Recently, there has been a renewed interest in avoidance behavior, and its applicability to clinical conditions such as anxiety, posttraumatic stress disorder (PTSD) and addiction. In a computer-based avoidance task for humans, participants control an on-screen spaceship, shoot at enemy targets (appetitive cue) to gain points, and learn to respond to an on-screen warning signal (WS) by entering safe "hiding" areas to escape/avoid an aversive event (point loss and on-screen explosion) paired with an on-screen aversive cue (bomb). Prior research on active avoidance in rodents suggests that avoidance learning is facilitated if the response also terminates the WS. Here, we adapted the computer-based task to investigate this idea in healthy humans. Two hundred and twenty-two young adults completed one of three conditions of the task: a non-contingent condition, where hiding caused omission/avoidance of the aversive event but did not terminate the WS; a fully-contingent condition, where hiding also caused omission/termination of all on-screen appetitive and aversive cues as well as terminating the WS; and a partially-contingent condition where hiding caused omission of the appetitive and aversive cues, but did not affect the WS. Both contingency manipulations decreased escape/avoidance behavior, as compared to the non-contingent condition where the WS and other cues are not affected by the avoidance behavior. This study has implications for the basic understanding of the mechanisms that affect avoidance behavior in humans.
Michael Todd Allen and Jony Sheynin have equal contribution.
Avoidance is the performance or withholding of a specific response to prevent an imminent aversive event. While often adaptive, avoidance can become pathological when the avoidance response persists even when the aversive event no longer occurs. Such pathological avoidance is a key feature of anxiety disorders, posttraumatic stress disorder (PTSD), depression and addiction disorders [[
One task that has been used to assess avoidance acquisition and extinction in rodents is a lever-press task in which a warning signal (WS) such as a light or a tone precedes and predicts an aversive event such as an electric foot shock. The rodent can learn to terminate the shock by making a behavioral response such as lever press (escape response), and to avert it altogether by responding during the WS but before the onset of the aversive event, which results in complete omission of the shock for that trial (avoidance response). While an increasing number of human avoidance tasks involve an electric shock as the aversive event, most do not directly measure avoidance learning, since the participant is typically told of the contingency between responding and shock termination/prevention; i.e., these tasks measure performance, rather than acquisition of the avoidance response [[
As an alternative to these approaches, computer-based avoidance tasks have been developed and involve a loss of points/money or a negative affective stimulus (e.g. unpleasant photo or sound), as the aversive event to be avoided [[
Since its development, we have used the spaceship task to demonstrate specific patterns of greater escape/avoidance behavior in various healthy and clinical populations. First, we demonstrated that two vulnerability factors for anxiety disorders (female sex and behaviorally-inhibited personality) were associated with greater avoidance behavior in healthy individuals, suggesting that this could be a behavioral mechanism that mediates vulnerability to anxiety in humans [[
The spaceship task therefore may provide a testbed to better understand the mechanisms underlying both normal and exaggerated acquisition, as well as ways in which individual's tendency to acquire and maintain avoidance can be manipulated; this might, in turn, suggest therapeutic approaches. For example, research in animals has suggested that avoidance learning is most efficient when the subject's avoidance response immediately terminates the WS, as well as prevents the upcoming shock [[
The question then arises of whether such contingency manipulations may similarly affect how humans acquire avoidance responses. One early study by Meyer [[
Therefore, the current study aimed to better understand whether stimulus contingencies could affect learning as human participants transition from escape to avoidance. We compared a control "non-contingent" condition, in which the WS occurred for a fixed duration regardless of whether an avoidance response was made, with a "fully-contingent" condition in which the avoidance response also caused immediate termination of the WS and all other on-screen events including the aversive (bomb) and appetitive cues (enemy targets) for the duration of the hiding response. We also considered an intermediate "partially-contingent" condition in which the avoidance response caused termination of all on-screen events except the WS. Our initial expectation, consistent with prior rat data, was that termination of the WS in the fully-contingent condition should enhance acquisition of avoidance responses, but that there might also be some facilitation of acquisition learning in the partially-contingent condition, since removal of the appetitive events could also provide informational feedback "marking" the successful avoidance response.
Two hundred twenty-two undergraduate students completed the study. The study was conducted at two sites, a large Northeastern University in an urban setting (n = 115) where participants were compensated $20 for one hour, and a large Western university, including both rural and urban components (n = 107), where participants received credit towards an undergraduate psychology research requirement. Data collection at the Northeastern site preceded data collection at the Western site. Procedures at each site were approved by the corresponding institutional review boards and followed federal guidelines for the protection of human subjects. After providing a written informed consent for participation, participants completed a short questionnaire about their demographic information including gender, age, years of education, and race/ethnicity.
All participants were administered a computer-based task that took the form of a spaceship videogame, which was a modification of a previously published task [[
Graph: Fig. 1Screenshots of the spaceship task during the acquisition phase, showing possible events across the warning period (left), the bomb period (middle) and the inter-trial interval (ITI; right). A Screen events for the warning, bomb, and inter-trial interval are the same in all three task conditions, if the participant's ship is not hiding. During the warning period, the warning signal (WS) appears as a blue light at the top of the screen. During the bomb period, an aversive cue (bomb) appears on the screen and there are periodic aversive events (explosions and point loss). During the ITI, the participant can fire at enemy targets to gain points. The three task conditions differ in what is visible when the participant's ship is in hiding. B In the "non-contingent" condition, all the cues including the WS continue to appear while the participant's ship is in hiding (here, in the box at lower right of screen). Specifically, during the warning period, the WS and enemy targets remain visible; during the bomb period, the bomb remains on the screen; during the inter-trial interval, enemy targets continue to appear. C In the partially-contingent condition, the appetitive and aversive cues are omitted if the participant is in hiding, but the WS remains visible throughout the warning period. D In the fully-contingent condition, no cues (WS, enemy targets, or bomb) appear if the participant's ship is in hiding. Thus, in the fully-contingent condition, the WS—along with the other on-screen events—is contingent on the participant's response
During the acquisition phase, presence of the WS for 5 s (warning period) was followed by appearance of a bomb, which served as the aversive cue. The bomb remained on-screen for another 5 s (bomb period). The bomb period was divided into six segments of equal duration; during each segment there was an explosion of the participant's spaceship and a loss of 5 points, to a maximum of six explosions and 30 points lost. At the bottom corners of the screen, there were two box-shaped areas representing "safe areas". Moving the participant's spaceship into one of those boxes was defined as "hiding." While hiding, the participant's spaceship could not be destroyed and no points could be lost, but neither could the participant shoot enemy targets and gain points. Hiding during the bomb period was defined as an "escape hiding" and terminated the ongoing aversive outcome (accumulating point loss and explosions) for any time segment that the participant stayed inside the box. Hiding during the warning period was defined as "avoidance hiding", and it completely prevented the aversive outcome if the participant remained inside the box through the rest of the warning period and through the bomb period. Of note, the warning period presented an approach-avoidance conflict, where both appetitive and aversive cues were displayed when the participants stayed outside the safe areas [[
The participant was given no explicit instructions about the safe areas or the hiding response. The only instructions received at the start of the task were: "You are about to play a game in which you will be piloting a spaceship. You may use LEFT and RIGHT keys to move your spaceship, and press the FIRE key to fire lasers. Your goal is to score as many points as you can. The number of points will appear on the top of the screen. Good Luck!" Participants were then given 1 min of practice time, during which they could shoot the enemy targets but no WS or bomb appeared. Twelve acquisition trials followed, each defined by the appearance of the WS; the start of a trial was not explicitly signaled to the participants. Each trial was followed by a 10-s inter-trial interval (ITI), during which participants could gain points by shooting at targets, without any risk of aversive events. A running tally at the top of the screen showed the current points accumulated. This tally was not allowed to fall below zero, to minimize frustration among participants, but even when the tally was equal to zero, if the participant failed to hide during the bomb period, point loss and explosion of the participant's spaceship were still displayed on the screen. The acquisition phase was followed by an extinction phase, consisting of 12 trials with the WS but no aversive cues (bombs) or aversive outcomes (point loss and explosions).
Participants were randomly assigned to one of three conditions: A non-contingent condition as described above, or "partially-contingent" or "fully-contingent" conditions, which altered the consequences of the hiding behavior. These conditions are visualized in Fig. 1 and summarized in Fig. 2. In the non-contingent condition, hiding during the bomb period (escape response) terminated the explosions and point loss but not the visual bomb cue; hiding during the warning period averted the explosions and point loss but did not terminate the WS nor affect the appetitive targets nor the visual bomb cue (Fig. 2A). In the partially-contingent condition, hiding triggered omission of both the aversive cue (bomb) and appetitive cue (enemy target) for the duration of the hiding response (Fig. 2B). However, the WS remained present for the duration of the warning period, regardless of the participant's behavior—i.e. the presence of the WS was not contingent on the participant's response. Finally, in the fully-contingent condition, hiding triggered omission of the aversive and appetitive cues as well as the WS (Fig. 2C); i.e., the WS duration was contingent on the participant's responses. Of note, the avoidance response in the fully-contingent condition can be considered a "WS-escape" response, which could be reinforced by the termination of the aversive properties of the WS ("two-factor theory" of avoidance) [[
Graph: Fig. 2Schematics of the spaceship task on three example trials during the acquisition phase. A Non-contingent condition. Here, trial 1 begins with a 5-s warning period, during which a warning signal (WS) appears and targets continue to appear, which the participant can shoot for points; Next, during a 5-s bomb period, the bomb cue appears and there are periodic explosions of the participant's ship accompanied by point losses; Finally, during a 10-s intertrial interval (ITI), targets continue to appear and can be shot for points. On trial 2, the participant enters hiding during the bomb period (and remains there for the duration of the bomb period); this is scored as an escape response (ER, blue arrow) and causes termination of the explosions and point loss, although the bomb cue remains on screen. On trial 3, the participant enters hiding during the warning period (and remains there for the duration of the warning and bomb periods); this is scored as an avoidance response (AR, blue arrow) and causes omission of the explosions and point loss, although the duration of the WS is unaffected. On both trials 2 and 3, the reappearance of targets at the start of the ITI thus provides a visual cue that it is "safe" to emerge from hiding (which occurs on trials 2 and 3 after the start of the ITI, schematized here by green arrows). B Partially-contingent condition. Here, an escape response is made on trial 2, which causes termination of the bomb cue as well as of the explosions and point loss; An avoidance response is made on trial 3 which causes termination of the targets (though not the WS). On trials 2 and 3, targets do not reappear until the participant re-emerges from hiding (green arrows), thus removing the informational cue for when it is safe to emerge from hiding. C Fully-contingent condition. This condition is similar to the partially-contingent condition except that an avoidance response also causes termination of the WS, i.e. there are no on-screen cues at any point while the subject is in hiding
Approximately every 100 ms, the program recorded if the participant's spaceship was inside or outside one of the boxes. For each trial, the program computed the percent of time the participant spent hiding during the 5-s warning period ("avoidance hiding"), as well as during the 5 s that followed the warning period ("escape hiding"). ITI hiding was computed as the percent of time spent hiding during the 10-s ITI. Avoidance, escape and ITI hiding were the dependent variables, and they were analyzed across all trials in all three task conditions, regardless of what cues were shown on screen; thus, for example, "escape hiding" was scored if participants spent time in hiding during the 5-s following the warning period, regardless of whether the bomb actually appeared on that trial. Statistical analyses were conducted using SPSS version 17.0 (SPSS Inc., Chicago, IL). Since the sample size was large (n = 222), normality of groups was not expected to affect analyses. The repeated measures ANOVA included Mauchly's test of sphericity. The assumption of sphericity was violated and thus the Greenhouse–Geisser correction was completed. Post-hoc tests were executed when appropriate with Tukey's HSD. Alpha was set to 0.05. The sample size was determined based on an a-priori power analysis conducted with G*Power [[
The demographics for the two sites and the three task conditions are shown in Table 1. Overall, the sample from the Northeastern site was more diverse in terms of ethnicity than the Western site. Participants at the Northeastern site self-reported race/ethnicity as Caucasian (n = 35), Asian or Pacific Islander (n = 33), Hispanic (n = 20), African–American (n = 14), multi-racial (n = 1) and other (n = 8). Participants at the Western site self-reported race/ethnicity as Caucasian (n = 69), followed by Hispanic (n = 17), African–American (n = 6), Asian or Pacific Islander (n = 4), multi-racial (n = 1) and other (n = 4). There were also differences in age and education level between the two sites, such that participants at the Northeastern site had higher mean age (t(
Table 1 Demographic information for the two sites and the three task conditions
Control condition Partially-contingent condition Fully-contingent condition Sample size 39 40 36 Gender (% female) 41% 53% 47% Mean age (SD) in years 20.2 (2.0) 21.0 (3.1) 20.1 (3.1) Education (SD) in years 12.3 (1.0) 12.4 (1.6) 12.1 (1.6) Sample size 36 36 35 Gender (% female) 53% 53% 46% Mean age (SD) in years 19.0 (1.5) 18.8 (2.3) 19.1 (1.0) Education (SD) in years 12.8 (1.1) 12.8 (1.8) 12.7 (1.0)
There was a significant difference in total scores across the three task conditions (F(
In all three task conditions, hiding in the safe areas during the 5 s that followed the warning period increased during the acquisition phase and decreased during the extinction phase. Specifically, during acquisition, participants learned to escape from (terminate) the aversive outcome by hiding in the safe areas during the bomb period (Fig. 3A). Such escape hiding increased over the 12 acquisition trials for all three conditions of the task, as suggested by a main effect of trial (mixed ANOVA, F(6.995,1532.009) = 208.336, p < 0.001). Escape hiding differed among the three conditions, as suggested by a main effect of condition (F(
Graph: Fig. 3Escape hiding, expressed as percent of the 5-s bomb period where the participant's ship is in hiding. Error bars represent standard error of the mean. A Escape hiding increased across the 12 acquisition trials in all three groups, but was greater in the non-contingent group than in the partially-contingent and fully-contingent groups, which did not differ. B Escape hiding decreased across the 12 extinction trials with a difference between the non-contingent group and the fully-contingent group
Pairwise comparisons among the three conditions revealed that escape hiding was reduced in fully-contingent condition compared to the non-contingent condition (p = 0.002), but the partially-contingent condition did not differ from either the non-contingent or fully-contingent condition (both p's > 0.05). To further examine this interaction, post-hoc independent samples t-tests with alpha corrected to 0.0042 to protect significance levels were conducted. These analyses revealed that the three conditions started at the same level on the first acquisition trial but non-contingent group expressed significantly more escape responding than both the partially-contingent group from the fourth trial on, and the fully-contingent group from the second trial on (all p's < 0.004).
During extinction when the aversive event no longer followed the WS, participants reduced hiding during the 5 s that followed the warning period (Fig. 3B). As expected, given the absence of any aversive event during these trials in all task conditions, there was a sharp decrease in escape hiding that was apparent from the first extinction trial. Interestingly, this was also the case for the two contingent conditions (where aversive events were already omitted during hiding), possibly due to their lower hiding on the last acquisition trial which increased their exposure to the new contingency of extinction trials. Specifically, there was a main effect of trial (F(6.902,1511.569) = 23.588, p < 0.001) as well as a main effect of condition (F(
Further examination of this trial x condition interaction revealed that the non-contingent group expressed significantly fewer escape responses from the second to eleventh trials than did the partially-contingent group. The non-contingent group had more escape responses on the first extinction trial than did the fully-contingent group, but fewer than the fully-contingent group on the fourth and eighth trials (all p's < 0.003). Post-hoc pairwise comparisons indicated that there was a significant difference between the non-contingent condition and the partially-contingent condition (p = 0.002), but not between the fully-contingent condition and either of the other two (both p's > 0.05).
Participants learned to not just escape, but to completely avoid the aversive outcome by initiating the hiding response during the WS, before the start of the bomb period. Avoidance hiding increased over the 12 acquisition trials for all three conditions, as suggested by a main effect of trial (mixed-design ANOVA, F(5.497,1203.932) = 208.336, p < 0.001; Fig. 4A). However, avoidance hiding was considerably higher in the non-contingent than in the two contingent conditions. This observation was supported by a main effect of condition (F(
Graph: Fig. 4Avoidance hiding, expressed as percent of the 5-s warning period during which the participant's ship is in hiding. Error bars represent standard error of the mean. A Avoidance hiding increased across the 12 acquisition trials in all three groups but was higher in the non-contingent group than in the non-contingent and partially-contingent groups, which did not differ. B Avoidance hiding decreased across the 12 extinction trials but was faster to extinguishin in the non-contingent group than the two contingent groups
Participants exhibited a decrease in avoidance hiding across the 12 extinction trials, as suggested by a main effect of trial (F(5.814,1135.249) = 44.213, p < 0.001). There was a main effect of condition for extinction hiding (F(
Lastly, as in previous analyses of the spaceship task [[
Overall, rates of ITI hiding during acquisition were low (Fig. 5A). There was no main effect of trial (p = 0.086) but there was a main effect of condition (F(
Graph: Fig. 5ITI hiding, expressed as percent of the 10-s inter-trial interval (ITI) during which the participant's ship is in hiding. Error bars represent standard error of the mean. A ITI hiding was low across the 12 acquisition trials, but was greater in the partially- and fully-contingent groups than the non-contingent group. B ITI hiding decreased across the 12 extinction trials in all three groups
Rates of ITI hiding during extinction were also quite low (Fig. 5B). There were main effects of condition (F(
The purpose of the current study was to test how stimulus contingencies, particularly the omission of the warning signal (WS), affect escape/avoidance behavior in humans. Participants were tested with a computer-based task previously shown to capture several features of avoidance paradigms in rodents. Specifically, we compared a non-contingent condition in which hiding resulted in termination or prevention of the aversive event (point loss and on-screen explosions) with two new conditions: a fully-contingent condition where hiding caused termination of all on-screen events including the WS, and a partially-contingent condition where hiding caused omission of the aversive cue (bomb) and the appetitive cue (enemy targets), but did not affect the duration of the WS. The current results showed that, while all conditions showed acquisition and extinction of escape/avoidance hiding, there was less acquisition of avoidance and escape hiding in the two contingent conditions compared to the non-contingent condition. We discuss these results and their implications further below.
One motivation for the two contingent conditions in the current study was to consider a version of the human task which more closely resembles the rodent lever-press task. Specifically, in the standard version of the rodent lever-press task, the subjects are given no information about how to predict, escape, or avoid the aversive foot shock, and if a lever-press occurs during the warning period, the rodent receives no informational feedback about whether that response caused omission of the aversive event, or if the event would have been omitted anyway (e.g. during an extinction trial).
In the "standard" non-contingent condition of the human task, the presence of the bomb cue without explosions and point loss provides visual confirmation that the aversive event (explosions and point loss) has been avoided or escaped, while also signaling that it is not yet safe to emerge from hiding. In the two new contingent conditions of the task, this on-screen information is omitted during hiding, which could cause even a well-trained participant to occasionally emerge from hiding to "check" whether it was now safe—reducing total hiding time. This interpretation is consistent with the lower percentage of escape hiding and avoidance hiding, as well as with the very low avoidance rates in the partially- and fully-contingent conditions, relative to the non-contingent condition. It is also interesting to note the lower escape hiding in both contingent conditions was in spite of the increased presence of warning cues (due to less avoidance hiding), which further strengthens these findings.
On the other hand, we expected that in the fully-contingent condition, the additional removal of the WS during hiding would have facilitated learning of the avoidance response. This would have been consistent with prior findings in rats [[
Further, the removal of appetitive targets during the warning period in the two contingent conditions was expected to reduce the incentive to emerge from the safe areas and try to gain points during those periods. In fact, this manipulation appeared to have the opposite effect. One possibility is that the targets functioned as safety signals in the non-contingent group, since they appear during warning period and ITI but not during the bomb period; thus, reappearance of targets following the bomb period serves as informational feedback that it is now "safe" to emerge from hiding. This may have allowed participants in the non-contingent condition to remain in hiding for the duration of the bomb period, rather than need to emerge early to "check" if the threat period had passed. Accordingly, it also allowed these participants to spend less time hiding during the ITI period, when feedback regarding "safety" was provided.
The current study did not manipulate the aversive and appetitive cues separately, but prior work with the spaceship avoidance task has selectively considered the role of safety signals by adding an explicit on-screen safety signal that appeared during the ITI [[
Both fully- and partially-contingent conditions showed lower avoidance responding during the first few trials of extinction, and slower extinction overall, relative to the non-contingent condition. The current finding is a novel effect since, to our knowledge, extinction has not been tested previously in the rodent lever-press avoidance paradigm with contingent or non-contingent WS termination. However, the finding is generally consonant with a prior rodent study indicating that omission of safety signals can slow extinction [[
Understanding the factors that increase or decrease extinction is potentially important; maintaining a previously-useful avoidance behavior, even when the WS is no longer followed by the aversive event, is maladaptive. In fact, pathological avoidance is not typically defined based on whether the individual learns to avoid threat, but rather on the persistence of the avoidance response when the threat is no longer present, i.e. impaired extinction. In the literature, sustained avoidance responding during extinction is related to anxiety and PTSD [[
Future work could continue to explore task manipulations to further understand how contingencies between behavior and stimuli affect avoidance behavior. In the current study, we tested how the termination of both aversive and appetitive cues by the hiding response alters escape or avoidance response. In light of prior evidence that appearance of appetitive cues affects avoidance behavior (e.g., [
Uncertainty has been implicated as a major factor in anxiety disorders [[
A limitation of this study and prior findings is that the participants' knowledge of the contingencies was not directly assessed and could only be inferred based on participants' actions. It would be of interest in the future to include a post-task questionnaire in which the participants report their awareness of the contingencies. Such self-reported knowledge could be examined in relation to patterns of escape and avoidance behavior as well as overall strategies in attempting to maximize points in the task. Another limitation of the current study is that, although an effort was made to recruit participants from two geographically distinct sites within the U.S., representing a number of ethnicities and races, it is still the case that the study sample represents a population that exemplifies what have been termed Western, Educated, Industrialized, Rich and Democratic (WEIRD) societies [[
Prior work with the spaceship avoidance task also included a computational model using a reinforcement-learning algorithm [[
Although not all the results in the current study were as expected, the basic finding confirms that, in a task of human escape/avoidance learning, manipulation of cue contingencies does affect acquisition and extinction of the avoidance response. To the extent that accelerated avoidance is part of the profile in several human disorders, a better understanding of this issue has direct clinical implications. Both escape and avoidance responses are key components of coping behaviors in the face of stressors. On the one hand, the ability to avoid aversive events is often highly adaptive, but it can become maladaptive when an individual first finds a way to escape from a stressful situation and then generalizes that strategy to avoid the situation altogether—a policy which can become dysfunctional if the warning signal is only weakly predictive of the stressor or if there is subsequent unpairing of the warning signal and stressor without extinction of the avoidance response. The ability to separately study escape and avoidance responses, as well as the transition from escape to avoidance, parallels rodent avoidance paradigms and is expected to facilitate our understanding of the symptomology in stress-related psychiatric conditions [[
The current study revealed that avoidance learning can be affected by the omission of aversive and appetitive cues, or omission of both those cues and the WS that predicts the upcoming aversive event. These omissions reduced both escape and avoidance hiding, indicating that in humans as in rodent models, omission contingencies can affect avoidance learning. In addition, the consistency of findings across two sites with diverse undergraduate samples supports the strong reliability of this computer-based task, as well as its utility as an avoidance task for use with human participants.
We would like to thank Nicole Anastasides, Asma Jafri and Michelle Garcia for assistance with data collection.
The contents do not represent the views of the U.S. Department of Veterans Affairs or the United States Government.
Design and conceptualization of the study: MTA, JS, CEM. Recruitment, data collection, data processing and statistical analysis: MTA, JS. All authors contributed to the drafting of the manuscript and approved the final version.
Not applicable.
Available from the corresponding author on reasonable request.
Approval was obtained from the ethics committee of New Jersey Institute of Technology, University of Medicine and Dentistry of New Jersey, and University of Northern Colorado. The procedures used in this study adhere to the tenets of the Declaration of Helsinki.
Informed consent was obtained from all individual participants included in the study.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
By Michael Todd Allen; Jony Sheynin and Catherine E. Myers
Reported by Author; Author; Author