Human‐operant evaluation of synthesized alternative reinforcement and resurgence

Abstract

Basic research suggests that differential reinforcement with synthesized alternative reinforcement (i.e., delivery of two or more qualitatively different reinforcers contingent on a single alternative response) increases resurgence during disruptions to the contingency. We evaluated whether these findings extended to human participants using a gamified computer task within a three‐phase resurgence evaluation across three experimental groups. In Phase 1, target responses produced isolated reinforcement across all groups. In Phase 2, target responses were placed on extinction across all groups but alternative responses produced either synthesized, isolated, or no reinforcement based on group assignment. In Phase 3, target and alternative responses were placed on extinction across all groups. Resurgence occurred for groups experiencing synthesized and isolated alternative reinforcement, and resurgence was greater following synthesized reinforcement. Unlike previous human‐operant research, extinction also reliably produced reductions in both target and alternative responding without other programmed contingencies for these responses. These findings may have important implications for clinical practice and future translational and applied research.

Differential reinforcement of alternative behavior (DRA) is highly effective at decreasing target behavior (Petscher et al., 2009). However, research has demonstrated that target behavior is prone to recur when there are changes to the conditions that originally produced the reduction. For example, resurgence is the recurrence of a target response caused by a worsening of reinforcement conditions (Lattal et al., 2017), and this type of behavioral lapse occurs during DRA‐schedule thinning in over 76% of treatment applications (e.g., Briggs et al., 2018; Falligant et al., 2022; Marshall et al., 2025; Muething et al., 2021). Such lapses could have substantial deleterious effects. For example, if the target behavior is aggression or self‐injurious behavior, each lapse could produce physical harm. Further, each lapse has the potential to contact reinforcement and lead to rapid reacquisition of the target response. Such outcomes could erode confidence in the treatment's efficacy and the effectiveness of behavior‐analytic interventions more generally.

These concerns have promoted extensive research on resurgence (Podlesnik et al., 2023), which is typically studied using a three‐phase behavioral assay that is analogous to clinical practice. During Phase 1, a target response produces reinforcement, like a history of reinforcement for challenging behavior in clinical situations. Then, DRA is implemented during Phase 2, wherein an alternative response produces reinforcement and the target response produces relatively less reinforcement (often, it is placed on extinction). During Phase 3, the reinforcement conditions for the target response remain unchanged while the conditions of alternative reinforcement worsen, typically by introducing omissions of alternative reinforcer deliveries or by placing the alternative response on extinction. Extensive research has shown that although the contingency for target responses remains unchanged, target responding tends to recur when reinforcement for the alternative response worsens (Podlesnik et al., 2023).

Research on resurgence has led to two major conceptual explanations of resurgence. One is resurgence as choice (RaC; Shahan & Craig, 2017) and extensions thereof (e.g., resurgence as choice in context: Shahan, Browning, Nist, et al., 2020; temporally weighted matching law: Shahan, 2022), which suggests that resurgence is a function of the same principles governing choice identified within the concatenated matching law (Baum & Rachlin, 1969). RaC explains how these variables continue to influence behavior across time using a scaled temporal‐weighting rule (for reviews, see Devenport & Devenport, 1994; Mazur, 1996). Ultimately, RaC claims that resurgence is due to shifts in the relative value associated with target and alternative responses. When the value of reinforcement associated with alternative reinforcement during DRA decreases, this causes a relative increase in value associated with target responding, thereby causing target responding to recur (i.e., resurgence; see Greer & Shahan, 2019, for a more comprehensive summary).

The other major conceptual explanation of resurgence is context theory (Bouton, 2019; Bouton et al., 2012). In short, context theory states that all learning is context dependent and that more recent events are more context dependent than more temporally distal events (Nelson, 2002). In terms of resurgence, there is a temporally distal context wherein target responses produced reinforcement, then DRA produces a novel, more recent context with a history of target–response inhibition while alternative responses produce reinforcement. When the conditions of reinforcement for alternative responding worsen, this contingency change serves as a change in context (Bouton et al., 2012; Trask et al., 2015). The recent history of target‐response inhibition, which is relatively more context dependent, does not generalize to this new context to the same degree as the more temporally distal history of reinforcement for target responding, which causes the target response to recur (i.e., resurgence).

Although RaC and context theory provide different explanations of resurgence, they have some important convergent predictions. One of these convergent predictions is that more resurgence of target responding will occur when synthesized alternative reinforcement (i.e., delivery of two or more qualitatively different reinforcers contingent on a single response) is suspended than when isolated alternative reinforcement (i.e., delivery of a single reinforcer contingent on a single response) is suspended (Smith et al., 2024). According to RaC, synthesized alternative reinforcement should produce more resurgence because it involves delivery of a larger overall amount of reinforcement contingent on a single response, so suspension of this larger amount of reinforcement would produce a larger shift in relative reinforcement, thereby producing more resurgence (a finding demonstrated by substantial basic and applied research; e.g., Craig et al., 2017; Craig & Shahan, 2016; Falligant et al., 2022; Greer et al., 2024; Shahan, Browning, Nist, et al., 2020; Shahan & Greer, 2021). According to context theory, suspension of synthesized reinforcement produces more resurgence because suspending the two reinforcers associated with synthesized reinforcement (i.e., going from 2 to 0 reinforcers) constitutes a larger context change than suspending the one reinforcer associated with isolated reinforcement (i.e., going from 1 to 0 reinforcers). This more substantial context change produces less generalization of target‐response inhibition, thereby producing more resurgence of target responding. Several studies have shown that larger context changes produce more recurrence of target responding (e.g., Balooch & Neumann, 2011; Havermans et al., 2005; Thomas et al., 2003; Todd et al., 2012).

Smith et al. (2024) conducted an empirical study evaluating differences between synthesized and isolated alternative reinforcement within a three‐phase resurgence evaluation with rats. In Phase 1, rats' target‐lever pressing produced isolated reinforcement (i.e., food only or sucrose solution only). In Phase 2, target responding was placed on extinction and alternative responses produced reinforcement. In one condition, alternative reinforcement produced isolated reinforcement. In the other condition, alternative reinforcement produced synthesized reinforcement (i.e., both food and sucrose). In Phase 3, target and alternative responses were placed on extinction across both conditions. Smith et al. found that rats demonstrated more resurgence following the synthesized reinforcement condition.

This finding is important for at least two reasons. First, this empirical work confirms predictions made by both prominent conceptual explanations of resurgence. Although the exact behavioral processes causing more resurgence following synthesized alternative reinforcement remain unclear (e.g., whether resurgence was greater following synthesized alternative reinforcement because of a larger relative change in alternative reinforcement magnitude, a larger change in stimulus conditions, or a combination thereof), researchers have noted that basic research is most likely to translate to applied situations when multiple competing theories converge to make the same prediction (Fisher et al., 2022). Thus, further translational research evaluating the effects of synthesized alternative reinforcement is more likely to produce similar effects.

Second, this finding from the basic laboratory could have substantial implications for clinical practice. Slaton and Hanley (2018) noted that research on synthesized contingencies has been increasing rapidly, but there has been limited research on the longer term effects of interventions using synthesized reinforcement. Although some research has demonstrated positive longer term effects (e.g., Slaton et al., 2024), these studies have not evaluated the effects of treatments with synthesized alternative reinforcement during decrements in treatment integrity, like those that occur routinely in more applied contexts. The aforementioned finding from the basic laboratory, which suggests that extinction may cause more resurgence following treatments that arrange synthesized alternative reinforcement, warrants further investigation.

This type of investigation is especially important because clinicians may be susceptible to programming synthesized alternative reinforcement during behavior‐reduction treatments when it is not necessary. For example, an individual's challenging behavior may be maintained by access to a specific reinforcer quality, like attention from an adult. In this case, a DRA‐based treatment would involve delivery of attention contingent on an appropriate alternative response, and extensive research has demonstrated that this type of treatment successfully reduces individuals' challenging behavior (Greer et al., 2016; Petscher et al., 2009). This suggests that synthesized alternative reinforcement may not be necessary for this clinical presentation (even though it may be indicated for other clinical presentations, like those whose challenging behavior is maintained by synthesized reinforcement; see McCabe et al., 2025, for review and discussion). There is a growing trend, however, for clinicians to use interview‐informed, synthesized contingency analyses and practical functional assessments to inform their behavior‐reduction treatments (Hanley et al., 2014; Jessel et al., 2016; Rajaraman et al., 2022), even though research has demonstrated that 50% of applications of these assessments may produce false‐positive results (i.e., six of 12 clients in the consecutive case series conducted by Greer et al., 2020; see also Fisher et al., 2016; Kranak & Briggs, 2025; McCabe et al., 2025; and Tiger & Effertz, 2021, for additional empirical data and further discussion). Because basic research suggests that synthesized alternative reinforcement makes treatments more prone to resurgence (Smith et al., 2024, 2026), clinicians may be developing treatments that are unnecessarily susceptible to relapse. Thus, research must evaluate whether human participants are also more susceptible to relapse following synthesized alternative reinforcement, which may guide clinicians away from using synthesized alternative reinforcement within their treatments when it is not necessary.

Human‐operant research using computerized tasks has been useful for translating research with nonhuman animals to human participants (e.g., findings from Shahan, Browning, & Nall, 2020, were translated to humans with computer‐based tasks by Smith & Greer, 2023, and these findings are now being translated to applied situations by Irwin Helvey et al., 2025); however, it is not without its limitations. For example, Saini et al. (2021) underscored a common finding in human‐operant research whereby participants are relatively unresponsive to extinction contingencies—they continue to engage in responses that do not produce programmed consequences (see Bolívar et al., 2017; Cox et al., 2019; Kuroda et al., 2016; Marsteller & St. Peter, 2012; Smith & Greer, 2022, 2023; and Sweeney & Shahan, 2016, for empirical examples of unresponsiveness to extinction during computer‐based tasks or discussions thereof). This poses issues for evaluating resurgence with computer‐based arrangements because three‐phase resurgence evaluations typically program extinction in Phase 3; however, participants often continue to engage in potentially indiscriminate responding in these situations. This may make it difficult for researchers to obtain differentiated results across conditions, obscuring the interpretation of what behavioral processes are affecting behavior.

There are numerous potential explanations for participants' unresponsiveness to extinction in human‐operant experiments. For example, the participants may experience demand characteristics associated with the overall experimental context that led them to engage with response options despite receiving no programmed reinforcement for doing so (Baron & Perone, 1982). If they only have target and alternative response operanda available (e.g., Smith & Greer, 2022, 2023), they may be compelled to engage in these two responses even when both responses are on extinction, especially because the participants may have no other sources of reinforcement toward which they can allocate their responding. Furthermore, these experiments often program arbitrary or low‐value reinforcers, so participants may be susceptible to responding according to rules offered by the experimenter or rules that the participant develops independently rather than responding according to the programmed contingencies (see Craig et al., 2024, for discussion). This could be especially true when there are no stimuli signaling the availability or unavailability of reinforcement, whereas stimulus changes associated with extinction could help reduce behavior more rapidly (Bouton et al., 2012) and help maintain those reductions (e.g., Fuhrman et al., 2016; Fisher et al., 2020; Gámez & Bernal‐Gamboa, 2019). Thus, researchers could potentially improve human‐operant experiments involving extinction contingencies by (a) providing participants with additional response options, (b) programming consequences that are valuable within the experimental context, and (c) signaling the presence and absence of reinforcer availability.

Ultimately, some researchers may question the validity of human‐operant arrangements if they do not produce response patterns consistent with findings in more basic or applied arrangements (see Craig et al., 2024, for discussion). Thus, more research is needed to develop human‐operant arrangements that produce response patterns consistent with basic and applied research in a way that preserves the basic behavioral processes involved.

The purpose of this research was twofold. First, we sought to replicate Smith et al. (2024) by evaluating whether suspension of synthesized alternative reinforcement produces more resurgence than suspension of isolated alternative reinforcement within a three‐phase resurgence evaluation. Second, we developed a novel, computer‐based, human‐operant task to evaluate this primary goal while implementing procedural enhancements to produce response patterns more consistent with basic and applied research on the effects of extinction.

METHOD

Participants and recruitment

All procedures were approved by our universities' institutional review boards (IRBs) and adhered to their ethical guidelines. Researchers recruited participants by directly approaching individuals and asking if they would be interested in participating in an experiment involving playing a computer game for approximately 10 min. We actively recruited participants with affiliations to our respective universities (i.e., students, coworkers, volunteers). We also included participants who learned about the experiment through word of mouth and expressed interest in participating. Participants had to be adults (i.e., 18 or older). We did not offer or provide compensation for participation.

We recruited participants until we had 16 participants for each of our three experimental groups (N = 48). We determined this sample size by conducting a power analysis with data from Smith et al. (2024), who compared resurgence following synthesized and isolated alternative reinforcement with nonhuman animals and obtained an interaction effect size of η² = .070. Our power analysis determined we would need 12 participants per group (N = 36) to achieve .80 power with this anticipated effect size. We conservatively recruited 16 participants per group because data from translational research with humans are often more variable than data from basic research with nonhuman animals, and 16 participants per group has often been sufficient to demonstrate experimental effects in research with human participants using similar procedures, independent variables, and numbers of experimental groups (e.g., Smith & Greer, 2022, 2023).

Our research was exempt from full IRB review, so participants were not required to provide demographic information. We received demographic information from 26 of our 48 participants. These participants' mean age was 31.4 (SD = 14.7, range: 19–68). Eighteen identified as female, eight identified as male, and nobody selected or indicated other options. Four identified as Black, 13 as White or Caucasian, three as Latino or Hispanic, four as Asian, one as both White and Asian, one preferred not to answer, and nobody selected or indicated other options. For our participants' highest level of education, three had completed a master's program unrelated to behavior analysis, 10 were currently enrolled in a master's program for behavior analysis, nine had completed a bachelor's degree, and four had completed high school.

Design and independent variables

We developed computer software to implement a three‐phase resurgence evaluation that was similar to a typical computer game (described in detail below). We used a between‐group design to evaluate the effects of different contingencies during Phase 2 of the three‐phase resurgence evaluation on target and alternative responding in Phases 2 and 3. We evaluated three different contingencies for alternative responses across the three groups: extinction, isolated alternative reinforcement, and synthesized alternative reinforcement (extinction, isolated, and synthesized groups, respectively).

Apparatus, experimental task, and dependent variables

The computer software was coded by the second author using Python, PyCharm. We used a secure Microsoft OneDrive folder to store all relevant files and data. The experimental task was loaded onto Dell Latitude 5540 laptops using Windows 11 in a 1400‐ x 800‐pixel window. The software recorded keystrokes, mouse clicks, and the consequences of these responses in a text file that was automatically stored according to a randomly generated participant identification number in the secure OneDrive folder. The software recorded all keystrokes that produced participants' movement or action. The second author obtained all assets and visual resources using open‐sourced websites (e.g., Itch.io).

The general experimental task consisted of the participant manipulating a spaceship in a space‐themed interface to shoot enemy spaceships. Depending on the experimental phase (see below), one or two enemy spaceships were present on the screen. One spaceship served as a “target” spaceship, and shooting this spaceship served as the target response (i.e., the primary dependent variable). The other spaceship was the “alternative” spaceship, and shooting this one was the alternative response (i.e., the secondary dependent variable). When a target or alternative spaceship appeared, it moved down the screen at 120 pixels per second. They moved down the screen for 1–5 s (determined by a function that randomly selected a random integer between 1 and 5 s with replacement¹), where the spaceship stopped moving vertically and began to move horizontally back and forth across their half of the screen at 180 pixels per second. The target and alternative spaceships always remained on the left and right halves of the interface, respectively, to minimize accidental target or alternative responses. The target and alternative spaceships were also differentiated based on their color and shape (see Figure A1 in the Appendix for images of the spaceships).

Each participant controlled their own spaceship using the laptop's keyboard. The “W,” “A,” “S,” and “D” keys made the participant's spaceship move up, left, down, and right on the experimental interface, respectively, at 360 pixels per second (36 cm per second on the laptops used during this experiment). The participant could shoot a laser vertically by pressing the spacebar key or clicking the left mouse button. Each press of the spacebar key or mouse click would produce only one shot (i.e., holding down the spacebar key or mouse button would not produce multiple shots) to ensure the participant would engage in discrete responses throughout the experiment. A laser could be shot a maximum of three times per second to minimize indiscriminate shooting. Although each keyboard press or mouse click was recorded by our software, shots that hit the target or alternative enemy spaceships were separately recorded as target or alternative responses.

Reinforcement was delivered for target and alternative responses depending on the phase of the experiment (see below). Reinforcement was always signaled by explosion of the enemy spaceship. Whenever a spaceship exploded, a new spaceship immediately flew back into the interface (e.g., exploding a target spaceship caused a new target spaceship to fly into view) at an x‐axis coordinate determined by a function that chose a random integer between the middle of the interface and the left (target ship) or right (alternative ship) boundary so that there was no interruption of opportunities to engage in target or alternative responses.

Two different qualities of reinforcement could be delivered when an enemy spaceship exploded: points or keys. Neither points nor keys were associated with compensation or prizes within or outside the experimental context. The quality of reinforcement varied based on experimental group and phase. When points were delivered, the value of the running total of obtained points, which was continuously displayed in the bottom‐center portion of the interface, increased by 10. When keys were delivered, a key would appear near the location of the exploded enemy spaceship and slowly move toward the participant's spaceship and then disappear upon contact with the spaceship. Then, the running total of accumulated keys would increase by one. The running total for keys was always displayed next to an image of a treasure chest in the lower right corner of the interface.

Procedural enhancements

We implemented several procedural enhancements within this computer‐based task designed to increase the likelihood of observing extinction effects within our three‐phase resurgence evaluation.

First, we created a gamified interface so that we could deliver reinforcers that would have value within the experimental context. Computer‐based, human‐operant experiments often program arbitrary and phylogenetically unimportant reinforcers (e.g., point deliveries; Smith & Greer 2022, 2023), but creating a gamified experiment may address this by creating an overarching context that helps establish value associated with earning these otherwise arbitrary reinforcers.

Second, we programmed stimulus changes that signaled extinction for target and alternative responses. When participants shot target and alternative spaceships when a reinforcement schedule was programmed for the response but reinforcement was not yet available according to the schedule of reinforcement in place (e.g., before the interval elapsed in a VI 3‐s schedule), a translucent forcefield briefly appeared around the enemy spaceship for 0.33 s. When extinction was programmed, this forcefield did not appear after target or alternative responses. We included this stimulus change to signal extinction to help mitigate response persistence during extinction because previous research has demonstrated that stimulus changes occurring simultaneously with the onset of extinction can cause more rapid reductions in responding (e.g., Bouton et al., 2012).

Third, we included additional operanda that served as opportunities to engage in “extraneous responses” and receive “extraneous reinforcement.” Throughout the entire experiment, gems and powerups (i.e., stimuli that caused changes in the experimental interface without affecting the contingencies within the experiment, e.g., making the participant's spaceship spin, changing the color of certain aspects of the interface; see Table A1 in the Appendix for specific descriptions) descended straight down at the speed of 60 pixels per second (6 cm/s on the laptops used during this experiment) from random coordinates across the top of the interface determined by a random‐integer function, which selected a random x‐axis coordinate from a range based on the width of the screen. A gem appeared approximately every 2 s (range: 1–3 s, as determined by a random‐integer‐generating function). To increase the gamification of the software, different powerups were generated at different rates. Some powerups appeared an average of every 15 s (range: 10–20 s), whereas others appeared an average of every 30 s (range 15–45 s), both of which were determined by random‐integer‐generating functions. Participants could collect these gems and powerups by navigating their spaceship to make contact with these stimuli, which would cause the stimuli to disappear. Contacting gems produced no consequences except to increase the value of a gem counter in the lower left corner of the interface (see Appendix Figure A1). Gems could be yellow, blue, or red, and the cumulative total that had been collected of each color was displayed separately. Additionally, there were various powerups, signaled by different images, each associated with different effects (see Appendix Table A1 for detailed list of all powerups and their effects). Notably, the powerups were specifically designed to have no effect on the participant's (a) ability to engage in target or alternative responses or (b) rate of obtaining reinforcement. That is, the powerups did not affect reinforcement rates, enemies' movement, participant's firing rate, participant's firing accuracy, participant's laser strength, or enemies' shield strength. Instead, powerups had the following effects: they changed the color of gems, the participant's spaceship, or the flames propelling the spaceship; briefly (5 s) caused gems to appear at twice the standard rate; caused the participant's spaceship to rotate 360 degrees; caused the participant's spaceship to do a barrel roll (i.e., rotation on both its longitudinal and lateral axes); and doubled the speed of the participant's spaceship for 5 s. Although a counter in the lower left corner of the interface depicted the cumulative number of gems each participant collected, there was no counter depicting the number of powerups each participant collected. The rate at which gems and powerups appeared remained consistent throughout the entire experiment (except for when a powerup briefly increased the gem production rate; see Appendix Table A1 for details).

We included these additional operanda with extraneous sources of reinforcement (i.e., gems, powerups) available because certain formulations of the matching law include sources of extraneous reinforcement as a relevant variable for predicting behavior (e.g., Herrnstein, 1970, Equation 13). Indeed, in clinical settings and in basic experiments, participants and subjects are always able to engage in behaviors irrelevant to experimenter‐programmed contingencies and, presumably, these behaviors have some reinforcing value that competes with the value of engaging in target and alternative responses during extinction. For example, during extinction procedures in a basic lab, rats may groom themselves or explore the experimental chamber. Similarly, during extinction in clinical settings, participants may walk around the room and engage with other materials in the room. In human‐operant arrangements, however, there may be demand characteristics that preclude participants from engaging in extraneous behaviors. For example, many crowdsourcing platforms implement “attention checks” to test whether participants are continuing to engage with their task, and participants can be penalized for failing these checks (e.g., Hauser & Schwarz, 2016). Another possible cause for persistent responding during extinction with human‐operant tasks is that humans tend to prefer doing low‐effort tasks instead of doing nothing at all (Wu et al., 2023). Thus, we created explicit operanda to serve as experimentally permitted extraneous responses for our participants to engage with throughout this experiment.

Procedure

Our universities' IRBs determined that our experiment was minimal risk and exempt, so informed consent was not necessary for participation. We approached prospective participants and asked if they would be interested in participating in an experiment. If prospective participants agreed to participate and were 18 years of age or older, we used block randomization to assign each participant to an experimental group, brought the participant to a quiet space with minimal distractions, and presented them with the experimental software programmed according to their group assignment loaded on the researcher's laptop computer. There was no specified time of day when we approached prospective participants or had them participate in the experiment. The computer software presented the entire experiment, but a researcher (i.e., an author or a research assistant) remained with the participant throughout. If participants had questions, the researcher referred the participant back to the software (e.g., “follow the prompts on the screen,” “the game will tell you when the experiment is over”).

Instructions, experiential learning, and comprehension quiz

The first thing the interface displayed was the instructions for the experiment. The instructions varied across participants in one way: The reinforcer quality described within the instructions matched the quality of the reinforcer the participant would receive during the experiment (i.e., points only, keys only, combinations thereof). The instructions for groups producing an isolated or extinction contingency were as follows:

The goal of the game is to [earn as many points / collect as many keys possible to open the prize box] as possible. You [earn points / collect keys] by destroying enemy spaceships. Use W to move up the screen, S to move down, D to move right, A to move left, and Spacebar or Left Mouse Button to fire the Lasers. Press Spacebar to begin practicing.

The instructions for the groups producing a synthesized contingency were as follows:

The goal of the game is to earn as many points and collect as many keys as possible. If you destroy a spaceship, you may earn points and sometimes you may earn a key. Sometimes destroying spaceships will give you points and a key. Use W to move up the screen, S to move down, D to move right, A to move left, and Spacebar or Left Mouse Button to fire the Lasers. Press Spacebar to begin practicing.

Next, the software presented the participant with the exact interface that would be used in the three‐phase resurgence evaluation. This phase ensured that each participant could successfully engage in the responses necessary to complete the experimental task presented in the three‐phase resurgence evaluation. Thus, participants were required to navigate their spaceship and successfully destroy one enemy spaceship. After destroying one spaceship, the experiential learning phase ended and the software presented the participant with the following instructions:

Nice job! Remember, the point of the game is to [earn as many points as possible and/or earn enough keys to open the prize box]. You are almost ready to start the game. But first, a quick quiz so I am sure you understand the game. Press Spacebar to continue.

After participants completed the experiential learning component, the software presented them with a two‐question, multiple‐choice quiz to evaluate whether the participant understood the instructions. The questions were “What is the goal of the game?” and “How do you [earn points and/or collect keys]?” If a participant selected an incorrect answer for one or both questions, the software informed them that they got a question wrong and returned them to the page displaying the instructions. After this additional opportunity to read the instructions, the participant was required to complete the experiential learning component again before being able to retake the quiz. There was no limit to how many times a participant could retake the quiz. Participants advanced to the experiment after answering both multiple‐choice questions correctly. All participants completed the experiential learning component and quiz components successfully and advanced to the experiment.

Three‐phase resurgence evaluation

Phase 1

During Phase 1, the target spaceship was present. Target responses (i.e., shooting the target spaceship) produced reinforcement according to a VI 3‐s schedule (range: 1–5 s). This and all other VI schedules were generated with a random‐integer‐generating function, which selected the duration of each interval from a range between 1 and 5 s with replacement. Phase 1 lasted 4 min and ended irrespective of participants' response patterns, a duration similar to previous research evaluating resurgence with computer‐based tasks (e.g., Ritchey et al., 2023, used a 5‐min duration).

Similar to Smith et al. (2024), target responding produced isolated reinforcement (i.e., points only or keys only) across all groups and the quality of reinforcement delivered was counterbalanced across participants within each of the three experimental groups (e.g., in the extinction group, eight participants received points and eight received keys). Isolated reinforcement was provided because this is analogous to clinical situations when a client has an isolated function (e.g., access to tangibles only, access to attention only) for their challenging behavior. In this way, we designed this translational experiment to evaluate whether unnecessary use of synthesized reinforcement during DRA may make treatments more prone to resurgence.

Phase 2

After 4 min in Phase 1, the software automatically proceeded to Phase 2 and an alternative spaceship flew into view on the interface. The target spaceship remained present, but shooting the spaceship (i.e., the target response) no longer produced reinforcement (i.e., the spaceship did not explode, points/keys were not delivered, brief forcefield appearances no longer occurred). Phase 2 lasted 3 min regardless of participant's response patterns. This phase duration was somewhat shorter than in previous research evaluating resurgence with computer‐based tasks (e.g., Ritchey et al., 2023, used a 5‐min duration); however, some research has shown that shorter Phase 2 durations may be associated with slightly larger resurgence effects (e.g., Shahan, Browning, & Nall, 2020; Smith & Greer, 2022), so we used a relatively short Phase 2 duration to increase the magnitude of resurgence across all groups and potentially increase the possibility of obtaining differentiation across groups during Phase 3.

Different contingencies were in place for shooting the alternative spaceship (i.e., the alternative response) across experimental groups as follows.

Extinction. Alternative responses did not produce reinforcement (i.e., the spaceship did not explode, points/keys were not delivered, brief forcefield appearances did not occur).

Isolated. Alternative responses produced reinforcement according to a VI 3‐s schedule (range: 1–5 s). Reinforcement was signaled by the explosion of the alternative spaceship, and reinforcement consisted of deliveries of points only or keys only throughout the entire phase. The alternative reinforcer quality (i.e., points or keys) always matched the quality of the reinforcer delivered in Phase 1, which is analogous to function‐based treatments (Fisher & Bouxsein, 2011).

Synthesized. Alternative responses produced reinforcement according to a VI 3‐s schedule (range: 1–5 s). Reinforcement was signaled by the explosion of the alternative spaceship, and reinforcement consisted of deliveries of both points and keys throughout the entire phase regardless of reinforcer quality during Phase 1. Recall that points increased the cumulative total point value immediately, whereas keys would be generated immediately, slowly move toward the participant's spaceship, disappear upon contacting the spaceship, and then increase the cumulative total value of keys collected. This ensured the reinforcers were qualitatively different (e.g., not simply increasing two separate counters) and created a slightly staggered reinforcement delivery to help ensure participants would detect delivery of both reinforcer qualities. This slightly staggered delivery is similar to the synthesized reinforcement deliveries in Smith et al. (2024, 2026), where a dipper of sucrose was presented then immediately followed by a food delivery, or vice versa. This also reflects what may occur in applied situations, where a treatment implementer may need to deliver one reinforcer then pick up another reinforcer and deliver the second reinforcer slightly after (e.g., providing escape as a reinforcer by removing nonpreferred instructional materials immediately, then reaching toward a toy and delivering the toy slightly after the instructional materials have been removed).

Phase 3

Across all groups and participants, neither target nor alternative responses produced reinforcement (i.e., the spaceships did not explode, points/keys were not delivered, brief forcefield appearances no longer occurred). Phase 3 lasted 3 min regardless of participants' response patterns. This phase duration was somewhat shorter than previous research evaluating resurgence with computer‐based tasks (e.g., Ritchey et al., 2023, used a 5‐min duration); however, resurgence is typically defined as changes in target responding from the end of Phase 2 to the beginning of Phase 3 rather than the entirety of Phase 3, decreasing the need to conduct a prolonged Phase 3 duration.

Data analysis

Similar to previous resurgence research conducted with computer‐based tasks (e.g., Ritchey et al., 2023), we graphed and analyzed target and alternative response rates in 1‐min bins and analyzed data on a minute‐by‐minute basis. We supplemented visual analysis with five additional statistical analyses. All statistical analyses were considered significant at α = .05, and Greenhouse–Geisser corrections were applied whenever the assumption of sphericity was violated for within‐subject factors in ANOVA. For our t test, we report effect sizes as Cohen's d. For all ANOVAs, we report effect sizes as partial eta squared (${\mathrm{\eta }}_p^2$).

First, we evaluated differences in target response rates in the last minute of Phase 1 using a one‐way ANOVA. We conducted this analysis to determine whether there were significant differences in target response rates in Phase 1 that might necessitate data transformations (i.e., calculating proportion‐of‐baseline response rates) to compare data across groups.

Second, we conducted a t test to evaluate whether there were significant differences in reinforcement rates between the isolated and synthesized groups at the end of Phase 2. We did not include the extinction group in this statistical analysis because this group did not receive reinforcement in Phase 2 and clearly had a rate of obtained reinforcement rate (i.e., 0 reinforcers per minute) that was different from the other groups. Excluding this group from this statistical analysis increases the statistical power for detecting differences in the groups that were intended to have similar reinforcement rates (i.e., the isolated and synthesized groups). We tested for differences between these two groups because a difference could potentially confound subsequent analyses because different obtained rates of alternative reinforcement in Phase 2 can have substantial effects on resurgence during changes to alternative reinforcement (Craig & Shahan, 2016; Falligant et al., 2022; Greer et al., 2024; Shahan, Browning, Nist, et al., 2020; Shahan & Greer, 2021).

Third, we conducted a 2 × 3 (Phase × Group) mixed‐model ANOVA on target response rates in the last minute of Phase 2 and the first minute of Phase 3. The primary purpose of this analysis was to evaluate whether the amount of change in target response rates during this transition differed across groups (i.e., whether different amounts of resurgence occurred across groups, which would be demonstrated by a statistically significant Phase × Group interaction effect). This analysis strategy is similar to previous research on resurgence, which often focuses on changes in target response rates from the last session of Phase 2 to the first session of Phase 3, except we analyzed data on a minute‐by‐minute basis because 1‐min bins were our primary unit of analysis.

Fourth, we conducted a 2 × 3 (Phase × Group) mixed‐model ANOVA on alternative response rates in the last minute of Phase 2 and the first minute of Phase 3. The primary purpose of this analysis was to evaluate whether the amount of change in alternative response rates during this transition differed across groups (i.e., whether resistance to extinction differed across groups, which would be demonstrated by a statistically significant Phase × Group interaction).

Fifth, we conducted a similar 2 × 3 (Phase × Group) mixed‐model ANOVA of extraneous response rates (i.e., rates of obtaining gems and powerups) during Phases 2 and 3. However, visual inspection of these data showed gradual changes in the rates of extraneous responses after contingency (i.e., phase) changes. Thus, we included data from throughout Phases 2 and 3 for each participant by calculating each participant's mean extraneous response rates for each phase and including these data in our ANOVA. For example, for Participant 1, we calculated the mean extraneous response rate during Minutes 5, 6, and 7 of the experiment (i.e., mean Phase 2 rate) and Minutes 8, 9, and 10 (i.e., mean Phase 3 rate). We calculated these means for each participant and conducted the mixed‐model ANOVA with these data. The primary purpose of this analysis was to evaluate whether the amount of change in extraneous response rates differed across groups. This evaluated whether the groups that experienced a contingency change (i.e., the isolated and synthesized groups) showed amounts of changes in extraneous responses that were different from those in a group that did not experience a contingency change from Phase 2 to 3 (i.e., the extinction group). This would be demonstrated by a statistically significant Phase × Group interaction.

RESULTS

Table 1 summarizes the mean and standard deviation of reinforcement deliveries per minute from the last minute of Phases 1 and 2. The delivery of multiple reinforcers within a single reinforcement interval during synthesized reinforcement was counted as a single reinforcement delivery. Data from the last minute of Phase 3 were not obtained for one participant from the extinction group because the battery on the computer died during the last minute of the experiment. The only statistical analysis that included data from the last minute of Phase 3 was the mixed‐model ANOVA of extraneous response phase means. For this participant, we calculated the Phase 3 mean with the data from the first 2 min of the phase and included this mean in the statistical analysis. Data on reinforcement rate for one additional participant in the isolated group were not obtained due to a programming error.

TABLE 1.

Reinforcement rate summary.

	Reinforcement deliveries per minute
	Extinction		Isolated		Synthesized
	Mean	SEM	Mean	SEM	Mean	SEM
Phase 1 (Target)	10.50	0.67	11.00	0.67	11.00	0.46
Phase 2 (Alternative)	–	–	11.60	0.92	10.38	0.63

Note: Values represent the mean and standard error of the mean (SEM) of reinforcement deliveries per minute. The delivery of multiple reinforcers within a single reinforcement interval during synthesized reinforcement was counted as a single reinforcement delivery. Data are from the last minute of Phase 1 and the last minute of Phase 2.

The data in Table 1 show similar rates of obtained reinforcement across groups at the end of Phase 1. In the last minute of Phase 2, rates of alternative reinforcement were nominally greater in the isolated group than in the synthesized group. Because rates of alternative reinforcement can have substantial effects on resurgence during disruptions to alternative reinforcement (Craig & Shahan, 2016; Falligant et al., 2022; Greer et al., 2024; Shahan, Browning, Nist, et al., 2020; Shahan & Greer, 2021), we conducted a t test to evaluate whether there were significant differences in reinforcement rates between the isolated and synthesized groups at the end of Phase 2. The t test did not show a significant difference in obtained reinforcement rate between these groups, t(29) = 1.09, p = .283, d = .391.

Target responses

Figure 1 depicts mean target response rates for each group throughout the entire three‐phase resurgence evaluation. During Phase 1, mean target responses increased similarly across groups. Appendix A2, A3, A4 show that target responding was elevated for all participants across all groups by the end of Phase 1, with a few exceptions (e.g., one and two participants in the synthesized and extinction groups, respectively, engaged in fewer than 10 target responses in the last minute of Phase 1). We conducted a one‐way ANOVA of target responses in the last minute of Phase 1, which did not show significant differences across groups, F(2, 45) = 0.004, p = .996, ${\mathrm{\eta }}_p^2$ < .001.

FIGURE 1.

Mean target responses in each minute of the three‐phase resurgence evaluation across experimental groups. Error bars represent standard error of the mean.

Figure 1 shows that mean target responses decreased across all groups during Phase 2. Appendix A2, A3, A4 show that target responses decreased with the onset of Phase 2 for most participants, with a few exceptions: Two and four participants in the isolated and extinction groups, respectively, did not demonstrate decreases in target responding with the onset of Phase 2. It is possible, however, that the increases we observed for these participants represent extinction bursts.

Figure 1 shows that mean target responses decreased more rapidly and to a greater extent across Phase 2 in the groups that experienced DRA (i.e., the isolated and synthesized groups) than in the group that experienced extinction only. Mean target responding was slightly lower in the synthesized group than in the isolated group throughout Phase 2; however, the difference between these groups was small. Appendix A2, A3, A4 show that most participants engaged in fewer target responses by the end of Phase 2 relative to their own target response rate at the end of Phase 1, with a few exceptions: One and three participants in the isolated and extinction groups, respectively, did not demonstrate decreases in target responding by the end of Phase 2 relative to their own rate at the end of Phase 1.

Figure 1 shows that mean target responding was highest in the synthesized group in the first minute of Phase 3 (i.e., the resurgence test), demonstrating the greatest increase from the last minute of Phase 2 to the beginning of Phase 3: the largest resurgence effect among all groups. The isolated group had the next highest rate of mean target responding, which was also a substantial increase from responding during the last minute of Phase 2 and indicative of a smaller resurgence effect. Mean target responding was lowest in the extinction group and showed a decrease relative to the last minute of Phase 2, demonstrating the absence of resurgence.

Appendix Figures A3 and A4 show that target response rates increased from the last minute of Phase 2 to the first minute of Phase 3 for most participants in the isolated and synthesized groups, with a few exceptions: One and two participants in the isolated and synthesized groups, respectively, did not engage in more target responses in the first minute of Phase 3 than in the last minute of Phase 2. Appendix Figure A2 shows individual data for participants in the extinction group. These participants did not experience any changes during the transition from Phase 2 to 3, and the numbers of participants who engaged in more or fewer target responses in the first minute of Phase 3 were seven and nine, respectively, representing a relatively equal probability of either pattern. Notably, of the participants in the extinction group who demonstrated an increased target response rate in the first minute of Phase 3, most showed relatively small increases. Only one participant's response rate increased by more than three responses per minute.

To supplement the visual analysis, we conducted a 2 × 3 (Phase × Group) mixed‐model ANOVA of target response rates in the last minute of Phase 2 and the first minute of Phase 3. This ANOVA indicated a significant main effect of phase, F(1, 45) = 11.260, p = .002, ${\mathrm{\eta }}_p^2$ = .088, and a significant Phase × Group interaction, F(2, 45) = 6.80, p = .003, ${\mathrm{\eta }}_p^2$ = .106. These results support the visual analysis that resurgence occurred, and the amount of resurgence differed across groups, respectively, with the greatest amount of resurgence occurring in the synthesized group. This ANOVA also indicated that the main effect of group, F(2, 45) = 0.399, p = .673, ${\mathrm{\eta }}_p^2$ = .008, was not significant.

Figure 1 also depicts mean target response rates during the rest of Phase 3. Mean target responding tended to decrease throughout Phase 3 for all groups, which is consistent with the expected effects of extinction. Appendix A2, A3, A4 show that target response rates remained similar to the rate at the end of Phase 2 (i.e., consistent with continued exposure to extinction and the absence of resurgence) or decreased across Phase 3 (i.e., consistent with resurgence and continued exposure to extinction) for 16, 11, and 15 participants in the extinction, isolated, and synthesized groups, respectively.

Alternative responses

Figure 2 depicts mean alternative response rates for each group throughout the entire three‐phase resurgence evaluation. During Phase 1, the alternative response was unavailable (i.e., the alternative spaceship was absent), so no data are depicted. During Phase 2, mean alternative responses increased in the groups who experienced DRA (i.e., the isolated and synthesized groups). Visual inspection shows that mean alternative responding increased more in the isolated group than in the synthesized group. Appendix A3, A4 show that alternative responses increased during Phase 2 for most participants in the isolated and synthesized groups, with a few exceptions: Two and three participants in the isolated and synthesized groups, respectively, demonstrated decreasing trends in alternative response rates across Phase 2, and one participant in the synthesized group engaged in fewer than 10 responses in the last minute of Phase 2.

FIGURE 2.

Mean alternative responses in each minute of the three‐phase resurgence evaluation across experimental groups. Error bars represent standard error of the mean.

Figure 2 shows that mean alternative responses were slightly elevated when the alternative response became available at the beginning of the phase for the extinction group but decreased throughout the phase with exposure to the extinction contingency, which is consistent with the expected effects of extinction. Appendix Figure A2 shows that all except two participants in the extinction group demonstrated decreasing trends in alternative response rates across Phase 2.

Figure 2 shows that mean alternative response rates decreased during the transition from Phase 2 to 3 (i.e., the resurgence test) for the isolated and synthesized groups and that mean alternative response rates remained similar during this transition for the extinction group, which is consistent with the onset and continuation of extinction, respectively. Appendix A3, A4 show that alternative responses decreased during this transition for most participants in the isolated and synthesized groups, with a few exceptions: Three and two participants in the isolated and synthesized groups, respectively, did not demonstrate decreases in alternative response rates during this transition. It is possible, however, that the increases we observed for these participants represent extinction bursts. Appendix Figure A2 shows that alternative responses increased, remained the same, or decreased for five, six, and four participants, respectively, for participants in the extinction group. Observed changes in response rates tended to be small, with 11 participants demonstrating a change in alternative response that did not exceed an absolute value of five, which is consistent with the continuation of previous contingencies.

We conducted a 2 × 3 (phase × group) mixed‐model ANOVA of alternative response rates in the last minute of Phase 2 and the first minute of Phase 3 to compare resistance to extinction across groups. This ANOVA indicated a significant effect of group, F(2, 45) = 12.630, p < .001, ${\mathrm{\eta }}_p^2$ = .298, and phase, F(1, 45) = 16.120, p < .001, ${\mathrm{\eta }}_p^2$ = .039, and a Phase × Group interaction, F(2, 45) = 4.810, p = .013, ${\mathrm{\eta }}_p^2$ = .023, suggesting that rates of alternative responding differed across groups and differed between the end of Phase 2 and beginning of Phase 3 and that the amount that alternative responding changed during this transition differed across groups.

Figure 2 shows that mean alternative responses decreased from high to low rates across Phase 3 for the groups who had experienced DRA during Phase 2 (i.e., the isolated and synthesized groups). Appendix Figures A3 and A4 show decreasing trends in alternative response rates across Phase 3 for most participants, with a few exceptions: Two participants in each of the isolated and synthesized groups did not demonstrate decreasing trends in alternative response rates across Phase 3. The decreasing trend for most participants is consistent with the expected effects of extinction.

Figure 2 shows that mean alternative responding for the extinction group remained low throughout Phase 3. Appendix Figure A2 shows similar or decreasing alternative response rates across Phase 3 for all participants in the extinction group, which is consistent with the expected effects of continued exposure to extinction.

Extraneous responses

Figure 3 depicts mean extraneous response rates for each group throughout the entire three‐phase resurgence evaluation. When target reinforcement was available during Phase 1, mean extraneous response rates initially increased and were similar across all three groups by the end of Phase 1. Appendix A2, A3, A4 show that target response rates exceeded extraneous response rates by the end of Phase 1 for most participants, with a few exceptions: Two, four, and four participants in the extinction, isolated, and synthesized groups, respectively, engaged in higher rates of extraneous responses.

FIGURE 3.

Mean extraneous responses (i.e., collecting gems and powerups) in each minute of the three‐phase resurgence evaluation across experimental groups. Error bars represent standard error of the mean.

During Phase 2, mean extraneous response rates in the extinction group, which did not receive reinforcement for alternative responses, were visually elevated relative to the groups that received reinforcement for alternative responses (i.e., the isolated and synthesized groups). Appendix Figure A2 shows that extraneous response rates exceeded alternative response rates by the end of Phase 2 for all participants in the extinction group. Appendix Figures A3 and A4 display extraneous response rates for individual participants in the groups experiencing DRA (i.e., isolated and synthesized groups, respectively), which tended to demonstrate the opposite effect. All but one and three participants in the isolated and synthesized groups, respectively, demonstrated higher rates of alternative responses relative to extraneous responses by the end of Phase 2.

Figure 3 shows that during Phase 3, mean extraneous response rates in the extinction group remained similar to rates for this group at the end of Phase 2, whereas these responses increased across Phase 3 for the isolated and synthesized groups. For each participant, we calculated phasewide extraneous response rate means. Phasewide extraneous response rate means demonstrated that extraneous responses occurred at overall higher rates in Phase 3 relative to Phase 2 for only two participants in the extinction group, whereas extraneous responses occurred at higher rates in Phase 3 than in Phase 2 for all except two and three participants in the isolated and synthesized groups, respectively. This demonstrates a general tendency for extraneous response rates to increase with the onset of extinction for target and alternative responses (i.e., frequent increases in phasewide means for the isolated and synthesized groups) but not with the mere continuation of extinction (i.e., infrequent increases in phasewide means for the extinction group).

To supplement individual‐participant data analysis, we used the phasewide means to conduct a 2 × 3 (phase × group) mixed‐model ANOVA of extraneous response rates in Phases 2 and 3 to test for groupwide differences. This ANOVA indicated a significant main effect of phase, F(1, 45) = 31.610, p < .001, ${\mathrm{\eta }}_p^2$ = .106, and a significant Phase × Group interaction effect, F(2, 45) = 3.830, p = .029, ${\mathrm{\eta }}_p^2$ = .026. These results suggest that there was a difference in the individuals' mean rates of extraneous responses between Phases 2 and 3 and that the amount of this difference varied across groups. The main effect of group was not significant, F(2, 45) = 1.832, p = .172, ${\mathrm{\eta }}_p^2$ = .054.

DISCUSSION

We examined whether synthesized alternative reinforcement produced more resurgence than isolated alternative reinforcement within a novel, gamified, human‐operant three‐phase resurgence evaluation. Our results generally replicated those obtained with nonhuman animals (Smith et al., 2024) and were consistent with predictions made by both RaC (Shahan & Craig 2017) and context theory (Bouton, 2019; Bouton et al., 2012): Exposure to extinction after DRA with synthesized reinforcement produced more resurgence than DRA with isolated reinforcement. Furthermore, the software we used to conduct our human‐operant three‐phase resurgence evaluation produced response patterns consistent with extinction, an effect that researchers have had difficulty producing previously (Saini et al., 2021; see also Craig et al., 2024, for discussion).

These results have important implications for applied researchers and potentially for clinical practice. Slaton and Hanley (2018) identified an increase in research on synthesized contingencies, and our research suggests that including synthesized alternative reinforcement may cause more resurgence than treatments with isolated alternative reinforcement. Increased resurgence could have substantial negative effects. If the behavior targeted for reduction is potentially harmful, each instance of resurgence could produce injury, and increased resurgence suggests an increased probability of these injuries. Increased resurgence could also increase the probability that the dangerous behavior could produce reinforcement and cause rapid reacquisition. Additionally, each recurrence would likely diminish stakeholders' confidence in the treatment and perhaps behavior‐analytic interventions in general. Thus, our finding that synthesized alternative reinforcement produces more resurgence than isolated reinforcement with human participants warrants further investigation with clinical populations and situations.

This need is emphasized by additional translational research being conducted with nonhuman animals. For example, Smith et al. (2026) recently demonstrated that resurgence may occur when only one of the reinforcers within a synthesized alternative reinforcement contingency is withheld. In their series of experiments, rats experienced isolated reinforcement (delivery of food only or sucrose only) for target responses, then rats experienced synthesized reinforcement (delivery of both food and sucrose) contingent on alternative responses. Last, Smith et al. removed either food or sucrose from the synthesized contingency while alternative responses continued to produce the reinforcer that had previously maintained target responding. The removal of a single reinforcer from the synthesized contingency produced resurgence in 70% of the rats. In their experiment, the addition of a second reinforcer during treatment created a novel situation for a reinforcer to become unavailable and cause resurgence. This finding is consistent with RaC and context theory and should also be translated to human participants because synthesized alternative reinforcement may have problematic effects beyond those evaluated herein.

The present experiment also has important implications for translational research with computer‐based, human‐operant tasks. It is well documented that similar research paradigms have failed to produce response patterns consistent with basic or applied research, especially in terms of producing response patterns consistent with extinction contingencies (Saini et al., 2021). Previous research has attempted to address this, for example, by increasing response effort (e.g., Cox et al., 2019; Smith & Greer, 2022, 2023) or including response‐cost procedures (e.g., Ritchey et al., 2023); however, the aforementioned response‐effort manipulations still failed to produce extinction, and inclusion of punishment contingencies introduces additional behavior processes that may make the procedure less analogous to clinical situations.

In this experiment, we attempted to produce an extinction effect at different points during the experiment (i.e., Phase 2 for the extinction group, Phase 3 for the isolated and synthesized groups; akin to a multiple‐baseline design). In general, our results demonstrated extinction effects for most participants. Target and alternative responses tended to decrease when and only when extinction was introduced. Furthermore, these decreases produced a decelerating curve, consistent with the characteristic “extinction curve” established in both basic and applied research (Fisher et al., 2023).

Our procedures differed from previous human‐operant research along several dimensions to produce this effect. We gamified our experimental task, we programmed explicit extraneous responses and reinforcement, and we signaled extinction. Any of these modifications may have helped produce extinction effects. For example, the reinforcers in human‐operant tasks are often arbitrary and phylogenetically unimportant, but gamification may help address this by creating an experiential context whereby the reinforcers are valuable for the participant during the experiment per se. Creating a situation where the programmed reinforcers have real value for the participant within the experimental context as opposed to using delayed backup reinforcers (e.g., small amounts of money) after the experiment may permit better detection of the effects of nonreinforcement in human‐operant experiments.

Inclusion of extraneous operanda may have helped produce extinction effects by creating opportunities to allocate time and responding toward other activities during extinction. This manipulation may have influenced responding through a variety of behavioral processes. For example, human‐operant participants may experience demand characteristics such that they feel compelled to continue to engage with the experimental task even when the task does not program contingencies to support their continued engagement. Furthermore, participants are often explicitly instructed to respond during experiments, which may produce behavior that is rule governed rather than contingency governed and potentially enhance the effects of these demand characteristics. Then, if engagement with the task is also restricted to target and alternative responses, participants may continue to engage in these responses because they feel compelled to meet the expectations of the experimenter or follow the rules they have been given or that they imposed on their own behavior. However, programming extraneous responses may allow participants to redistribute their responding toward other operanda while remaining engaged with the experiment. This procedural enhancement may provide participants with a way to continue to meet the experimenter's expectations and follow the rules without needing to continue to engage in target and alternative responses, thereby leading to lower target and alternative response rates during extinction than in arrangements without these additional operanda.

Additionally, these other operanda may have provided extraneous sources of reinforcement. The matching law has demonstrated that including parameters to account for extraneous sources of reinforcement can improve predictions of behavior (Herrnstein, 1970), so it is possible that explicitly including these sources of reinforcement may cause patterns of human‐operant behavior to align more closely to behavior in other situations when these sources of extraneous reinforcement (e.g., grooming, looking out the window) are inherently present. Similarly, previous research has demonstrated that human participants prefer to engage in menial tasks rather than doing nothing (Wu et al., 2023), so programming extraneous operanda and reinforcement may have provided participants with an opportunity to engage in a menial task other than repetitively engaging with nonfunctional operanda. Regardless of the exact mechanisms by which the extraneous responses/reinforcement may have helped produce response patterns more consistent with extinction, the increase in extraneous responses when and only when (a) extinction was implemented for both target and alternative responses and (b) target and alternative responses decreased suggests that this procedural modification may have played an important role.

Our procedures also programmed stimulus changes that signaled extinction. When extinction was in effect, target and alternative spaceships no longer exploded and a forcefield that previously signaled the presence of a reinforcement schedule (despite the response being unreinforced) no longer appeared. This is similar to clinical practice, during which there are often stimulus changes associated with extinction. For example, a caregiver speaking on the phone or with another person may signal extinction for requests for attention, or a caregiver placing an iPad in a drawer and closing it may signal extinction for requests for the iPad. Notably, it is possible that explicit extinction signals enhance the effects of extinction. For example, research has demonstrated that introducing extinction at the same time as a change in context will extinguish responding more rapidly (Bouton et al., 2012). Furthermore, stimuli associated with extinction also help maintain extinction effects across time and situations. For example, signals associated with extinction can mitigate both renewal (e.g., Gámez & Bernal‐Gamboa, 2019) and resurgence (e.g., Fisher et al., 2020; Fuhrman et al., 2016). Thus, our inclusion of signals associated with the extinction contingency may have helped produce response patterns consistent with extinction more rapidly and completely within the relatively short duration of our experiment.

Discussion of these three different factors that may have helped produce extinction effects also highlights a primary limitation of our experiment—we did not isolate the effects of any of these variables. The relative effects of each procedural enhancement could certainly be elucidated in future research; however, our experiment still demonstrates a positive effect related to our procedural enhancements that could be beneficial for other translational researchers to use in their research. Moreover, demonstration of this effect was secondary to the primary purpose of our study, which was to translate the effects of synthesized alternative reinforcement on resurgence to human participants.

There are several notable limitations to our experiment in relation to this purpose. For example, our research was conducted with a nonclinical population with analogues to socially significant behavior specifically to test our hypotheses in a context that minimized risk of harm before conducting an experiment in the context of clinical care (Greer et al., 2022), but this also means it remains unclear whether the effects we observed will generalize to more applied situations. However, our experiment, in conjunction with empirical findings from the basic laboratory and conceptual support from well‐established behavioral frameworks (i.e., RaC, context theory), demonstrate a clear need for researchers to evaluate this in more applied situations. Another limitation is that our research always programmed isolated reinforcement for target responses during Phase 1. There may be situations when multiple qualitatively different reinforcers maintain target responding, so future research should explore arrangements that account for this type of clinical presentation. Additionally, our research only programmed sources of positive reinforcement; however, research has shown that treatments with synthesized contingencies often include a negative reinforcement contingency. There has been some recent research on treatment lapses in relation to negative reinforcement contingencies (e.g., Alessandri et al., 2015; Bruzek et al., 2009; Mitteer et al., 2018; Velasquez et al., 2025; Williams et al., 2023), and researchers should consider including similar contingencies in evaluations of treatment lapses following synthesized reinforcement.

This experiment also highlights future directions for basic research. Neither this experiment nor previous research (Smith et al., 2024, 2026) differentiates the exact behavioral processes that cause synthesized alternative reinforcement to produce more resurgence during disruptions to the alternative reinforcement contingency. It is possible that synthesized alternative reinforcement makes behavior more susceptible to relapse because synthesized reinforcement produces higher value reinforcement than isolated reinforcement. This would be consistent with prior research demonstrating that disruptions to higher value (e.g., larger magnitude) alternative reinforcement tends to produce more resurgence (e.g., Craig et al., 2017) and RaC's conceptualization of the basic processes that cause resurgence (Shahan & Craig, 2017). It is also possible that suspension of multiple, qualitatively different stimuli (as with synthesized reinforcement) is a larger context change than suspension of a single stimulus (as with isolated reinforcement), which could cause less generalization of the treatment and more resurgence. This interpretation is consistent with previous empirical research (e.g., Balooch & Neumann, 2011; Havermans et al., 2005; Thomas et al., 2003; Todd et al., 2012) and the conceptualization of relapse advanced by context theory (Bouton, 2019; Bouton et al., 2012).

In future basic research, researchers could directly compare the relative effects of changes in reinforcer value (e.g., magnitude) with those of changes in reinforcer type (e.g., quality). For example, researchers could compare the extent to which relapse occurs due to changes in reinforcer (a) type while keeping the overall reinforcement value equal and (b) value while the type of reinforcer remains constant. To do this, researchers could identify values of food and sucrose reinforcement that are approximately equal for rats and deliver equal values. Then one group of rats would receive both food and sucrose during Phase 2 of a three‐phase resurgence evaluation followed by a switch to delivering the same overall value reinforcement with a single reinforcer type in Phase 3. These Phase 2 and 3 conditions could be compared with a group that receives a single reinforcer type during Phase 2 followed by a decrease in reinforcement value (e.g., two to one reinforcer per reinforcement delivery) during the transition to Phase 3. Additional groups of rats could be included that experience combined changes to reinforcement type and value (i.e., similar to the synthesized group from Smith et al., 2024). Such comparisons could isolate the effects of changes to reinforcer value and type while also evaluating how these changes may interact to produce the greater amount of resurgence observed in this experiment.

Conclusions

Extinction following synthesized alternative reinforcement produced more resurgence than extinction following isolated alternative reinforcement. This finding is consistent with basic research and prominent, well‐established conceptual explanations of behavior. If this finding generalizes to more applied situations, it suggests that synthesized alternative reinforcement could have downsides that clinicians should consider. Future research should evaluate whether this effect generalizes to more applied situations.

Our human‐operant preparation demonstrated a basic behavioral process (i.e., extinction) that has been difficult to produce with other human‐operant arrangements. To do this, we included several procedural variables that have not been included in other arrangements. Future researchers could consider evaluating the specific roles of such procedural variations. Continued refinement of such behavioral assays may enhance the translation of research from basic laboratories to applied contexts. Future basic research should also be directed toward discerning the relative effects of magnitude and quality manipulations and interactions thereof on treatment relapse.

AUTHOR CONTRIBUTIONS

Sean W. Smith: Conceptualization, investigation, funding acquisition, writing original draft, review, and revisions, methodology, validation, visualization, formal analysis, project administration, data curation, supervision, resources.

Courtney R. Mauzy IV: Investigation, funding acquisition, methodology, validation, visualization, review and revisions, software, formal analysis, project administration, data curation, supervision, resources.

Casey Irwin Helvey: Investigation, methodology, visualization, formal analysis, project administration, data curation, supervision, resources.

Brian D. Greer: Investigation, methodology, visualization, review, revisions, formal analysis, project administration, data curation, supervision, resources.

William E. Sullivan: Funding acquisition, methodology.

Henry S. Roane: Funding acquisition, review, and revisions.

Andrew R. Craig: Review, revisions, funding acquisition, methodology, formal analysis, visualization, supervision.

CONFLICT OF INTEREST STATEMENT

Drs. Smith and Mauzy disclose as a potential conflict of interest that they developed and own the intellectual property of the software used in this study.

ETHICS APPROVAL

This research was conducted with approval from the SUNY Upstate Medical University and Rutgers University IRBs.

APPENDIX A.

TABLE A1.

Extraneous responses.

Type	Effect	Appearance schedule
Gems	Collectible counter	VT 2 s
Spin	360‐degree spin animation	VT 15 s
Roll	Barrel‐roll animation	VT 15 s
Speed Boost	Increased ship speed to 14 pixel/frame for 5 s	VT 30 s
Flame Color Changer	Changed flame color to red or blue	VT 30 s
Gem Boost	Increased gem appearance frequency	VT 30 s
Gem Color Changer	Changed gem color to blue, red, green	VT 30 s

Note: Variable‐time (VT) intervals were selected by a function that randomly selected an integer between a range of 1–3 s (VT 2 s), 10–20s (VT 15 s), or 20–40s (VT 30 s).

FIGURE A1.

Experimental interface. The top panel displays stimuli on the experimental interface during Phase 1. The bottom panel displays stimuli on the experimental interface during Phases 2 and 3. In both panels, the spaceship closest to the bottom of the image is the participant's spaceship. Above that spaceship, the target spaceship is on the left side of the interface (top and bottom panels), and the alternative spaceship is on the right side of the interface (bottom panel only). In both panels, the counter displaying the cumulative number of gems of each color collected by the participant appears in the bottom‐left corner, the counter displaying cumulative obtained points appears in the bottom middle, and the counter displaying cumulative obtained keys (“0 / 150” indicates “0” keys were obtained and the “150” implies that up to 150 could be obtained, although this was not actually possible) appears next to a treasure chest image in the bottom‐right corner (intended to imply that collecting keys would eventually produce treasure). In both panels, small yellow prisms depict yellow gems. Gems could also be blue or red (not displayed). All other stimuli represent various powerups that could be obtained.

FIGURE A2.

Individual participant response rates: Extinction group. Each panel represents target, alternative, and extraneous responses represented by closed circles, open circles, and open squares, respectively, for individual participants within the extinction group.

FIGURE A3.

Individual participant response rates: Isolated group. Each panel represents target, alternative, and extraneous responses represented by closed circles, open circles, and open squares, respectively, for individual participants within the isolated group.

FIGURE A4.

Individual participant response rates: synthesized group. Each panel represents target, alternative, and extraneous responses represented by closed circles, open circles, and open squares, respectively, for individual participants within the synthesized group.

Smith, S. W. , Mauzy, C. R. IV , Helvey, C. I. , Greer, B. D. , Sullivan, W. E. , Roane, H. S. , & Craig, A. R. (2026). Human‐operant evaluation of synthesized alternative reinforcement and resurgence. Journal of the Experimental Analysis of Behavior, 126(2), e70134. 10.1002/jeab.70134

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.