency of the conditioned reinforcer is relatively unimportant as well (Stagner et al., 2012). That is, the probability of winning is relatively unimportant. This finding, as well, has implications for human gambling behavior. If the probability of the appearance of the conditioned reinforcer is relatively unimportant, it provides a plausible reason for why humans gamble when the odds of losing are very high (lotteries). Furthermore, those who run casinos and lotteries have found a way to get people to gamble, even though they may very rarely win, by drawing attention to winning by others (bells ringing and lights flashing when there is a slot-machine winner in a casino and an announcement on TV when there is a lottery jackpot winner). By doing this, they make it appear that winning is much more likely than it is (the availability heuristic; Tversky & Kahneman, 1973).

Why the relative unimportance of losing exists in humans and other animals is not clear, but in nature it may have had an evolutionary adaptive value. If food is scarce, there may be many more failures than successes to find food. But developing inhibition to searching, generally would not be adaptive. Thus, in nature, it may be more adaptive to disregard or at least de-emphasize losses.

A second reason that animals may be attracted to low-probability but high-valued outcomes is that their attraction, which generally takes the form of approach behavior, is likely to have an effect on later outcomes. As an animal approaches the edge of a foraging patch, it may occasionally encounter a high-valued outcome, but entering the patch may result in an increase in the probability of those outcomes. Thus, in nature, attraction to a low-probability but high-valued outcome may increase that probability. That would not be true, of course, in commercial gambling, where the probability of reinforcement is independent of prior choices.

Treatment Implications

The present research suggests that one approach to the treatment of problem gambling may be to make wins less salient and, perhaps more important and easier to accomplish, make losses more salient. The present research also suggests that changes in environment conditions may affect gambling. It may be difficult to overcome the greater tendency to gamble by those humans with lower socioeconomic status because, although the real cost of gambling for those people is relatively higher than for those with higher socioeconomic status, the possibility of winning a jackpot would presumably represent a greater improvement in lifestyle for those who are poor. Alternatively, it may be possible to affect the tendency to gamble by making other changes in the environment. Although it is not clear whether problem gamblers spend as much time as they do gambling because they have few outside interests or that problem gambling results in having few outside interests, however, the finding with pigeons that the manipulation of environmental enrichment can reduce the attraction to the suboptimal alternative suggests the possibility that exposing humans to other enriching activities may also serve to reduce their attraction to gambling.

Conclusion

The fact that several examples of bias and suboptimal choice by humans can also be found in pigeons suggests that the mechanisms responsible for those choices by humans do not depend on culture (in the case of sunk cost, “don’t waste”) or complex cognitions (in the case of justification of effort, the need to be consistent in beliefs and actions) or even its entertainment value (in the case of gambling, “it’s fun”). Instead, these behaviors appear to be associated with basic behavioral processes that are likely to have evolved because they have had adaptive value in natural environments.

References

Alessandri, J., Darcheville, J.-C., Delevoye-Turrell, Y., & Zentall, T. R. (2008). Preference for rewards that follow greater effort and greater delay. Learning & Behavior, 36, 352–358. doi:10.3758/LB.36.4.352

Alessandri, J., Darcheville, J.-C., & Zentall, T. R. (2008). Cognitive dissonance in children: Justification of effort or contrast? Psychonomic Bulletin & Review, 15, 673–677. doi:10.3758/PBR.15.3.673

Allais, M. (1953). Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école Américaine, Econometrica, 21, 503–546.

Arantes, J., & Grace, R. C. (2008). Failure to obtain value enhancement by within-trial contrast in simultaneous and successive discriminations. Learning & Behavior, 36, 1–11. doi:10.3758/LB.36.1.1

Ariely, D. (2010). Predictably irrational: The hidden forces that shape our decisions. New York: Harper Perennial.

Arkes, H. R., & Ayton, P. (1999). The sunk cost and Concorde effects: Are humans less rational than lower animals? Psychological Bulletin, 125, 591–600. doi:10.1037/0033-2909.125.5.591

Arkes, H. R., & Blumer, C. (1985). The psychology of sunk costs. Organizational Behavior and Human Decision Processes, 35, 124–140. doi:10.1016/0749-5978(85)90049-4

Aronson, E., & Mills, J. (1959). The effect of severity of initiation on liking for a group. Journal of Abnormal and Social Psychology, 59, 177–181. doi:10.1037/h0047195

Belke, T. W., & Spetch, M. L. (1994). Choice between reliable and unreliable reinforcement alternatives revisited: Preference for unreliable reinforcement. Journal of the Experimental Analysis of Behavior, 62, 353–366. doi:10.1901/jeab.1994.62-353

Bem, D. J. (1967). Self-perception: An alternative interpretation of cognitive dissonance phenomena. Psychological Review, 74, 183–200. doi:10.1037/h0024835

Blanchard, T. C., Hayden, B. Y., & Bromberg-Martin, E. S. (2015). Orbitofrontal cortex uses separate codes for different choice attributes in a curiosity task. Neuron, 85, 602–614.

Blanco, C., Ibañez, A., Sáiz-Ruiz, J., Blanco-Jerez, C., & Nunes, E. (2000). Epidemiology, pathophysiology and treatment of pathological gambling. CNS Drugs, 13, 397–407. doi:10.2165/00023210-200013060-00002

Bromberg-Martin, E. S., & Hikosaka, O. (2009). Midbrain dopamine neurons signal preference for advance information about upcoming rewards. Neuron, 63, 119–126. doi:10.1016/j.neuron.2009.06.009

Brunborg, G. S., Johnsen, B. J., Pallesen, S., Molde, H., Mentzoni, R. A., & Myrseth, H. (2010). The relationship between aversive conditioning and risk-avoidance in gambling. Journal of Gambling Studies, 26, 545–559. doi:10.1007/s10899-010-9178-0

Caraco, T. (1981). Energy budgets, risks and foraging preferences in dark-eyed juncos. Behavioral Ecology and Sociobiology, 8, 213–217. doi:10.1007/BF00299833

Clement, T. S., Feltus, J., Kaiser, D. H., & Zentall, T. R. (2000). ‘Work ethic’ in pigeons: Reward value is directly related to the effort or time required to obtain the reward. Psychonomic Bulletin & Review, 7, 100–106. doi:10.3758/BF03210727

Dawkins, R., & Brockmann, H. J. (1980). Do digger wasps commit the concorde fallacy? Animal Behaviour, 28, 892–896. doi:10.1016/S0003-3472(80)80149-7

DiGian, K. A., Friedrich, A. M., & Zentall, T. R. (2004). Discriminative stimuli that follow a delay have added value for pigeons. Psychonomic Bulletin & Review, 11, 889–895. doi:10.3758/BF03196717

Dijksterhuis, A. (2004). Think different: The merits of unconscious thought in preference development and decision making. Journal of Personality and Social Psychology, 87, 586–598. doi:10.1037/0022-3514.87.5.586

Dinsmoor, J. A. (1983). Observing and conditioned reinforcement. Behavioral and Brain Science, 6, 693–728. doi:10.1017/S0140525X00017969

Dunn, R., & Spetch, M. L. (1990). Choice with uncertain outcomes: Conditioned reinforcement effects. Journal of the Experimental Analysis of Behavior, 53, 201–218. doi:10.1901/jeab.1990.53-201

Evans, J. St. B. T. (2003). In two minds: Dual process accounts of reasoning. Trends in Cognitive Science, 7, 454–459. doi:10.1016/j.tics.2003.08.012

Fantino, E. (1977). Conditioned reinforcement: Choice and information. In W. K. Honig & J. E. R. Staddon (Eds.), Handbook of operant behavior (pp. 313–339). Englewood Cliffs, NJ: Prentice–Hall.

Fantino, E., & Abarca, N. (1985). Choice, optimal foraging, and the delay-reduction hypothesis. Behavioral and Brain Science, 8, 315–330. doi:10.1017/S0140525X00020847

Fantino, E., Dunn, R., & Meck, W. (1979). Percentage reinforcement and choice. Journal of the Experimental Analysis of Behavior, 32 335–340. doi:10.1901/jeab.1979.32-335

Festinger, L., & Carlsmith, J. M. (1959). Cognitive consequences of forced compliance. Journal of Abnormal and Social Psychology. 58, 203–210. doi:10.1037/h0041593

Friedrich, A. M., Clement, T. S., & Zentall, T. R. (2005). Discriminative stimuli that follow the absence of reinforcement are preferred by pigeons over those that follow reinforcement. Learning & Behavior, 33, 337–342. doi:10.3758/BF03192862

Friedrich, A. M., & Zentall, T. R. (2004). Pigeons shift their preference toward locations of food that take more effort to obtain. Behavioural Processes, 67, 405–415. doi:10.1016/j.beproc.2004.07.001

Gibbon, J., & Church, R. M. (1981). Time left: Linear versus logarithmic subjective time. Journal of Experimental Psychology: Animal Behavior Processes, 7, 87–108. doi:10.1037/0097-7403.7.2.87

Gipson, C. D., Alessandri, J. D., Miller, H. C., & Zentall, T. R. (2009). Preference for 50% reinforcement over 75% reinforcement by pigeons. Learning & Behavior, 37, 289–298. doi:10.3758/LB.37.4.289

Hearst, E., Besley, S., & Farthing, G.W. (1970). Inhibition and the stimulus control of operant behavior. Journal of the Experimental Analysis of Behavior, 14, 373–409. doi:10.1901/jeab.1970.14-s373

Jones, G. H., Marsden, C. A., & Robbins, T. W. (1990). Increased sensitivity to amphetamine and reward-related stimuli following social isolation in rats: Possible disruption of dopamine-dependent mechanisms of the nucleus accumbens. Psychopharmacology, 3, 364–372. doi:10.1007/BF02244105

Kacelnik, A. & Bateson, M. (1996). Risky theories—the effects of variance on foraging decisions. American Zoologist, 36, 402–434. doi:10.1093/icb/36.4.402

Kacelnik, A., & Marsh, B. (2002). Cost can increase preference in starlings. Animal Behaviour, 63, 245–250. doi:10.1006/anbe.2001.1900

Kahneman, D. (2013). Thinking, fast and slow. New York: Farrar, Straus & Giroux.

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–91. doi:10.2307/1914185

Khan, B. A., Salter, S. B., & Sharp, D. J. (2000). Throwing good money after bad: A comparison of Asian and developed country managerial decision-making in sunk cost situations. Asian Pacific Development Journal, 7, 105–120.

Klaczynski, P. A. (2005). Metacognition and cognitive variability: A two-process model of decision making and its development. In J. E. Jacobs & P. A. Klaczynski (Eds.), The development of decision making in children and adolescents (pp. 39–76). Mahwah, NJ: Erlbaum.

Klein, E. D., Bhatt, R. S., & Zentall, T. R. (2005). Contrast and the justification of effort. Psychonomic Bulletin & Review, 12, 335–339. doi:10.3758/BF03196381

Ladouceur, R., Mayrand, M., & Tourigny, Y. (1987). Risk-Taking behavior in gamblers and non-gamblers during prolonged exposure. Journal of Gambling Behavior, 3, 115–122. doi:10.1007/BF01043450

Laude, J. R., Pattison, K. F., Zentall, T. R. (2012). Hungry pigeons make suboptimal choices, less hungry pigeons do not. Psychonomic Bulletin & Review, 19, 884–891. doi:10.3758/s13423-012-0282-2

Laude, J. R., Beckmann, J. S., Daniels, C. W., & Zentall, T. R. (2014). Impulsivity affects gambling-like choice by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 40, 2–11. doi:10.1037/xan0000001

Laude, J. R., Stagner, J. P., & Zentall, T. R. (2014). Suboptimal choice by pigeons may result from the diminishing effect of nonreinforcement. Journal of Experimental Psychology: Animal Behavior Processes, 40, 12–21. doi:10.1037/xan0000010

Lionello-DeNolf, K. M., & Urcuioli, P. J. (2000). Transfer of pigeons’ matching to sample to novel sample locations. Journal of the Experimental Analysis of Behavior, 73, 141–161. doi:10.1901/jeab.2000.73-141

Lyk-Jensen, S. V. (2010). New evidence from the grey area: Danish results for at-risk gambling. Journal of Gambling Studies, 26, 455–467. doi:10.1007/s10899-009-9173-5

Macaskill, A., & Hackenberg, T. D. (2012). The sunk cost effect with pigeons: Some determinants of decisions about persistence. Journal of the Experimental Analysis of Behavior, 97, 85–100. doi:10.1901/jeab.2012.97-85

Magalhães, P., & White, K. G. (2014). The effect of a prior investment on choice: The sunk cost effect. Journal of Experimental Psychology: Animal Learning and Cognition, 40, 22–37. doi:10.1037/xan0000007

Marsh, B., Schuck-Paim, C., & Kacelnik, A. (2004). Energetic state during learning affects foraging choices in starlings. Behavioral Ecology, 15, 396–399. doi:10.1093/beheco/arh034

Mazur, J. E. (1996). Choice with certain and uncertain reinforcers in an adjusting delay procedure. Journal of the Experimental Analysis of Behavior, 66, 63–73. doi:10.1901/jeab.1996.66-63

McAfee, R. P., Mialon, H. M., & Mialon, S. H. (2010). Do sunk costs matter? Economic Inquiry, 48, 323–336. doi:10.1111/j.1465-7295.2008.00184.x

McDevitt, M. A., Spetch, M. L., & Dunn, R. (1997). Contiguity and conditioned reinforcement in probabilistic choice. Journal of the Experimental Analysis of Behavior, 68(3), 317–327.
doi:10.1901/jeab.1997.68-317

McGonigle, B. O., & Chalmers, M. (1977). Are monkeys logical? Nature, 267, 694–696. doi:10.1038/267694a0

Michalczuk, R., Bowden-Jones, H., Verdejo-Garcia, A., & Clark, L. (2011). Impulsivity and cognitive distortions in pathological gamblers attending the UK National Problem Gambling Clinic: A preliminary report. Psychological Medicine, 41, 2625–2635. doi:10.1017/S003329171100095X

Navarro, A. D., & Fantino, E. (2005). The sunk cost effect in pigeons and humans. Journal of the Experimental Analysis of Behavior, 83, 1–13. doi:10.1901/jeab.2005.21-04

Neilsen, M., Subiaul, F., Galef, B. G., Zentall, T. R., & Whiten, A. (Eds.). (2012). Social learning in humans and non-human animals: Theoretical and empirical dissections: Special Issue. Journal of Comparative Psychology, 126, 109–205. doi:10.1037/a0027758

Nower, L., & Blaszczynski, A. (2006). Impulsivity and pathological gambling: A descriptive model. International Gambling Studies, 6, 61–75. doi:10.1080/14459790600644192

Pattison, K. F., Zentall, T. R., & Watanabe, S. (2012). Sunk cost: Pigeons (Columba livia) too show bias to complete a task rather than shift to another. Journal of Comparative Psychology, 126, 1–9. doi:10.1037/a0023826

Pattison, K. F., Laude, J. R., & Zentall, T. R. (2013). Social enrichment affects suboptimal, risky, gambling-like choice by pigeons. Animal Cognition, 16, 429–434. doi:10.1007/s10071-012-0583-x

Perry, J. L., & Carroll, M. E. (2008). The role of impulsive behavior in drug abuse. Psychopharmacology, 200, 1–26. doi:10.1007/s00213-008-1173-0

Pompilio, L., Kacelnik, A., & Behmer, S.T. (2006). State-dependent learned valuation drives choice in an invertebrate. Science, 311, 1613–1615. doi:10.1126/science.1123924

Potenza, M. N. (2008). The neurobiology of pathological gambling and drug addiction: An overview and new findings. Philosophical Transactions of the Royal Society: B, 363, 3181–3189. doi:10.1098/rstb.2008.0100

Prokasy, W. F. (1956). The acquisition of observing responses in the absence of differential external reinforcement. Journal of Comparative and Physiological Psychology, 49, 131–134. doi:10.1037/h0046740

Roper, K. L., & Zentall, T. R. (1999). Observing behavior in pigeons: The effect of reinforcement probability and response cost using a symmetrical choice procedure. Learning and Motivation, 30, 201–220. doi:10.1006/lmot.1999.1030

Shafir, S., Reich, T., Tsur, E., Erev, I., & Lotem, A. (2008). Perceptual accuracy and conflicting effects of certainty on risk-taking behavior. Nature, 453, 917–921. doi:10.1038/nature06841

Singer, R. A., & Zentall, T. R. (2011). Preference for the outcome that follows a relative aversive event: Contrast or delay reduction? Learning and Motivation, 42, 255–271. doi:10.1016/j.lmot.2011.06.001

Smith, A. P., & Zentall, T. R. (in press). Suboptimal choice in pigeons: Choice is based on the value of the conditioned reinforcer not its frequency.

Spetch, M. L., Belke, T. W., Barnet, R. C., Dunn, R., & Pierce, W. D. (1990). Suboptimal choice in a percentage-reinforcement procedure: Effects of signal condition and terminal link length. Journal of the Experimental Analysis of Behavior, 53, 219–234. doi:10.1901/jeab.1990.53-219

Spetch, M. L., Mondloch, M. V., Belke, T. W., & Dunn, R. (1994). Determinants of pigeons’ choice between certain and probabilistic outcomes. Animal Learning & Behavior 22, 239–251. doi:10.3758/BF03209832

Stagner, J. P., Laude, J. R., & Zentall, T. R. (2012). Pigeons prefer discriminative stimuli independently of the overall probability of reinforcement and of the number of presentations of the conditioned reinforcer. Journal of Experimental Psychology: Animal Behavior Processes, 38, 446–452. doi:10.1037/a0030321

Stagner, J. P., & Zentall, T. R. (2010). Suboptimal choice behavior by pigeons. Psychonomic Bulletin & Review, 17, 412–416. doi:10.3758/PBR.17.3.412

Stairs, D. J., & Bardo, M. T. (2009). Neurobehavioral effects of environmental enrichment and drug abuse vulnerability. Pharmacology Biochemistry and Behavior, 92, 377–382. doi:10.1016/j.pbb.2009.01.016

Staw, B. M. (1976). Knee-deep in the big muddy: A study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance, 16, 27–44. doi:10.1016/0030-5073(76)90005-2

Staw, B. M. (1981). The escalation of commitment to a course of action. Academy of Management Review, 6, 577–587. doi:10.5465/AMR.1981.4285694

Staw, B. M., & Ross, J. (1978). Commitment to a policy decision: A multi-theoretical perspective. Administrative Science Quarterly, 23, 40–64. doi:10.2307/2392433

Steel, Z., & Blaszczynski, A. (1998). Impulsivity, personality disorders and pathological gambling severity. Addiction, 93, 895–905. doi:10.1046/j.1360-0443.1998.93689511.x

Stephens, D. W, (1981). The logic of risk-sensitive foraging preferences. Animal Behavior, 29, 628–629. doi:10.1016/S0003-3472(81)80128-5

Stephens, D. W. & Krebs, J. R. (1986). Foraging Theory. Princeton University Press: Princeton, New Jersey.

Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 677–695. doi:10.1016/0010-0285(73)90033-9

Vasconcelos, M., & Urcuioli, P. J. (2008). Deprivation level and choice in pigeons: A test of within-trial contrast. Learning & Behavior, 36, 12–18. doi:10.3758/LB.36.1.12

Vasconcelos, M., & Urcuioli, P. J. (2009). Extensive training is insufficient to produce the work-ethic effect in pigeons. Journal of the Experimental Analysis of Behavior, 91, 143–152. doi:10.1901/jeab.2009.91-143

Vasconcelos, M., Urcuioli, P. J., & Lionello-DeNolf, K. M. (2007). When is a failure to replicate not a Type II error? Journal of the Experimental Analysis of Behavior, 87, 405–407. doi:10.1901/jeab.2007.10-07

Wasserman, E., Brooks, D., & McMurray, B. (2015). Pigeons acquire multiple categories in parallel via associative learning: A parallel to human word learning? Cognition, 136, 99–122. doi:10.1016/j.cognition.2014.11.020

Wasserman, E. A., & Zentall, T. R. (Eds.) (2006). Comparative cognition: Experimental explorations of animal intelligence. New York: Oxford University Press.

Worthington, A. C. (2001). Implicit finance in gambling expenditures: Australian evidence on socioeconomic and demographic tax. Public Finance Review, 29, 326–342. doi:10.1177/109114210102900403

Yechiam, E., Busemeyer, J. R, Stout, J. C., & Bechara, A. (2005). Using cognitive models to map relations between neuropsychological disorders and human decision making deficits. Psychological Science, 16, 973–978. doi:10.1111/j.1467-9280.2005.01646.x

Zentall, T. R., & Stagner, J. P. (2011a). Maladaptive choice behavior by pigeons: An animal analog of gambling (sub-optimal human decision making behavior). Proceedings of the Royal Society B: Biological Sciences, 278, 1203–1208. doi:10.1098/rspb.2010.1607

Zentall, T. R., & Stagner, J. P. (2011b). Sub-optimal choice by pigeons: Failure to support the Allais paradox. Learning and Motivation, 42, 245–254. doi:10.1016/j.lmot.2011.03.002

Zentall, T. R., & Wasserman, E. A. (Eds.). (2012). Oxford handbook of comparative cognition. New York: Oxford University Press.

Zhou, W., Hohmann, A. G., & Crystal, J. D. (2012). Rats answer an unexpected question after incidental encoding. Current Biology, 22, 1149–1153. doi:10.1016/j.cub.2012.04.040

A Social History of the Founding of the Conference on Comparative Cognition and the Comparative Cognition Society

Reading Options:

Continue reading below, or:
Read/Download PDF| Add to Endnote

Ronald G. Weisman
Queen’s University

Mark E. Bouton
University of Vermont

Marcia L. Spetch
University of Alberta

Edward A. Wasserman
University of Iowa

---

This memoir is a brief history of the founding of the Conference on Comparative Cognition and the Comparative Cognition Society. The text represents the best recollections of the authors, who together founded the Conference. In the 1980s, Ron Weisman visited Melbourne, Florida, regularly to enjoy the warm weather in March and to visit friends at Florida Tech. Over time, he began thinking about sharing the Melbourne experience with other comparative cognition scientists. He discussed the idea with Mark Bouton, Marcia Spetch, and Ed Wasserman in the late 1980s: it is probably no accident that all four taught on campuses that experience harsh winters. By the early 1990s, the group began planning the meetings in earnest. Together, all four became known as the steering committee—or “steers” for short. The steering committee began meeting as a group and in pairs over the next few years to plan the conference. They decided on a name (the Conference on Comparative Cognition), and Ed Wasserman provided the acronym, CO3. Suzy Bouton did the wonderful logo. The committee discussed the lengths of the talks (5, 10, and 20 minutes). Mark Bouton suggested including very short, 5-minute talks, borrowed from the Winter Conference on the Neurobiology of Learning and Memory (Park City, Utah). At CO3, the 5-minute talks evolved into “spoken posters,” complete in themselves, practiced, polished, and informative. Graduate students were allowed to present these brief talks from the beginning, and faculty members were encouraged to do likewise. However, in practice, faculty give 10-minute talks, or, less commonly, 20-minute talks. Other more pragmatic, but important, decisions dealt with providing snacks and drinks at the meetings and when to schedule sessions.

Most important, the steering committee discussed what the meeting was going to be about. They decided that CO3 was to be about comparative cognition in the broadest sense, with encouragement to report on the standard laboratory species and on more naturally occurring species. By defining cognition broadly, they were able to avoid the squabbles then current between more behaviorally oriented and more cognitively oriented scientists.

CO3 first met March 17–20, 1994, at the Holiday Inn on the Ocean in Melbourne. Slightly fewer than three dozen scientists attended, but CO3 was off to a promising start. The attendees liked the meeting, and more than half said they would attend often if not every year. In those early years, more pelicans attended, and more alcohol was consumed.

Noise issues in the normal meeting rooms prompted the hotel to move us to a swank, oceanfront, two-story, glass-walled penthouse. For a time, the luxurious penthouse was perfect for meetings and great evening parties. But eventually rising room rates and noise from the Holiday Inn’s oceanfront entertainment drove CO3 a few miles north to the Hilton Hotel, which served CO3 well until the conference outgrew the limited, but excellent, accommodations. Conveniently, the Radisson Hotel directly up the beach was larger and a better business partner.

While the group met at the Hilton, in 1997, Suzanne MacDonald began to design and sell shirts commemorating the meetings. The designs always gracefully depicted an animal native to Florida. Suzanne joined the steering committee soon thereafter. That the meetings were held in warm, sunny Florida was always important. Mike Brown, later a member of the steering committee, gave the meeting its informal description: “It’s surf and science,” he said with a smile. Melbourne is a relatively quiet town and over the years Mark Bouton and then Marcia Spetch and Suzanne MacDonald sought out other venues; nevertheless, in direct comparisons, Melbourne always offered greater value than better-known towns. Cost has remained important because the meeting attracts so many graduate students.

In 1997, it became obvious that CO3 was growing into what we had all hoped it would become—a successful annual meeting of like-minded scientists. Just to be clear, CO3 was governed by an unelected steering committee, a sort of benign dictatorship. There was no plan for succession. The rapid growth in the number of presenters made it clear that the same few people could not be expected to continue to carry the responsibility, year after year, for a meeting projected to soon include 200+ scientists. At the request of Ed Wasserman, the steering committee met and discussed the next phase: incorporation as a scientific society, the Comparative Cognition Society. The overwhelming majority of presenters favored incorporation, but a few, including a member of the steering committee, did not. Ron Weisman responded to the majority opinion and carried through the incorporation in Florida, where the fees and responsibilities of running a nonprofit corporation are not onerous. We met as a society for the first time in 1999. By that time, we had assembled a rotating cadre of colleagues who worked hard to make sure that the meeting ran smoothly. After incorporation, we set up elected governance to take the Comparative Cognition Society forward into the new century.

Since incorporation, the Society has thrived. Members of the executive board (an expanded and elected version of the steering committee) have included both original members and new faces: Mike Brown, Bob Cook, Jon Crystal, Jeff Katz, Suzanne MacDonald, Marcia Spetch, Ed Wasserman, Ron Weisman, and Tom Zentall. Bob Cook has published two cyber books under the Comparative Cognition Society’s imprint. Ron Weisman and Bob Cook founded and took the first six-year turn editing the Society’s successful online journal, Comparative Cognition and Behavior Reviews. As first suggested by Ed Wasserman, leaders in the field of comparative cognition have been honored with a CCS Distinguished Researcher award, a Master Lecture, and a special issue of Behavioral Processes.

Over the years, CO3 has included talks on over 100 species by scientists from over a dozen countries. Through the political and economic crises of the early 21st century and the Society’s incorporation, attendance at CO3 has grown and remained strong. We are pleased to report that each year brings both returning and new researchers, together with students ready to become the future of our field.

In Memory of Ronald G. Weisman

(September 14, 1937 – January 27, 2015)

Reading Options:

Continue reading below, or:
Read/Download PDF| Add to Endnote

Christopher B. Sturdy & Marcia L. Spetch
University of Alberta

With his family by his side, Ron Weisman, Professor Emeritus, Departments of Psychology and Biology, Queen’s University, died at home on 27 January 2015. Ron obtained his Ph.D. from Michigan State University in 1964 and was hired as Assistant Professor of Psychology at Queen’s in that same year. Ron was promoted to Associate Professor in 1970, Professor in 1977, cross appointed to the Department of Biology in 1993, and finally promoted to Professor Emeritus in 2000. In sum, Ron was a professor at Queen’s for over 50 years. He is well known for his numerous significant contributions to our understanding of animal learning, cognition, and behaviour. Maybe more important, but not so easily tallied with facts and numbers, are the more qualitative and impactful contributions that Ron made to the research areas in which he was so totally and passionately invested during his long and productive career but that escape the accountant’s ledger. Of these less quantifiable, but absolutely important contributions, one cannot hope to produce a comprehensive report here. And Ron himself would not want such a thing. “Too many words that no one is likely to read or care about” would probably be his quip in response to such an idea. No, the manner in which Ron operated and conducted himself is best described using the words of those who have commented about Ron’s influence in the days since his passing. Strong themes like “force of nature”, “intellectually challenging”, “passionate”, “inspiring” are a constant in Ron’s colleagues’ narratives shared in conversations, social media, and emails. Never one to back down from a challenge, Ron reinvented his research career from the ground up when he realized an opportunity to pursue new more challenging but meaningful problems. This categorical change came when Ron was at a point in his career in which most people would be happy to simply maintain the currently successful status quo until retirement. Not Ron. Instead, and in spite of, or perhaps, because of, the fear of the unknown, Ron forged a second, even more well-known career for himself, combining research in learning, cognition, ethology, and neuroscience in a manner not often done, certainly not with the same effect. While on this new path, Ron continued to make significant contributions to the scientific literature and to the field through the founding of the Comparative Cognition Society, and their flagship online and open access journal, Comparative Cognition & Behavior Reviews. Perhaps Ron’s most enduring legacy will be of the contributions that he made to the mentorship and encouragement of young scientists. Many successful scientists owe their “academic legs” to Ron’s strong and generous support and wisdom. Ron posed challenging questions and championed points of view that were sometimes controversial and always aimed at pushing back the darkness to, as Ron put it, “Explain nature”. Ron always managed to be engaging, encouraging, and able to coax the absolute best out of everyone who was willing to meet his enthusiasm and level of commitment to science. Ron’s enthusiasm, wit, candor, compassion, and his huge smile will be sorely missed by all who had the pleasure of knowing him. What a guy.

Experimental Divergences in the Visual Cognition of Birds and Mammals

Muhammad A. J. Qadri & Robert G. Cook
Department of Psychology, Tufts University

Reading Options:

Continue reading below, or:
Read/Download PDF | Add to Endnote

---

Abstract

The comparative analysis of visual cognition across classes of animals yields important information regarding underlying cognitive and neural mechanisms involved with this foundational aspect of behavior. Birds, and pigeons specifically, have been an important source and model for this comparison, especially in relation to mammals. During these investigations, an extensive number of experiments have found divergent results in how pigeons and humans process visual information. Four areas of these divergences are collected, reviewed, and analyzed. We examine the potential contribution and limitations of experimental, spatial, and attentional factors in the interpretation of these findings and their implications for mechanisms of visual cognition in birds and mammals. Recommendations are made to help advance these comparisons in service of understanding the general principles by which different classes and species generate representations of the visual world.

Keywords: Pigeons; Humans; Visual cognition; Spatial attention; Perceptual grouping; Perceptual completion; Visual illusions

Corresponding author:
Dr. Robert G. Cook Department of Psychology; Tufts University 490 Boston Ave.; Medford, MA 02155, USA; Phone: 617 627-2456, Fax: 617 627-3181; E-mail: Robert.Cook@tufts.edu

This research was supported by National Eye Institute grant EY022655.

---

Visual cognition is critical to the behavior of complex animals. It generates the working internal cognitive representations of the external world that guide action, orientation, and navigation. The extensive study of the human animal has dominated the theoretical and empirical investigations of vision and visual cognition (Palmer, 1999). In comparison, the psychological investigation of visual cognition in other animals has received far less attention. Not surprisingly, the examinations of nonhuman primates have been of most interest precisely because their visual system most closely resembles our own. Despite this focus on primates, there is a long and distinguished record of comparative research with non-primate species that has profoundly enhanced our understanding of vision and its underlying mechanisms (e.g., Hartline & Ratliff, 1957; Hubel & Wiesel, 1962; Lettvin, Maturana, McCulloch, & Pitts, 1959; Reichardt, 1987). An appreciation of the entire spectrum of visually driven cognitive systems and how vision is implemented in different nervous systems is key to a complete and general understanding of the evolution, operations, and functions of vision and its role in cognition and intelligent behavior (Cook, 2001; Cook, Qadri, & Keller, in press; Lazareva, Shimizu, & Wasserman, 2012; Marr, 1982).

One of the most fruitful investigations of these comparative questions has focused on the visual behavior of birds, especially in comparison to mammals (Cook, 2000, 2001; Zeigler & Bischof, 1993). There is no question of the importance of the visual modality for these highly mobile creatures. Beginning with their origins within the lineage of feathered theropod dinosaurs (Alonso, Milner, Ketcham, Cookson, & Rowe, 2004; Corwin, 2010; Lautenschlager, Witmer, Altangerel, & Rayfield, 2013; Sereno, 1999), birds have subsequently and rapidly evolved on a number of fronts, including pulmonary physiology, the development of endothermy, distinctive strategies for reproduction and growth, and their central neuroanatomy (Balanoff, Bever, & Norell, 2014; Xu et al., 2014). Over that time, birds have evolved central and visual systems that are well suited for high-speed flight within the restrictions of muscle-powered transport. While quite large relative to birds’ body size, the avian brain is still small compared to primates’ in absolute size. Given the computational complexity and problems associated with vision, the difficulties of building flexible and accurate optically based machine vision systems, and the considerable and large portions of the primate brain devoted to visual cognition, the small size and visual excellence of the avian brain presents an interesting challenge and scientific opportunity. Given their high visual functionality and small absolute brain size, birds provide an excellent model system for guiding the practical and efficient engineering of small visual prostheses, while simultaneously advancing our general theoretical understanding of visual cognition.

The ancestors of modern-day birds and mammals followed contrasting diurnal and nocturnal evolutionarily pathways during the Mesozoic era, and as a result, these two major classes of vertebrate have evolved to rely more heavily on structurally different portions of their nervous systems to mediate visually guided behavior (Cook et al., in press). Most likely because of their nocturnal origins, mammals have evolved solutions to the challenges of vision that developed into numerous lemnothalamic cortical mechanisms and areas that primarily mediate visual cognition (Felleman & Van Essen, 1991; Homman-Ludiye & Bourne, 2014; Kaas, 2013). In contrast, birds use a collothalamically dominant vision system, mediated by the tectum and related structures, to process visual information. From one perspective, birds may represent the evolutionary zenith of the animals that rely on this ancient primary ascending pathway for vision. The complementary pathway present in both animal classes, however, is still critical to visual function. The collothalamic pathway involving the superior colliculus and pulvinar have well established and important visual and attentional functions in mammals (Müller, Philiastides, & Newsome, 2005; Petersen, Robinson, & Morris, 1987; Robinson, 1972), while the visual Wulst in the avian lemnothalamic pathway may play similar roles in birds (Shimizu & Hodos, 1989). Given these differences in the relative weighting and possible functions of these different pathways for each class, the direct comparison of these two types of vision systems provides theoretically revealing comparative information regarding the implementation and role of general, specific, and alternative routes to representing and understanding the visual world (Marr, 1982).

Pigeons have been the dominant avian model and focus species for this comparison. Years of intensive study have resulted in this bird’s visual, cognitive, and neural systems being the best understood of any avian species (Cook, 2001; Honig & Fetterman, 1992; Spetch & Friedman, 2006a; Zeigler & Bischof, 1993). Because the study of visual cognition in mammals has been dominated by studies of humans, the outcomes of the laboratory studies of pigeons have naturally and frequently been compared with our own visual behavior. More important, the extensive theoretical concepts developed from research on human visual cognition have regularly served as a guide for developing investigations of avian visual cognition. Combined, these forces have produced an extensive number of studies in which these two contrasting vertebrate species have been tested with identical or highly similar visual stimuli.

What is the current status of this scientific comparison of pigeon and human visual cognition? Moreover, what similarities and differences have been established regarding how these different classes of animals solve the challenging problems of visually navigating and acting in an object-filled world? On one front, a number of similarities have been established. For example, humans and pigeons discriminate letters of the English alphabet in highly analogous ways, suggesting that shape processing across these species may share similarities (D. S. Blough, 1982; D. S. Blough & Blough, 1997). Looking more deeply at the mechanisms underlying such findings, the early processes responsible for dimensional grouping appear to share similar organizational principles, with their combination, use, and recognition of color, shape, and relative illumination operating in ways that appear comparable (Cook, 1992a, 1992b, 1993; Cook, Cavoto, Katz, & Cavoto, 1997; Cook, Cavoto, & Cavoto, 1996; Cook & Hagmann, 2012; Cook, Qadri, Kieres, & Commons-Miller, 2012). The investigation of visual search behavior has suggested that the search for targets in noise is governed by the same basic parameters across species (D. S. Blough, 1977, 1990, 1992, 1993; P. M. Blough, 1984, 1989; Cook & Qadri, 2013). Extensive research examining the pictorial discrimination of various objects derived from “geons” has suggested that pigeons and humans share commonalities in their processing of these stimuli as well (Kirkpatrick-Steger, Wasserman, & Biederman, 1996, 1998; Van Hamme, Wasserman, & Biederman, 1992; Wasserman & Biederman, 2012; Wasserman, Kirkpatrick-Steger, Van Hamme, & Biederman, 1993; Young, Peissig, Wasserman, & Biederman, 2001). These different parallels carry the important theoretical implication that the natural structure of the visual world may restrict the classes of computational solutions to a fairly small set of mechanisms, even across quite different biological visual systems. Thus, whether an animal is using a collothalamic- (birds) or lemnothalamic-dominant (mammals) visual system, they may operate using similar computational and processing principles because of the structure of the visual world (J. J. Gibson, 1979; Marr, 1982).

Despite the existence of these numerous experimental parallels, this “representational equivalence” hypothesis is surely not a comprehensive description. One might reasonably question given just the simple disparity in absolute size and internal organization of the brains of birds and mammals how these visual systems could function comparably. All of our additional cortical areas and tissue must allow us some enhanced functionality, such as mental imagery or manipulation. Consistent with this line of thinking, a number of experimental findings create problems for such a “representational equivalence” hypothesis. These findings include discrepancies, anomalies, or divergences in the apparent perceptual behaviors of these two species across many experiments. These divergences have not been just one or two isolated occurrences in a few limited settings, which may be overlooked, brushed aside, or dismissed. To the contrary, many occur in persistent clusters in theoretically relevant areas. The purpose of this article is to collect and evaluate lines of these divergent findings and their implications for theories of comparative visual cognition.

Conceptual Overview and Framework

There are a number of areas in which divergent findings have been reported involving pigeons and humans. Examining such divergences is an important way to evaluate the scope and limitations of representational equivalence and to identify potential functional differences. Precisely identifying and isolating where avian and mammalian visual systems differ and where they share commonalities is crucial to reverse engineering their computational mechanisms and evaluating all of the different possible alternative routes to visual representation.

The quintessential outcome of any one of these studies is that the perceptual responses of the pigeons fail to mimic those of humans (or vice versa depending on your taxonomic affection). For example, a number of psychophysical investigations have found that pigeons have poorer acuity and motion thresholds, lower flicker fusion thresholds, and differences in their processing of color relative to humans (Bischof, Reid, Wylie, & Spetch, 1999; P. M. Blough, 1971; Hendricks, 1966; Wright & Cumming, 1971). Beyond these differences in basic visual sensory function between pigeon and human, another difference is the pigeon’s strong propensity to attend to smaller local features or portions of stimuli rather than grasp the larger global form (Aust & Huber, 2001, 2003; K. K. Cavoto & Cook, 2001; Cerella, 1977, 1986; Emmerton & Renner, 2009; Kelly, Bischof, Wong-Wylie, & Spetch, 2001; Lea, Goto, Osthaus, & Ryan, 2006; Navon, 1977, 1981; Vallortigara, 2006). This same local precedence has also been evidenced to a certain degree outside of the operant chamber, during spatial cognition investigations of landmark use (Kelly, Spetch, & Heth, 1998; Spetch, 1995; Spetch & Edwards, 1988). While pigeons are able in the right circumstances to process global information, processing information at larger spatial scales seems far more difficult for them (Cook, Goto, & Brooks, 2005; Fremouw, Herbranson, & Shimp, 2002; Kelly et al., 2001). Such psychophysical and attentional differences are noteworthy and significant and likely play important roles in resolving some of the findings considered in more detail below.

To make this review manageable, however, we have restricted our considerations to four topics that have generated a larger corpus of divergent findings in the domain of visual cognition. Specifically, we look at the discrimination of different arrangements of line-based shapes, the grouping and integration of dot-based perceptual information, the perceptual completion of spatially separated information, and the perception of geometric visual illusions. These are selected because they each represent persistent areas of experimental divergences that have centered on processes that are fundamental to visual cognition theoretically.

The pivotal issue in each case centers around whether any dissimilarities between avian and primate perception reflect a true qualitative difference in how the two classes of animals visually perceive and process the world or instead reflect experimental artifacts or limitations that do not represent the true, underlying state of affairs. Despite the best intentions of the experimenters to nominally investigate the same question across these species by testing similar or identical stimuli, many different variables and procedural issues could potentially produce a given divergent result.

Some of these complicating issues may be related to the experimental or discriminative procedures involved with testing different species. Differences in visual angle, subject training, previous experience, or experimental instructions are all examples of this type of issue. For instance, humans are often explicitly instructed about what features are relevant during testing. In marked contrast, pigeons always have to discover the relevant visual features on their own based on differential reinforcement. Consequently, the two species may ultimately perceive or attend to different features or aspects of superficially identical displays. If the latter occurs, this naturally limits any implications for our deeper understanding of visual processing. To draw the strongest conclusions, both species must attend to the same features in the experimental displays.

Likewise, discrepancies may stem from other attentional or cognitive differences between these species. As mentioned, pigeons regularly exhibit a bias to attend to spatially local information in preference to more globally available information in visual discriminations. Humans contrastingly appear to be much more global in their allocation of attention. Because of its potential impact, a framework for thinking about how animals might spatially attend to stimuli is worth considering at this juncture.

Figure 1 shows two important facets of spatially controlled attention. The first is the size of the area processed in a single visual scan. This might be best thought of as a visual “aperture,” an adjustable area, presumably circular, over which information is processed without any additional eye fixations. The second important component is the size of the spatial search area that is analyzed or integrated over using a series of successive fixations of this visual aperture around the display. This is also adjustable, and presumably operates over a larger region to integrate information. This search area could be as expansive as the whole operant chamber, or it could be limited to the regions of the display critical for correct discrimination. Both of these spatial components, search aperture and area, likely can vary independently, although a trade-off necessarily occurs between them. When a small visual aperture is employed, for example, greater scanning over a display may be strongly encouraged.

The combination of these two attentional attributes results in four modes by which information can be extracted from a visual display. One means of globally processing information over a spatial area would be to employ a large visual aperture and reduce the need for much successive scanning. This is much like what occurs in parallel search or perceptual grouping, especially in humans, where information is extracted or discriminated rapidly over a large spatial extent. A second way is to employ a smaller or more local visual aperture with numerous successive scans of a display that are then combined in later computations. This would yield a process similar to serial visual search and is a different way that animals could successively integrate information over a spatial extent. We think it is important to distinguish between these alternatives when thinking about their impact on global processing. We use the term global perception to indicate the use of a single, large spatial aperture, while the term sequential integration will be used to indicate the hybrid use of a smaller local aperture with a global scanning and integrating strategy. The third approach would be to use a smaller visual aperture with a spatially restricted scanning strategy, without integrating information from separate scans. This would lead to a more particulate perception of the display. We will use the term restricted local processing to describe this attentional approach, and to distinguish it from the sequential integration strategy that may have a similarly sized aperture, but a more expansive scanning strategy. The logical fourth alternative is one that combines a large visual aperture distributed widely over a large spatial area employing a large number of scans or fixations. This last method has similarities to how we experience and navigate the natural world. Given the restricted spatial scale of the typical operant setting, we think this mode plays a less prominent role in the findings below (but merits considerable more research attention). We think these different processing distinctions are worth keeping in mind when evaluating the results collected here.

Line-Based Shape Processing

Overall, the review is divided into four sections covering each broad topic area, followed by a discussion that integrates the interim conclusions of each section in service of answering the larger theoretical question of how avian and mammalian visual cognition are similar and different. This first section examines divergent findings involved with the processing of shape discriminations by pigeons. Because the motivations, stimuli, and tasks are different from each other, they do not easily form a shared theoretical focus. Thus, the possibility that we are combining different underlying phenomena by grouping them should be held in mind. They do share, perhaps importantly, the common feature of using stimuli comprised of different complex arrangements of line segments.

Stimulus Configuration

One important visual outcome in humans is a set of findings classified as configural superiority effects. Here, humans perform better when the arrangement or context of simple elements create configural or emergent properties that facilitate discrimination. The important theoretical idea captured by such results is that the emergent or holistic features of some stimuli precede or dominate the processing of their component elements (Kimchi & Bloch, 1998; Pomerantz, 2003; Pomerantz & Pristach, 1989).

The classic example involves a simple discrimination of the diagonal tilt of two lines, as shown in Figure 2A. With the addition of a redundant “L” context to the tilted lines, this transforms into a discrimination of a “triangle” versus an “arrow,” facilitating performance in people (Pomerantz, Sager, & Stoever, 1977). Because of its importance to the visual mechanisms of holistic and analytic processing, this same type of visual phenomenon has been examined in pigeons.

D.S. Blough (1984) reported the first results testing configural-like stimuli with pigeons. He found mixed results in the two experiments briefly described in that chapter. Using a simultaneous discrimination and three highly experienced birds familiar with making letter discriminations, he reported the results of a discrimination with pattern-producing configural contexts. These consisted of a contextual L that produced an emergent “triangle” or “arrow” for rightward or leftward diagonal lines, or a “U” or “sideways U” as added to horizontal and vertical lines. Through a series of reacquisitions, the configural patterns were learned more quickly and responded to faster than the context-free line discriminations, suggesting that pigeons experience the same kind of configural superiority effect as humans. Subsequent investigations of this type of effect were not as encouraging, however.

In the same chapter, Blough also reported tests with line stimuli in which the added context resulted in a figure that formed a possible 3D object, as well as equally complex contexts that had no obvious 3D interpretation (see Figure 16.4 in Blough, 1984 for examples). For humans, the configural 3D figures were far easier to discriminate because they formed different global objects. For pigeons, this form of configural discrimination was reported as difficult to learn, regardless of the potential 3D interpretation. This suggests that the different line placements did not produce configural depth relations or objects in pigeons, or if they did, the resulting figures did not aid in the discriminability of the display.

In a more extensive investigation of this general issue with a larger number of pigeons, Donis and Heinemann (1993) also trained their animals to discriminate between rightward and leftward sloped diagonal lines in isolation or with the same addition of an L context (i.e., the classic “triangle” vs. “arrow”). In Donis and Heinemann’s study, however, the pigeons showed reduced accuracy when the discriminated lines were embedded in the configural L context. Unlike humans, the pigeons were more accurate when discriminating the lines in isolation. There are notable methodological differences that could have contributed to this difference from Blough’s brief description. Whereas the pigeons in D.S. Blough (1984) pecked directly at the correct stimulus, the pigeons in Donis and Heinemann had a subsequent spatial choice after pecking the discriminative stimulus. Also, the stimuli in Donis and Heinemann’s investigation were about four times larger than those in D.S. Blough (1984), raising the possibility that limits of processing might be related to the size of the stimuli. This latter possibility would suggest that Blough’s pigeons may have employed a global processing strategy, while Donis and Heinemann’s pigeons could have relied on sequential integration or restricted local processing. Donis and Heinemann’s results are not unique, however.

Kelly and Cook (2003) conducted three experiments with different groups of pigeons, examining the role of contextual information on the discrimination of diagonal lines and a mirror-reversed L discrimination. One group was tested in a target localization task using texture displays made from either repeated lines or configural patterns. The second group of pigeons was tested in an oddity-based same/different task. Similar to Donis and Heinemann (1993), the pigeons exhibited a reduction in target localization or same/different choice accuracy with the pattern-producing stimuli in comparison to the simple discrimination of diagonal lines. This was true across presentations using both small and large sizes of the stimuli, suggesting that visual angle was not particularly important. Furthermore, a second pair of configural stimuli involving a “positional discrimination” was also tested. Here the “featural” stimuli consisted of an L versus a mirror-reversed L, and the redundant context that permitted emergent features was a diagonal line (see Figure 1B in Kelly and Cook, 2003, for examples of these stimuli). This type of stimulus also showed no differences compared to the elemental and configural conditions in the same/different task, although it did reveal a configural superiority effect in the target localization task. This configural superiority effect may have been caused by the high similarity of the textured regions produced by the mirror-reversed elements. Nevertheless, in general, the pigeons in this study were typically better when the discriminative line stimuli were presented in isolation than in a configural organization.

A different configural effect found with humans involves stimuli using component line elements arranged to form a human face. Testing stimuli that produce a face superiority effect in humans, Donis, Chase, and Heinemann (2005) found that their pigeons’ capacity to discriminate the feature of a U (the ”smile”) versus its flipped counterpart ∩ (the “frown”) was impaired when a triplet of dots arranged as “eyes” and a “nose” was placed above it. The pigeons were further impaired when these features were enclosed within a larger ellipse, a condition in which humans see a clear and readily discriminable face. Thus, instead of promoting discrimination with the addition of these configuration-producing elements, for the pigeons, these additional features obscured the critical portion of the discrimination. Given that the pigeons failed to learn to discriminate these complete, configural displays even with extensive training, these authors suggested that the additional context increased the similarity between the configural stimuli for the pigeons instead of accentuating or producing new featural differences as it appears to in humans.

Thus, despite the promising start, the preponderance of the evidence suggests that pigeons tend not to exhibit the same configural superiority effects as observed in humans. As accessed by several experiments using different discrimination approaches, pigeons do not consistently benefit from the addition of contextual or configural information in these stimuli that humans find beneficial. Instead, the more typical result seems to be that the pigeons exhibit a form of configural inferiority effect, where the added elements interfere with discrimination by increasing the similarity of figures. This suggests that the two species are deriving or attending to different features within these stimuli.

Search Asymmetry

Another line of divergent research in this area is related to visual search asymmetries. In multiple investigations with humans, Treisman and various colleagues have found that not all sets of features produce identical modes of visual search (Treisman & Gormican, 1988; Treisman & Souther, 1985). In particular, Treisman and Souther found that some shapes could produce parallel search (like a single Q embedded in Os; see Figure 2B), while a reversal of the same features would result in serial-like search (a single O embedded in Qs). Treisman’s theoretical analysis of these search asymmetries focused on the fact that distinctive visual features were selectively activated for one type of search but not the other, such as detecting the presence of the singular line when a Q was the target in a set of Os.

Because of its theoretically revealing nature, Allan and Blough (1989) examined visual search in pigeons with variations of two types of search asymmetry stimuli previously established with humans. Their search displays included the search between O and Q and between triangles with and without a gap along one edge. Overall, they found no search differences depending on which line shapes were the target or distractors; the presence or absence of a feature in the target generally did not affect their search speed or accuracy for either the added line or gap stimuli. This divergence from humans—the apparent absence of a feature search ­asymmetry—is likely not due to pigeons being unable to exhibit such asymmetries in search tasks. Using search displays made up of groups of smaller squares of differing colors, Pearce and George (2003) found that pigeons did show asymmetries in accuracy when distinctive color features are located in the target rather than the distractors. This suggests that the divergence between humans and pigeons found by Allan and Blough may be specifically tied to the dimensional or featural processing of lines or shapes.

In later work, D. S. Blough (1993) examined the use of cues in stimuli that were square-like and contained an additional line and/or gap. On any given trial, an array of 32 stimuli in four rows was displayed, and the pigeon had to identify which stimulus within the array was unique. In this visual search task, Blough analyzed how reaction time varied according to the specific stimulus-pairs tested. He found that the pigeons appeared to independently attend to the presence of the additional lines in the different stimuli and to either location at the top or bottom of the shapes. The gap in the stimulus appeared not to capture attention, consistent with Allan and Blough’s (1989) earlier results. Importantly, the consistent and measurable within-stimulus effect highlights how even small differences in spatial attention directed toward different local areas of stimuli may be a potentially important concern in the analysis of stimuli.

Vertices and edges

For humans, one important outcome of studying visual cognition is our reliance on information at the junctions or vertices of objects for their recognition. Biederman (1987), for example, has found that the equivalent deletion of the vertices of an object is far more detrimental to its subsequent recognition by humans than the deletion of contour information in the middle of line segments (see Figure 2C). The analysis and prioritization of vertices as a critical feature also plays a classic and prominent role in object recognition algorithms by computers (e.g., Harris & Stephens, 1988; Trajković & Hedley, 1998). One possible reason for this reliance is that junctions contain greater information content to aid in deriving the non-accidental structural relations of an object’s surfaces as compared to edges. Focusing on vertices thus reduces the probability of accidental visual properties causing misperception of objects. It is natural to extend this question to whether pigeons use this feature in the recognition of objects.

Several investigations have suggested that pigeons might differ from humans in this regard. The earliest, most prominent, and most complete example was reported by Rilling, De Marse, and La Claire (1993). They trained pigeons to discriminate different shapes using both 2D (outlined square vs. triangle) and 3D (outlined cube vs. prism) figures. They then tested the pigeons by deleting different portions of the shapes at either the center of the lines or at their junctions. Across both sets of stimuli, the pigeons were more disrupted by the elimination of the line segments midway between the vertices than at the vertices. This outcome contrasts markedly with the typical human finding.

Several unpublished observations have since been consistent with Rilling et al.’s (1993) findings. An unpublished experiment from the Comparative Cognition Laboratory at the University of Iowa (Wasserman, personal communication) further suggested that vertices made less of a contribution to the recognition of geons by pigeons. Here segments were removed from complex line drawings of objects at the vertices and between the vertices. Similar to Rilling et al, the pigeons performed better in the former conditions. This outcome was complicated by additional differences between the conditions, however, as it was difficult to equate the length of the remaining line segments between these different conditions because of their complexity. In our own lab, we have found something similar in a target visual search task using texture stimuli. Cook (1993) reported that linear arrangements of distractors were more interfering than either randomized or spaced distractors in such a task (see examples of these stimuli in Figure 14.1 of Cook, 1993, p. 247). In subsequent unreported experiments, we found that edge-like linear distractors interfered more with target search than distractors made from the same number of elements but forming vertex-like right angles. This outcome hints that the edges of the square target regions were more critical to their identification by the pigeons than the corners.

Using a different approach, Peissig, Young, Wasserman, and Biederman (2005) examined how similarly pigeons process shaded complex objects and line drawings of the same objects. Using a variety of training and transfer designs, it appeared that the pigeons did not use the common edges or edge relations across shaded and line objects to mediate their discrimination, as their transfer was poor between these sets of stimuli. The results suggest instead that the pigeons used different representations of each group of stimuli, perhaps based on the availability of surface characteristics. Peissig et al. suggested that their pigeons may have placed greater importance on surface features than edges in learning these discriminations. Not all studies have found this type of result.

In contrast, a recent experiment by Gibson, Lazareva, Gosselin, Schyns, and Wasserman (2007) suggested instead that line co-termination in complex stimuli is an important factor for pigeons. In their experiment, pigeons were trained to discriminate four shaded geons in a choice task. They used a “bubbles” technique, where differing amounts of information were randomly visible through a set of Gaussian apertures placed over the stimuli. Accuracy was averaged across these varying amounts and locations of visible information to identify which portions of the images were most important to the pigeons’ discrimination. Statistical pixel-based analyses of the resulting classification images indicated that both pigeons and humans used pixels near vertices more than edges or surfaces. Visual inspection of the classification images for the individual pigeons, however, do suggest that line segment, edge, and surface information may have also made important contributions to each bird’s idiosyncratic solution. Nonetheless, this work provides the best evidence yet that the vertices or co-terminations of the objects carry more weight than edges for the pigeons.

Taken all together, these sundry lines of investigation based on various aspects of processing line-based stimuli suggest that pigeons do not always readily reproduce the same visual phenomena observed in humans. Pigeons frequently show configural effects that are different from or possibly opposite those in humans. Pigeons appear not to exhibit the same feature-based search asymmetries as humans with seemingly comparable shape-based stimuli. Finally, pigeons do not always consistently prioritize junctions and co-terminations in shape discriminations like humans. While each of these conclusions involves a different type of discrimination, the one common feature across these investigations is the discrimination of simple lines and their relations. As outlined previously, the key question is whether these findings are truly capturing a difference in processing or represent some kind of experimental by-product or artifact.

One possibility is that any emergent structures from simple lines have a greater meaningfulness to humans. Perhaps these more impoverished stimuli require some abstraction to recognize their relation or correspondence to real objects. Humans may have this capacity, but the pigeon visual system may require more complete and realistic stimuli to properly process elements and their configurations (B. R. Cavoto & Cook, 2006). One reason that Gibson et al.’s (2007) results might differ, for instance, is that their object stimuli were more complete and realistic because of the presence of surface shading information. Furthermore, in humans, these types of stimuli can take advantage of linguistic labels or our greater experience at reading with complex line shapes. Each of these experiential factors could provide an advantage in processing lines and their configurations. This line of reasoning would suggest that examining arrangements that are more naturally salient to pigeons (such as those related to food or mating) could reveal processing more similar to that in humans.

Alternatively, pigeons and humans could have simply focused on different aspects of the stimuli because of experiential, instructional, or other cognitive differences. In humans, global perception of the stimuli allows all of the display’s features to create new unitary forms that are easy to discriminate. This might not be the case for pigeons, where local details of the stimuli might dominate their perception. There are several ways that this difference could have manifested in these experiments. One possibility is the use of experimental conditions that do not promote the use of larger visual apertures or global perception strategies by the pigeons. This stems from the fact that the majority of the pigeon studies varied neither stimulus size nor stimulus location during training. Fixed-size and fixed-position procedures are likely the best conditions for supporting restricted local processing strategies. Additional concerns in the same vein can also be raised regarding the differences in visual angle of the stimuli experienced by both species. Furthermore, humans often receive explicit instructions as to what to attend to, whereas the pigeons do not. Thus, before concluding the theoretical question of whether pigeons visually process line information or employ features in shape processing differently than humans, it would be valuable to consider and alleviate the possibility that the results are artifacts of restricted attentional strategies, stimulus size, or instruction.

If we ignore these concerns for a moment, the pattern of results raises the possibility that humans and pigeons process line-based visual features in different fashions. The machine vision literature is replete with different ways to combine the visual features corresponding to the edges and vertices of objects, as well as other higher-level shape features, to generate representations of objects in space. If there are differences in the way these line-based shape features are processed, despite the apparent similarities in the behavior of pigeons and humans in many settings, it would give rise to questions about how such features coalesce into representations that produce similar actions (Pomerantz, 2003). This larger issue is returned to subsequently in the general discussion.

Dot-Based Perceptual Grouping

Moving beyond the “simpler” line stimuli of the first section, the next area examines more complex stimuli perhaps best placed under the heading of perceptual grouping. Broadly conceived, perceptual grouping involves grouping identical, disconnected local elements into larger, global configurations. Some investigations of grouping using humans and pigeons have shown similar or overlapping patterns of responding. For example, as investigated by texture segregation, the early perceptual grouping of color and shape has generally been found to be similar across the two species (Cook, 1992b, 1992c, 2000; Cook et al., 1997; Cook et al., 1996; Cook, Katz, & Blaisdell, 2012). Based on such evidence, we have suggested that early vision is organized along highly similar lines, perhaps because of the importance of determining the extent and relations of object surfaces. Nevertheless, there are several areas where pigeons have consistently deviated from humans with stimuli that presumably involve similar grouping processes. A place to start is with larger global stimuli built from localized smaller dots.

Glass Patterns

In an important study in this area, Kelly et al. (2001) examined how pigeons and humans process Glass patterns. Glass patterns are theoretically revealing stimuli created by taking randomly placed dots, offsetting them appropriately, and superimposing the transposed result on the original stimulus (Glass, 1969; see Figure 3A). Humans readily perceive the global organization of the resultant Glass patterns from these dotted dipoles. Furthermore, humans detect circular or radial Glass patterns through random noise more easily than either translational or spiral patterns (Kelly et al., 2001; Wilson & Wilkinson, 1998). A similar sensitivity to circular information has been found with gratings in nonhuman primates (Gallant, Braun, & Van Essen, 1993). It has been hypothesized that this particular pattern superiority is caused by specialized concentric form detectors that might be the precursors to cortical face processing (Wilson, Wilkinson, & Asaad, 1997).

Using these types of dotted displays, Kelly et al. (2001) trained pigeons to discriminate vertical, horizontal, circular, and radial Glass patterns from a comparable number of randomly placed dots. Besides being generally more difficult for the pigeons, they found no differences in accuracy across the different global patterns regardless of their organization. When different numbers of the dots were placed randomly, creating noise in the displays, the pigeons continued to exhibit equivalent performance among the display types, unlike humans who showed the typical benefits of circular-like patterns. Kelly et al. suggested that this difference between species was potentially driven by the neurological differences between avian and primate visual systems. Consistent with this analysis, we recently extended these findings to a new species of birds, starlings (Qadri & Cook, 2014). Using Glass patterns that duplicated those tested with pigeons, the starlings’ behavior was highly similar to the pigeons’ with no differences found among the patterns.

Biological Motion

Another important area of visual cognition research involves the study of biological motion (Johansson, 1973). Coordinated moving points or dots that correspond to the motions of different articulated behaviors, such as in point-light displays (PLDs), powerfully invoke the perception of acting agents in humans (see Figure 3B). Humans can recognize a variety of actions and socially relevant features (e.g., age, gender, emotion) from these coordinated motions (Blake & Shiffrar, 2007). As a result, the study of action in humans has historically relied on these form-impoverished displays because they isolate motion-related contributions to action recognition. Because of their dominance in the study of action in humans, attempts have been made to examine action recognition in animals using PLDs with varying degrees of success (Blake, 1993; J. Brown, Kaplan, Rogers, & Vallortigara, 2010; Oram & Perrett, 1994; Parron, Deruelle, & Fagot, 2007; Puce & Perrett, 2003; Tomonaga, 2001). These include several investigations testing birds (Dittrich, Lea, Barrett, & Gurr, 1998; Regolin, Tommasi, & Vallortigara, 2000; Troje & Aust, 2013; Vallortigara, Regolin, & Marconato, 2005).

The earliest attempt to examine action perception by pigeons was conducted by Dittrich et al. (1998). They trained pigeons to discriminate between videos showing pigeons engaged in pecking and non-pecking behaviors using either full-figured, complete videos or PLDs. While pigeons trained on full-figured displays showed some transfer to PLDs, those trained with PLDs failed to show any transfer to full-figured displays. Differences in the background between the videos and PLDs, resulting from the recording settings needed to generate PLD videos, may have been a complicating factor that interfered with transfer of this action discrimination across the conditions.

That the processing of PLDs and full-figured complete displays is not equivalent is further supported by studies of action recognition using digital models (Asen & Cook, 2012; Qadri, Asen, & Cook, 2014; Qadri, Sayde, & Cook, 2014). After training pigeons to discriminate complete, digitally rendered animal models engaged in either walking or running actions (Asen & Cook, 2012), Qadri, Asen, and Cook (2014) found that the discrimination of these action categories did not transfer to PLD models that were built using the same articulated structure, motion, and background as the trained actions. This result persisted across differences in the size of the defining dots and changes in the overall visual angle of the display, both changes designed to promote the perceptual grouping of the separated dots.

Because of the problems associated with getting good transfer back and forth between complete models and PLD representations of the same actions, Troje and Aust (2013) instead trained pigeons to discriminate PLDs from the beginning of their experiment. Eight pigeons were trained to discriminate leftward from rightward walking in pigeon and human PLDs using a choice task. After learning the task, the pigeons were tested using globally and locally inconsistent displays. These stimulus analytic tests suggested six of the eight pigeons were attending primarily to the local motion of the dots to perform the discrimination, most often of the dots corresponding to the feet. Two pigeons were seemingly responding based on the overall global walking direction of the models. Thus, while the majority of pigeons seemed to locally process only a subset of the elements from these displays, a limited number of the pigeons did seem capable of attending and responding to the larger global organization of the motions. While this last investigation is perhaps promising, unlike with humans, the processing of PLDs does not appear to readily generate a global perceptual representation in pigeons of a behaving animal in the same way as full-featured videos of the same behavior.

Other studies have examined the perception of dot-based stimuli in motion using simpler patterns than these complex biological motion studies. For example, Nankoo, Madan, Spetch, and Wylie (2014) presented pigeons and humans with updating randomized dot patterns that created the impression of rotational, radial, or spiral motion. By updating only a subset of the stimuli across frames, they were able to vary the degree of motion coherence. Using a simultaneous choice task, both species discriminated an organized motion display from a randomized display. Humans were much better at the task than the pigeons, needing only 16–20% coherence to discriminate organized motion, while the pigeons needed more than 90% coherence to perform comparably. Furthermore, the species differed in their relative ease of discriminating the different types of motion. Humans were equally good with both rotational and radial motions, while the pigeons were best at discriminating rotational motion. Humans found the spiral motion displays most difficult to detect, while radial motion was poorest for the pigeons. Such results seem to suggest basic differences in the motion perception systems of these species.

The above experiments all suggest that when pigeons are required to group separated dotted elements into a global pattern, they have considerable difficulty doing so, or they process the displays in ways that differ from humans. Putting aside concerns about their naturalness, the pigeons did not perceive Glass patterns in ways that mimic humans. When dot arrays were placed in coordinated motion, as in studies of biological motion, the ready and apparent global perception of “behavior” from these dots is seemingly absent in the pigeons, unless perhaps specifically trained. When combined with their generally greater difficulty at detecting coordinated, dot-based motion, it seems reasonable at the current moment to conclude that neither static nor moving dots readily produce the same type of perceptual representation in pigeons as generated in humans. One possibility is that these stimuli are difficult to discriminate because they resist being perceptually grouped into larger configurations due to the spatial separation between the elements.

Again, it is necessary to raise the possibility that this difference originates in the established attentional bias that pigeons have against globally perceiving stimuli rather than a limitation on perceptual grouping. Sequential integration and restricted local processing strategies both would be specifically challenged by these types of dotted stimuli because their small identical components contain little local information to solve the task. The availability of information at these smaller levels (i.e., dipole spacing or orientation) may prevent the pigeons from seeing the larger patterns in these stimuli. The considerable appeal of the dotted stimuli in this section is that they require some analysis of extended areas for any level of discrimination. Somewhat surprisingly, the size and flexibility of pigeons’ visual or attentional aperture has not been experimentally determined, and its control mechanisms are poorly understood. This is a key oversight and an important area for future investigation. If the above difficulties with dotted stimuli are associated with spatial separation, then the next topic on perceptual grouping is likely directly related.

Perceptual Completion

The real world regularly requires the nervous system to make inferences about incomplete or overlapping information. For instance, when one object occludes a portion of another object behind it, as when an animal is moving behind the trees, humans effortlessly see one continuous and unified object over time. This human capacity toward figural completeness from perceptual fragments contributes importantly to our perception of a coherent visual world (Kellman & Shipley, 1991). Correctly connecting the separated edges and surfaces across such gaps makes this one of vision’s more challenging computational problems (Drori, Cohen-Or, & Yeshurun, 2003; Williams & Hanson, 1994; Williams & Jacobs, 1997; Zhang, Marszałek, Lazebnik, & Schmid, 2007). Because of its critical nature to understanding visual processing and the considerable number of anomalous results found with pigeons, the examination of perceptual completion has been recurrently investigated in this species.

One of the first studies to examine the issue of perceptual completion in pigeons was conducted by Cerella (1980). Using a shape discrimination task, pigeons were trained to peck at an outline of a triangle versus a set of other geometric shapes. After learning, the pigeons were then tested with partial triangles and ones in which increasingly more of the triangle was “covered” by a black “occluder.” Cerella found that as the occluder increasingly covered the S+ triangle, responding decreased. Interestingly, the partial figures supported more responding than the occluded condition. He suggested the possibility that this decreased responding was due to the pigeons not completing the invisible parts of the triangle behind the occluder, although neophobia to this new display element may have also been a factor. Subsequent studies in the same report had pigeons discriminate among Peanuts characters. These found that an occluded figure was responded to at levels similar to complete figures. The results also indicated that local features, rather than the entire figure, controlled the discrimination. As a result, any evidence for “completion” is reduced in that light.

Sekuler, Lee, and Shettleworth (1996) conducted a more complete and demanding test with pigeons using a different shape discrimination to index completion. Pigeons were first trained to respond differentially to full and partial circles (Pac-Man figures—partial circles with a 90° pie-piece removed; see Figure 4A). This training was conducted with a rectangle separated by a short distance from the Pac-Man figure. During the critical test, the partial circles were placed to appear as if they were a complete circle being partially occluded by the rectangle. The pigeons consistently responded as if they only saw an incomplete figure, and not completed inferred circles. The test was then repeated using an elongated ellipse partially occluding a rectangle, suspecting that completing a smaller area might be easier. The same result was found, with the pigeons reporting seeing “incomplete” figures. In some sense, however, the pigeons reported exactly what they had been trained do for the Pac-Man-like figure. It suggests that any good continuity potentially available across the edges of such stimuli was not sufficient to produce the same amodal perception of these figures as in humans.

Fujita and Ushitani (2005) examined the same fundamental question using a visual search procedure. Here, the pigeons were trained to visually search for a square red target with a small notch taken out of it among a set of distractors of complete squares. This training included preparation for the future occluder condition by having a white square in various nonadjacent positions around the target to familiarize the pigeons with its presence. When subsequently tested with new configurations of the elements, such that the notched target was adjacent to the white square, creating the perceptual possibility of appearing to be behind the occluder, the pigeons exhibited no accuracy or reaction time differences. This suggests that they did not see the critical configuration as forming a “complete” figure that would have instead impaired or slowed responding. Tests with humans using the same displays, on the other hand, confirmed that such conditions supported this perceptual inference as exhibited in slowed reaction times. Such failures to find evidence of completion engendered a number of experiments attempting to determine whether some additional factor was preventing the pigeons from evidencing perceptual completion.

One factor that has been examined is whether common motion might enhance the capacity of the pigeons to see complete stimuli. To help the birds better understand the demands of the task, Ushitani, Fujita, and Yamanaka (2001) had the fragmented elements of the display moving together in a synchronized fashion consistent with their potential connectedness. Following matching-to-sample training with moving elements that consisted of one object or two aligned objects in common motion, they then tested the pigeons with the occluded version of the stimuli. In the latter condition, however, the pigeons still reported perceiving two separated stimuli instead of a single completed one. Additional experiments with modifications of the moving stimuli designed to further enhance their potential perceptual unification did not alter this basic result.

Another concern that might be raised about these experiments is the relative naturalness or ecological validity of the stimuli tested in such completion experiments. As a result, several investigations have asked whether the use of more species-appropriate stimuli might produce perceptual completion. Watanabe and Furuya (1997) used a go/no-go task to test pigeons with televised images of full-colored conspecifics. They found that pigeons trained to detect the presence or absence of a pigeon behind an occluder showed no greater transfer to a complete image of the pigeon than a partial one. Their results suggest the birds were not seeing the partial images of the conspecifics used in training as “complete” pigeons. Shimizu (1998) also tested the perception of conspecifics by pigeons. In that study, Shimizu measured the elicited courtship behavior of male pigeons toward live and video-recorded female conspecifics. In one test of these experiments, they occluded either the upper or lower half of one of the videos. Their results suggested that the head, rather than the lower portions of the body contained the critical stimulus for generating courtship behavior, although neither was as effective as the complete stimulus. While not designed to test perceptual completion, these results are consistent with the hypothesis that pigeons do not see the partially occluded conspecifics as identical to complete ones.

Continuing in the same vein, Aust and Huber (2006) tested this issue using higher quality, photograph-based stimuli of pigeons. Pigeons were trained to discriminate between fragmented pictures of conspecifics in a go/no-go task. In these experiments, the occluder used in the training and test resembled a tree trunk (e.g., Figure 4B). After training, seven of the ten pigeons tested with occluded versions of the photos showed evidence suggestive of perceptual completion. Subsequent tests revealed, however, that that this outcome was a byproduct of the pigeons using simple visual features that correlated with complete and fragmented images during acquisition, suggesting that the initially promising results were not products of true completion. After further training, one pigeon was able to learn the discrimination independent of these secondary feature cues, but it responded to the critical occluded test exemplar as if it were an incomplete figure.

Ushitani and Fujita (2005) used a nonsocial approach to examine the potential contribution of ecological validity to perceptual completion. In this study, they trained pigeons to visually search and discriminate between small images of grains and non-grains (e.g., Figure 4C). After learning this task, the pigeons were tested with mixed displays having images of grains occluded by a feather or equivalent truncated or deleted photos of the grains mixed in. Perhaps not surprisingly, they found that the complete, unoccluded grains were selected first from the display. More critically, the images of truncated grains were selected before the occluded ones. If the pigeons had been seeing the occluded grains as completed objects, one might have expected them to be pecked off prior to the truncated ones. To rule out any neophobia of the occluder, they conducted further tests in which they familiarized the pigeons with the occluder before testing, but this made no difference in the order in which the occluded and truncated stimuli were selected. Thus, across these several different experiments, each attempting to increase the ecological validity of the stimuli in different ways, the pigeons revealed no better evidence of perceptual completion than the original demonstrations using more artificial stimuli.

Another approach to this general question is to use a discrimination where completion is not the direct source of responding, but inferred from a different type of outcome. Fujita (2001) investigated a line length estimation task to tap into the hypothesized completion process that occurs when a line meets an edge. For primates, this task indexes such completion by showing that there is a systematic error when a line abruptly ends at an adjacent figure. Humans judge such lines to be slightly longer than veridical measurement. This “illusory continuation” is presumably due to an inferred extrapolation of the line behind the adjacent figure. After training three pigeons to discriminate a range of lines as being either “short” or “long” in a choice task, Fujita found no similar continuation effect in the birds. If anything, they seemed to report the lines as longer when more distant from the adjacent figure.

Given the wide number of different stimuli, the different approaches involved, and the additional factors varied, the above results suggest that pigeons may not experience perceptual completion in the same way as primates. In fact, the pigeons frequently react quite literally, faithfully reporting exactly what is on the display regardless of its alternative perceptual possibilities. Whatever is going on here, it is certainly the case that this type of completion phenomenon is not easily produced in pigeons, unlike with human observers. Most of these tests have involved presumed “occlusion” by other stimuli in their testing procedure. The pigeon’s response to “occlusion” in other circumstances is not always so straightforward, however.

DiPietro, Wasserman, and Young (2002) tested the recognition of three-dimensionally depicted drawings of different objects by pigeons. In their test, pigeons had to discriminate which of four different objects had been presented. In the critical test, a familiar and adjacent brick wall–like occluder was then placed in different arrangements relative to a present object. When the occluder was placed in front of the objects, the pigeon’s recognition correspondingly decreased. If they had completed the objects then accuracy should not have decreased, but given the previous results, this reduction is not so surprising. They also included a novel control, however, in which the “occluder” was placed behind the object. Even though the objects were still fully visible, this condition also reduced the pigeon’s ability to recognize the objects. Further research showed that training with these conditions could reduce, but not eliminate, this “behind” interference effect (Lazareva, Wasserman, & Biederman, 2007).

We have observed similar results when we placed an occluder either in front or behind of a discriminative object (Koban & Cook, 2009; Qadri, Asen, et al., 2014). In both of these discriminations the pigeons had to discriminate among different kinds of moving stimuli. In the first case, these were different rotating 3D shapes, and in the second they were digital animal models performing different articulated actions. In both cases, we found similar interference effects to DiPietro, Wasserman, and Young (2002). When an occluder was simply placed behind the critical, and in our case moving, information, the pigeons showed a reduced capacity to perform the learned discrimination. The origins and conditions of this “behind” interference effect are yet to be determined. It appears that some type of masking or interference effect occurs when new edges or surfaces are created or added to previously learned depictions of objects. Other disparities between humans and pigeons have also been noted when pigeons have been asked to make explicit judgments of figure and ground in various types of complex images containing overlapping elements (Lazareva & Wasserman, 2012). Together these different interference results, like the various completion studies and the studies on feature use, indicate that the processing of lines and edges at the intersection and boundaries of multiple visual elements is not well understood in pigeons.

Not all of the results of investigations into completion are negative, however. Nagasaka and his colleagues have conducted three different experiments that they suggest indicate that pigeons can perceptually complete occluded and fragmented objects. In the first of these reports, Nagasaka, Hori, and Osada (2005) trained pigeons to discriminate the depth ordering of three lines that were arranged in the form of an “H.” On any trial, one of the two vertical bars was in front of the horizontal bar and one behind, but both were placed within the horizontal extent of the horizontal bar. The horizontal bar had a consistent and intermediate level of brightness, while the two vertical bars varied between black and a light gray. The depth ordering of the three bars could be changed by independently placing each of the vertical bars either in front of or behind the horizontal one. Four pigeons learned either to identify the nearer (unoccluded) or further (occluded) vertical bar (two birds each) by pecking at its location in the display.

The pigeons were then transferred to configurations that had the overlapped region between the vertical and horizontal bars varied in brightness to simulate transparency. To the human eye, the depth ordering of the bars could still be perceived with an apparent continuity of the upper and lower portions of each vertical bar. The pigeons’ responding to these stimuli was in accord with a transparency-conveyed depth ordering, suggesting that they were completing the bars. This highly interesting result critically rests on the previously untested assumption that pigeons perceived transparency in this context. A number of additional tests would have been interesting to further examine this claim. For example, could the pigeons have continued to perform the task if the upper and lower portions of the vertical bars were removed? A completion account suggests this would have been unlikely, if not impossible. Another test would have been to misalign or rotate the vertical bar segments to see how disruption of continuity affected accuracy. Finally, using varied gradients or textures to modulate the availability of simple edge relationships would have been an effective way to test how the local edge cues contributed to the overall discrimination. While we find this result intriguing for its profound implications for both the perception of completion and transparency in birds, we would like better evidence that the pigeons are not attending to some other set of cues to mediate this quite clever manipulation. This stimulating and important finding deserves wider investigation.

Nagasaka, Lazareva, and Wasserman (2007) reported another line of potentially positive results using a three-item choice task. Here the pigeons were trained in multiple stages to peck at a target shape that would be occluded by a darker adjacent shape. This was combined with comparison distractors that were either complete or incomplete versions of the target shape. They found that the errors to the different distractors were initially evenly divided, but with experience, errors gradually accumulated more frequently to the complete distractor than the incomplete one. Furthermore, the presence of a monocular perspective context cue to depth had no impact on this error rate. The authors suggest that this differential error rate to the distractors stems from the pigeons perceptually completing the occluded target shape, but other alternatives can account for these results. The most obvious alternative is that the pigeons learned a form of relative size discrimination, since the target shape and occluder always form the largest area, to which the complete distractor would be most similar. The authors spend considerable time attempting to rule this alternative out from post-hoc examinations of the data. While the overall effect is in the right direction, it is also clear that additional controls are needed before this study’s outcomes persuade.

Finally, Nagasaka and Wasserman (2008) used object motion in a highly original design to possibly capture evidence of perceptual completion in pigeons. In their first experiment, they trained four pigeons to choose one spatial choice alternative for a square moving in a circular trajectory and the other choice alternative when a set of four separated line segments moved in a synchronous pattern that looks very different. The segment-comprised pattern, however, has the appearance of what a complete, but occluded, square would look like if its vertices were being hidden behind a set of circles. After learning, they tested the pigeons with gray, circular occluders added to the displays, “covering” the vertices that were previously deleted in the training condition. Because of the design, if the pigeons were seeing the occluders as superimposed on a moving and complete square, they should have chosen that response alternative at high rates. However, in their first experiment, all four pigeons strongly responded as if they were seeing line segments instead. Three additional experiments involved further training and various improvements in the testing situations (increasing the occluder contrast, familiarizing the occluders, using a circular target form, filling the target form). In each experiment, one or two pigeons responded to the “complete” alternative at levels consistent with seeing the stimuli in that manner. However, a different combination of pigeons exhibited this “completion” result in each experiment, such that no single pigeon showed it consistently across all the experiments. Thus, a “completion” report by one bird in one experiment would disappear in the next, for example, despite changes in the displays designed to enhance the completion effect. This is a puzzling outcome. If the results were highly consistent across birds and experiments, this would be an excellent demonstration of perceptual completion in pigeons. In total, these various ingenious designs purported to have shown completion in pigeons deserve high marks for originality. They have produced the best evidence yet that pigeons might perceptually complete figures. That said, a number of reservations and additional conditions limit this evidence at the moment as providing proof that pigeons can perceptually complete or connect parts of occluded objects.

Taken together, the considerable number of experiments in this section all seem to point to one consistent and undeniable fact. It has just not been easy to get evidence of perceptual completion in pigeons. This easily produced phenomenon in humans is not readily reproduced in pigeons, despite numerous attempts with different and sometimes complicated approaches. The majority of experimental tests have produced negative results, while the few that seem more promising have reservations suggesting further research is needed. In the majority of cases, the pigeons either found alternative cues to the discrimination or accurately reported exactly what was being presented to them. While there have been varied attempts to address the issue of “naturalness” in several of these studies, the general concern over whether the pigeons globally perceive the displays remains a recurring issue. Much as with dot-based stimuli, however, these results on their surface suggest the pigeons may not have the processes needed to connect separated elements into larger configurations (but see Kirkpatrick, Wilkinson, & Johnston, 2007). Despite a natural world that seems to require the ability to complete occluded and disconnected edges and surfaces, this visual capacity remains an elusive phenomenon to elicit in the laboratory with pigeons.

Geometric Visual Illusions

Visual illusions are stable, non-veridical perceptions of the world by the visual system. Besides being fun to experience, these reliable misperceptions provide psychological insight into the contribution of the nervous system to the act of perception. The large number of identified illusions affecting human perception has contributed substantially to our understanding of the mechanisms of perception. Presumably, such illusory perceptions are the by-products of processes that have evolved over time to allow observers to effectively and quickly process the natural world, despite the lost fidelity when encountering the specific, often artificial, circumstances present in illusions.

Because of these considerations, the examination of visual illusions in animals has been of long-standing interest (Fujita, Nakamura, Sakai, Watanabe, & Ushitani, 2012; Malott, Malott, & Pokrzywinski, 1967; Révész, 1924; Warden & Baar, 1929). If animals experience visual illusions as we do, it would be good evidence that the underlying processes and representations are functionally the same, since illusions directly capture the influence and action of neural processes. If animals do not experience them as we do, it would suggest that different neural organizations are involved in their processing of the elements of these displays. Furthermore, these different mechanisms would be alternative solutions to the “visual problem” presumably addressed by the creation of illusions in the human visual system.

Likely because they are easy to create, geometric visual illusions have been the most common type of illusion examined in animals. In pigeons, four illusions have attracted the most attention. These are the Ponzo, Müller-Lyer, Ebbinghaus-Titchener, and Zöllner illusions. Examples of each of these four illusions can be seen in Figure 5. In each case, a basic psychophysical discrimination, such as a line length or circle size judgment, is tested with inducing contexts that shift or bias responding in humans, despite there being no requirement to use or consider the context when making the judgments. These illusions in humans nonetheless highlight the automatic context-dependence of such judgments. The story for pigeons is more complicated.

Several well-designed studies have suggested that pigeons may share a common perception of the Ponzo illusion. In this illusion, the inducing context consists of two converging lines that alter the length judgment of a centrally positioned line (see Figure 5A). Fujita, Blough, and Blough (1991) found evidence that pigeons seem to experience this illusion in a similar manner as humans. Pigeons were trained to discriminate the length of a centralized horizontal line, making a choice to one alternative for three shorter lines and to the other choice alternative for the three longer lines (i.e., trained to categorize lines as “short” and “long”). To familiarize the pigeons with the surrounding context, this training was conducted with parallel lines in the surrounding context and with the target line placed at three different positions within this context (high, medium, and low). After learning the discrimination, the pigeons were tested with illusion-inducing contexts produced by making the irrelevant lines non-parallel and converging toward the top. This inducing context produced an asymmetric biasing effect, with a very large “long” effect on lines placed near the converging top of the context and a smaller, but consistent, “shorter” effect on lines placed near the bottom diverging end of the context. They also tested varying degrees of context-generated depth perspective, but this did not affect the pigeons’ responding. Thus, it appeared not to matter whether the inducing context portrayed “depth” or not; simply appearing convergent was sufficient. Follow-up experiments with additional pigeons found that this biasing effect was generally true over a variety of line lengths and different converging angles of the inducing context (Fujita, Blough, & Blough, 1993). The latter research also found that the gap between the inducing context and line made important contributions to the discrimination by the pigeons. Together, these systematic biases are consistent with the pigeons’ possibly experiencing the induction of a Ponzo-like illusion.

The Müller-Lyer illusion is another classic illusion investigated in pigeons. In this illusion, an inducing context of inward and outward facing “arrows” at the endpoints of a line segment alters the length judgment of the line (see top of Figure 5B). The results from different experiments have been mixed for this display. Malott et al. (1967) and Malott and Malott (1970) trained pigeons to respond to a horizontal bar with vertical end lines. When subsequently tested for generalization with inward or outward inducing arrows on lines of varying length, response rates changed for outward arrows consistent with the perception of the illusion. The inward arrows, however, appeared not to affect responding.

More recently, Nakamura, Fujita, Ushitani, and Miyata (2006) explored this same illusion using a choice procedure. They examined both the Müller-Lyer illusion and the reversed Müller-Lyer illusion. In the latter, a small gap is inserted between the arrows at the end and the interior line, and this typically reverses the illusion in humans (see bottom of Figure 5B). After successfully training three out of four pigeons to indicate whether a target line was “short” or “long” with arrows present but facing in the same directions, they tested non-differentially reinforced probe tests with illusion-inducing placements of the arrowheads. For the standard Müller-Lyer stimuli, the pigeons shifted their line judgment in the same way as humans; with inward pointing arrowheads increasing “long” responses and outward pointing arrowheads increasing “short” responses. In contrast, the pigeons showed no effect of the reversed Müller-Lyer illusion, unlike the humans tested with these figures. Further investigations with improved reversed Müller-Lyer figures, at least according to human judgments, proved ineffective at inducing this form of the illusion (Nakamura, Watanabe, & Fujita, 2009).

The third type of geometric illusion examined with pigeons is the Ebbinghaus-Titchener illusion. In this illusion, the perceived size of an interior circle is altered by the placement of larger or smaller circles around it (see Figure 5C). In humans, this inducing context of larger circles makes the interior circle appear smaller and vice versa. Nakamura, Watanabe, and Fujita (2008) investigated whether pigeons similarly experience this illusion. After training pigeons to report three sizes of circles as “small” and three sizes of circles as “large” in a choice task, a surrounding context of intermediate-sized (i.e., neither “large” nor “small”) circles was slowly faded in over training. After the pigeons learned to discriminate the displays, the authors varied the size of the inducing circles during probe trials. They found an effect the reverse of that in humans. Smaller inducing circles caused the pigeons to respond as if the interior circle were smaller and the presence of larger inducing circles caused them to respond with “larger” responses. Concerned that their pigeons may have been responding to some weighted combination of information based on the relevant target circle and irrelevant inducing circles, isolated target-only trials were re-introduced into baseline for one or two sessions and the observations were repeated. Only two of the four pigeons responded as though they were insensitive to a weighted combination of the inducers and the target. Although these individual differences complicate the results, at least as tested here, the pigeons showed no evidence of experiencing the perceptual illusion in the same manner as humans.

Finally, Watanabe, Nakamura, and Fujita (2011) recently tested pigeons with the Zöllner illusion. Humans perceive the parallel lines in this illusion as converging toward each other (or diverging away) when short, inducing crosshatches are added to the lines (see Figure 5D). With red squares used as choice alternatives initially superimposed at either end of the two non-parallel lines, six pigeons learned to peck toward the converging end of these two lines. The red choice areas were faded away over the course of training and randomly directed crosshatching on the lines was faded in as the pigeons maintained this “convergence” judgment. The angle of these crosshatches was the same within a line, but random across the lines. The pigeons were then tested with parallel lines with “Zöllner-inducing” crosshatching added. The pigeons’ responses were again the opposite of that reported by humans, with the pigeons choosing the end that humans perceive as diverging as their “converging” one.

Several concerns need resolution before concluding that pigeons differ in their perception of the Zöllner illusion, however. The most critical is the possibility that the pigeons were being influenced by local cues during the test of the discrimination. When the test lines are eliminated as a source of information by making them parallel, the only remaining convergence information resides with the local directional features of the inducing crosshatching. In this case, they point toward a direction opposite that of the human illusion. If the pigeons were looking for any type of orientation information consistent with their training, they perhaps should have responded in the way they did. The authors argue that such local cuing is unlikely because of the large number of irrelevant orientations used during training. This may be the case, but tests evaluating the direct and local effects of the crosshatching would have been desirable.

Other data have been reported that pigeons’ perception of the closely related Herringbone illusion is consistent with human illusory perception (Güntürkün, 1997b). In this study, which was briefly described within a larger report, pigeons were trained to discriminate between square and trapezoidal line figures with irrelevant interior lines (see Figure 2A of Güntürkün, 1997b). Pigeons were then tested with interior lines oriented in a single direction or in two directions that converged toward the middle of the figures. The latter configuration biases humans to see the square boundary as trapezoidal. Among the pigeons that were not bothered by the new orientations, the illusory configuration did bias the pigeons in the same way as humans. Thus, the effect of oriented inducing lines on angle-based discriminations is mixed. Similar concerns, however, can be raised about this study as for the Zöllner study. It is not clear, for instance, how the pigeons were using the oriented inducers to judge the boundary of the figure. Were the oriented inducing lines again providing local cues that were the cause of the observed bias?

As with the other three large topic areas considered in this review, the reactions of the pigeons to these different geometric illusions have not always mimicked those of humans. The results for tests of the Müller-Lyer, Zöllner and Ebbinghaus-Titchener illusions have all either been mixed or exhibited a reversal. The best case for a similarity is for the Ponzo illusion. However, the testing of illusory perception in animals has theoretical complexities that need further examination.

One essential issue is how the various inducing contexts used to produce illusions are being integrated or assimilated into the responding of the pigeons. The key question is whether the context is actually producing a true perceptual alteration. This is what happens in humans. A second possibility, however, is that these contexts have an indirect discriminative biasing effect that is related to the learned response rule and one not based on perception. For humans, even top-down information that everything is equivalent does not alter one’s misperception of the stimulus. What is not clear is which of these two alternatives is true for pigeons.

The outcome of the Ebbinghaus-Titchener illusion provides a nice illustration of this issue and these alternatives. The human perceptual illusion is that the surrounding context of larger elements makes the internal circle appear smaller. The pigeons react in the opposite way, as if this internal circle is “larger.” One possibility is that the pigeons perceptually experience something that is the opposite of humans. Alternatively, however, the pigeons were trained to report “large” to larger circles as their solution or rule to the discrimination. Thus, when large circles are present in the surround, the pigeons simply are more biased to report large (and vice versa for smaller inducers). The reversed nature of the illusion makes it hard to know whether this is a true perceptual reversal or the result of the nature of training (about which the authors appropriately worried, as well). Identifying specifically which of these alternatives is the case is critical. To do so, one has to establish exactly what the pigeons are doing in the original baseline and illusion tests and confirm that the discriminative bases of responding accords with that in humans (i.e., the size of the center circle, exclusively). Stimulus analytic tests to determine the nature of the controlling features and effect of the inducing context itself are really the only route to consider.

One nice property of the Ebbinghaus-Titchener illusion for study is that these expected human perceptual effects and any trained discriminative effects (at least for pigeons) are in opposite directions. This helps to raise and isolate this key issue. Consider next the Ponzo illusion, however, where the evidence is thought best for pigeons experiencing the illusion. In this case these two alternatives parallel one another. The inducing contexts both make the line at the top appear perceptually longer, but also add potential discriminative biasing effects that make the line appear longer because of the spatial proximity of the inducing lines or by shortening the gap between the line and inducers. If the pigeons had learned to use the gap between the discriminative line and the inducers as part of their “length” discrimination (and there is some evidence of that; see Fujita et al., 1993) then the results are possibly equally explained by discriminative biasing rather than the direct illusory perception of the displays. Both would bias responding in the same direction.

Consequently, better understanding and separating such perceptual and discriminative effects is critical to using illusions as a means to revealing the mechanisms of visual cognition in birds and other animals. Effective investigation in this area requires a series of stimulus analytic tests that isolate and pinpoint how the animals are actually performing the discrimination and how the inducing context affects responding. So far, the evidence for a similar or different perception of illusions by pigeons is frequently not compelling in either direction. Given the private nature of illusions, the burden of proof is clearly and appropriately far greater for those arguing for any type of perceptual account (Wasserman, 2012). That said, the exploration of illusions of all types is a fruitful endeavor for future comparative research.

Discussion

Collectively, the above analyses suggest that there are at least four clusters of experimental differences regarding how pigeons and humans react to a number of different, theoretically relevant stimuli. Over all of these clusters, different line- and dot-based stimuli often produced results suggesting that pigeons do not experience the same stimulus configurations as reported by humans. Furthermore, these outcomes were often quite persistent, despite the best efforts of experimenters across different approaches. The question of understanding the visual and attentional mechanisms of both species pivots on the source of these differences. Is this just smoke or is there a real fire? Are these just experimental detritus and artifacts or markers of a more fundamental underlying truth? It seems unlikely, given the diversity of the outcomes across the different topics, that a single unified account of the observed differences can be identified. Nonetheless, considering several such possibilities is instructive.

One possibility is that these divergences are procedural in their origins. This account argues these results are artifactual or unrelated to the underlying motivating question of the mechanisms of vision and action. There are several variations of this account. All are concerned with the idea that pigeons are not processing or attending to the stimuli in the same manner as humans. If the two species learn to discriminate or attend to different features or parts of the stimuli, then the divergent outcomes may not have meaningful implications for the mechanisms of visual processing.

A concern we raised in reviewing these findings was an uncertainty over whether the pigeons were globally processing the entire, larger configurations of the displays. For humans, this global perception of the entire configuration is an essential property for virtually all of the perceptual phenomena examined. The perception of configural stimuli, the integration of dot-based stimuli, the completion of occluded or disconnected elements, and the influence of various inducing contexts to illusions all require the observer to integrate information from an extended spatial extent. Humans integrate this information naturally and without much explicit instruction. It is not so clear that this is always the case for pigeons. They may often instead rely on sequential integration or local processing strategies, which may present serious problems and limitations in contrast with global perception.

Two direct physical and experimental concerns stand out. The first is related to stimulus size. Between the proximity of the pigeons to the stimuli for response purposes and the human-designed resolution of computer displays, the tests with pigeons routinely display the stimuli at larger visual angles than with humans. The complex stimulus displays tested above are likely designed more often to support directed pecking behavior, human intuitions, and/or human aesthetics rather than promote global perception by the birds. The limited availability of information about the appropriate size to ensure global perception strategies by pigeons is a shortcoming that may hamper progress toward removing this procedural issue.

A second related concern regards the limited variation in the sizes and locations of the stimulus displays used in the different experiments. These two spatial properties are often fixed over the course of a specific experiment, but this lack of spatial variation permits, and perhaps promotes, restricted local processing strategies by boosting their effectiveness. Pigeons can clearly direct pecking and processing to smaller portions of displays. Several experiments have found that directed pecking or attention to small locational differences can have important impacts on feature and compound stimulus processing (D. S. Blough, 1993; M. F. Brown, Cook, Lamb, & Riley, 1984; Cook, Riley, & Brown, 1992). Experiments employing stimuli of varied size and variable location would enhance the probability that pigeons process displays more globally because of the need to localize them prior to their identification.

These physical attributes surely interact with a more psychological concern. Even when the size and location of the stimuli are varied, pigeons may still process local information before or in preference to global information. The tendency of pigeons to process stimuli locally has been repeatedly observed and was discussed for and in a number of the papers reviewed here. Thus, by this account, the divergences between pigeon and human vision stem from differences in attentional bias to different features of the stimuli rather than physical issues. If pigeons are prone to attend to smaller, local features when available, there are many reasons to be concerned that the reviewed experiments may not have generated equivalent visual processing demands for each species. We therefore have to reconsider the divergences in the light of these potential attentional accounts.

One possibility is that pigeons’ spatial aperture is limited or tuned to a local scale by the experimental contingences. If so, then seeing the larger configuration of the displays is difficult. Ensuring that the visual or attentional aperture employed for each experiment is sufficiently large to extract global information is important. Alternatively, instead of having a broadly tuned spatial aperture, another variation of this type of differential attentional account assumes that pigeons exhibit global-like stimulus control by gathering information from multiple, successive, local fixations of the display. In this scheme, perhaps pigeons are psychologically challenged by the area of information that they can attend to and integrate over at any one time. Similar types of aperture problems have been reported in a human with visual agnosia who required a more feature-by-feature approach to object recognition (Semmes & De Bleser, 1992). While pigeons may be able to flexibly adjust the size of their aperture over a limited range, this area may be constrained and therefore require multiple fixations. This makes for a greater reliance on memory and greater opportunities for integrative errors as a result. As a consequence, stimuli requiring the completion of separated elements over extents in the display might have difficulty being cognitively integrated. This kind of account would allow pigeons to integrate small portions of Glass patterns, permitting them to perform at above chance levels, while still making it difficult for them to see the larger configural patterns present in them. Without assuring equivalent attention to the same discriminative features or patterns in these various complex displays, these different factors or accounts would suggest it is premature to conclude that humans and pigeons differ in visual cognition. One important element for future experiments is to consider the addition of more analytic tests to reveal and confirm which features of the displays are controlling the actions of each species.

Setting aside these experimental and attentional concerns for the moment, taken at face value these different experiments all point toward qualitative differences in how pigeons visually process and perceive these stimuli. If this is the case, several possible psychological implications are raised regarding the mechanisms underlying visual cognition in pigeons. These are considered next.

One implication is that the fundamental building blocks of complex visual objects are somehow different in pigeons. While there is good physiological evidence that pigeons are sensitive to color, spatial frequency, brightness, and other fundamental features related to surfaces and shape, the relative weighting of these features may be different than in humans soon after their initial registration. Precisely specifying these features is difficult. There are scattered results from paradigms thought to capture feature processing in which pigeons may not be weighting features in the same way as humans, although several studies also suggest that these weightings can be highly similar, too (D. S. Blough & Blough, 1997). If this differential feature or weighting hypothesis is true, these processing differences seem likely located somewhere between initial sensory input and the subsequent layers that produce internal shape representations. These kinds of intermediate possibilities are raised by the differences in the processing of certain patterns or configurations. Pomerantz (2003) has suggested that human configural superiority effects may be due to the existence of intermediate level features or channels. The lack of configural superiority in pigeons could reflect the absence of similar intermediate channels, even if the simpler line features are detected in the same way.

Besides differences in bottom-up to intermediate processing, later stages in processing provide other possible alternatives. One important point to consider is whether the stimuli tested here are sufficiently stimulating to accurately drive the pigeon’s visual and cognitive systems. Virtually all of the stimuli tested here are controlled and highly abstract—black and white, line- or dot-based configurations with few enriching details. These stimuli have been highly revealing in humans for precisely those properties, making them ideal for controlled experimentation. That being said, these stimuli are also quite impoverished. The pigeon’s systems may require a more complete and realistic depiction of the world’s patterns to function at its best. At a perceptual level, the simplistic quality of these lines and dots may not drive their visual system properly. Perhaps the intermediate or additional integration of several other types of information from surfaces, texture, or shading are needed for a suitable working visual representation to be generated in these animals. While humans can cognitively cope with deriving “meaning” from them, the impoverished stimuli may be too limited for the pigeons, and then they may be too abstract for later cognitive mechanisms to compensate. Consistent with this line of thinking, more realistic and complete stimuli have often proved to be successful in demonstrating various types of complex stimulus control in pigeons (B. R. Cavoto & Cook, 2006; Cook, Qadri, et al., 2012; B. M. Gibson et al., 2007; Herrnstein & Loveland, 1964; Spetch & Friedman, 2006b).

A third account of these differences is that pigeons visually process spatially extended and disconnected information more poorly than humans. This is different from the previous integrative account in that the limitation is linked to the connective or grouping processes themselves rather than attentional factors. Specifically, the mechanisms by which edges, contours and surfaces are fashioned in avian visual cognition do not function over large spatial distances or gaps, perhaps because they require continuous edges to effectively function. As a result, judgments of separated elements are difficult. As mentioned, many of the experimental findings above do require this type of integration. A related limitation in computing and assigning foreground and background surface and edge relations may also factor into the anomalous results of some occlusion studies. This might result in a more fragmented visual experience for pigeons. In this sense, they may share some of the characteristics of individuals with brain damage that result in various types of integrative agnosias or also with some developmental disorders (Avidan, Tanzer, & Behrmann, 2011; Behrmann & Williams, 2007; Farran, 2005; Farran & Brosnan, 2011; Kaiser & Shiffrar, 2009; Riddoch et al., 2008).

The proposal that pigeons exist in a fragmented world is not a new one (Ushitani & Fujita, 2005; Vallortigara, 2006). Our own anthropocentric view of the world finds this difficult to imagine, but it might not be as challenging as it first seems. The ecology of the pigeon may be such that completing and grouping separated objects is not all that essential. The presumed benefit of perceptual completion is that it allows observers to make inferences about partially occluded objects and other situations where information about continuous edges cannot be directly extracted. One question to ask is whether the pigeons have any ethological demand for such completion. Grain is sufficiently small and numerous that, when visible, it is something to eat and likely never occluded. Similarly, the smaller features of any visible portion of a predator or mate might be sufficient to activate avoidance or mating behavior, respectively, without sufficient risk or depletion of resources when the resulting behavior is a false alarm. Any looming edge, fragmented or not, should likely be avoided during flight (Sun & Frost, 1998). Edges and surfaces for perching after flight probably only need to be completed sufficiently to provide evidence of their adequacy for support or suitability for landing. Perhaps not committing neural resources to this computation is a valuable way to reduce the processing load on pigeons’ more limited visual machinery. Given these different alternatives, where do these various lines of thinking leave us with respect to visual processing in other bird species, besides pigeons? Are they representative of birds in general or more limited to the widely explored pigeon model?

Comparisons with Other Birds

There is an unfortunate lack of corresponding research with the same degree of detail, coverage, and precision on visual cognition in other birds. For instance, passerines are the largest order of birds. They are often better studied than pigeons with regard to many aspects of bird behavior, except in the area of visual cognition. The vast majority of research has typically focused on peripheral sensory mechanisms related to the eye, its anatomy, various psychophysical sensitivities, and visual field organization (Endler, Westcott, Madden, & Robson, 2005; Hart, 2001; Jones, Pierce, & Ward, 2007; Martin, 2007; Zeigler & Bischof, 1993). Thus, beyond properties of the eye, there is a large theoretical lacuna in our knowledge about how passerines and other birds process complex visual information. The extant literature involving complex stimuli is mixed and the experimental questions and procedures are different enough that direct comparison is an issue. Nevertheless, there are hints and allegations of differences between how pigeons and other bird species perceive the world.

Much of this research has been conducted with chickens. For instance, studies with hens have produced results more indicative of figural completion and the global perception of separated elements (Forkman, 1998; Forkman & Vallortigara, 1999). Regolin and Vallortigara (1995) found that, when tested early in their development, chicks in an imprinting paradigm seemed to complete figures presented behind occluders. These results were then replicated using moving stimuli (e.g., common fate; Lea, Slater, & Ryan, 1996) for comparative strength, and they were also replicated to evaluate the hemispheric lateralization of the effect (Regolin, Marconato, & Vallortigara, 2004). Young chicks have also been successfully tested with biological motion animations and have been shown to exhibit some perception of biological-type motion, though it is not clear whether a fully articulated figure is perceived or necessary for the differences that have been observed. (Regolin, Tommasi, & Vallortigara, 1999; Regolin et al., 2000; Vallortigara et al., 2005). An investigation of Ebbinghaus-Titchener illusions suggested that four-day-old domestic chicks saw the illusion in accordance with human perception (Salva, Rugani, Cavazzana, Regolin, & Vallortigara, 2013). While their design avoids the problem of mistakenly reporting the inducers by giving the target a visually distinct color (cf. Pepperberg, Vicinay, & Cavanagh, 2007), the illusion controls are arguably weaker in this study as the authors did not control the distance between the elements or the count of inducers between conditions.

More naturalistic investigations with passerine birds also suggest that these birds perceptually complete occluded objects. Using the same video methodology as used with pigeons (Shimizu, 1998; Watanabe & Furuya, 1997), Bengalese finches behaved as if they preferred completed conspecifics (Takahasi & Okanoya, 2013). In a more ethological study, Tvardíková and Fuchs (2010) showed that tits would approach a feeder with a proximally located pigeon dummy over one with an amputated or occluded hawk dummy. More interestingly, they would approach an amputated or partially occluded hawk dummy with a higher frequency than a complete one. In comparing these conditions, the partially amputated hawk was found to be less aversive than the occluded hawk, suggesting that the tits might have amodally completed the occluded model.

Other studies have found results more in keeping with the divergences found in pigeons. The most directly comparable work has been conducted with bantams (Nakamura, Watanabe, Betsuyaku, & Fujita, 2010, 2011; Nakamura, Watanabe, & Fujita, 2014; Watanabe, Nakamura, & Fujita, 2013). Nakamura et al. (2010) tested bantams with the same stimuli as Fujita and Ushitani (2005) and found little evidence of perceptual completion. Consistent with this, bantams also show no “continuation illusion” as well (Nakamura et al., 2011). Bantam performance also matches pigeon results for investigations of the Ebbinghaus-Titchener illusion (Nakamura et al., 2014) and Zöllner illusion (Watanabe et al., 2013). As already mentioned, we have tested starlings with Glass patterns similar to those previously tested with pigeons (Qadri & Cook, 2014). Despite our trying to promote global perception by varying the size of the stimuli, the outcome with the starlings was virtually identical to that observed with pigeons. Starlings have also exhibited mixed results in other settings that require global integration, such as in the detection of symmetry in an image (Swaddle, Che, & Clelland, 2004; Swaddle & Pruett-Jones, 2001; Swaddle & Ruff, 2004).

A simple summary of these comparative outcomes is challenging. Because of the considerable differences between the species tested, the mixed outcomes, and the differences in stimuli and procedures used to test them, we are just not positioned to judge whether there exists visual or attentional differences among bird species. The gap in our knowledge is sizable enough that even a simple conclusion is elusive. The prototypical answer in science is to say that more research is needed, but in this particular case, it is desperately needed. A broader comparative examination of visual and attentional processing in other types of birds using the same sophisticated approaches developed with pigeons is critical to determining the scope of any similarities and differences across species. Conversely, extending our knowledge of pigeon visual cognition beyond the touchscreen, either by using real objects in laboratory contexts or using the broader, open-field tests similar to those conducted with tits above, may also contribute critical information regarding comparative visual processes, as well as providing valuable ecological validity (cf. Qadri, Romero, & Cook, 2014; Rowland, Cuthill, Harvey, Speed, & Ruxton, 2008). In addition to better controlling the experimental methods applied to each species, selecting species according to the visual cognition necessary for their ecological niche or natural history would further strengthen evaluations of unique and general avian visual mechanisms. Comparison species could be distantly related but clearly occupying similar visual ecologies, or closely related species whose ecologies or behaviors importantly differ. For instance, comparing coastal-dwelling birds who generally have unobstructed views during navigation and foraging with forest-dwelling birds whose visual environment is incredibly noisy would be highly informative regarding the role of any visual completion processes. Ultimately, the key is establishing how many visual and cognitive profiles need to be considered to resolve the comparative issue.

Recommendations

Besides a call for broader examinations of more carefully chosen bird species, we have several recommendations for advancing the investigation of these general questions. From the review, it is clear that understanding the role and contribution of spatial attention and integration is critical to any advance. Visual and attentional mechanisms are clearly linked, and separating them is not always easy. Nonetheless, it is important that we have procedures in place to ensure that we properly address whether pigeons, and other birds, are integrating and perceiving all of the elements of the displays. Virtually all of the interesting theoretical effects reviewed above require such integration. It is only when we have a comparative situation that allows us to ascertain such integration (or its absence) that we will be able to tell if and how birds and mammals differ in the computational or representational mechanisms of vision, attention, or both.

Several immediate and concrete experimental improvements can be made. These enhance and promote the possibility of global perception and integration and simultaneously discourage the use of local or featural biases necessary for the success of restricted local processing or global sequential integration strategies. First, researchers should reduce the visual angle of the stimuli or patterns tested with birds. In general, the visual angle of the displays tested with pigeons is consistently larger than those with humans, perhaps because they look the “right size” to us. The easiest and simplest solution would be to collect comparison data from humans with stimuli that mimic the visual angles tested with birds. The other alternative is to test smaller stimuli with the birds. The physical resolution of computer screens may limit this approach, however. Another simple strategy that accomplishes much the same goal is to recess the computer display farther back behind the touchscreen. While the highly directed responses of choice tasks are more difficult to execute with such a setup, go/no-go procedures can be conducted using this arrangement (cf. Asen & Cook, 2012; Cook, Qadri, et al., 2012).

A second important set of methodological improvements involves placing the informative features at different locations around the display. Given the apparent capacity of pigeons to attend to absolute location and local features, stimuli with fixed locations are likely more prone to having only a portion of them processed. If this portion contains relevant or co-varying discriminative information, then a restricted local processing strategy is efficient. Moving the stimuli around the screen discourages the use of this strategy. If nothing else, it ensures that at least a global fixation is needed first, prior to potentially attending more locally at features of the display.

Furthermore, before reaching conclusions about the similarity or dissimilarity of avian and human perception, stimulus analytic tests need to be conducted to understand and isolate the nature of discriminative control in the pigeons. Without understanding what features of the display are integral in the pigeons’ discrimination, we will not be able to easily assign differences to effects of visual, attentional, or discrimination learning. Without such analytic tests to determine what the subjects in these experiments are responding to, the key assumption that the pigeons and humans are processing the displays in the same way as we intended will continue to frustrate our understanding. Such evidence is more easily requested than collected. One new method we have recently started developing is to determine which features are critical or relevant by using genetic algorithms to isolate and extract the best stimulus configurations and features as identified by the selection behavior of the birds (Cook & Qadri, 2013, 2014). Regardless of the analytic tool employed, our history with studying pigeons has shown that these are efficient and “clever” problem solvers, regularly finding unanticipated solutions to our discriminative tasks. Simply duplicating the stimuli tested with humans is insufficient; it is critical that we determine how the birds really are processing them to know how to scientifically categorize the outcomes. Similarly, it is important to test humans with the same information-impoverished learning conditions experienced by the pigeons. While providing explicit or implicit attentional and strategic instructions is experimentally convenient, better comparisons can be generated when humans also have to discover their solutions via reinforcement contingences, with little instruction or information beyond how to advance a trial and maximize an outcome signal. The final behavior and performance of the humans should be the critical metric, and introspective reports should be treated with caution.

It is also important to recognize that the pigeon visual system is designed to process the real world. While the artificial stimuli that psychologists have used to isolate aspects of visual processing have been valuable, their power frequently comes from being highly controlled, abstract, and impoverished. While point-light displays work remarkably well at producing perception of human action and motion for humans, this appears not to be the case for pigeons. One possibility to consider is that pigeons require more visual support to accurately perceive the world. The different avian subsystems that function to divide the work of vision may not be so capable when placed in isolation. As a result, unlike humans, pigeons may not be as readily able to perform with highly abstract or restricted stimuli. If this speculation is true, that would be an important comparative difference to establish. As a step forward, making stimuli more complete and realistic or directly tied to visual problems encountered by pigeons would likely provide theoretically revealing and new information about the operations of their visual and attentional systems.

Finally, it would be valuable to move beyond looking at just behavioral outcomes by combining them with investigations of the neuroscience of avian visual cognition. Coordinating behavioral experiments with manipulations of the different ascending and descending visual and attentional pathways represents an important direction for future work (e.g., Cook & Hagmann, 2012; Cook, Patton, & Shimizu, 2013; Nguyen et al., 2004). More investigations of lateralization and the role of differential hemispheric contributions are also needed (e.g., Güntürkün, 1997a; Güntürkün, Hellmann, Melsbach, & Prior, 1998; Vallortigara, 2000). Finally, investigating these behavioral outcomes in combination with manipulations of the frontal and lateral visual fields of birds is another important direction that needs further exploration (e.g., Bloch & Martinoya, 1983; Roberts, Phelps, Macuda, Brodbeck, & Russ, 1996).

In ending, it can be safely said that the comparative analysis of different species, especially pigeons, has yielded important new information about vision, attention, and their mechanisms (Cook, 2000, 2001; Nielsen & Rainer, 2007; Soto & Wasserman, 2010). Nonetheless, how these remarkably small, but highly capable, visual systems function remains a deep and unresolved puzzle. The frequency and regularity of the divergent outcomes reviewed here from our theoretical and anthropocentric expectations indicate that we do not yet fully understand visual cognition in this important class of animal. An improved understanding will represent an important scientific advance toward a unified general theory of vision, representation, and cognition.

References

Allan, S. E., & Blough, D. S. (1989). Feature-based search asymmetries in pigeons and humans. Perception & Psychophysics, 46(5), 456–464. doi:10.3758/BF03210860

Alonso, P. D., Milner, A. C., Ketcham, R. A., Cookson, M. J., & Rowe, T. B. (2004). The avian nature of the brain and inner ear of Archaeopteryx. Nature, 430, 666–669. doi:10.1038/nature02706

Asen, Y., & Cook, R. G. (2012). Discrimination and categorization of actions by pigeons. Psychological Science, 23, 617–624. doi:10.1177/0956797611433333

Aust, U., & Huber, L. (2001). The role of item- and category-specific information in the discrimination of people- versus nonpeople images by pigeons. Animal Learning & Behavior, 29(2), 107–119. doi:10.3758/Bf03192820

Aust, U., & Huber, L. (2003). Elemental versus configural perception in a people-present/people-absent discrimination task by pigeons. Learning & Behavior, 31(3), 213–224. doi:10.3758/BF03195984

Aust, U., & Huber, L. (2006). Does the use of natural stimuli facilitate amodal completion in pigeons? Perception, 35, 333–349. doi:10.1068/p5233

Avidan, G., Tanzer, M., & Behrmann, M. (2011). Impaired holistic processing in congenital prosopagnosia. Neuropsychologia, 49(9), 2541–2552. doi:10.1016/j.neuropsychologia.2011.05.002

Balanoff, A. M., Bever, G. S., & Norell, M. A. (2014). Reconsidering the avian nature of the oviraptorosaur brain (Dinosauria: theropoda). PLoS ONE, 9(12), e113559. doi:10.1371/journal.pone.0113559

Behrmann, M., & Williams, P. (2007). Impairments in part–whole representations of objects in two cases of integrative visual agnosia. Cognitive Neuropsychology, 24(7), 701–730. doi:10.1080/02643290701672764

Biederman, I. (1987). Recognition-by-components: A theory of human image understanding. Psychological Review, 94(2), 115–147. doi:10.1037/0033-295X.94.2.115

Bischof, W. F., Reid, S. L., Wylie, D. R. W., & Spetch, M. L. (1999). Perception of coherent motion in random dot displays by pigeons and humans. Perception & Psychophysics, 61(6), 1089–1101. doi:10.3758/BF03207616

Blake, R. (1993). Cats perceive biological motion. Psychological Science, 4(1), 54–57. doi:10.1111/j.1467-9280.1993.tb00557.x

Blake, R., & Shiffrar, M. (2007). Perception of human motion. Annual Review of Psychology, 58, 47–73. doi:10.1146/annurev.psych.57.102904.190152

Bloch, S., & Martinoya, C. (1983). Specialization of visual functions for different retinal areas in the pigeon. In J. P. Ewert (Ed.), Advances in vertebrate neuroethology (pp. 359–368). Springer. doi:10.1007/978-1-4684-4412-4_18

Blough, D. S. (1977). Visual search in the pigeon: Hunt and peck method. Science, 196(4293), 1013–1014. doi:10.1126/science.860129

Blough, D. S. (1982). Pigeon perception of letters of the alphabet. Science, 218(4570), 397–398. doi:10.1126/science.7123242

Blough, D. S. (1984). Form recognition in pigeons. In H. L. Roitblat, T. E. Bever, & H. S. Terrace (Eds.), Animal cognition (pp. 277–290). New York: Columbia University Press.

Blough, D. S. (1990). Form similarity and categorization in pigeon visual search. In M. L. Commons, R. J. Herrnstein, S. M. Kosslyn, & D. B. Mumford (Eds.), Behavioral approaches to pattern recognition and concept formation: Quantitative analyses of behavior (Vol. 8, pp. 129–143). Hillsdale, NJ: Erlbaum.

Blough, D. S. (1992). Features of forms in pigeon perception. In W. K. Honig & J. Fetterman (Eds.), Cognitive Aspects of Stimulus Control (pp. 263–277). Hillsdale, NJ: Erlbaum.

Blough, D. S. (1993). Reaction time drifts identify objects of attention in pigeon visual search. Journal of Experimental Psychology: Animal Behavior Processes, 19(2), 107–120. doi:10.1037/0097-7403.19.2.107

Blough, D. S., & Blough, P. M. (1997). Form perception and attention in pigeons. Animal Learning & Behavior, 25(1), 1–20. doi:10.3758/Bf03199020

Blough, P. M. (1971). The visual acuity of the pigeon for distant targets. Journal of the Experimental Analysis of Behavior, 15, 57–67. doi:10.1901/jeab.1971.15-57

Blough, P. M. (1984). Visual search in pigeons: Effects of memory set size and display variables. Perception & Psychophysics, 35(4), 344–352. doi:10.3758/BF03206338

Blough, P. M. (1989). Attentional priming and visual search in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 15(4), 358–365. doi:10.1037/0097-7403.15.4.358

Brown, J., Kaplan, G., Rogers, L. J., & Vallortigara, G. (2010). Perception of biological motion in common marmosets (Callithrix jacchus): By females only. Animal Cognition, 13(3), 555–564. doi:10.1007/s10071-009-0306-0

Brown, M. F., Cook, R. G., Lamb, M. R., & Riley, D. A. (1984). The relation between response and attentional shifts in pigeon compound matching-to-sample performance. Animal Learning & Behavior, 12, 41–49. doi:10.3758/BF03199811

Cavoto, B. R., & Cook, R. G. (2006). The contribution of monocular depth cues to scene perception by pigeons. Psychological Science, 17, 628–634. doi:10.1111/j.1467-9280.2006.01755.x

Cavoto, K. K., & Cook, R. G. (2001). Cognitive precedence for local information in hierarchical stimulus processing by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 27, 3–16. doi:10.1037/0097-7403.27.1.3

Cerella, J. (1977). Absence of perspective processing in the pigeon. Pattern Recognition, 9, 65–68. doi:10.1016/0031-3203(77)90016-4

Cerella, J. (1980). The pigeon’s analysis of pictures. Pattern Recognition, 9, 1–6. doi:10.1016/0031-3203(80)90048-5

Cerella, J. (1986). Pigeons and perceptrons. Pattern Recognition, 19(6), 431–438. doi:10.1016/0031-3203(86)90041-5

Cook, R. G. (1992a). Acquisition and transfer of visual texture discriminations by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 18, 341–353. doi:10.1037/0097-7403.18.4.341

Cook, R. G. (1992b). Dimensional organization and texture discrimination in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 18, 354–363. doi:10.1037/0097-7403.18.4.354

Cook, R. G. (1992c). The visual perception and processing of textures by pigeons. In W. K. Honig & J. G. Fetterman (Eds.), Cognitive aspects of stimulus control (pp. 279–299). Hillsdale, NJ: Erlbaum.

Cook, R. G. (1993). Gestalt contributions to visual texture discriminations by pigeons. In T. Zentall (Ed.), Animal cognition: A tribute to Donald A. Riley (pp. 251–269). Hillsdale, NJ: Erlbaum.

Cook, R. G. (2000). The comparative psychology of avian visual cognition. Current Directions in Psychological Science, 9(3), 83–89. doi:10.1111/1467-8721.00066

Cook, R. G. (2001). Avian Visual Cognition. Retrieved from www.pigeon.psy.tufts.edu/avc

Cook, R. G., Cavoto, B. R., Katz, J. S., & Cavoto, K. K. (1997). Pigeon perception and discrimination of rapidly changing texture stimuli. Journal of Experimental Psychology: Animal Behavior Processes, 23, 390–400. doi:10.1037/0097-7403.23.4.390

Cook, R. G., Cavoto, K. K., & Cavoto, B. R. (1996). Mechanisms of multidimensional grouping, fusion, and search in avian texture discrimination. Animal Learning & Behavior, 24, 150–167. doi:10.3758/BF03198963

Cook, R. G., Goto, K., & Brooks, D. I. (2005). Avian detection and identification of perceptual organization in random noise. Behavioural Processes, 69, 79–95. doi:10.1016/j.beproc.2005.01.006

Cook, R. G., & Hagmann, C. E. (2012). Grouping and early visual processing in avian vision. In O. F. Lazareva, T. Shimizu, & E. A. Wasserman (Eds.), How animals see the world: Behavior, biology, and evolution of vision. London: Oxford University Press. doi:10.1093/acprof:oso/9780195334654.003.0004

Cook, R. G., Katz, J. S., & Blaisdell, A. P. (2012). Temporal properties of visual search in pigeon target localization. Journal of Experimental Psychology: Animal Behavior Processes, 38(2), 209–216. doi:10.1037/A0026496

Cook, R. G., Patton, T. B., & Shimizu, T. (2013). Functional segregation of the entopallium in pigeons. Philosophy (London), 130, 59–86.

Cook, R. G., & Qadri, M. A. J. (2013). The adaptive analysis of visual cognition using genetic algorithms. Journal of Experimental Psychology: Animal Behavior Processes, 39(4), 20. doi:10.1037/a0034074

Cook, R. G., & Qadri, M. A. J. (2014). Visualizing search behavior with adaptive discriminations. Behavioural Processes, 102, 40–50. doi:10.1016/j.beproc.2013.12.007

Cook, R. G., Qadri, M. A. J., & Keller, A. M. (in press). The analysis of visual cognition in birds: Implications for evolution, mechanism, and representation. Psychology of Learning and Motivation, 63, 1–38.

Cook, R. G., Qadri, M. A. J., Kieres, A., & Commons-Miller, N. (2012). Shape from shading in pigeons. Cognition, 124, 284–303. doi:10.1016/j.cognition.2012.05.007

Cook, R. G., Riley, D. A., & Brown, M. F. (1992). Spatial and configural factors in compound stimulus processing by pigeons. Animal Learning & Behavior, 20, 41–55. doi:10.3758/BF03199945

Corwin, S. (2010). The asymmetry of the carpal joint and the evolution of wing folding in maniraptoran theropod dinosaurs. Proceedings of the Royal Society B: Biological Sciences, 277, 2027–2033. doi:10.1098/rspb.2009.2281

DiPietro, N. T., Wasserman, E. A., & Young, M. E. (2002). Effects of occlusion on pigeons’ visual object recognition. Perception, 31(11), 1299–1312. doi:10.1068/p3441

Dittrich, W. H., Lea, S. E. G., Barrett, J., & Gurr, P. R. (1998). Categorization of natural movements by pigeons: Visual concept discrimination and biological motion. Journal of the Experimental Analysis of Behavior, 70(3), 281–299. doi:10.1901/jeab.1998.70-281

Donis, F. J., Chase, S., & Heinemann, E. G. (2005). Effects of identical context on visual pattern recognition by pigeons. Learning & Behavior, 33(1), 90–98. doi:10.3758/BF03196053

Donis, F. J., & Heinemann, E. G. (1993). The object-line inferiority effect in pigeons. Perception and Psychophysics, 53(1), 117–122. doi:10.3758/BF03211720

Drori, I., Cohen-Or, D., & Yeshurun, H. (2003). Fragment-based image completion. ACM Transactions on Graphics, 22(3), 303–312. doi:10.1145/882262.882267

Emmerton, J., & Renner, J. C. (2009). Local rather than global processing of visual arrays in numerosity discrimination by pigeons (Columba livia). Animal Cognition, 12(3), 511–526. doi:10.1007/s10071-009-0212-5

Endler, J. A., Westcott, D. A., Madden, J. R., & Robson, T. (2005). Animal visual systems and the evolution of color patterns: Sensory processing illuminates signal evolution. Evolution, 59(8), 1795–1818. doi:10.1111/j.0014-3820.2005.tb01827.x

Farran, E. K. (2005). Perceptual grouping ability in Williams syndrome: Evidence for deviant patterns of performance. Neuropsychologia, 43(5), 815–822. doi:10.1016/j.neuropsychologia.2004.09.001

Farran, E. K., & Brosnan, M. J. (2011). Perceptual grouping abilities in individuals with autism spectrum disorder; Exploring patterns of ability in relation to grouping type and levels of development. Autism Research, 4(4), 283–292. doi:10.1002/aur.202

Felleman, D. J., & Van Essen, D. C. (1991). Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex, 1(1), 1–47. doi:10.1093/cercor/1.1.1

Forkman, B. (1998). Hens use occlusion to judge depth in a two-dimensional picture. Perception, 27(7), 861–867. doi:10.1068/P270861

Forkman, B., & Vallortigara, G. (1999). Minimization of modal contours: An essential cross-species strategy in disambiguating relative depth. Animal Cognition, 2, 181–185. doi:10.1007/s100710050038

Fremouw, T., Herbranson, W. T., & Shimp, C. P. (2002). Dynamic shifts of pigeon local/global attention. Animal Cognition, 5(4), 233–243. doi:10.1007/s10071-002-0152-9

Fujita, K. (2001). Perceptual completion in rhesus monkeys (Macaca mulatta) and pigeons (Columba livia). Perception & Psychophysics, 63(1), 115–125. doi:10.3758/Bf03200507

Fujita, K., Blough, D. S., & Blough, P. M. (1991). Pigeons see the Ponzo illusion. Animal Learning & Behavior, 19(3), 283–293. doi:10.3758/Bf03197888

Fujita, K., Blough, D. S., & Blough, P. M. (1993). Effects of the inclination of context lines on perception of the Ponzo illusion by pigeons. Animal Learning & Behavior, 21(1), 29–34. doi:10.3758/Bf03197972

Fujita, K., Nakamura, N., Sakai, A., Watanabe, S., & Ushitani, T. (2012). Amodal completion and illusory perception in birds and primates. In O. F. Lazareva, T. Shimizu, & E. A. Wasserman (Eds.), How animals see the world: Comparative behavior, biology, and evolution of vision. London: Oxford University Press. doi:10.1093/acprof:oso/9780195334654.003.0008

Fujita, K., & Ushitani, T. (2005). Better living by not completing: A wonderful peculiarity of pigeon vision. Behavioural Processes, 69, 59–66. doi:10.1016/j.beproc.2005.01.003

Gallant, J. L., Braun, J., & Van Essen, D. C. (1993). Selectivity for polar, hyperbolic, and Cartesian gratings in macaque visual cortex. Science, 259(5091), 100–103. doi:10.1126/science.8418487

Gibson, B. M., Lazareva, O. F., Gosselin, F., Schyns, P. G., & Wasserman, E. A. (2007). Nonaccidental properties underlie shape recognition in mammalian and nonmammalian vision. Current Biology, 17(4), 336–340. doi:10.1016/j.cub.2006.12.025

Gibson, J. J. (1979). The ecological approach to visual perception. Boston: Houghton Mifflin.

Glass, L. (1969). Moire effect from random dots. Nature, 223(5206), 578–580. doi:10.1038/223578a0

Güntürkün, O. (1997a). Avian visual lateralization: A review. Neuroreport: An International Journal for the Rapid Communication of Research in Neuroscience, 8(6), iii-xi.

Güntürkün, O. (1997b). Visual lateralization in birds: From neurotrophins to cognition? European Journal of Morphology, 35(4), 290–302. doi:10.1076/ejom.35.4.290.13075

Güntürkün, O., Hellmann, B., Melsbach, G., & Prior, H. (1998). Asymmetries of representation in the visual system of pigeons. Neuroreport, 9(18), 4127–4130.

Harris, C., & Stephens, M. (1988). A combined corner and edge detector. Paper presented at the Alvey vision conference (Manchester, United Kingdom). doi:10.5244/C.2.23

Hart, N. S. (2001). The visual ecology of avian photoreceptors. Progress in Retinal and Eye Research, 20(5), 675–703. doi:10.1016/S1350-9462(01)00009-X

Hartline, H. K., & Ratliff, F. (1957). Inhibitory interaction of receptor units in the eye of Limulus. The Journal of General Physiology, 40(3), 357–376. doi:10.1085/jgp.40.3.357

Hendricks, J. (1966). Flicker thresholds as determined by a modified conditioned suppression procedure. Journal of the Experimental Analysis of Behavior, 9(5), 501–506. doi:10.1901/jeab.1966.9-501

Herrnstein, R. J., & Loveland, D. H. (1964). Complex visual concept in the pigeon. Science, 146(3643), 549–551. doi:10.1126/science.146.3643.549

Homman-Ludiye, J., & Bourne, J. A. (2014). Mapping arealisation of the visual cortex of non-primate species: Lessons for development and evolution. Frontiers in Neural Circuits, 8. doi:10.3389/fncir.2014.00079

Honig, W. K., & Fetterman, J. G. (1992). Cognitive Aspects of Stimulus Control. Hillsdale, NJ: Erlbaum.

Hubel, D. H., & Wiesel, T. N. (1962). Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. The Journal of Physiology, 160, 106–154. doi:10.1113/jphysiol.1962.sp006837

Johansson, G. (1973). Visual perception of biological motion and a model of its analysis. Perception and Psychophysics, 14(2), 201–211. doi:10.3758/Bf03212378

Jones, M. P., Pierce, K. E., & Ward, D. (2007). Avian vision: A review of form and function with special consideration to birds of prey. Journal of Exotic Pet Medicine, 16(2), 69–87. doi:10.1053/j.jepm.2007.03.012

Kaas, J. H. (2013). The evolution of brains from early mammals to humans. Wiley Interdisciplinary Reviews: Cognitive Science, 4(1), 33–45. doi:10.1002/wcs.1206

Kaiser, M. D., & Shiffrar, M. (2009). The visual perception of motion by observers with autism spectrum disorders: A review and synthesis. Psychonomic Bulletin & Review, 16(5), 761–777. doi:10.3758/pbr.16.5.761

Kellman, P., & Shipley, T. (1991). A theory of visual interpolation in object perception. Cognitive Psychology, 23(2), 141–221. doi:10.1016/0010-0285(91)90009-D

Kelly, D. M., Bischof, W. F., Wong-Wylie, D. R., & Spetch, M. L. (2001). Detection of glass patterns by pigeons and humans: Implications for differences in higher-level processing. Psychological Science, 12(4), 338–342. doi:10.1111/1467-9280.00362

Kelly, D. M., & Cook, R. G. (2003). Differential effects of visual context on pattern discrimination by pigeons (Columba livia) and humans (Homo sapiens). Journal of Comparative Psychology, 117(2), 200–208. doi:10.1037/0735-7036.117.2.200

Kelly, D. M., Spetch, M. L., & Heth, C. D. (1998). Pigeons’ (Columba livia) encoding of geometric and featural properties of a spatial environment. Journal of Comparative Psychology, 112(3), 259–269. doi:10.1037/0735-7036.112.3.259

Kimchi, R., & Bloch, B. (1998). Dominance of configural properties in visual form perception. Psychonomic Bulletin & Review, 5(1), 135–139. doi:10.3758/Bf03209469

Kirkpatrick-Steger, K., Wasserman, E. A., & Biederman, I. (1996). Effects of spatial rearrangement of object components on picture recognition in pigeons. Journal of the Experimental Analysis of Behavior, 65(2), 465–475. doi:10.1901/jeab.1996.65-465

Kirkpatrick-Steger, K., Wasserman, E. A., & Biederman, I. (1998). Effects of geon deletion, scrambling, and movement on picture recognition in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 24(1), 34–46. doi:10.1037/0097-7403.24.1.34

Kirkpatrick, K., Wilkinson, A., & Johnston, S. (2007). Pigeons discriminate continuous versus discontinuous line segments. Journal of Experimental Psychology: Animal Behavior Processes, 33(3), 273–286. doi:10.1037/0097-7403.33.3.273

Koban, A., & Cook, R. G. (2009). Rotational object discrimination by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 35, 250–265. doi:10.1037/a0013874

Lautenschlager, S., Witmer, L. M., Altangerel, P., & Rayfield, E. J. (2013). Edentulism, beaks, and biomechanical innovations in the evolution of theropod dinosaurs. Proceedings of the National Academy of Sciences of the United States of America, 110, 20657–20662. doi:10.1073/pnas.1310711110

Lazareva, O. F., Shimizu, T., & Wasserman, E. A. (2012). How animals see the world: Comparative behavior, biology, and evolution of vision. Oxford University Press. doi:10.1093/acprof:oso/9780195334654.001.0001

Lazareva, O. F., & Wasserman, E. A. (2012). Figure–ground segregation and object-based attention in pigeons. In O. F. Lazareva, T. Shimizu, & E. A. Wasserman (Eds.), How animals see the world: Comparative behavior, biology, and evolution of vision. Oxford University Press. doi:10.1093/acprof:oso/9780195334654.003.0005

Lazareva, O. F., Wasserman, E. A., & Biederman, I. (2007). Pigeons’ recognition of partially occluded objects depends on specific training experience. Perception, 36(1), 33–48. doi:10.1068/p5583

Lea, S. E. G., Goto, K., Osthaus, B., & Ryan, C. M. E. (2006). The logic of the stimulus. Animal Cognition, 9, 247–256. doi:10.1007/s10071-006-0038-3

Lea, S. E. G., Slater, A. M., & Ryan, C. M. E. (1996). Perception of object unity in chicks: A comparison with the human infant. Infant Behavior and Development, 19, 501–504. doi:10.1016/S0163-6383(96)90010-7

Lettvin, J. Y., Maturana, H. R., McCulloch, W. S., & Pitts, W. H. (1959). What the frog’s eye tells the frog’s brain. Proceedings of the IRE, 47, 1940–1951. doi:10.1109/JRPROC.1959.287207

Malott, R. W., & Malott, M. K. (1970). Perception and stimulus generalization. In W. Stebbins (Ed.), Animal psychophysics: The design and conduct of sensory experiments (pp. 363–400). Springer. doi:10.1007/978-1-4757-4514-6_17

Malott, R. W., Malott, M. K., & Pokrzywinski, J. (1967). The effects of outward-pointing arrowheads on the Mueller-Lyer illusion in pigeons. Psychonomic Science, 9(1), 55–56. doi:10.3758/BF03330756

Marr, D. (1982). Vision. San Francisco: Freeman.

Martin, G. (2007). Visual fields and their functions in birds. Journal of Ornithology, 148, 547–562. doi:10.1007/s10336-007-0213-6

Müller, J. R., Philiastides, M. G., & Newsome, W. T. (2005). Microstimulation of the superior colliculus focuses attention without moving the eyes. Proceedings of the National Academy of Sciences of the United States of America, 102(3), 524–529. doi:10.1073/pnas.0408311101

Nagasaka, Y., Hori, K., & Osada, Y. (2005). Perceptual grouping in pigeons. Perception, 34(5), 625–632. doi:10.1068/p5402

Nagasaka, Y., Lazareva, O. F., & Wasserman, E. A. (2007). Prior experience affects amodal completion in pigeons. Perception & Psychophysics, 69(4), 596–605. doi:10.3758/BF03193917

Nagasaka, Y., & Wasserman, E. A. (2008). Amodal completion of moving objects by pigeons. Perception, 37(4), 557. doi:10.1068/p5899

Nakamura, N., Fujita, K., Ushitani, T., & Miyata, H. (2006). Perception of the standard and the reversed Müller-Lyer figures in pigeons (Columba livia) and humans (Homo sapiens). Journal of Comparative Psychology, 120(3), 252–261. doi:10.1037/0735-7036.120.3.252

Nakamura, N., Watanabe, S., Betsuyaku, T., & Fujita, K. (2010). Do bantams (Gallus gallus domesticus) experience amodal completion? An analysis of visual search performance. Journal of Comparative Psychology, 124(3), 331–335. doi:10.1037/a0019459

Nakamura, N., Watanabe, S., Betsuyaku, T., & Fujita, K. (2011). Do bantams (Gallus gallus domesticus) amodally complete partly occluded lines? An analysis of line classification performance. Journal of Comparative Psychology, 125(4), 411–419. doi:10.1037/a0024629

Nakamura, N., Watanabe, S., & Fujita, K. (2008). Pigeons perceive the Ebbinghaus-Titchener circles as an assimilation illusion. Journal of Experimental Psychology: Animal Behavior Processes, 34(3), 375–387. doi:10.1037/0097-7403.34.3.375

Nakamura, N., Watanabe, S., & Fujita, K. (2009). Further analysis of perception of the standard Müller-Lyer figures in pigeons (Columba livia) and humans (Homo sapiens): Effects of length of brackets. Journal of Comparative Psychology, 123(3), 287–294. doi:10.1037/a0016215

Nakamura, N., Watanabe, S., & Fujita, K. (2014). A reversed Ebbinghaus–Titchener illusion in bantams (Gallus gallus domesticus). Animal Cognition, 17(2), 471–481. doi:10.1007/s10071-013-0679-y

Nankoo, J., Madan, C., Spetch, M., & Wylie, D. (2014). Perception of complex motion in humans and pigeons (Columba livia). Experimental Brain Research, 1–11. doi:10.1007/s00221-014-3876-2

Navon, D. (1977). Forest before trees: The precedence of global features in visual perception. Cognitive Psychology, 9, 353–383. doi:10.1016/0010-0285(77)90012-3

Navon, D. (1981). The forest revisted: More on global precedence. Psychological Review, 43, 1–32. doi:10.1007/BF00309635

Nguyen, A. P., Spetch, M. L., Crowder, N. A., Winship, I. R., Hurd, P. L., & Wylie, D. R. (2004). A dissociation of motion and spatial-pattern vision in the avian telencephalon: Implications for the evolution of “visual streams.” The Journal of Neuroscience, 24, 4962–4970. doi:10.1523/JNEUROSCI.0146-04.2004

Nielsen, K. J., & Rainer, G. (2007). Object recognition: Similar visual strategies of birds and mammals. Current Biology, 17(5), 174–176. doi:10.1016/j.cub.2007.01.014

Oram, M., & Perrett, D. (1994). Responses of anterior superior temporal polysensory (STPa) neurons to “biological motion” stimuli. Journal of Cognitive Neuroscience, 6(2), 99–116. doi:10.1162/jocn.1994.6.2.99

Palmer, S. E. (1999). Vision science: Photons to phenomenology. Cambridge, MA: MIT Press.

Parron, C., Deruelle, C., & Fagot, J. (2007). Processing of biological motion point-light displays by baboons (Papio papio). Journal of Experimental Psychology: Animal Behavior Processes, 33(4), 381. doi:10.1037/0097-7403.33.4.381

Pearce, J. M., & George, D. N. (2003). Virtual search asymmetry in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 29(2), 118–129. doi:10.1037/0097-7403.29.2.118

Peissig, J. J., Young, M. E., Wasserman, E. A., & Biederman, I. (2005). The role of edges in object recognition by pigeons. Perception, 34(11), 1353–1374. doi:10.1068/p5427

Pepperberg, I., Vicinay, J., & Cavanagh, P. (2007). Processing of the Müller-Lyer illusion by a Grey Parrot (Psittacus erithacus). Journal of Vision, 7(9), 601. doi:10.1167/7.9.601

Petersen, S. E., Robinson, D. L., & Morris, J. D. (1987). Contributions of the pulvinar to visual spatial attention. Neuropsychologia, 25(1A), 97–105. doi:10.1016/0028-3932(87)90046-7

Pomerantz, J. R. (2003). Wholes, holes, and basic features in vision. Trends in Cognitive Sciences, 7(11), 471–473. doi:10.1016/j.tics.2003.09.007

Pomerantz, J. R., & Pristach, E. A. (1989). Emergent features, attention, and perceptual glue in visual form perception. Journal of Experimental Psychology: Human Perception and Performance, 15(4), 635. doi:10.1037/0096-1523.15.4.635

Pomerantz, J. R., Sager, L. C., & Stoever, R. J. (1977). Perception of wholes and their component parts: Some configural superiority effects. Journal of Experimental Psychology: Human Perception and Performance, 3(3), 422–435. doi:10.1037/0096-1523.3.3.422

Puce, A., & Perrett, D. (2003). Electrophysiology and brain imaging of biological motion. Philosophical Transactions of the Royal Society B: Biological Sciences, 358(1431), 435–445. doi:10.1098/rstb.2002.1221

Qadri, M. A. J., Asen, Y., & Cook, R. G. (2014). Visual control of an action discrimination in pigeons. Journal of Vision, 14, 1–19. doi:10.1167/14.5.16

Qadri, M. A. J., & Cook, R. G. (2014). The perception of Glass patterns by starlings (Sturnus vulgaris). Psychonomic Bulletin & Review, 1–7. doi:10.3758/s13423-014-0709-z

Qadri, M. A. J., Romero, L. M., & Cook, R. G. (2014). Shape-from-shading in European starlings (Sturnus vulgaris). Journal of Comparative Psychology, 128(4), 343–356. doi:10.1037/a0036848

Qadri, M. A. J., Sayde, J. M., & Cook, R. G. (2014). Discrimination of complex human Behavior by pigeons (Columba livia) and humans. PLoS ONE, 9(11), e112342. doi:10.1371/journal.pone.0112342

Regolin, L., Marconato, F., & Vallortigara, G. (2004). Hemispheric differences in the recognition of partly occluded objects by newly hatched domestic chicks (Gallus gallus). Animal Cognition, 7(3), 162–170. doi:10.1007/s10071-004-0208-0

Regolin, L., Tommasi, L., & Vallortigara, G. (1999). Discrimination of point-light animation sequences (Johansson’s biological-motion displays) by newborn chicks. Paper presented at the ECVP (Trieste, Italy).

Regolin, L., Tommasi, L., & Vallortigara, G. (2000). Visual perception of biological motion in newly hatched chicks as revealed by an imprinting procedure. Animal Cognition, 3(1), 53–60. doi:10.1007/s100710050050

Regolin, L., & Vallortigara, G. (1995). Perception of partly occluded objects by young chicks. Perception & Psychophysics, 57(7), 971–976. doi:10.3758/BF03205456

Reichardt, W. (1987). Evaluation of optical motion information by movement detectors. Journal of Comparative Physiology A: Neuroethology, Sensory, Neural, and Behavioral Physiology, 161, 533–547. doi:10.1007/BF00603660

Révész, G. (1924). Experiments on animal space perception. British Journal of Psychology. General Section, 14(4), 387–414. doi:10.1111/j.2044-8295.1924.tb00151.x

Riddoch, M. J., Humphreys, G. W., Akhtar, N., Allen, H., Bracewell, R. M., & Schofield, A. J. (2008). A tale of two agnosias: Distinctions between form and integrative agnosia. Cognitive Neuropsychology, 25(1), 56–92. doi:10.1080/02643290701848901

Rilling, M., De Marse, T., & La Claire, L. (1993). Contour deletion as a method for identifying the weights of features underlying object recognition. The Quarterly Journal of Experimental Psychology, 46(1), 43–61. doi:10.1080/14640749308401094

Roberts, W. A., Phelps, M. T., Macuda, T., Brodbeck, D. R., & Russ, T. (1996). Intraocular transfer and simultaneous processing of stimuli presented in different visual fields of the pigeon. Behavioral Neuroscience, 110(2), 290–299. doi:10.1037/0735-7044.110.2.290

Robinson, D. A. (1972). Eye movements evoked by collicular stimulation in the alert monkey. Vision Research, 12(11), 1795–1808. doi:10.1016/0042-6989(72)90070-3

Rowland, H. M., Cuthill, I. C., Harvey, I. F., Speed, M. P., & Ruxton, G. D. (2008). Can’t tell the caterpillars from the trees: Countershading enhances survival in a woodland. Proceedings of the Royal Society B: Biological Sciences, 275, 2539–2545. doi:10.1098/rspb.2008.0812

Salva, O. R., Rugani, R., Cavazzana, A., Regolin, L., & Vallortigara, G. (2013). Perception of the Ebbinghaus illusion in four-day-old domestic chicks (Gallus gallus). Animal Cognition, 16(6), 895–906. doi:10.1007/s10071-013-0622-2

Sekuler, A. B., Lee, J. A. J., & Shettleworth, S. J. (1996). Pigeons do not complete partly occluded figures. Perception, 25, 1109–1120. doi:10.1068/p251109

Semmes, J., & De Bleser, R. (1992). Visual agnosia: A case of reduced attentional “spotlight”? Cortex, 28(4), 601–621. doi:10.1016/S0010-9452(13)80230-4

Sereno, P. C. (1999). The evolution of dinosaurs. Science, 284(5423), 2137–2147. doi:10.1126/science.284.5423.2137

Shimizu, T. (1998). Conspecific recognition in pigeons (Columba livia) using dynamic video images. Behaviour, 135, 43–53. doi:10.1163/156853998793066429

Shimizu, T., & Hodos, W. (1989). Reversal learning in pigeons: Effects of selective lesions of the Wulst. Behavioral Neuroscience, 103(2), 262–272. doi:10.1037/0735-7044.103.2.262

Soto, F. A., & Wasserman, E. A. (2010). Comparative vision science: Seeing eye to eye? Comparative Cognition & Behavior Reviews, 5, 148. doi:10.3819/ccbr.2010.50011

Spetch, M. L. (1995). Overshadowing in landmark learning—Touch-screen studies with pigeons and humans. Journal of Experimental Psychology: Animal Behavior Processes, 21(2), 166–181. doi:10.1037/0097-7403.21.2.166

Spetch, M. L., & Edwards, C. A. (1988). Pigeons’, Columba Livia, use of global and local cues for spatial memory. Animal Behaviour, 36, 293–296. doi:10.1016/S0003-3472(88)80274-4

Spetch, M. L., & Friedman, A. (2006a). Comparative cognition of object recognition. Comparative Cognition & Behavior Reviews, 1, 12–35. doi:10.3819/ccbr.2008.10002

Spetch, M. L., & Friedman, A. (2006b). Pigeons see correspondence between objects and their pictures. Psychological Science, 17(11), 966. doi:10.1111/j.1467-9280.2006.01814.x

Sun, H., & Frost, B. J. (1998). Computation of different optical variables of looming objects in pigeon nucleus rotundus neurons. Nature Neuroscience, 1, 296–303. doi:10.1038/1110

Swaddle, J. P., Che, J. P. K., & Clelland, R. E. (2004). Symmetry preference as a cognitive by-product in starlings. Behaviour, 141(4), 469–478. doi:10.1163/156853904323066748

Swaddle, J. P., & Pruett-Jones, S. (2001). Starlings can categorize symmetry differences in dot displays. American Naturalist, 158(3), 300–307. doi:10.1086/321323

Swaddle, J. P., & Ruff, D. A. (2004). Starlings have difficulty in detecting dot symmetry: Implications for studying fluctuating asymmetry. Behaviour, 141, 29–40. doi:10.1163/156853904772746583

Takahasi, M., & Okanoya, K. (2013). An invisible sign stimulus: completion of occluded visual images in the Bengalese finch in an ecological context. Neuroreport, 24(7), 370–374. doi:10.1097/WNR.0b013e328360ba32

Tomonaga, M. (2001). Visual search for biological motion patterns in chimpanzees (Pan troglodytes). Psychologia: An International Journal of Psychology in the Orient, 44(1), 46–59.

Trajković, M., & Hedley, M. (1998). Fast corner detection. Image and Vision computing, 16(2), 75–87. doi:10.1016/S0262-8856(97)00056-5

Treisman, A., & Gormican, S. (1988). Feature analysis in early vision: Evidence from search asymmetries. Psychological Review, 95(1), 15–48. doi:10.1037/0033-295X.95.1.15

Treisman, A., & Souther, J. (1985). Search asymmetry: A diagnostic for preattentive processing of separable features. Journal of Experimental Psychology: General, 114(3), 285–310. doi:10.1037/0096-3445.114.3.285

Troje, N. F., & Aust, U. (2013). What do you mean with “direction”? Local and global cues to biological motion perception in pigeons. Vision Research, 79(0), 47–55. doi:10.1016/j.visres.2013.01.002

Tvardíková, K., & Fuchs, R. (2010). Tits use amodal completion in predator recognition: a field experiment. Animal Cognition, 13, 609–615. doi:10.1007/s10071-010-0311-3

Ushitani, T., & Fujita, K. (2005). Pigeons do not perceptually complete partly occluded photos of food: An ecological approach to the “pigeon problem.” Behavioural Processes, 69, 67–78. doi:10.1016/j.beproc.2005.01.002

Ushitani, T., Fujita, K., & Yamanaka, R. (2001). Do pigeons (Columba livia) perceive object unity? Animal Cognition, 4, 153–161. doi:10.1007/s100710100088

Vallortigara, G. (2000). Comparative neuropsychology of the dual brain: A stroll through animals’ left and right perceptual worlds. Brain and Language, 73(2), 189–219. doi:10.1006/brln.2000.2303

Vallortigara, G. (2006). The cognitive chicken: Visual and spatial cognition in a nonmammalian brain. In E. A. Wasserman & T. R. Zentall (Eds.), Comparative cognition: Experimental explorations of animal intelligence (pp. 71–86). New York: Oxford University Press.

Vallortigara, G., Regolin, L., & Marconato, F. (2005). Visually inexperienced chicks exhibit spontaneous preference for biological motion patterns. PLoS Biology, 3(7), 1312–1316. doi:10.1371/journal.pbio.0030208

Van Hamme, L. J., Wasserman, E. A., & Biederman, I. (1992). Discrimination of contour-deleted images by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 18(4), 387–399. doi:10.1037/0097-7403.18.4.387

Warden, C. J., & Baar, J. (1929). The Müller-Lyer illusion in the ring dove, Turtur risorius. Journal of Comparative Psychology, 9(4), 275–292. doi:10.1037/h0071052

Wasserman, E. A. (2012). Illusory perception in animals. In O. F. Lazareva, T. Shimizu, & E. A. Wasserman (Eds.), How animals see the world: Comparative behavior, biology, and evolution of vision. Oxford University Press. doi:10.1093/acprof:oso/9780195334654.003.0007

Wasserman, E. A., & Biederman, I. (2012). Recognition-by-components: A bird’s eye view. In O. F. Lazareva, T. Shimizu, & E. A. Wasserman (Eds.), How animals see the world: Comparative behavior, biology, and evolution of vision. Oxford University Press. doi:10.1093/acprof:oso/9780195334654.003.0012

Wasserman, E. A., Kirkpatrick-Steger, K., Van Hamme, L. J., & Biederman, I. (1993). Pigeons are sensitive to the spatial organization of complex visual stimuli. Psychological Science, 4(5), 226–341. doi:10.1111/j.1467-9280.1993.tb00575.x

Watanabe, S., & Furuya, I. (1997). Video display for study of avian visual cognition: From psychophysics to sign language. International Journal of Comparative Psychology, 10, 111–127.

Watanabe, S., Nakamura, N., & Fujita, K. (2011). Pigeons perceive a reversed Zöllner illusion. Cognition, 119(1), 137–141. doi:10.1016/j.cognition.2010.10.020

Watanabe, S., Nakamura, N., & Fujita, K. (2013). Bantams (Gallus gallus domesticus) also perceive a reversed Zöllner illusion. Animal Cognition, 16(1), 109–115. doi:10.1007/s10071-012-0556-0

Williams, L. R., & Hanson, A. R. (1994, June). Perceptual completion of occluded surfaces. Paper presented at the Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 1994 (Seattle, Washington, USA). doi:10.1109/CVPR.1994.323803

Williams, L. R., & Jacobs, D. W. (1997). Stochastic completion fields: A neural model of illusory contour shape and salience. Neural Computation, 9(4), 837–858. doi:10.1162/neco.1997.9.4.837

Wilson, H. R., & Wilkinson, F. (1998). Detection of global structure in Glass patterns: Implications for form vision. Vision Research, 38(19), 2933–2947. doi:10.1016/S0042-6989(98)00109-6

Wilson, H. R., Wilkinson, F., & Asaad, W. (1997). Concentric orientation summation in human form vision. Vision Research, 37(17), 2325–2330. doi:10.1016/S0042-6989(97)00104-1

Wright, A. A., & Cumming, W. W. (1971). Color-naming functions for the pigeon. Journal of the Experimental Analysis of Behavior, 15(1), 7–17. doi:10.1901/jeab.1971.15-7

Xu, X., Zhou, Z., Dudley, R., Mackem, S., Chuong, C.-M., Erickson, G. M., & Varricchio, D. J. (2014). An integrative approach to understanding bird origins. Science, 346(6215). doi:10.1126/science.1253293

Young, M. E., Peissig, J. J., Wasserman, E. A., & Biederman, I. (2001). Discrimination of geons by pigeons: The effects of variations in surface depiction. Animal Learning & Behavior, 29(2), 97–106. doi:10.3758/Bf03192819

Zeigler, H. P., & Bischof, W. F. (1993). Vision, brain, and behavior in birds. Cambridge, MA: MIT Press.

Zhang, J., Marszałek, M., Lazebnik, S., & Schmid, C. (2007). Local features and kernels for classification of texture and object categories: A comprehensive study. International Journal of Computer Vision, 73(2), 213–238. doi:10.1007/s11263-006-9794-4

Mechanisms of Individual Differences in Impulsive and Risky Choice in Rats

Kimberly Kirkpatrick
Department of Psychological Sciences, Kansas State University

Andrew T. Marshall
Department of Psychological Sciences, Kansas State University

Aaron P. Smith
Department of Psychology, University of Kentucky

Reading Options:

Continue reading below, or:
Read/Download PDF| Add to Endnote

---

Abstract

Individual differences in impulsive and risky choice are key risk factors for a variety of maladaptive behaviors such as drug abuse, gambling, and obesity. In our rat model, ordered individual differences are stable across choice parameters and months of testing, and span a broad spectrum, suggesting that rats, like humans, exhibit trait-level impulsive and risky choice behaviors. In addition, impulsive and risky choices are highly correlated, suggesting a degree of correspondence between these two traits. An examination of the underlying cognitive mechanisms has suggested an important role for timing processes in impulsive choice. In addition, in an examination of genetic factors in impulsive choice, the Lewis rat strain emerged as a possible animal model for studying disordered impulsive choice, with this strain demonstrating deficient delay processing. Early rearing environment also affected impulsive behaviors, with rearing in an enriched environment promoting adaptable and more self-controlled choices. The combined results with impulsive choice suggest an important role for timing and reward sensitivity in moderating impulsive behaviors. Relative reward valuation also affects risky choice, with manipulation of objective reward value (relative to an alternative reference point) resulting in loss chasing behaviors that predicted overall risky choice behaviors. The combined results are discussed in relation to domain-specific versus domain-general subjective reward valuation processes and the potential neural substrates of impulsive and risky choice.

Keywords: impulsive choice; risky choice; discounting; individual differences; rat

Author Note: Kimberly Kirkpatrick, Department of Psychological Sciences, 492 Bluemont Hall, Kansas State University, Manhattan, KS 66506; Andrew T. Marshall, Department of Psychological Sciences, 492 Bluemont Hall, Kansas State University, Manhattan, KS 66506; Aaron P. Smith, Department of Psychology, University of Kentucky, 106B Kastle Hall, Lexington, KY 40506.

Correspondence concerning this article should be addressed to Kimberly Kirkpatrick at kirkpatr@ksu.edu.

---

Impulsive choice is measured by presenting a choice between a smaller reward that is available sooner (the SS) and a larger reward that is available later (the LL). Thus, the impulsive choice paradigm pits reward magnitude against delay to reward by essentially asking whether an individual is willing to wait longer to receive a better outcome (Mazur, 1987, 2007). Impulsive choice is indicated by preferences for the SS, particularly when those choices lead to less overall reward earning, and are thus maladaptive, whereas choices of the LL (when it is more objectively valuable) are indicative of greater self-control. Individual differences in impulsive choice are associated with numerous maladaptive behaviors and disorders such as: attention deficit hyperactivity disorder (ADHD; Barkley, Edwards, Laneri, Fletcher, & Metevia, 2001; Solanto et al., 2001; Sonuga-Barke, 2002; Sonuga-Barke, Taylor, Sembi, & Smith, 1992), pathological gambling (Alessi & Petry, 2003; MacKillop et al., 2011; Reynolds, Ortengren, Richards, & de Wit, 2006), obesity (Davis, Patte, Curtis, & Reid, 2010), and substance abuse (Bickel & Marsch, 2001). Additionally, impulsive choice has also been posited as a primary risk factor (MacKillop et al., 2011; Verdejo-García, Lawrence, & Clark, 2008) and predictor of treatment outcomes (Broos, Diergaarde, Schoffelmeer, Pattij, & DeVries, 2012; Krishnan-Sarin et al., 2007; Yoon et al., 2007) for drug abuse.

Risky choice behavior has traditionally been studied by giving individuals repeated choices between a certain, smaller reward and a risky, larger reward (Mazur, 1988; Rachlin, Raineri, & Cross, 1991). The risky outcome usually consists of a larger reward that occurs with some probability, including the possibility of gaining no reward. For example, a rat could be offered a choice between receiving 2 pellets 100% of the time (the certain, smaller option) versus 4 pellets 50% of the time (the larger, risky option), with the possibility of gaining 0 pellets the other 50% of the time. Thus, the risky choice paradigm pits amount of reward against probability (or risk) of reward omission by essentially asking how much risk will an individual endure to receive a better reward. As the probability of receiving the risky reward decreases, it is chosen less often; this process is known as probability discounting (Rachlin et al., 1991) and has been demonstrated in both human and nonhuman animals (e.g., Mazur, 1988; Myerson, Green, Hanson, Holt, & Estle, 2003). Individual differences in risky choice behavior are related to cigarette smoking (Reynolds, Richards, Horn, & Karraker, 2004) and pathological gambling (Madden, Petry, & Johnson, 2009; Myerson et al., 2003). Specifically, gamblers discount probabilistic rewards less steeply than control subjects (Holt, Green, & Myerson, 2003; Madden et al., 2009; also see Weatherly & Derenne, 2012) and continue to make risky choices despite the experience of repeated losses (Linnet, Røjskjær, Nygaard, & Maher, 2006). Accordingly, a thorough understanding of the mechanisms driving individual differences offers critical insight into questions such as why some individuals continue to gamble despite having experienced a series of consecutive losses (Rachlin, 1990).

Recently, much of the work from our laboratory has been focused on the assessment of individual differences in impulsive and risky choice and the underlying cognitive and neural mechanisms in rats (Galtress, Garcia, & Kirkpatrick, 2012; Garcia & Kirkpatrick, 2013; Kirkpatrick, Marshall, Clarke, & Cain, 2013; Kirkpatrick, Marshall, Smith, Koci, & Park, 2014; Marshall & Kirkpatrick, 2013, 2015; Marshall, Smith, & Kirkpatrick, 2014; Smith, Marshall, & Kirkpatrick, 2015), which will be the primary focus of this review. Here, we will discuss mechanisms of impulsive and risky choice and their relationship. Within each section we will describe factors that influence the nature of individual differences and moderators of those individual differences to provide a potential window into the underlying cognitive mechanisms. These moderators include genetic factors, early rearing environment, and relative subjective reward valuation manipulations. Finally, we will close by discussing possible neural mechanisms within the domain-specific and domain-general reward valuation systems to provide a possible framework for interpreting and integrating the results of the different manipulations of impulsive and risky choice and their role in individual differences.

Mechanisms of Impulsive Choice

Traditionally, impulsive choice has been interpreted within the theoretical framework of delay discounting (Mazur, 1987). Delay discounting refers to the phenomenon in which a temporally distant reward is subjectively devalued due to its delayed occurrence. This loss of subjective value can be modeled using Equation 1:

    (1)

in which V refers to a reward’s subjective value that is determined by A, the reward’s objective amount, divided by D, the delay to the reward, and k, the discounting parameter that has been proposed as an individual difference variable (Odum, 2011a). We have adopted a somewhat different focus of viewing amount and delay not as objective parameters, but as subjective ones, consistent with a long history of research on the psychophysics of amount and delay perception. Specifically, differences in the perception of or sensitivity to amount, delay, or their interaction may influence impulsive choice behavior. Accordingly, we have employed multiple tasks to investigate individual differences in both reward amount/magnitude sensitivity (e.g., reward magnitude discrimination) and temporal sensitivity (e.g., temporal bisection), as more thoroughly described below. Finally, we have examined stable individual differences in impulsive choice across various experimental manipulations. For these analyses, we have parsed out measures of bias and sensitivity, which are both captured by k-values in Equation 1. Bias in impulsive choice is measured using the mean choice across several parameters, which provides an index of overall preference for one outcome over another. Alternatively, the slope of the function assesses sensitivity to changes in choice parameters, which may relate to the adaptability of choice behavior. The slope of the function is an index of how much individuals change their choice behavior when there is a change in delay or magnitude of one of the options. Bias (mean choice) and sensitivity (the slope of the choice function) usually have little to no correlation, indicating that they may be orthogonal measures of behavior.

Individual Differences

Several studies have examined individual differences in choice behavior in rats, discovering that rats exhibit substantial individual differences that are stable across different choice parameters (Galtress et al., 2012; Garcia & Kirkpatrick, 2013). More recently, we examined timing and reward processing differences as potential correlates of individual differences in impulsive choice. Marshall, Smith, and Kirkpatrick (2014) trained rats using a procedure adapted from Green and Estle (2003) with manipulations of the SS delay while also assessing timing and delay tolerance in separate tasks. The SS was 1 pellet after either 30, 10, 5, or 2.5 s across phases, and the LL was 2 pellets after 30 s. The rats were subsequently tested on a temporal bisection task (Church & Deluty, 1977) to examine individual differences in temporal discrimination. In this task, a houselight cue lasted either 4 or 12 s, after which two levers were inserted into the box corresponding to the ‘short’ or ‘long’ duration levers; food was delivered for correct responses. After the rats had achieved 80% accuracy, they received test sessions in which the houselight was illuminated for 4, 5.26, 6.04, 6.93, 7.94, 9.12 and 12 s. This procedure yields ogive-shaped psychophysical functions. Each individual rat’s psychophysical function was fit with a cumulative logistic function and the parameters of the mean (a measure of timing accuracy) and the standard deviation (a measure of timing precision) of the function were determined. Finally, the rats completed a progressive interval (PI) task to examine individual differences in delay tolerance. The rats received PI schedules of 2.5, 5, 10, and 30 s. In the PI schedule, the delay for the first reward is equal to the PI (e.g., 2.5 s) and then increases by the PI duration for each successive reward (e.g., 5, 7.5, 10, etc.). If the rat ceased responding for 10 min, then the last PI completed is recorded as the breakpoint. Longer breakpoints should be indicative of greater delay tolerance.

The results, shown in Figure 1, disclosed strong individual differences in all three tasks consistent with our previous studies. In impulsive choice, the rats decreased their impulsive choices as the delay to the SS increased, but the rats that were more impulsive with the shorter delay generally remained more impulsive. In the bisection task, the percentage of long responses increased with the stimulus duration, and the psychophysical functions showed the characteristic ogive form. However, there were substantial individual differences, with some rats displaying much steeper psychophysical functions than others; the steeper psychophysical functions are associated with lower standard deviations. In the PI task, the breakpoints increased as the PI duration increased, and again there were fairly substantial and stable individual differences. Assessments of internal reliability using a Cronbach’s alpha test, which measures the cross-correlation of multiple observations, revealed moderate to strong consistency in impulsive choice (a = .91), bisection (a = .73) and PI (a = .68) tasks. This indicated that the rats were generally consistent in their behaviors when tested across parameters in each task.

An examination of the correlation of individual differences across tasks revealed a significant positive correlation (r = .73) between the standard deviation of the bisection function (a measure of timing precision) and the mean of the impulsive choice function (a measure of choice bias) and a negative correlation (r = −.63) between the PI breakpoint (a measure of delay tolerance) and mean impulsive choice. These relationships, diagrammed in Figure 2, each accounted for approximately half of the variance in choice behavior. There also was a negative correlation between the bisection standard deviation and the PI breakpoint (r = −.59). The correlational pattern indicates that the rats with more precise timing (steeper bisection psychophysical functions) and greater delay tolerance (later breakpoints) showed greater LL preference (self-control) in the impulsive choice task. Due to the correlational nature of these results, we cannot determine whether timing precision, delay tolerance, and/or self-control possess causal relationships, but some additional recent work from our laboratory examining time-based interventions to improve self-control suggests that timing processes may have a causal relationship with impulsive choice (Smith et al., 2015).

In addition to examining the potential role of timing processes in impulsive choice, Marshall et al. (2014) also examined reward magnitude sensitivity in a separate group of rats. The magnitude group was tested on an impulsive choice task in which the SS delivered 1 pellet after 10 s, and the LL delivered either 1, 2, 3, or 4 pellets after 30 s across phases. The rats then completed a reward magnitude sensitivity task where each lever delivered reinforcement on a random interval (RI) 30 s schedule. The small lever always delivered 1 pellet and the large lever delivered 1, 2, 3, or 4 pellets across phases. Discrimination ratios were calculated using the rats’ response rates to determine whether greater responding occurred on the LL lever when it delivered greater magnitudes. Finally, the magnitude group completed a progressive ratio (PR) 3 task where the response requirement began at 3 and increased by 3 responses per reward earned. The PR3 delivered 1, 2, 3, or 4 pellets of food across phases and a breakpoint was determined for each magnitude. The PR task is frequently used in behavioral economics as a measure of motivation to work for different rewards (e.g., Richardson & Roberts, 1996), and in this case provided an assessment of motivation to work for different magnitudes of reward. The results again showed strong and stable individual differences in the impulsive choice (a = .86), reward magnitude discrimination (a = .80), and PR (a = .85) tasks. However, the only significant correlation was between the PR breakpoint and the magnitude discrimination ratio (r = −.72), but neither measure correlated with impulsive choice behavior (data not shown).

Overall, this study, coupled with the results from our previous studies (Galtress et al., 2012; Garcia & Kirkpatrick, 2013), indicated stable and substantial individual differences in rats, suggesting that impulsive choice may be a trait variable in rats similar to what has been shown in humans (Jimura et al., 2011; Kirby, 2009; Matusiewicz, Carter, Landes, & Yi, 2013; Odum, 2011a, 2011b; Odum & Baumann, 2010; Ohmura, Takahashi, Kitamura, & Wehr, 2006; Peters & Büchel, 2009). In addition, timing processes may exhibit stronger control over impulsive choice than reward magnitude processes (Marshall et al., 2014), but further research will be needed to verify that possibility. The correlations between timing and choice behavior do, however, corroborate other studies showing that more impulsive humans tend to overestimate interval durations (Baumann & Odum, 2012) and display poorer temporal discrimination capabilities (Van den Broek, Bradshaw, & Szabadi, 1987), and more impulsive rats show greater variability in timing on the peak procedure (McClure, Podos, & Richardson, 2014).

Moderating Impulsive Choice

Strain differences. While much of our work has examined impulsive choice in outbred populations, we have also assessed impulsive choice in inbred strains of rats that are potential animal models of ADHD (Garcia & Kirkpatrick, 2013). The spontaneously hypertensive (SHR) and Lewis strains have been derived from their respective control strains, the Wistar Kyotos (WKY) and Wistars, and both have been reported to demonstrate possible markers of increased impulsive choice in previous studies (Anderson & Diller, 2010; Anderson & Woolverton, 2005; Bizot et al., 2007; Fox, Hand, & Reilly, 2008; García-Lecumberri et al., 2010; Hand, Fox, & Reilly, 2009; Huskinson, Krebs, & Anderson, 2012; Madden, Smith, Brewer, Pinkston, & Johnson, 2008; Stein, Pinkston, Brewer, Francisco, & Madden, 2012).

Garcia and Kirkpatrick (2013) sought to potentially isolate the source of impulsive choice behaviors to either deficits in delay or magnitude sensitivity by delivery of two different impulsive choice tasks modeled after previous research (Galtress & Kirkpatrick, 2010; Roesch, Takahashi, Gugsa, Bissonette, & Schoenbaum, 2007). The four strains of rats were given an impulsive choice task of 1 pellet after 10 s (the SS) or two pellets after 30 s (the LL) to establish a baseline. Subsequently, all rats in each strain experienced an LL magnitude increase to 3 and 4 pellets and an SS delay increase to 15 and 20 s across phases in a counterbalanced order. Additionally, in between the LL magnitude and SS delay phases, all rats returned to baseline.

The WKY and SHR strains were similar in their choice behavior in both tasks (data not shown), suggesting that the SHR strain may not be a suitable model of disordered impulsive choice. While this finding does contrast with some literature (e.g., Fox et al., 2008; Russell, Sagvolden, & Johansen, 2005), our results corroborate other findings that SHR rats do not always show heightened impulsivity across tasks (van den Bergh et al., 2006), with inconsistencies perhaps due to the observation that they are a heterogeneous strain (Adriani, Caprioli, Granstrem, Carli, & Laviola, 2003). The Lewis rats did, however, show greater impulsive choices compared to the Wistar control strain in both tasks with larger effects in the SS delay manipulations (see Figure 3). In addition, the Lewis rats displayed delay aversion that developed over the course of the session in the SS delay manipulation, and this may be an important factor in their increased impulsive choice. These results substantiate the Lewis strain as a possible model for ADHD (see also García-Lecumberri et al., 2010; Stein et al., 2012; Wilhelm & Mitchell, 2009).

Early rearing environment. In addition to genetic moderators, we have also assessed environmental moderators of impulsive choice. In one experiment (Kirkpatrick et al., 2013), rats were split into either an enriched condition (EC) that involved a large cage, several conspecifics, daily handling, and daily toy changes, or an isolated condition (IC) that involved single housing in a small hanging wire cage without any toys or handling. The rats were reared in these conditions from post-natal day 21 for 30 days, after which they were tested on impulsive choice and reward challenge tasks. For the impulsive choice task, the rats were given a choice between 1 pellet after 10 s (SS) or 2 pellets after 30 s (LL). For the reward challenge task, the delay to the SS and LL were both 30 s, but the magnitudes remained at 1 versus 2 pellets. Finally, after completing both tasks, the rats were given a test for impulsive actions using a differential reinforcement of low rates (DRL) schedule with criterion values of 30 and 60 s in separate phases. In the DRL task, the rats had to wait for a duration greater than or equal to the criterion time between successive responses to receive food. Premature responses reset the required waiting time.

The top panel of Figure 4 demonstrates the results from the impulsive choice and reward challenge tasks. The IC rats (red triangles) were slightly more likely to choose the SS in the impulsive choice task. However, the IC rats chose the LL alternative more often in the reward challenge when the SS and LL delays were equal, indicating that the IC rats were more sensitive to the magnitude differences between the two alternatives. An analysis of their latencies to initiate forced choice trials during the impulsive choice task (middle panel of Figure 4) suggested that the isolated rats displayed greater subjective valuation of the SS outcome due to their shorter latencies to initiated SS forced choice trials compared to LL forced choice trials (see Kacelnik, Vasconcelos, Monteiro, & Aw, 2011; Shapiro, Siller, & Kacelnik, 2008 for further information on forced choice latencies as a metric of subjective reward valuation). On the other hand, EC rats demonstrated similar latencies to initiate both SS and LL forced choice trials, suggesting similar subjective valuation of the two options. Finally, in the DRL task, the IC rats were more efficient at earning rewards in the 30-s criterion task, requiring fewer responses to earn rewards (bottom panel of Figure 4), but there were no group differences at 60 s.

This finding was somewhat counterintuitive in that the IC rats tended to be more impulsive in the choice task, but showed more efficient performance in the DRL task, which has been interpreted as less impulsive (Pizzo, Kirkpatrick, & Blundell, 2009). However, both the impulsive choice and DRL findings are consistent with multiple other reports in the literature (Dalley, Theobald, Periera, Li, & Robbins, 2002; Hill, Covarrubias, Terry, & Sanabria, 2012; Kirkpatrick et al., 2014; Marusich & Bardo, 2009; Perry, Stairs, & Bardo, 2008; Zeeb, Wong, & Winstanley, 2013). One potential mechanism that could explain this pattern of results is that the increased reward sensitivity in the IC rats may have produced greater sensitivity to local rates of reward, which would lead to momentary maximizing. This would presumably enhance performance on tasks such as DRL and reward challenge, but would skew subjective reward valuation toward delays associated with higher local rates of reward (i.e., the SS). This hypothesis was further supported by a positive correlation (r = .53) between the reward challenge mean and the responses/reward in the DRL 30 task that is diagrammed in Figure 5. This relationship demonstrates that the rats that performed more poorly on the reward challenge (showing more SS responses) also performed more poorly on the DRL 30 task, suggesting that intrinsic reward sensitivity may be related to the ability to successfully inhibit responding on the DRL task. This pattern is intriguing given that increases in reward magnitude on DRL tasks typically lead to increased impulsivity (Doughty & Richards, 2002). This suggests a possible differentiation between extrinsic reward magnitude changes and intrinsic reward valuation processes that may interact differently with impulsive behaviors. Further research is needed to disentangle the different aspects of reward sensitivity in relation to impulsive choice and impulsive action behaviors.

While there was an indication of increased subjective valuation of the impulsive outcome by IC rats, the findings were only expressed in the latencies on forced choice trials rather than directly in choice behavior. To further assess the potential effects of rearing environment on impulsive choice, Kirkpatrick et al. (2014) compared EC and IC rats’ choice behavior across a wider range of choice parameters. Rats received choices between an SS of 1 pellet after 10 s and an LL of 1, 2, or 3 pellets after 30 s, with LL magnitude manipulated across phases. Under these conditions, differential rearing exerted a significant effect on impulsive choice (Figure 6), corroborating the findings of the previous studies with IC rats displaying greater impulsive choice behaviors.

Additionally, in order to better understand the relationship between reward sensitivity and impulsivity, Kirkpatrick et al. (2013) conducted a second experiment that also included a standard rearing condition (SC) in addition to the IC and EC groups. SC rats were pair-housed and handled daily but were not provided with any novel objects. The rats were presented with the same reward discrimination task used by Marshall et al. (2014) described above with the magnitudes of 1:1, 1:2, 1:3, 2:3, and 2:4 on the small and large levers. In this experiment, the IC rats in the baseline 1:1 condition showed significantly higher response rates to both levers than the SC and EC rats. Additionally, as the large reward increased, the EC and SC rats showed increased responding to both the large and small levers. Even when the small lever magnitude remained at 1 pellet, the EC and SC rats increased their responding on the small lever when the large lever magnitude increased, demonstrating generalization of responding to the small lever. The IC rats, however, did not generalize, and instead showed significantly lower responding on the SS lever compared to SC and EC rats, suggestive of potentially greater reward discriminability (consistent with the previous findings from the reward challenge task).

Overall, the combined findings of the two experiments are consistent with previous research showing that rearing environment moderates the assignment of incentive value to stimuli associated with rewards (Beckmann & Bardo, 2012). The IC rats overall showed greater SS preference in the impulsive choice task and greater valuation of the SS alternative as indicated through their shortened forced choice latencies. The IC rats also, however, showed an increased ability to discriminate between the SS and LL rewards as indicated in their differentiated response rates, showed greater LL preference in the reward challenge task, and showed greater efficiency in the DRL task, another widely used measure of impulsivity. Importantly, environmental enrichment does not seem to affect interval timing within a choice environment (Marshall & Kirkpatrick, 2012), suggesting that differences between EC and IC rats are not driven by enrichment-induced differences in temporal processing. Thus, it appears as though changes in reward discrimination and/or reward sensitivity may explain the rearing condition differences, although further research will be needed to determine the nature of these effects and their relationship with impulsive behaviors.

Mechanisms of Risky Choice

In conjunction with our research on impulsive choice behavior, we have also been examining factors that impact risky choice behaviors (e.g., Kirkpatrick et al., 2014). Risky choice can also be modeled using Equation 1 by substituting odds against reward (q) in place of delay to reward, indicating that subjective value decreases as a function of the odds against reward delivery:

    (2)

This effect is known as probability discounting because it reflects the loss of subjective value that occurs as the probability of a reward decreases (or as the odds against reward increases).

Individual Differences

Mirroring our research on impulsive choice, we have been examining the cognitive mechanisms of risky choice behavior and how individual differences in risky choice may be identified and moderated to alleviate problematic risky decision making behaviors. Previous research has examined sensitivity to reward probability and magnitude as key factors that govern individual differences in risky decision making in humans (see Myerson, Green, & Morris, 2011). Until more recently, one consistent omission from the human choice literature was the absence of decision feedback following different types of choices (see Hertwig & Erev, 2009; Lane, Cherek, Pietras, & Tcheremissine, 2003). Theoretically, as such decisions are neither rewarded nor punished, consecutive choices should be relatively independent. While such independence has been suggested within the animal literature (e.g., Caraco, 1981), humans are indeed affected by whether choices occur in isolation or in succession (see Kalenscher & van Wingerden, 2011; Keren & Wagenaar, 1987). Therefore, in contrast to the traditional molar analyses of reward and probability sensitivity, we have focused on the local influences on choice behavior in terms of the effect of recent outcomes on subsequent choices.

Our first risky choice experiment sought to determine the effects of previous outcomes on risky choice behavior (Marshall & Kirkpatrick, 2013). We offered rats choices between a certain outcome that always delivered either 1 or 3 pellets (p = .5) and a risky outcome that probabilistically delivered either 3 or 9 pellets following each risky choice. Thus, the certain and risky outcomes each involved variable reward magnitudes. The probability of risky-outcome delivery varied across phases: .10, .33, .67, and .90, so that the probability of reward omission was .90, .67, .33, and .10, respectively.

As expected, we observed an increase in risky choice with increases in risky food probability (Figure 7). Similar to our results from impulsive choice tasks (e.g., Marshall et al., 2014), the individual differences were relatively stable across probabilities (a = .68), suggesting that risk-taking, like impulsivity, may be a trait variable in rats. At the local level, we found that the previous outcome of a choice had a significant impact on the subsequent choice made (Figure 8). There was a greater prevalence of risky choices following rewarded risky choices (uncertain-small, U-S, and uncertain-large, U-L) than following reward omission (uncertain-zero, U-Z), indicating win-stay/lose-shift behavior. Moreover, the individual differences in risky choice behavior as a function of previous outcome were stable across outcomes, a = .62, suggesting a relatively consistent choice pattern across previous outcomes. These findings further support the trait nature of risk-taking and indicate that this attribute is present in local choices as well as global choice behavior.

Moderating Individual Differences

Early rearing environment. The stability of individual differences raises the question of whether risky choice behavior can be moderated. As various subpopulations that may be characterized as “unhealthy” exhibit elevated propensities to make risky choices (e.g., Reynolds et al., 2004), early assessment of risky choice tendencies followed by corresponding targeted therapies to reduce such maladaptive behaviors may ultimately attenuate corresponding risk-related substance and behavioral addiction.

One manipulation that has been shown to moderate individual differences in a variety of behavioral paradigms is the rat’s rearing/housing environment (Simpson & Kelly, 2011). Accordingly, we were interested in determining whether environmental rearing moderates risky choice (Kirkpatrick et al., 2014). Rats were reared in EC and IC conditions described above and then tested with a risky choice task from Marshall and Kirkpatrick (2013) with risky food probabilities of .17, .33, .5, and .67. As shown in Figure 9, there was an increase in risky choices as the probability of risky food increased, and there were substantial and stable individual differences in risky choice, but there were no significant differences between rearing conditions. These results stand in contrast to recent research in pigeons using a suboptimal choice task, which have found a decreased speed of attraction to risky behaviors in pigeons reared in enriched environments (Pattison, Laude, & Zentall, 2013) and also research using an analog of an Iowa Gambling Task in rats demonstrating increased risky behavior in IC rats (Zeeb et al., 2013). The source of these differences in results may be due to the different task demands across the studies, but this remains to be determined.

Even though rearing environment did not significantly impact risky choice, several other factors have been hypothesized to affect risky decision making. For example, proposed psychological correlates of risky choice include sensitivity to reward magnitude (Myerson et al., 2011) and probability (Rachlin et al., 1991), the subjective integration of recent rewards with previous computations/expectations of subjective reward value (Sutton & Barto, 1998), and sensitivity to experienced and prospective gains and losses (Kahneman & Tversky, 1979). Therefore, it may be that these factors are potential targets to be addressed in future research.

Manipulations of subjective value. Sensitivity to the objective value of rewards depends on the encoding of outcomes as gains and losses relative to some reference point. Furthermore, as humans have been proposed to be more sensitive to losses than they are to gains (Kahneman & Tversky, 1979), sensitivity to reward magnitude and probability would therefore depend on whether experienced outcomes are regarded as gains or losses. Indeed, individuals will behave considerably differently if they are facing prospective gains or prospective losses (Kahneman & Tversky, 1979; Levin et al., 2012). Thus, the most critical factor in understanding idiosyncrasies in risky choice may be the mechanisms by which individuals encode differential outcomes in a relative fashion as opposed to absolute value encoding.

It has been well established that humans employ subjective criterions known as reference points when they encode and evaluate differential outcomes (e.g., Wang & Johnson, 2012). Specifically, outcomes that are greater than the reference point are gains, and outcomes that are less than the reference point are losses. Until recently, the possibility that nonhuman animals employ some type of reference-point criterion was open for investigation, even though previous reports have considered the possibility that animals may in fact use heuristics in decision making (Marsh, 2002). If reference point use can be determined and subsequently manipulated, it may be possible to effectively optimize decision making across the populations of individuals prone to behave suboptimally.

Accordingly, Marshall (2013) investigated reference point use in rats (also see Bhatti, Jang, Kralik, & Jeong, 2014; Marshall & Kirkpatrick, 2015). We hypothesized that rats may use at least one of three possible reference points to encode risky choice outcomes: the expected value of the risky outcome, the zero-outcome value, or the expected value of the certain outcome. In accordance with linear-operator models of subjective reward valuation (Sutton & Barto, 1998), rats may use the learned expected value of the risky choice, such that outcomes greater than the expected value are gains and outcomes less than the expected value are losses. Alternatively, rats may regard any nonzero outcome as a gain, such that the only loss experienced is that of zero pellets. Last, in reference to research on regret following losses (e.g., Connolly & Zeelenberg, 2002), rats may encode gains and losses relative to what could have been received had a different choice been made (see Steiner & Redish, 2014). Marshall (2013) found that rats appeared to use the expected value of the certain outcome as a reference point for risky choices. In a follow-up study, Marshall and Kirkpatrick (2015), presented rats with a certain choice that delivered an average of 3 pellets (2 or 4, 1 or 5; certain-small, C-S, and certain-large, C-L, outcomes, respectively) and a risky choice that delivered 0 (uncertain-zero), 1 (­uncertain-small), or 11 pellets (uncertain-large). The between-subjects factor was the outcome values associated with certain choices (2 or 4 for Group 2-4, 1 or 5 for Group 1-5) in order to determine whether it was the individual outcome values or the expected value of the certain choice that more greatly drove behavior. The probability of receiving 0 [P(0)] or 1 pellet [P(1)] following a risky choice was manipulated in separate phases, with all rats receiving both manipulations in a counterbalanced order. The probability of zero pellets, P(0), was .9, .5, and .1 and P(1) and P(11) were each equal to .05, .25, and .45, respectively. Similarly, the probability of one pellet, P(1), was equal to .9, .5, and .1 and P(0) and P(11) were each equal to .05, .25, and .45, respectively.

As seen in Figure 10, the P(0) choice function was steeper than the P(1) condition and the individual differences showed good internal reliability across different conditions/probabilities (a = .85), further supporting the trait nature of risk-taking in rats. There were no differences between Groups 2-4 and 1-5 (the data in Figure 10 are collapsed across groups) indicating that the rats were not sensitive to the individual values making up the certain outcome.

As shown previously, rats will make more risky choices after being rewarded for a risky choice than after not being rewarded, a phenomenon known as win-stay/lose-shift behavior (Evenden & Robbins, 1984; Heilbronner & Hayden, 2013; Marshall & Kirkpatrick, 2013). Figure 11 shows the log odds of risky choices following uncertain-zero (U-Z) and uncertain-small (U-S) outcomes in the P(0) and P(1) conditions. In the P(0) condition, the rats were more likely to make risky choices following uncertain-small compared to uncertain-zero outcomes, consistent with win-stay/lose-shift behavior. However, in the P(1) conditions, the rats made more risky choices following the uncertain-zero outcome than the uncertain-small outcome (i.e., a violation of win-stay/lose-shift behavior). This behavior is indicative of elevated loss chasing following the zero outcomes (i.e., a tendency to make risky choices following risky losses; see Linnet et al., 2006), and may relate to a relative subjective devaluation of the 1-pellet outcome when it is the source of the probability manipulation.

In addition, we also found a relationship between the local choice behavior and overall choice behavior in the P(1) condition that suggested a possible role of the loss chasing behavior in overall risky choices. We assessed whether loss chasing tendency [i.e., making more risky choices following an uncertain-zero than an uncertain-small outcome in the P(1) condition] was related to overall risky choice behavior. For this analysis, we subtracted post uncertain-small risky choice behavior from post uncertain-zero risky choice behavior and correlated this difference score with overall risky choice behavior in the P(1) condition. As seen in Figure 12, while the majority of the rats made more risky choices following uncertain-zero than following uncertain-small outcomes (the loss chasers), the loss-averse rats made more risky choices following uncertain-small than uncertain-zero outcomes; the loss-averse rats also were less likely to exhibit risky choices overall. The results suggest that the rats that were riskier were also those that were more likely to chase losses (i.e., make more risky choices after uncertain-zero than uncertain-small outcomes).

These results may have implications for understanding why some individuals continue to gamble (i.e., make risky choices) despite the experience of repeated losses while other individuals do not (see Rachlin, 1990), but further research is needed to verify this possibility. For example, if there are individual differences in reference point use, then such differences could predict which outcomes are regarded as gains and losses. Specifically, if an individual regards a wider variety of outcomes as gains, then subsequent win-stay behavior would be greater, compared to an individual who is more conservative with his or her gain/loss distinctions. Thus, it is possible that the onset of pathological gambling in some individuals, but not others, may be at least partially caused by individual differences in reference point use or the subjective weighting of different reference points (see Linnet et al., 2006). Ultimately, moderating individual differences via reference point use may be the critical factor in adjusting subjective tendencies to make too many (or not enough) risky choices. This could be a fruitful area for further research on individual differences in the onset of gambling behavior.

Correlations of Impulsive and Risky Choice

As discussed above, impulsive and risky choice behaviors have been identified as potential trait variables in humans (Jimura et al., 2011; Kirby, 2009; Matusiewicz et al., 2013; Odum, 2011a, 2011b; Odum & Baumann, 2010; Ohmura et al., 2006; Peters & Büchel, 2009) and in rats (Galtress et al., 2012; Garcia & Kirkpatrick, 2013; Marshall et al., 2014). In addition, the fact that these stable individual differences have been identified as predictors of substance abuse and pathological gambling (e.g., Bickel & Marsch, 2001; Carroll, Anker, & Perry, 2009; de Wit, 2008; Perry & Carroll, 2008) suggests that there may be a correlation between impulsive and risky behaviors.

Individual Differences

The few examinations of correlations in individual differences in impulsive and risky choice have revealed inconsistent results, with weak to moderate correlations in humans (Baumann & Odum, 2012; Myerson et al., 2003; Peters & Büchel, 2009; Richards, Zhang, Mitchell, & De Wit, 1999), and moderately strong correlations in pigeons (Laude, Beckman, Daniels, & Zentall, 2014), but to our knowledge no observations had been undertaken in rats. In addition, the studies in humans have used varying methods, which may be a source of the discrepancies in results. The recent study by Kirkpatrick et al. (2014), discussed above in the early rearing environment sections, sought to rectify this issue. Rats were trained on both impulsive and risky choice tasks and assessments of the correlations of their behavioral patterns were conducted. For the impulsive choice task, the rats were given a choice between an SS of 1 pellet after a 10-s delay versus an LL of 1, 2, or 3 pellets after a 30-s delay, with LL magnitude manipulated across phases. For the risky choice task, rats were given a choice between a certain outcome that averaged 2 pellets (p = 1) and a risky outcome that delivered an average of 6 pellets with a probability of .17, .33, .5 or .67 across phases. The delay to reward was 20 s for both certain and risky outcomes. Figure 13 displays the results from the two tasks for the individual rats. (Note that these are the same data from Figures 6 and 9, but here plotted collapsed across rearing condition to highlight the relationship.) There were substantial individual differences in choice behavior across the rats. In addition, there were 8 rats (red dots) that represented a subpopulation that were “impulsive and risky,” or I/R rats.

To assess the relationship between impulsive and risky choices, two measures were extracted from the choice functions for each rat: (a) the mean overall log odds impulsive and risky choices as an index of bias; and (b) the slope of the choice functions as a measure of sensitivity. As seen in Figure 14, there was a strong positive relationship between the mean choice on the two tasks (r = .83), indicating that the rats that were the most impulsive (high impulsive mean choices) were also the most risky (high risky mean choices). The 8 I/R rats displayed a strong co-occurrence in the mean choice, indicating a clear convergence in their choice biases. The choice correlation is higher than most reports in the literature (but see Laude et al., 2014), and may be due to a notable difference in methodology, which is the use of delayed reward deliveries in the risky choice task. This was designed to engage anticipatory processes between the time of choice and food delivery in risky choice that would mimic those processes in impulsive choice. This is particularly important for promoting the activation of brain areas such as the nucleus accumbens core (NAC) that are believed to be more heavily involved in processing delayed rewards (Cardinal, Pennicott, Sugathapala, Robbins, & Everitt, 2001), as discussed below. There also was a significant positive relationship (r = .68) between the slope of the impulsive and risky choice functions, indicating that the rats that were the most sensitive to changes in choice parameters in one task were generally more sensitive to changes in the other task (Figure 14, bottom panel). Interestingly, only one I/R rat displayed poor sensitivity in both tasks in the face of changing parameters, indicating that response perseveration is unlikely to serve as the sole explanation for the biases in their choice behavior. When these rats were confronted with more extreme choice parameters, they did often change their behavior (see also Figure 13). Also note that a simple “choose larger” or “choose smaller” bias cannot explain the relationship in Figure 14 because high impulsive mean scores were associated with the smaller outcome, whereas high risky mean scores were associated with the larger outcome.

Understanding the patterns of individual differences is an important and relatively overlooked area of research. The rats in Figure 14 varied in their patterns, with some showing the I/R co-occurrence pattern, some showing deficits in impulsive or risky choice alone, and some showing low levels of impulsive and risky choices. By understanding the factors that uniquely affect impulsive and risky choice and factors that drive correlations, we can potentially gain deeper insights into processes that produce vulnerabilities to different disease patterns. For example, drug abuse and other addictive behaviors (e.g., gambling) are associated with deficiencies in both impulsive and risky choice (e.g., de Wit, 2008; Kreek, Nielsen, Butelman, & LaForge, 2005; Perry & Carroll, 2008), suggesting that addictive diseases may emerge from shared neural substrates. In contrast, obesity appears to be primarily associated with disordered impulsive choice (Braet, Claus, Verbeken, & Van Vlierberghe, 2007; Bruce et al., 2011; Duckworth, Tsukayama, & Geier, 2010; Nederkoorn, Braet, Van Eijs, Tanghe, & Jansen, 2006; Nederkoorn, Jansena, Mulkensa, & Jansena, 2007; Verdejo-Garcia et al., 2010; Weller, Cook, Avsar, & Cox, 2008). Understanding the behavioral phenotypes that may predict different disease patterns is particularly important because individual differences in traits such as impulsive choice are expressed at an early age and remain relatively stable during development (e.g., Mischel et al., 2011; Mischel, Shoda, & Rodriguez, 1989). Identifying causes of impulsive and risky choice could potentially lead to opportunities for early interventions to moderate individual differences in these traits and potentially mitigate later disease development. These efforts are in their early stages, and the picture is still developing, so the understanding of factors involved in individual differences in impulsive and risky choice will undoubtedly evolve over time.

Moderating Individual Differences in Impulsive and Risky Choice

Early rearing environment. Due to the paucity of research on the correlation of impulsive and risky choice, there is relatively poor understanding of potential moderators of the correlations. As reviewed in previous sections, environmental rearing has been emerging as a possible moderator of impulsive choice. However, it is not clear whether rearing environment would moderate the correlation between impulsive and risky choice, an issue that was examined in the individual differences analyses by Kirkpatrick et al. (2014) and their interaction with the rearing environment manipulations featured in Figures 6 and 9. As noted in previous sections, isolation rearing relative to enriched rearing increased impulsive choice, but had no effect on risky choices. In addition, rearing environment did not appear to moderate the individual differences correlations as these were still intact when collapsing across rearing condition in the analysis above (Figure 14) and also when examining the correlations within each rearing group (EC: r = .87, IC: r = .91, for impulsive–risky mean correlations). This suggests that rearing environment did not moderate the relationship between impulsive and risky choice and instead exerted its effects solely on impulsive choice.

Domain-General and Domain-Specific Valuation Processes

Individual differences in impulsive and risky choice most likely operate through domain-specific processes involved in probability, magnitude, and delay sensitivity, along with domain-general processes involved in overall reward value computations, incentive valuation, and action valuation. The general idea of domain-general versus domain-specific processes has been applied to a wide range of cognitive processes, but only more recently have these concepts been invoked to explain impulsive and risky choice by Peters and Büchel (2009). Their exposition of these processes was relatively limited, so we attempt to expand on this general idea here by providing a general conceptualization of these processes (see Figure 15). The proposed model is derived from a range of cognitive, behavioral, and neurobiological evidence related to impulsive and risky choice, to expand on the original idea proposed by Peters and Büchel.

Domain-specific processes refer to specialized cognitive processes that operate within a restricted cognitive system. Most likely, there are separate domain-specific processes for determining probability, magnitude, and delay to reward. An example of domain-specific processes related to impulsive choice is the observation of the relationship between timing processes and impulsive choice and the possible role of poor timing processes in promoting delay aversion and potentially amplifying impulsive choices. With risky choice, examples of domain-specific processes include sensitivity to the previous outcome and its effects on subsequent choice behavior, and sensitivity to relative outcomes (gains versus losses). These effects most likely reflect the role of domain-specific processes involved in processing information about reward omission and/or the magnitude of the rewards delivered in risky choice tasks. Domain-specific processes may also explain divergences between impulsive and risky choice (Green & Myerson, 2010). For example, variations in the magnitude of reward in monetary discounting tasks in humans produce opposite effects: in impulsive choice, smaller amounts are discounted more steeply, whereas in risky choice, smaller amounts are discounted less steeply (Green & Myerson, 2004). These patterns may reflect differences in the way that magnitude information is processed within impulsive and risky choice tasks.

Domain-general processes are cognitive processes that result in global knowledge that has an impact on a wide range of behaviors. We propose that there is a domain-general system that includes three components related to the overall value of the outcomes in impulsive and risky choice. (a) Overall reward value is the subjective value that an individual subscribes to an outcome, and this encompasses information about the delay, magnitude, and probability of reward within impulsive and risky choice tasks. Overall reward value computation in impulsive and risky choice is often proposed to follow the hyperbolic rule (Mazur, 2001; Myerson et al., 2011) given in Equations 1 and 2, with higher k-values resulting in steeper decay rates as a function of delay or odds against receipt of reward (Green, Myerson, & Ostaszewski, 1999; Myerson & Green, 1995; Odum, 2011a, 2011b; Odum & Baumann, 2010; Peters, Miedl, & Büchel, 2012). Assuming that perception of the inputs is veridical, then overall reward value would be determined by the k-value. For example, 2 pellets in 10 s for an individual rat with a k-value of .5 would have an overall subjective reward value of .33 [2 pellets / (1 + .5 ⋅ 10 s). However, it is possible that amount and/or delay could be misperceived in the domain-specific processing stage, in which case A and D would not be veridical in Equation 1, and this would provide an additional source of variation in the computation of the overall reward value. A number of results presented in this review indicate that A and D are not veridical, so these are likely to serve as factors in individual differences in overall reward value computations. (b) Overall reward value is proposed to be transformed into an incentive value signal, which encodes the hedonic properties of the outcome, which will drive the motivation to perform behaviors to receive the outcome. The key addition here is that the value of the reward in terms of its overall value is transformed into a motivationally relevant signal that will affect the desire for obtaining that reward. This provides an opportunity for the incentive motivational state of the animal to impose additional effects on choice behavior. For example, the overall reward value may be .33 for both a hungry rat and a sated rat, but the hungry rat will be more likely to work to obtain that outcome. This allows for factors such as energy budget to have an impact on choice behavior (Caraco, 1981). (c) An action value (or decision value) signal reflects the expected utility from gaining access to the outcome, and this will ultimately determine output variables such as impulsive and risky choice behavior. There is growing evidence from neuroimaging studies that the choice response value is encoded through distinct mechanisms from the value of the outcome (either overall reward value or incentive value; Camille, Tsuchida, & Fellows, 2011; Kable & Glimcher, 2009), indicating that action values carry their own unique significance. The inclusion of action value in the model also allows for explanation of phenomena such as framing effects (Marsh & Kacelnik, 2002), where choices are affected by other outcomes in the absence of any direct effects on overall reward value or incentive value processes.

There is evidence to support these three valuation processes as separate aspects of the domain-general system that may be subsumed by different neural substrates (Camille et al., 2011; Kable & Glimcher, 2007, 2009; Kalivas & Volkow, 2005; Lau & Glimcher, 2005; Rushworth, Kolling, Sallet, & Mars, 2012; Rushworth, Noonan, Boorman, Walton, & Behrens, 2011). The domain-general system is a primary target for understanding correlations between impulsive and risky choice. This system would also play a role in impulsive and risky choice in relation to general motivational and subjective valuation processes invoked by those tasks.

While our understanding of subjective valuation processes is still relatively in its infancy, there is sufficient understanding of the valuation network to speculate on the possible mechanisms that may drive individual differences in impulsive and risky choice. The following sections integrate information from a variety of methods (neuroimaging, lesions, and electrophysiology) and species (humans, primates, and rodents) to provide as complete a picture as possible given the current gaps in knowledge.

Domain-Specific Brain Mechanisms

Impulsive and risky choice tasks pit reward magnitude against delay and certainty of reward, respectively. Impulsive choice uniquely relies on delay processing, and interval timing processes have been implicated as playing an important role in impulsive choice (Baumann & Odum, 2012; Cooper, Kable, Kim, & Zauberman, 2013; Cui, 2011; Kim & Zauberman, 2009; Lucci, 2013; Marshall et al., 2014; Takahashi, 2005; Takahashi, Oono, & Radford, 2008; Wittmann & Paulus, 2008; Zauberman, Kim, Malkoc, & Bettman, 2009). The dorsal striatum (DS) is a key target for timing processes as it has been proposed to function as a “supramodal timer” (Coull, Cheng, & Meck, 2011) that is involved in encoding temporal durations (Coull & Nobre, 2008; Matell, Meck, & Nicolelis, 2003; Meck, 2006; Meck, Penney, & Pouthas, 2008). Thus, one would expect that poor functioning of the DS may be responsible for promoting impulsive choice behavior through the route of increased variability in timing (which may operate to decrease delay tolerance). There is no direct evidence to support this supposition, so future research should examine this possibility.

Risky choice, however, uniquely relies on reward omission and reward probability sensitivity. Here, the basolateral amygdala (BLA) contributes to the encoding of an omitted reward (Frank, 2006), so this structure is a likely candidate for processing reward probability information that contributes to the overall reward value computations in risky choice. Thus, the BLA should presumably play an important role in sensitivity to the previous outcomes, a possibility that remains to be tested.

Impulsive and risky choice also should conjointly rely on structures involved in reward magnitude processing, as reward magnitude is involved in overall reward value determination in both tasks. The BLA is involved in processing sensory aspects of rewards (Blundell, Hall, & Killcross, 2001), so this structure should be considered as a potential candidate for producing individual differences in reward magnitude sensitivity that would be relevant in both tasks. The orbitofrontal cortex (OFC) is also involved in encoding reward magnitude (da Costa Araújo et al., 2010), so this is another potential candidate structure for reward magnitude processing that could affect performance on both impulsive and risky choice tasks.

Domain-General Brain Mechanisms

The domain-general reward valuation processes, the key structures, and their specific roles are not very well understood. Based on the current literature, choices are likely driven by a determination of the action value of different outcomes, with a comparison of the action values resulting in the final choice (e.g., Lim, O’Doherty, & Rangel, 2011; Shapiro et al., 2008). Overall reward value computations are formed by integrating reward magnitude, delay, and/or probability (see Figure 15). Mounting evidence indicates that overall reward value is determined by the mesocorticolimbic structures, particularly the medial pre-frontal cortex (mPFC) and nucleus accumbens core (NAC; Peters & Büchel, 2010; Peters & Büchel, 2009). The NAC may be involved in the assignment of the overall value of rewards (Galtress & Kirkpatrick, 2010; Olausson et al., 2006; Peters & Büchel, 2011; Robbins & Everitt, 1996; Zhang, Balmadrid, & Kelley, 2003). As a result, it has been proposed as a possible target site for the integration of domain-specific information into an overall reward value signal (Gregorios-Pippas, Tobler, & Schultz, 2009; Kable & Glimcher, 2007). This idea is consistent with the importance of the NAC in choice behavior (Basar et al., 2010; Bezzina et al., 2008; Bezzina et al., 2007; Cardinal et al., 2001; da Costa Araújo et al., 2009; Galtress & Kirkpatrick, 2010; Kirkpatrick et al., 2014; Pothuizen, Jongen-Relo, Feldon, & Yee, 2005; Scheres, Milham, Knutson, & Castellanos, 2007; Winstanley, Baunez, Theobald, & Robbins, 2005).

The mPFC has been implicated in the representation of reward incentive value (Peters & Büchel, 2010; Peters & Büchel, 2011), the encoding of the magnitude of future rewards (Daw, O’Doherty, Dayan, Seymour, & Dolan, 2005), the determination of positive reinforcement values (Frank & Claus, 2006), the processing of immediate rewards (McClure, Laibson, Loewenstein, & Cohen, 2004), and the determination of cost and/or benefit information (Basten, Heekeren, & Fiebach, 2010; Cohen, McClure, & Yu, 2007). The orbitofrontal cortex (OFC) contributes to both impulsive and risky choice (da Costa Araújo et al., 2010; Mobini et al., 2002) and is a candidate for encoding the action value of a choice (Hare, O’Doherty, Camerer, Schultz, & Rangel, 2008; Kable & Glimcher, 2009; Kringlebach & Rolls, 2004; Peters & Büchel, 2010; Peters & Büchel, 2011; Schoenbaum, Roesch, Stalnaker, & Takahashi, 2009). While these structures form only a portion of the reward valuation system, they are likely to play a central role in the determination of domain-general valuation of rewards that guides impulsive and risky choice behavior and should drive the correlations.

Individual Differences Correlations

There has generally been little emphasis on neural correlates of individual differences in impulsive and risky choice in animals, particularly with regard to correlations between impulsive and risky choice. However, Kirkpatrick et al. (2014) recently examined the correlation of monoamines (norepinephrine, epinephrine, dopamine, and serotonin) and their metabolites in the NAC and mPFC with individual differences in impulsive and risky choice as a function of environmental rearing conditions. There were no effects of rearing condition on the neurotransmitter concentrations, and there were no correlations of mPFC monoamine concentrations with impulsive or risky choice behavior, but there were several significant correlations between NAC monoamine/metabolite concentrations and impulsive or risky choice behavior. The key neurotransmitter/metabolites were norepinephrine (NE) and serotonin (5-HT) and its metabolite 5-hydroxyindoleacetic acid (5-HIAA). The relationships between serotonergic concentrations and choice behavior are shown in Figure 16. For impulsive choice behavior, NAC 5-HIAA concentrations were positively correlated with the impulsive slope (r = .55) and for risky choice behavior, NAC 5-HT (r = −.43) concentrations were negatively correlated with the risky mean. Thus, serotonin turnover, an indicator of activity, was related to sensitivity in impulsive choice (measured by the slope) with lower concentrations leading to lower sensitivity. In risky choice, basal 5-HT levels were related to the risky mean (a measure of choice bias) indicating that rats with lower 5-HT levels were more risk prone. The I/R rats were generally characterized by lower basal 5-HT levels and lower metabolite concentrations suggesting that high levels of impulsive and risky choice may be driven by deficient serotonin homeostasis and metabolic processes in the NAC. In addition, individual rats displayed similar patterns with NE concentrations, which were negatively correlated with the impulsive mean (r = −.44), positively correlated with the impulsive slope (r = .45), and negatively correlated with the risky mean (r = −.44), as shown in Figure 17. The I/R rats, shown in red, demonstrated generally lower NE concentrations that were associated with higher impulsive and risky means and less sensitivity in impulsive choice. NE concentrations are generally indicative of arousal levels (e.g., Harley, 1987), and the results suggest that the more impulsive and risky rats may suffer from hypoactive arousal levels which could impact on incentive motivational valuation processes.

In relating the results in Figures 16 and 17 to the conceptual model in Figure 15, it is possible that NAC 5-HT and NE homeostatic and metabolic processes may be playing an important role in incentive value processes, which has been previously suggested particularly for serotonergic activity (Galtress & Kirkpatrick, 2010; Olausson et al., 2006; Peters & Büchel, 2011; Robbins & Everitt, 1996; Zhang et al., 2003). This suggests that the NAC is a potential source for domain-general reward valuation (Peters & Büchel, 2009), and could drive the correlations between impulsive and risky choice. Cleary, the NAC should be examined more extensively to understand the nature of its involvement in choice behavior, particularly in promoting correlations between impulsive and risky choice.

Summary and Conclusions

The present review has discussed a number of factors involved in individual differences in impulsive and risky choice and their correlation. Due to the importance of these traits as primary risk factors for a variety of maladaptive behaviors, understanding the factors that may produce and moderate individual differences is a critical problem. While our research on this subject is still in its early stages, we have discovered a few important clues to the cognitive and neural mechanisms of impulsive and risky choice.

In impulsive choice, timing and/or delay tolerance may be an important underlying determinant in both outbred and also Lewis rat strains, suggesting that deficient timing processes should be further examined as a potential causal factor in producing individual differences in impulsive choice. In addition, it appears that reward ­sensitivity/discrimination may be a factor in impulsive choice as well, due to the joint effect of isolation rearing (relative to enriched rearing) in promoting both reward discrimination and also increasing impulsive choices. Therefore, reward sensitivity should be further examined as a factor for interventions to decrease impulsive choices.

In risky choice, recent outcomes appear to play an important role in choice behavior and sensitivity to those outcomes may be a key variable in producing individual differences in risky choice behavior. In addition, rats appear to use the certain outcome as a reference point, gauging uncertain outcomes as gains versus losses relative to the certain outcome. This suggests that absolute reward magnitudes may be less important in risky choice. While absolute value may be less important, the relative subjective valuation processes in risky choice induced loss chasing when the probability of nonzero losses of one pellet was manipulated directly. Loss chasing predicted greater overall risky choices, suggesting that loss chasing may play a role in overall risky choice biases. Further research should examine loss chasing as a potential causal factor in risky choice behaviors.

Finally, an examination of the pattern of impulsive and risky choice revealed strong correlational patterns between impulsive and risky choice across the full spectrum of individual differences. As a result of the strong correlation, approximately one third of the rats (the I/R rats) demonstrated overly high impulsive and risky choices. The correlation of impulsive and risky choice was not moderated by environment rearing, which is not surprising given the lack of effects of rearing environment on risky choice behaviors. Further research should aim to determine factors that moderate the correlation between impulsive and risky behaviors. The examination of neurobiological correlates of impulsive and risky choice suggest possible targets of domain-general processes involved in subjective overall reward and incentive valuation in structures such as the NAC.

While the present review only provides some preliminary insights into the mechanisms of impulsive and risky choice and their correlational patterns, the consideration of the conceptual model in Figure 15 may provide an initial framework for interpreting these results and for motivating further work. The parsing out of domain-general versus domain-specific factors can provide a means of understanding both the shared (domain-general) and unique (domain-specific) processes involved in impulsive and risky choice. While this conceptual model will undoubtedly undergo some degree of metamorphosis as our understanding grows, the focus on domain-general and domain-specific factors is likely to motivate a plethora of future research in this area.

Acknowledgements

The authors would like to thank various members of the Reward, Timing, and Decision laboratory and our collaborators both past and present who contributed to this research including Mary Cain, Jacob Clarke, Tiffany Galtress, Ana Garcia, Juraj Koci, and Yoonseong Park. The research summarized in this article was supported by NIMH grant R01-MH085739 awarded to Kimberly Kirkpatrick and Kansas State University. Publication of this article was funded in part by the Kansas State University Open Access Publishing Fund.

References

Adriani, W., Caprioli, A., Granstrem, O., Carli, M., & Laviola, G. (2003). The spontaneously hypertensive-rat as an animal model of ADHD: Evidence for impulsive and nonimpulsive subpopulations. Neuroscience Biobehavioral Reviews, 27, 639–651. doi:10.1016/j.neubiorev.2003.08.007

Alessi, S. M., & Petry, N. M. (2003). Pathological gambling severity is associated with impulsivity in a delay discounting procedure. Behavioral Processes, 64, 345–354. doi:10.1016/S0376-6357(03)00150-5

Anderson, K. G., & Diller, J. W. (2010). Effects of acute and repeated nicotine administration on delay discounting in Lewis and Fischer 344 rats. Behavioural Pharmacology, 21(8), 754–764. doi:10.1097/FBP.0b013e328340a050

Anderson, K. G., & Woolverton, W. L. (2005). Effects of clomipramine on self-control choice in Lewis and Fischer 344 rats. Pharmacology, Biochemistry and Behavior, 80, 387–393. doi:10.1016/j.pbb.2004.11.015

Barkley, R. A., Edwards, G., Laneri, M., Fletcher, K., & Metevia, L. (2001). Executive functioning, temporal discounting, and sense of time in adolescents with attention deficit hyperactivity disorder (ADHD) and oppositional defiant disorder (ODD). Journal of Abnormal Child Psychology, 29(6), 541–556. doi:10.1023/A:1012233310098

Basar, K., Sesia, T., Groenewegen, H., Steinbusch, H. W. M., Visser-Vandewalle, V., & Temel, Y. (2010). Nucleus accumbens and impulsivity. Progress in Neurobiology, 92, 533–557. doi:10.1016/j.pneurobio.2010.08.007

Basten, U., Heekeren, H. R., & Fiebach, C. J. (2010). How the brain integrates costs and benefits during decision making. Proceedings of the National Academy of Sciences of the United States of America, 107, 21767–21772. doi:10.1073/pnas.0908104107

Baumann, A. A., & Odum, A. L. (2012). Impulsivity, risk taking and timing. Behavioural Processes, 90, 408–414. doi:10.1016/j.beproc.2012.04.005

Beckmann, J. S., & Bardo, M. T. (2012). Environmental enrichment reduces attribution of incentive salience to a food-associated stimulus. Behavioural Brain Research, 226, 331–334. doi:10.1016/j.bbr.2011.09.021

Bezzina, G., Body, S., Cheung, T. H., Hampson, C. L., Deakin, J. F. W., Anderson, I. M., et al. (2008). Effect of quinolinic acid-induced lesions of the nucleus accumbens core on performance on a progressive ratio schedule of reinforcement: Implications for inter-temporal choice. Psychopharmacology, 197(2), 339–350. doi:10.1007/s00213-007-1036-0

Bezzina, G., Cheung, T. H. C., Asgari, K., Hampson, C. L., Body, S., Bradshaw, C. M., et al. (2007). Effects of quinolinic acid-induced lesions of the nucleus accumbens core on inter-temporal choice: A quantitative analysis. Psychopharmacology (Berlin), 195, 71–84. doi:10.1007/s00213-007-0882-0

Bhatti, M., Jang, H., Kralik, J. D., & Jeong, J. (2014). Rats exhibit reference-dependent choice behavior. Behavioural Brain Research, 267, 26–32. doi:10.1016/j.bbr.2014.03.012

Bickel, W. K., & Marsch, L. A. (2001). Toward a behavioral economic understanding of drug dependence: Delay discounting processes. Addiction, 96(1), 73–86. doi:10.1046/j.1360-0443.2001.961736.x

Bizot, J.-C., Chenault, N., Houzé, B., Herpin, A., David, S., Pothion, S., et al. (2007). Methylphenidate reduces impulsive behaviour in juvenile Wistar rats, but not in adult Wistar, SHR and WKY rats. Psychopharmacology, 193(2), 215–223. doi:10.1007/s00213-007-0781-4

Blundell, P., Hall, G., & Killcross, A. S. (2001). Lesions of the basolateral amygdala disrupt selective aspects of reinforcer representation in rats. Journal of Neuroscience, 21(22), 9018–9026.

Braet, C., Claus, L., Verbeken, S., & Van Vlierberghe, L. (2007). Impulsivity in overweight children. European Child and Adolescent Psychiatry, 16(8), 473–483. doi:10.1007/s00787-007-0623-2

Broos, N., Diergaarde, L., Schoffelmeer, A. N. M., Pattij, T., & DeVries, T. J. (2012). Trait impulsive choice predicts resistance to extinction and propensity to relapse to cocaine seeking: A bidirectional investigation. Neuropsychopharmacology, 37, 1377–1386. doi:10.1038/npp.2011.323

Bruce, A. S., Black, W. R., Bruce, J. M., Daldalian, M., Martin, L. E., & Davis, A. M. (2011). Ability to delay gratification and BMI in preadolescence. Obesity (Silver Spring), 19(5), 1101–1102. doi:10.1038/oby.2010.297

Camille, N., Tsuchida, A., & Fellows, L. K. (2011). Double dissociation of stimulus-value and action-value learning in humans with orbitofrontal and anterior cingulate cortex damage. The Journal of Neuroscience, 31(42), 15048–15052. doi:10.1523/JNEUROSCI.3164-11.2011

Caraco, T. (1981). Energy budgets, risk and foraging preferences in dark-eyed juncos (Junco hyemalis). Behavioral Ecology and Sociobiology, 8(3), 213–217. doi:10.1007/BF00299833

Cardinal, R. N., Pennicott, D. R., Sugathapala, C. L., Robbins, T. W., & Everitt, B. J. (2001). Impulsive choice induced in rats by lesions of the nucleus accumbens core. Science, 292, 2499–2501. doi:10.1126/science.1060818

Carroll, M. E., Anker, J. J., & Perry, J. L. (2009). Modeling risk factors for nicotine and other drug abuse in the preclinical laboratory. Drug and Alcohol Dependence, 104, 70–78. doi:10.1016/j.drugalcdep.2008.11.011

Church, R. M., & Deluty, M. Z. (1977). Bisection of temporal intervals. Journal of Experimental Psychology: Animal Behavior Processes, 3(3), 216–228.

Cohen, J. D., McClure, S. M., & Yu, A. J. (2007). Should I stay or should I go? How the human brain manages the trade-off between exploitation and exploration. Philosophical Transactions of the Royal Society B: Biological Sciences, 362, 933–942. doi:10.1098/rstb.2007.2098

Connolly, T., & Zeelenberg, M. (2002). Regret in decision making. Current Directions in Psychological Science, 11(6), 212–216. doi:10.1111/1467-8721.00203

Cooper, N., Kable, J. W., Kim, B. K., & Zauberman, G. (2013). Brain activity in valuation regions while thinking about the future predicts individual discount rates. The Journal of Neuroscience, 33(32), 13150–13156. doi:10.1523/JNEUROSCI.0400-13.2013

Coull, J. T., Cheng, R.-K., & Meck, W. H. (2011). Neuroanatomical and neurochemical substrates of timing. Neuropsychopharmacology, 36, 3–25. doi:10.1038/npp.2010.113

Coull, J. T., & Nobre, A. (2008). Dissociating explicit timing from temporal expectation with fMRI. Current Opinion in Neurobiology, 18, 137–144. doi:10.1016/j.conb.2008.07.011

Cui, X. (2011). Hyperbolic discounting emerges from the scalar property of interval timing. Frontiers in Integrative Neuroscience, 5(24), 1–2. doi:10.3389/fnint.2011.00024

da Costa Araújo, S., Body, S., Hampson, C. L., Langley, R. W., Deakin, J. F. W., Anderson, I. M., et al. (2009). Effects of lesions of the nucleus accumbens core on inter-temporal choice: Further observations with an adjusting-delay procedure. Behavioural Brain Research, 202, 272–277. doi:10.1016/j.bbr.2009.04.003

da Costa Araújo, S., Body, S., Valencia-Torres, L., Olarte-Sanchez, C. M., Bak, V. K., Deakin, J. F. W., et al. (2010). Choice between reinforcer delays versus choice between reinforcer magnitudes: Differential Fos expression in the orbital prefrontal cortex and nucleus accumbens core. Behavioural Brain Research, 213, 269–277. doi:10.1016/j.bbr.2010.05.014

Dalley, J. W., Theobald, D. E., Pereira, E. A. C., Li, P. M. M. C., & Robbins, T. W. (2002). Specific abnormalities in serotonin release in the prefrontal cortex of isolation-reared rats measured during behavioural performance of a task assessing visuospatial attention and impulsivity. Psychopharmacology, 164, 329–340. doi:10.1007/s00213-002-1215-y

Davis, C., Patte, K., Curtis, C., & Reid, C. (2010). Immediate pleasures and future consequences: A neuropsychological study of binge eating and obesity. Appetite, 54, 208–213. doi:10.1016/j.appet.2009.11.002

Daw, N. D., O’Doherty, J. P., Dayan, P., Seymour, B., & Dolan, R. J. (2005). Cortical substrates for exploratory decisions in humans. Nature, 441(7095), 876–879. doi:10.1038/nature04766

de Wit, H. (2008). Impulsivity as a determinant and consequence of drug use: A review of underlying processes. Addiction Biology, 14, 22–31. doi:10.1111/j.1369-1600.2008.00129.x

Doughty, A. H., & Richards, J. B. (2002). Effects of reinforcer magnitude on responding under differential-reinforcement-of-low-rate schedules of rats and pigeons. Journal of the Experimental Analysis of Behavior, 78(1), 17–30. doi:10.1901/jeab.2002.78-17

Duckworth, A. L., Tsukayama, E., & Geier, A. B. (2010). Self-controlled children stay leaner in the transition to adolescence. Appetite, 54(2), 304–308. doi:10.1016/j.appet.2009.11.016

Evenden, J. L., & Robbins, T. W. (1984). Win-stay behaviour in the rat. Quarterly Journal of Experimental Psychology, 36B, 1–26. doi:10.1080/14640748408402190

Fox, A. T., Hand, D. J., & Reilly, M. P. (2008). Impulsive choice in a rodent model of attention-deficit/hyperactivity disorder. Behavioural Brain Research, 187(1), 146–152. doi:10.1016/j.bbr.2007.09.008

Frank, M. J. (2006). Hold your horses: A dynamic computational role for the subthalamic nucleus in decision making. Neural Networks, 10, 1120–1136. doi:10.1016/j.neunet.2006.03.006

Frank, M. J., & Claus, E. D. (2006). Anatomy of a decision: Striato-orbitofrontal interactions in reinforcement learning, decision making, and reversal. Psychological Review, 113(2), 300–326. doi:10.1037/0033-295X.113.2.300

Galtress, T., Garcia, A., & Kirkpatrick, K. (2012). Individual differences in impulsive choice and timing in rats. Journal of the Experimental Analysis of Behavior, 98(1), 65–87. doi:10.1901/jeab.2012.98-65

Galtress, T., & Kirkpatrick, K. (2010). The role of the nucleus accumbens core in impulsive choice, timing, and reward processing. Behavioral Neuroscience, 124(1), 26–43. doi:10.1037/a0018464

García-Lecumberri, C., Torres, I., Martín, S., Crespo, J. A., Miguéns, M., Nicanor, C., et al. (2010). Strain differences in the dose-response relationship for morphine self-administration and impulsive choice between Lewis and Fischer 344 rats. Journal of Psychopharmacology, 25(6), 783–791. doi:10.1177/0269881110367444

Green, L., & Estle, S. J. (2003). Preference reversals with food and water reinforcers in rats. Journal of the Experimental Analysis of Behavior, 79(2), 233–242. doi:10.1901/jeab.2003.79-233

Green, L., & Myerson, J. (2004). A discounting framework for choice with delayed and probabilistic rewards. Psychological Bulletin, 130(5), 769–792. doi:10.1037/0033-2909.130.5.769

Green, L., & Myerson, J. (2010). Experimental and correlational analyses of delay and probability discounting. In G. J. Madden & W. K. Bickel (Eds.), Impulsivity: The Behavioral and Neurological Science of Discounting (pp. 67–92). Washington, DC: American Psychological Association.

Green, L., Myerson, J., & Ostaszewski, P. (1999). Amount of reward has opposite effects on the discounting of delayed and probabilistic outcomes. Journal of Experimental Psychology: Learning, Memory, and Cognition 25(2), 418–427. doi:10.1037/0278-7393.25.2.418

Gregorios-Pippas, L., Tobler, P. N., & Schultz, W. (2009). Short-term temporal discounting of reward value in human ventral striatum. Journal of Neurophysiology, 101, 1507–1523. doi:10.1152/jn.90730.2008

Hand, D. J., Fox, A. T., & Reilly, M. P. (2009). Differential effects of d-amphetamine on impulsive choice in spontaneously hypertensive and Wistar-Kyoto rats. Behavioural Pharmacology, 20, 549–553. doi:10.1097/FBP.0b013e3283305ee1

Hare, T. A., O’Doherty, J., Camerer, C. F., Schultz, W., & Rangel, A. (2008). Dissociating the role of the orbitofrontal cortex and the striatum in the computation of goal values and prediction errors. The Journal of Neuroscience, 28(22), 5623–5630. doi:10.1523/JNEUROSCI.1309-08.2008

Harley, C. W. (1987). A role for norepinephrine in arousal, emotion and learning?: Limbic modulation by norepinephrine and the Kety hypothesis. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 11(4), 419–458. doi:10.1016/0278-5846(87)90015-7

Heilbronner, S. R., & Hayden, B. Y. (2013). Contextual factors explain risk-seeking preferences in rhesus monkeys. Frontiers in Neuroscience, 7(7), 1–7. doi:10.3389/fnins.2013.00007

Hertwig, R., & Erev, I. (2009). The description-experience gap in risky choice. Trends in Cognitive Sciences, 13(12), 517–523. doi:10.1016/j.tics.2009.09.004

Hill, J. C., Covarrubias, P., Terry, J., & Sanabria, F. (2012). The effect of methylphenidate and rearing environment on behavioral inhibition in adult male rats. Psychopharmacology, 219, 353–362. doi:10.1007/s00213-011-2552-5

Holt, D. D., Green, L., & Myerson, J. (2003). Is discounting impulsive? Evidence from temporal and probability discounting in gambling and non-gambling college students. Behavioural Processes, 64, 355–367. doi:10.1016/S0376-6357(03)00141-4

Huskinson, S. L., Krebs, C. A., & Anderson, K. G. (2012). Strain differences in delay discounting between Lewis and Fischer 344 rats at baseline and following acute and chronic administration of d-amphetamine. Pharmacology, Biochemistry and Behavior, 101(3), 403–416. doi:10.1016/j.pbb.2012.02.005

Jimura, K., Myerson, J., Hilgard, J., Keighley, J., Braver, T. S., & Green, L. (2011). Domain independence and stability in young and older adults’ discounting of delayed rewards. Behavioural Processes, 87(3), 253–259. doi:10.1016/j.beproc.2011.04.006

Kable, J. W., & Glimcher, P. W. (2007). The neural correlates of subjective value during intertemporal choice. Nature Neuroscience, 10(12), 1625–1633. doi:10.1038/nn2007

Kable, J. W., & Glimcher, P. W. (2009). The neurobiology of decision: Consensus and controversy. Neuron, 63, 733–745. doi:10.1016/j.neuron.2009.09.003

Kacelnik, A., Vasconcelos, M., Monteiro, T., & Aw, J. (2011). Darwin’s “tug-of-war” vs. starlings’ “horse-racing”: How adaptations for sequential encounters drive simultaneous choice. Behavioral Ecology and Sociobiology, 65, 547–558. doi:10.1007/s00265-010-1101-2

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291. doi:10.2307/1914185

Kalenscher, T., & van Wingerden, M. (2011). Why we should use animals to study economic decision making—A perspective. Frontiers in Neuroscience, 5(82), 1–11. doi:10.3389/fnins.2011.00082

Kalivas, P. W., & Volkow, N. D. (2005). The neural basis of addiction: A pathology of motivation and choice. American Journal of Psychiatry, 162(8), 1403–1413. doi:10.1176/appi.ajp.162.8.1403

Keren, G., & Wagenaar, W. A. (1987). Violation of utility theory in unique and repeated gambles. Journal of Experimental Psychology: Learning, Memory, and Cognition 13(3), 387–391.

Kim, B. K., & Zauberman, G. (2009). Perception of anticipatory time in temporal discounting. Journal of Neuroscience, Psychology, and Economics, 2(2), 91–101. doi:10.1037/a0017686

Kirby, K. N. (2009). One-year temporal stability of delay-discount rates. Psychonomic Bulletin & Review, 16(3), 457–462. doi:10.3758/PBR.16.3.457

Kirkpatrick, K., Marshall, A. T., Clarke, J., & Cain, M. E. (2013). Environmental rearing effects on impulsivity and reward sensitivity. Behavioral Neuroscience, 127(5), 712–724. doi:10.1037/a0034124

Kirkpatrick, K., Marshall, A. T., Smith, A. P., Koci, J., & Park, Y. (2014). Individual differences in impulsive and risky choice: Effects of environmental rearing conditions. Behavioural Brain Research, 269(115–127). doi:10.1016/j.bbr.2014.04.024

Kreek, M. J., Nielsen, D. A., Butelman, E. R., & LaForge, K. S. (2005). Genetic influences on impulsivity, risk taking, stress responsivity and vulnerability to drug abuse and addiction. Nature Neuroscience, 8(11), 1450–1457. doi:10.1038/nn1583

Kringlebach, M. L., & Rolls, E. T. (2004). The functional neuroanatomy of the human orbitofrontal cortex: Evidence from neuroimaging and neuropsychology. Progress in Neurobiology, 72, 341–372. doi:10.1016/j.pneurobio.2004.03.006

Krishnan-Sarin, S., Reynolds, B., Duhig, A. M., Smith, A., Liss, T., McFetridge, A., et al. (2007). Behavioral impulsivity predicts treatment outcome in a smoking cessation program for adolescent smokers. Drug and Alcohol Dependence, 88(1), 79–82. doi:10.1016/j.drugalcdep.2006.09.006

Lane, S. D., Cherek, D. R., Pietras, C. J., & Tcheremissine, O. V. (2003). Measurement of delay discounting using trial-by-trial consequences. Behavioural Processes, 64, 287–303. doi:10.1016/S0376-6357(03)00143-8

Lau, B., & Glimcher, P. W. (2005). Dynamic response-by-response models of matching behavior in rhesus monkeys. Journal of the Experimental Analysis of Behavior, 84(3), 555–579. doi:10.1901/jeab.2005.110-04

Laude, J. R., Beckman, J. S., Daniels, C. W., & Zentall, T. R. (2014). Impulsivity affects sub-optimal gambling-like choice by pigeons. Journal of Experimental Psychology: Animal Learning and Cognition, 40(1), 2–11. doi:10.1037/xan0000001

Levin, I. P., Xue, G., Weller, J. A., Reimann, M., Lauriola, M., & Bechara, A. (2012). A neuropsychological approach to understanding risk-taking for potential gains and losses. Frontiers in Neuroscience, 6(15), 1–11. doi:10.3389/fnins.2012.00015

Lim, S.-L., O’Doherty, J. P., & Rangel, A. (2011). The decision value computations in the vmPFC and striatum use a relative value code that is guided by visual attention. The Journal of Neuroscience, 31(37), 13214–13223. doi:10.1523/JNEUROSCI.1246-11.2011

Linnet, J., Røjskjær, S., Nygaard, J., & Maher, B. A. (2006). Episodic chasing in pathological gamblers using the Iowa gambling task. Scandinavian Journal of Psychology, 47, 43–49. doi:10.1111/j.1467-9450.2006.00491.x

Lucci, C. R. (2013). Time, self, and intertemporal choice. Frontiers in Neuroscience, 7(40), 1–5. doi:10.3389/fnins.2013.00040

MacKillop, J., Amlung, M. T., Few, L. R., Ray, L. A., Sweet, L. H., & Munafo, M. R. (2011). Delayed reward discounting and addictive behavior: A meta-analysis. Psychopharmacology (Berlin), 216, 305–321. doi:10.1007/s00213-011-2229-0

Madden, G. J., Petry, N. M., & Johnson, P. S. (2009). Pathological gamblers discount probabilistic rewards less steeply than matched controls. Experimental and Clinical Psychopharmacology, 17(5), 283–290. doi:10.1037/a0016806

Madden, G. J., Smith, N. G., Brewer, A. T., Pinkston, J. W., & Johnson, P. S. (2008). Steady-state assessment of impulsive choice in Lewis and Fischer 344 rats: Between-condition delay manipulations. Journal of the Experimental Analysis of Behavior, 90(3), 333–344. doi:10.1901/jeab.2008.90-333

Marsh, B. (2002). Do animals use heuristics? Journal of Bioeconomics, 4, 49–56. doi:10.1023/A:1020655022163

Marsh, B., & Kacelnik, A. (2002). Framing effects and risky decisions in starlings. Proceedings of the National Academy of Sciences of the United States of America, 99(5), 3352–3355. doi:10.1073/pnas.042491999

Marshall, A. T. (2013). Relative gains and losses in risky choice. Unpublished master’s thesis, Kansas State University. https://krex.k-state.edu/dspace/handle/2097/15631

Marshall, A. T., & Kirkpatrick, K. (2012). Analysis of interval timing in two discounting procedures. Paper presented at the 38th Annual Convention of the Association for Behavior Analysis International, Seattle, WA.

Marshall, A. T., & Kirkpatrick, K. (2013). The effects of the previous outcome on probabilistic choice in rats. Journal of Experimental Psychology: Animal Behavior Processes, 39(1), 24–38. doi:10.1037/a0030765

Marshall, A. T., & Kirkpatrick, K. (2015). Relative gains, losses, and reference points in probabilistic choice in rats. PLoS ONE, 10(2), e0117697. doi:10.1371/journal.pone.0117697

Marshall, A. T., Smith, A. P., & Kirkpatrick, K. (2014). Mechanisms of impulsive choice: I. Individual differences in interval timing and reward processing. Journal of the Experimental Analysis of Behavior, 102(1), 86–101. doi:10.1002/jeab.88

Marusich, J. A., & Bardo, M. T. (2009). Differences in impulsivity on a delay-discounting task predict self-administration of a low unit dose of methylphenidate in rats. Behavioural Pharmacology, 20(5–6), 447–454. doi:10.1097/FBP.0b013e328330ad6d

Matell, M. S., Meck, W. H., & Nicolelis, M. A. L. (2003). Interval timing and the encoding of signal duration by ensembles of cortical and striatal neurons. Behavioral Neuroscience, 117, 760–773. doi:10.1037/0735-7044.117.4.760

Matusiewicz, A. K., Carter, A. E., Landes, R. D., & Yi, R. (2013). Statistical equivalence and test-retest reliability of delay and probability discounting using real and hypothetical rewards. Behavioural Processes, 100, 116–122. doi:10.1016/j.beproc.2013.07.019

Mazur, J. E. (1987). An adjusting procedure for studying delayed reinforcement. In M. L. Commons, J. E. Mazur, J. A. Nevin, & H. Rachlin (Eds.), Quantitative Analyses of Behavior. Vol. 5. The Effect of Delay and of Intervening Events on Reinforcer Value (pp. 55–73). Hillsdale, NJ: Erlbaum.

Mazur, J. E. (1988). Choice between small certain and large uncertain reinforcers. Animal Learning and Behavior, 16(2), 199–205. doi:10.3758/BF03209066

Mazur, J. E. (2001). Hyperbolic value addition and general models of animal choice. Psychological Review, 108(1), 96–112. doi:10.1037/0033-295X.108.1.96

Mazur, J. E. (2007). Rats’ choices between one and two delayed reinforcers. Learning & Behavior, 35(3), 169–176. doi:10.3758/BF03193052

McClure, J., Podos, J., & Richardson, H. N. (2014). Isolating the delay component of impulsive choice in adolescent rats. Frontiers in Integrative Neuroscience, 8(3), 1–9. doi:10.3389/fnint.2014.00003

McClure, S. M., Laibson, D., Loewenstein, G., & Cohen, J. D. (2004). Separate neural systems value immediate and delayed monetary rewards. Science, 306(5695), 503–507. doi:10.1126/science.1100907

Meck, W. H. (2006). Neuroanatomical localization of an internal clock: A functional link between mesolimbic, nigrostriatal, and mesocortical dopaminergic systems. Brain Research, 1109(1), 93–107. doi:10.1016/j.brainres.2006.06.031

Meck, W. H., Penney, J. B., & Pouthas, V. (2008). Cortico-striatal representation of time in animals and humans. Current Opinion in Neurobiology, 18, 145–152. doi:10.1016/j.conb.2008.08.002

Mischel, W., Ayduk, O., Berman, M. G., Casey, B. J., Gotlib, I. H., Jonides, J., et al. (2011). ‘Willpower’ over the life span: Decomposing self-regulation. Social Cognitive and Affective Neuroscience, 6(2), 252–256. doi:10.1093/scan/nsq081

Mischel, W., Shoda, Y., & Rodriguez, M. L. (1989). Delay of gratification in children. Science, 244, 933–938. doi:10.1126/science.2658056

Mobini, S., Body, S., Ho, M. Y., Bradshaw, C. M., Szabadi, E., Deakin, J. F., et al. (2002). Effects of lesions of the orbitofrontal cortex on sensitivity to delayed and probabilistic reinforcement. Psychopharmacology, 160(3), 290–298. doi:10.1007/s00213-001-0983-0

Myerson, J., & Green, L. (1995). Discounting of delayed rewards: Models of individual choice. Journal of the Experimental Analysis of Behavior, 64, 263–276. doi:10.1901/jeab.1995.64-263

Myerson, J., Green, L., Hanson, J. S., Holt, D. D., & Estle, S. J. (2003). Discounting delayed and probabilistic rewards: Processes and traits. Journal of Economic Psychology, 24, 619–635. doi:10.1016/S0167-4870(03)00005-9

Myerson, J., Green, L., & Morris, J. (2011). Modeling the effect of reward amount on probability discounting. Journal of the Experimental Analysis of Behavior, 95(2), 175–187. doi:10.1901/jeab.2011.95-175

Nederkoorn, C., Braet, C., Van Eijs, Y., Tanghe, A., & Jansen, A. (2006). Why obese children cannot resist food: The role of impulsivity. Eating Behaviors, 7(4), 315–322. doi:10.1016/j.eatbeh.2005.11.005

Nederkoorn, C., Jansena, E., Mulkensa, S., & Jansena, A. (2007). Impulsivity predicts treatment outcome in obese children. Behaviour Research and Therapy, 45(5), 1071–1075. doi:10.1016/j.brat.2006.05.009

Odum, A. L. (2011a). Delay discounting: I’m a k, you’re a k. Journal of the Experimental Analysis of Behavior, 96(3), 427–439. doi:10.1901/jeab.2011.96-423

Odum, A. L. (2011b). Delay discounting: Trait variable? Behavioural Processes, 87(1), 1–9. doi:10.1016/j.beproc.2011.02.007

Odum, A. L., & Baumann, A. A. L. (2010). Delay discounting: State and trait variable. In G. J. Madden & W. K. Bickel (Eds.), Impulsivity: The Behavioral and Neurological Science of Discounting (pp. 39–65). Washington, DC: APA Books.

Ohmura, Y., Takahashi, T., Kitamura, N., & Wehr, P. (2006). Three-month stability of delay and probability discounting measures. Experimental and Clinical Psychopharmacology, 14, 318–328. doi:10.1037/1064-1297.14.3.318

Olausson, P., Jentsch, J. D., Tronson, N., Neve, R. L., Nestler, E. J., & Taylor, J. R. (2006). ΔFoSB in the nucleus accumbens regulates food-reinforced instrumental behavior and motivation. The Journal of Neuroscience, 26(36), 9196–9204. doi:10.1523/JNEUROSCI.1124-06.2006

Pattison, K. F., Laude, J. R., & Zentall, T. R. (2013). Environmental enrichment affects suboptimal, risky, gambling-like choice by pigeons. Animal Cognition, 16, 429–434. doi:10.1007/s10071-012-0583-x

Perry, J. L., & Carroll, M. E. (2008). The role of impulsive behavior in drug abuse. Psychopharmacology, 200, 1–26. doi:10.1007/s00213-008-1173-0

Perry, J. L., Stairs, D. J., & Bardo, M. T. (2008). Impulsive choice and environmental enrichment: Effects of d-amphetamine and methylphenidate. Behavioural Brain Research, 193(1), 48–54. doi:10.1016/j.bbr.2008.04.019

Peters, J., & Büchel, C. (2010). Neural representations of subjective reward value. Behavioural Brain Research, 213, 135–141. doi:10.1016/j.bbr.2010.04.031

Peters, J., & Büchel, C. (2011). The neural mechanisms of inter-temporal decision-making: Understanding variability. Trends in Cognitive Sciences, 15(5), 227–239. doi:10.1016/j.tics.2011.03.002

Peters, J., Miedl, S. F., & Büchel, C. (2012). Formal comparison of dual-parameter temporal discounting models in controls and pathological gamblers. PLoS ONE, 7(11), e47225. doi:10.1371/journal.pone.0047225

Pizzo, M. J., Kirkpatrick, K., & Blundell, P. J. (2009). The effect of changes in criterion value on differential reinforcement of low rate schedule performance. Journal of the Experimental Analysis of Behavior, 92, 181–198. doi:10.1901/jeab.2009.92-181

Pothuizen, H. H., Jongen-Relo, A. L., Feldon, J., & Yee, B. K. (2005). Double dissociation of the effects of selective nucleus accumbens core and shell lesions on impulsive-choice behaviour and salience learning in rats. European Journal of Neuroscience, 22, 2605–2616. doi:10.1111/j.1460-9568.2005.04388.x

Rachlin, H. (1990). Why do people gamble and keep gambling despite heavy losses? Psychological Science, 1(5), 294–297. doi:10.1111/j.1467-9280.1990.tb00220.x

Rachlin, H., Raineri, A., & Cross, D. (1991). Subjective probability and delay. Journal of the Experimental Analysis of Behavior, 55(2), 233–244. doi:10.1901/jeab.1991.55-233

Reynolds, B., Ortengren, A., Richards, J. B., & de Wit, H. (2006). Dimensions of impulsive behavior: Personality and behavioral measures. Personality and Individual Differences, 40(2), 305–315. doi:10.1016/j.paid.2005.03.024

Reynolds, B., Richards, J. B., Horn, K., & Karraker, K. (2004). Delay discounting and probability discounting as related to cigarette smoking status in adults. Behavioural Processes, 65, 35–42. doi:10.1016/S0376-6357(03)00109-8

Richards, J. B., Zhang, L., Mitchell, S. H., & De Wit, H. (1999). Delay or probability discounting in a model of impulsive behavior: Effect of alcohol. Journal of the Experimental Analysis of Behavior, 71(2), 121–143. doi:10.1901/jeab.1999.71-121

Richardson, N. R., & Roberts, D. C. S. (1996). Progressive ratio schedules in drug self-administration studies in rats: A method to evaluate reinforcing efficacy. Journal of Neuroscience Methods, 66(1), 1–11. doi:10.1016/0165-0270(95)00153-0

Robbins, T. W., & Everitt, B. J. (1996). Neurobehavioural mechanisms of reward and motivation. Current Opinion in Neurobiology, 6, 228–236. doi:10.1016/S0959-4388(96)80077-8

Roesch, M. R., Takahashi, Y., Gugsa, N., Bissonette, G. B., & Schoenbaum, G. (2007). Previous cocaine exposure makes rats hypersensitive to both delay and reward magnitude. The Journal of Neuroscience, 27(1), 245–250. doi:10.1523/JNEUROSCI.4080-06.2007

Rushworth, M. F. S., Kolling, N., Sallet, J., & Mars, R. B. (2012). Valuation and decision-making in frontal cortex: One or many serial or parallel systems? Current Opinion in Neurobiology, 22(6), 946–955. doi:10.1016/j.conb.2012.04.011

Rushworth, M. F. S., Noonan, M. P., Boorman, E., Walton, M. E., & Behrens, T. E. (2011). Frontal cortex and reward-guided learning and decision making. Neuron, 70, 1054–1069. doi:10.1016/j.neuron.2011.05.014

Russell, V. A., Sagvolden, T., & Johansen, E. B. (2005). Animal models of attention-deficit hyperactivity disorder. Behavioral and Brain Functions, 1(9), 1–17. doi:10.1186/1744-9081-1-9

Scheres, A., Milham, M. P., Knutson, B., & Castellanos, F. X. (2007). Ventral striatal hyporesponsiveness during reward anticipation in attention-deficit/hyperactivity disorder. Biological Psychiatry, 61(5), 720–724. doi:10.1016/j.biopsych.2006.04.042

Schoenbaum, G., Roesch, M. R., Stalnaker, T. A., & Takahashi, Y. K. (2009). A new perspective on the role of the orbitofrontal cortex in adaptive behavior. Nature Reviews Neuroscience, 10, 885–892. doi:10.1038/nrn2753

Shapiro, M. S., Siller, S., & Kacelnik, A. (2008). Simultaneous and sequential choice as a function of reward delay and magnitude: Normative, descriptive and process-based models tested in the European starling (Sturnus vulgaris). Journal of Experimental Psychology: Animal Behavior Processes, 34(1), 75–93. doi:10.1037/0097-7403.34.1.75

Simpson, J., & Kelly, J. P. (2011). The impact of environmental enrichment in laboratory rats—Behavioural and neurochemical aspects. Behavioural Brain Research, 222, 246–264. doi:10.1016/j.bbr.2011.04.002

Smith, A. P., Marshall, A. T., & Kirkpatrick, K. (2015). Mechanisms of impulsive choice: II. Sensitivity to time as a potential mechanism to increase self-control. Behavioural Processes. 112, 29-42. doi:10.1016/j.beproc.2014.10.010

Solanto, M. V., Abikoff, H., Sonuga-Barke, E., Schachar, R., Logan, G. D., Wigal, T., et al. (2001). The ecological validity of delay aversion and response inhibition as measures of impulsivity in AD/HD: A supplement to the NIMH multimodal treatment study of AD/HD. Journal of Abnormal Child Psychology, 29(3), 215–228. doi:10.1023/A:1010329714819

Sonuga-Barke, E. J. S. (2002). Psychological heterogeneity in AD/HD—A dual pathway model of behaviour and cognition. Behavioural Brain Research, 10, 29–36. doi:10.1016/S0166-4328(01)00432-6

Sonuga-Barke, E. J. S., Taylor, E., Sembi, S., & Smith, J. (1992). Hyperactivity and delay aversion. I. The effect of delay on choice. Journal of Child Psychology and Psychiatry, 33(2), 387–398. doi:10.1111/j.1469-7610.1992.tb00874.x

Stein, J. S., Pinkston, J. W., Brewer, A. T., Francisco, M. T., & Madden, G. J. (2012). Delay discounting in Lewis and Fischer 344 rats: Steady-state and rapid-determination adjusting-amount procedures. Journal of the Experimental Analysis of Behavior, 97(3), 305–321. doi:10.1901/jeab.2012.97-305

Steiner, A. P., & Redish, A. D. (2014). Behavioral and neurophysiological correlates of regret in rat decision-making on a neuroeconomic task. Nature Neuroscience, 17, 995–1002. doi:10.1038/nn.3740

Sutton, R. S., & Barto, A. G. (1998). Reinforcement Learning: An Introduction. Cambridge, MA: MIT Press.

Takahashi, T. (2005). Loss of self-control in intertemporal choice may be attributable to logarithmic time-perception. Medical Hypotheses, 65, 691–693. doi:10.1016/j.mehy.2005.04.040

Takahashi, T., Oono, H., & Radford, M. H. B. (2008). Psychophysics of time perception and intertemporal choice models. Physica A: Statistical and Theoretical Physics, 387(8), 2066–2074. doi:10.1016/j.physa.2007.11.047

van den Bergh, F. S., Bloemarts, E., Chan, J. S. W., Groenink, L., Olivier, B., & Oosting, R. S. (2006). Spontaneously hypertensive rats do not predict symptoms of attention-deficit hyperactivity disorder. Pharmacology Biochemistry and Behavior, 83(3), 380–390. doi:10.1016/j.pbb.2006.02.018

Van den Broek, M. D., Bradshaw, C. M., & Szabadi, E. (1987). Behaviour of “impulsive” and “non-impulsive” humans in a temporal differentiation schedule of reinforcement. Personality and Individual Differences, 8(2), 233. doi:10.1016/0191-8869(87)90179-6

Verdejo-García, A., Lawrence, A. J., & Clark, L. (2008). Impulsivity as a vulnerability marker for substance-use disorders: Review of findings from high-risk research, problem gamblers and genetic association studies. Neuroscience and Biobehavioral Reviews, 32, 777–810. doi:10.1016/j.neubiorev.2007.11.003

Verdejo-Garcia, A., Perez-Exposito, M., Schmidt-Rio-Valle, J., Fernandez-Serrano, M. J., Cruz, F., Perez-Garcia, M., et al. (2010). Selective alterations within executive functions in adolescents with excess weight. Obesity (Silver Spring), 18(8), 1572–1578. doi:10.1038/oby.2009.475

Wang, X. T., & Johnson, J. G. (2012). A tri-reference point theory of decision making under risk. Journal of Experimental Psychology: General, 141(4), 743–756. doi:10.1037/a0027415

Weatherly, J. N., & Derenne, A. (2012). Investigating the relationship between the contingencies that maintain gambling and probability discounting of gains and losses. European Journal of Behavior Analysis, 13(1), 39–46.

Weller, R. E., Cook, E. W., Avsar, K. B., & Cox, J. E. (2008). Obese women show greater delay discounting than healthy-weight women. Appetite, 51, 563–568. doi:10.1016/j.appet.2008.04.010

Wilhelm, C. J., & Mitchell, S. H. (2009). Strain differences in delay discounting using inbred rats. Genes, Brain and Behavior, 8, 426–434. doi:10.1111/j.1601-183X.2009.00484.x

Winstanley, C. A., Baunez, C., Theobald, D. E., & Robbins, T. W. (2005). Lesions to the subthalamic nucleus decrease impulsive choice but impair autoshaping in rats: The importance of the basal ganglia in Pavlovian conditioning and impulse control. European Journal of Neuroscience, 21(11), 3107–3116. doi:10.1111/j.1460-9568.2005.04143.x

Wittmann, M., & Paulus, M. P. (2008). Decision making, impulsivity and time perception. Trends in Cognitive Sciences, 12(1), 7–12. doi:10.1016/j.tics.2007.10.004

Yoon, J. H., Higgins, S. T., Heil, S. H., Sugarbaker, R. J., Thomas, C. S., & Badger, G. J. (2007). Delay discounting predicts postpartum relapse to cigarette smoking among pregnant women. Experimental and Clinical Psychopharmacology, 15(2), 176–186. doi:10.1037/1064-1297.15.2.186

Zauberman, G., Kim, B. K., Malkoc, S. A., & Bettman, J. R. (2009). Discounting time and time discounting: Subjective time perception and intertemporal preferences. Journal of Marketing Research, 46(4), 543–556. doi:10.1509/jmkr.46.4.543

Zeeb, F. D., Wong, A. C., & Winstanley, C. A. (2013). Differential effects of environmental enrichment, social-housing, and isolation-rearing on a rat gambling task: Dissociations between impulsive action and risky decision-making. Psychopharmacology (Berlin), 225, 381–395. doi:10.1007/s00213-012-2822-x

Zhang, M., Balmadrid, C., & Kelley, A. (2003). Nucleus accumbens opioid, GABAergic and dopaminergic modulation of palatable food motivation: Contrasting effects revealed by a progressive ratio study in the rat. Behavioral Neuroscience, 117(2), 202–211. doi:10.1037/0735-7044.117.2.202