Evaluating Extinction, Renewal, and Resurgence of Operant Behavior in Humans with Amazon Mechanical Turk

Abstract

Amazon Mechanical Turk (MTurk) is a crowdsourcing marketplace providing researchers with the opportunity to collect behavioral data from remote participants at a low cost. Recent research demonstrated reliable extinction effects, as well as renewal and resurgence of button pressing with MTurk participants. To further examine the generality of these findings, we replicated and extended these methods across six experiments arranging reinforcement and extinction of a target button press. In contrast to previous findings, we did not observe as reliable of decreases in button pressing during extinction (1) after training with VR or VI schedules of reinforcement, (2) in the presence or absence of context changes, or (3) with an added response cost for button pressing. However, we found that that a 1-point response cost for all button presses facilitated extinction to a greater extent than the absence of response cost. Nevertheless, we observed ABA renewal of button pressing when changing background contexts across phases and resurgence when extinguishing presses on an alternative button. Our findings suggest that MTurk could be a viable platform from which to ask and address questions about extinction and relapse processes, but further procedural refinements will be necessary to improve the replicability of control by experimental contingencies.

Keywords: Amazon Mechanical Turk; button press; extinction; humans; reinforcement.

Fig. 1.

Objects shown on browser during sessions. Objects on browser included (1) one or two workspaces (350-px by 350-px squares) depending on the experiment, (2) button(s) that were 100-px by 100 px squares with either a red heart or a black club symbol, each situated within its own workspace, (3) yellow star with green text (“+100”) above the target or alternative button indicating reinforcement, (4) red text below the button indicating response cost in some experiments, (5) a point bar with label “POINTS” below, (6) total number of points (e.g., 385), and (7) current monetary gain (e.g., US $0.01925).

Fig. 2.

Experiment 1 Results and Model Predictions. P1=Phase 1; P2=Phase 2. Mean responses per min for groups experiencing a VR or a VI schedule yoked to the mean reinforcement rate for the VR group. Black and gray lines show the predictions from the final mixed-effects model. Error bars represent standard errors of the mean.

Fig. 3.

Prevalence of Meeting an 80%-Reduction Criterion in Phase 2. Percentage of participants demonstrating an 80% reduction in target responding from each participant’s mean Phase-1 target response rate (1) in the last two consecutive bins by 180 s in Phase 2 (Last Two), (2) in the last two consecutive bins between 180 and 360 s in Phase 2 (Last Two > 180 s), or (3) in any two consecutive bins by 180 s in Phase 2 (Any Two).

Fig. 4.

Experiment 2 Results and Model Predictions. P1=Phase 1; P2=Phase 2. Mean responses per min for groups experiencing a contextual change (AB) versus no contextual change (AA) between phases. Black and gray lines show the predictions from the final mixed-effects model. Error bars represent standard errors of the mean.

Fig. 5.

Experiment 3 Results and Model Predictions. P1=Phase 1; P2=Phase 2; P3=Phase 3. Mean responses per min for groups experiencing a contextual change (ABA) versus no contextual change (ABB) between Phases 2 and 3. Data from all Phase-1 bins, the last five Phase-2 bins, and all Phase-3 bins are shown. Black and gray lines show the predictions from the final mixed-effects model. Error bars represent standard errors of the mean.

Fig. 6.

Experiment 4 Results and Model Predictions. P1=Phase 1; P2=Phase 2. Mean responses per min for groups experiencing no response cost (RC-0), a 1-point cost (RC-1), a 5-point cost (RC-5), or a 10-point cost (RC-10) for all responses in Phases 1 and 2. Black and gray lines show predictions from the final mixed-effects model. Error bars represent standard errors of the mean.

Fig. 7.

Experiment 5 Results and Model Predictions. P1=Phase 1; P2=Phase 2; P3=Phase 3. Mean responses per min for groups experiencing the presence or absence of a 1-point cost for each response in Phases 1–3 (RC-1 or NoRC, respectively), and the presence or absence of a context change (ABA or ABB, respectively). Data from all Phase-1 bins, the last five Phase-2 bins, and all Phase-3 bins are shown. Black and gray lines show the predictions from the final mixed-effects model. Error bars represent standard errors of the mean.

Fig. 8.

Experiment 6 Results and Model Predictions. P1=Phase 1; P2=Phase 2; P3=Phase 3. Mean target (top panel) and alternative (bottom panel) responses per min for one group experiencing no response cost (NoRC) and one group experiencing a 1-point cost for all responses in Phases 1–3 (RC-1). Data from all Phase-1 bins, the last five Phase-2 bins, and all Phase-3 bins are shown. Black and gray lines show the predictions from the final mixed-effects models. Error bars represent standard errors of the mean.