On the Efficacy of and Preference for Signaling Extinction in a Multiple Schedule

Abstract

Previous basic research has shown that signaling the extinction component of a compound schedule can be aversive and nonpreferred. However, such discriminative stimuli are common when thinning schedules of reinforcement in practice, and they provide several advantages to clinicians. A limitation of previous applied studies on different arrangements of discriminative stimuli is that researchers have used identical stimuli to signal the availability of reinforcement across conditions that do and do not signal extinction, often doubling exposure to the stimulus signaling the availability of reinforcement. The present experiments corrected this limitation by comparing multiple-schedule arrangements that do and do not signal extinction when unique stimuli signal each component across conditions. Results from three participants indicated that both multiple-schedule arrangements were similarly efficacious when teaching the successive discrimination. However, response patterns differed when testing under a concurrent-operants arrangement, suggesting different patterns of preference across various multiple-schedule arrangements.

The development of functional analysis methodology (Iwata et al., 1982/1994) enabled the identification of function-based interventions that target the variables maintaining destructive behavior. One common function-based intervention is differential reinforcement of alternative behavior (DRA) in which behavior analysts often place destructive behavior on extinction and teach the individual an alternative response (e.g., a functional communication response) to gain access to the reinforcer maintaining destructive behavior (i.e., functional communication training; Carr & Durand, 1985; Fisher et al., 1993). This approach to treatment is beneficial in that it teaches a socially appropriate response that replaces destructive behavior. However, applications of DRA begin with an impractically dense schedule of reinforcement. To make the intervention more practical for caregivers to implement in the home, school, or community, behavior analysts must thin the schedule of reinforcement maintaining the alternative response.

One effective way to thin schedules of reinforcement is through the use of multiple schedules (Greer et al., 2016; Saini et al., 2016). Multiple schedules are compound schedules of reinforcement that use schedule-correlated stimuli to signal periods in which reinforcement is available (in this case, for the alternative response) and periods in which the same response is on extinction. Individuals first learn to discriminate between the stimulus correlated with reinforcement (S^D) and the stimulus correlated with extinction (S^Δ). Then, the behavior analyst systematically increases the duration of the extinction component, relative to that of the reinforcement component, until the individual tolerates progressively longer periods of time without reinforcement. However, even with multiple schedules of reinforcement, recurrence of destructive behavior can occur during reinforcement schedule thinning (Briggs et al., 2018).

Applied researchers have studied the efficacy of and preference for including schedule-correlated stimuli in different multiple-schedule arrangements. For example, Tiger et al. (2006) examined the efficacy of compound schedules that included (a) an S^D and an S^Δ, (b) an S^D only, and (c) neither an S^D nor an S^Δ (i.e., a mixed schedule of reinforcement). Both multiple-schedule arrangements were effective in teaching children to discriminate when reinforcement was and was not available, while the mixed schedule was generally ineffective.

Tiger et al. (2008) extended this work by evaluating the use of brief vocal statements as schedule-correlated stimuli in different multiple-schedule arrangements. A brief verbal cue (i.e., “It is your time” or “It is my time”) delivered at the start of the reinforcement and extinction components served as the S^D and S^Δ, respectively. The S^D and S^Δ arrangement was more effective than an S^D-only arrangement in producing discriminated responding for three of the four participants.

A systematic replication of the Tiger et al. (2006) study conducted by Landa and Hanley (2016) produced somewhat discrepant results. Landa and Hanley evaluated different multiple-schedule arrangements for the treatment of high-rate requests when using the Picture Exchange Communication System (PECS; Bondy & Frost, 1993). Participants had a three-ring binder with pictures of different edibles they could request. Researchers placed colored pieces of paper under the binder to signal reinforcement and extinction components. For example, in the condition in which researchers signaled both the reinforcement and extinction components, a green piece of paper under the binder served as the S^D, and a red piece of paper under the binder served as the S^Δ. For both participants, the S^D-only arrangement was effective at teaching the multiple-schedule discrimination. The authors suggested that this finding could be the result of the saliency of the stimulus changes programmed in each condition. That is, the presentation and removal of the green paper (i.e., S^D only) may have been more salient than transitioning from one color of paper to another (i.e., S^D and S^Δ).

A limitation of previous applied research in this area is that researchers often programmed the same stimulus to serve as the S^D across experimental conditions that did and did not signal extinction explicitly, which may have increased the likelihood of carryover effects across conditions. That is, the establishment of stimulus control under one multiple-schedule arrangement (e.g., S^D and S^Δ) may be more likely to generalize to another multiple-schedule arrangement (e.g., S^D only) if both conditions share common stimulus features (e.g., a shared S^D). Correcting this issue is straightforward. Using three distinct schedule-correlated stimuli (i.e., one stimulus for each of the two S^Ds and a third for the S^Δ) would allow for a more independent evaluation of each multiple-schedule arrangement. Pizarro et al. (2020) recently evaluated whether programming unique discriminative stimuli across experimental arrangements improved the successive discrimination of a multiple schedule using a group design. Although these investigators found no difference between the two groups of participants in the time to master or the ability to discriminate the different multiple-schedule arrangements, the authors recommended continued research on using unique stimuli to reduce the possibility of carryover effects.

The objective of this research is to better understand optimal arrangements of schedule-correlated stimuli when used in application, where determining whether to signal extinction explicitly is an important consideration. In Experiment 1, we compared the efficacy of two multiple-schedule arrangements, one that signaled extinction and another that included no such signal, when all signals were distinct within and across the two arrangements. In Experiment 2, we examined preference across various multiple-schedule arrangements using a concurrent-operants procedure rather than the more traditional concurrent-chains procedure that applied researchers have exclusively used when evaluating preference for multiple-schedule arrangements.

Given the methodological differences across previous studies on this topic, we conducted sessions using analog responses to allow us to better focus on the basic principles affecting behavior under the different multiple-schedule arrangements and to minimize the likelihood that destructive behavior would interfere with implementation of the experimental procedures. Additionally, we programmed an analogue of destructive behavior because treatment of destructive behavior often involves programming multiple schedules (Greer et al., 2016). Incorporating a target response also helped extend prior research on this topic that has generally focused on teaching the discrimination of an appropriate communication response outside the context of treating problem behavior (Tiger et. al., 2006; 2008).

General Method

Participants and Setting

We recruited three participants from the Center for Autism Spectrum Disorders and the Pediatric Feeding Disorders Program at the University of Nebraska Medical Center’s Munroe-Meyer Institute. Tommy was an 8-year-old male, diagnosed with Autism Spectrum Disorder (ASD) who communicated using full sentences. Emmitt was a 3-year-old male who also had a diagnosis of ASD and who communicated using full sentences. Riley was a 7-year-old male who also had a diagnosis of ASD but who communicated using short phrases. All participants could manipulate an iPad independently and engaged in little to no destructive behavior.

Therapists conducted all sessions in 3-m by 3-m rooms that contained a table, two chairs, and any materials (e.g., target-response pad, communication card, reinforcers, schedule-correlated stimuli) necessary for the upcoming session. Session rooms also contained a one-way observation window, behind which individuals collected data and recorded videos of sessions. Each session room also contained a small window that let in natural light so that the participant could see the session materials when the overhead lights were turned off.

Materials

We used a 15-cm by 15-cm foam pad for the target response and an index card with a picture of preferred toys for the alternative response. The highest-preference item identified from a preference assessment (see below) constituted the programmed reinforcer throughout the study. We used Phillips Hue light bulbs to project different colors of light as the schedule-correlated stimuli and the Phillips Hue mobile-phone application to control the colors of the lights.

Pre-Experimental Assessments

Paired-Stimulus Preference Assessment

Therapists conducted a paired-stimulus preference assessment similar to that described by Fisher et al. (1992) to determine a high-preference item for each participant. A therapist presented the participant with two stimuli at a time and instructed the participant to “pick one.” After choosing an item, the therapist allowed the participant to play with the chosen item for 30 s. The therapist then removed the item and delivered the next pair of stimuli. After the therapist presented each stimulus with every other stimulus, we determined a preference hierarchy. We used each participant’s highest-preference item as the programmed reinforcer throughout the study.

Color Preference Assessment

Therapists conducted a color preference assessment similar to that conducted by Heal et al. (2009) to avoid the potential of a color bias influencing choices between the multiple-schedule arrangements. A therapist presented two colors at a time (via colored cards or colored lights) and instructed the participant to “pick one.” For Tommy, therapists used index cards of colored paper during the color preference assessment to determine which colors to program as schedule-correlated stimuli. For Emmitt and Riley, therapists used the Phillips Hue light bulbs during the color preference assessment. Therapists ensured that all colors were easily discriminable from one another by having the participants tact each color chosen in the preference assessment. No participants had difficulty discriminating the colors. After the therapist presented all colors with every other color, the therapists determined a preference hierarchy. We used three moderately preferred colors for each participant as the schedule-correlated stimuli for the multiple schedules.

Experiment 1

Method

Experiment 1 consisted of baseline, DRA, and multiple schedule (mult) DRA extinction (EXT). We used a reversal design to show the efficacy of DRA and a multielement design with an S^D-only condition and an S^D and S^Δ condition to demonstrate experimental control over the two mult DRA EXT conditions. All sessions lasted five min. We conducted sessions consecutively with approximately one minute between each. We conducted research with Tommy for two hours per day and conducted approximately 8–10 sessions per day. The study took place over approximately 2 weeks. We conducted research with Emmitt and Riley for 30 min per day and conducted approximately 4 sessions per day. Emmitt completed Experiment 1 in 4 weeks, and Riley completed Experiment 1 in 8 weeks.

Response Measurement

We collected data on the frequency of target responses (i.e., pad touch) and alternative responses (i.e., communication card exchanges) using laptop computers. Target responses entailed touching the entire palm onto a foam pad. Alternative responses entailed exchanging a communication card with the therapist. Correct alternative responses occurred when the participant, without already having access to the reinforcer, exchanged the communication card during the reinforcement component (e.g., in the presence of the S^D). Incorrect alternative responses included exchanging the communication card during the extinction component (e.g., in the presence of the S^Δ) or when the participant already had access to the reinforcer. We calculated the percentage correct alternative responses by dividing the frequency of correct alternative responses by the sum of all correct and incorrect alternative responses and multiplying by 100.

Interobserver Agreement

A second independent observer collected data simultaneously with the primary data collector to assess interobserver agreement (IOA). We divided each session into 10-s intervals and scored an agreement for each interval in which both observers measured the same number of responses (i.e., exact agreement within the interval). We then summed the number of agreement intervals and divided by the number of agreement intervals plus disagreement intervals. Finally, we multiplied by 100 to convert each quotient to a percentage. We collected IOA on 22% of Tommy’s sessions. Coefficients averaged 99% (range, 93%–100%) for correct alternative responses, 100% for incorrect alternative responses, and 97% (range, 86%–100%) for target responses. We collected IOA on 29% of Emmitt’s sessions. Coefficients averaged 95% (range, 63%–100%) for correct alternative responses, 99% (range, 93%–100%) for incorrect alternative responses, and 96% (range, 63%–100%) for target responses. We collected IOA on 21% of Riley’s sessions. Coefficients averaged 94% (range, 77%–100%) for correct alternative responses, 97% (range, 90%–100%) for incorrect alternative responses, and 100% for target responses.

Procedures

Baseline.

During baseline, the participant sat at a table with a foam pad (i.e., target response). When the participant touched the pad with their entire palm, the therapist provided the participant with the highest-preference item from their paired-stimulus preference assessment for 20 s according to a fixed-ratio (FR) 1 schedule. At the end of the 20 s, the therapist removed the reinforcer and said, “My turn.” Therapists conducted at least three baseline sessions and ended baseline upon meeting the following stability criteria: (a) the standard deviation of the rate of target responding was less than 50% of its mean and (b) response rates were either flat or increasing in trend across three sessions.

DRA.

Following baseline, we taught the alternative response. Tommy, Emmitt, and Riley all had previous histories of exchanging communication cards. For this reason, we conducted a probe prior to the first DRA session. During the probe, we removed the preferred item with the communication card present. If the participant correctly exchanged the communication card upon the first removal or exchanged the communication card correctly in 4 out of the first 5 removals, the therapist moved directly into the DRA phase. All three participants demonstrated mastery of the communication card exchange during the initial probe. Thus, we did not conduct additional DRA pretraining.

During DRA, the target response materials remained on the table, and the therapist placed the communication card next to the pad. Therapists randomly alternated the location of the target response relative to the location of the alternative response, such that in some sessions, the target response was to the left of the alternative response but the right of the alternative response in other sessions. We placed the target and alternative responses equidistant from one another and from the participant. Response materials for both responses remained on the table throughout the session. The target response did not result in any programmed consequence (i.e., extinction), and exchanges of the communication card resulted in 20-s access to the same high-preference item from baseline, again on an FR 1 schedule of reinforcement.

Therapists conducted a minimum of three DRA sessions. Following two consecutive sessions with an 80% reduction of the target response from baseline and high rates of the alternative response, we reversed to the baseline condition before initiating mult DRA EXT under two multiple-schedule arrangements.

Reinforcement Schedule Thinning.

After therapists re-established target responding in a reversal to baseline, we introduced schedule-correlated stimuli corresponding to two multiple-schedule arrangements (i.e., S^D only, S^D and S^Δ) and thinned the schedules of reinforcement in each context independently. Within all mult DRA EXT sessions, we sequenced the reinforcement and extinction components for the alternative response in a quasirandom order such that no more than two components of the same type occurred consecutively. Target responding resulted in extinction across both components of mult DRA EXT. We also sequenced the two conditions of mult DRA EXT in a quasirandom order such that no more than two sessions of the same condition type occurred consecutively. The reinforcement component always lasted 30 s.

We modified the mult DRA EXT procedures for Riley because he engaged in variable levels of correct alternative responding across several sessions of both conditions. Riley engaged in a ritualistic response that consisted of raising his hand while holding the communication card and lowering it slowly into the hand of the therapist. This response often lasted the duration of one S^Δ component, as though he was timing the start of the reinforcement component, and it occurred in both conditions. Therefore, beginning with Session 42, we increased the variability of the reinforcement- and extinction-component durations to decrease the likelihood that therapists would inadvertently reinforce Riley’s ritualistic behavior and increase the likelihood that the alternative response would come under the control of the programmed schedule-correlated stimuli. When we did not see a consistent increase in correct alternative responding, we added a rule statement prior to each session (described below).

S^D Only.

In the S^D-only condition, a pre-determined Phillips Hue light color served as the S^D (e.g., pink), and the absence of that light color indicated extinction (i.e., all lights were turned off in the session room). As noted above, a small window in the session room let in natural light so that the participant could see the session materials while the overhead lights were turned off.

In the presence of the S^D, emitting the alternative response produced reinforcement for the remainder of the reinforcement component. In the absence of the S^D (i.e., during the extinction component), target and alternative responses resulted in extinction. The duration of the extinction component increased following every two consecutive sessions with (a) an 80% or greater reduction in the mean rate of target responding from baseline, and (b) correct alternative responding had to be at least 80%. We increased the duration of the extinction component as follows: 0 s, 2 s, 4 s, 8 s, 15 s, and 30 s. Beginning with Session 60 for Riley, the therapist said, “When the lights are pink, you can hand me your card and play with your toys,” before each session.

S^D and S^Δ.

The S^D and S^Δ condition was identical to the S^D-only condition with the exception that we correlated a uniquely colored light with the extinction component of mult DRA EXT and another uniquely colored light with the reinforcement component. For example, a yellow light (i.e., the S^D) signaled reinforcement of the alternative response for Riley, and a purple light (i.e., the S^Δ) signaled extinction of the alternative response. These light colors differed from that used to signal the reinforcement component in the S^D-only condition. Beginning with Session 60 for Riley, the therapist said, “When the lights are yellow, you can hand me your card and play with your toys. When the lights are purple, even if you hand me your card, you can’t play with your toys.”

Results and Discussion

Figure 1 displays the results of Experiment 1 for Tommy, Figure 2 displays the results of Experiment 1 for Emmitt, and Figure 3 displays the results of Experiment 1 for Riley. Target responding remained elevated during baseline for each participant. During DRA, target responding decreased to near-zero levels, and although alternative responding for this phase is not plotted (i.e., no opportunity for an incorrect alternative response, yielding an undefined percentage of correct alternative responses), rates of alternative responding increased to elevated and stable levels for all participants. Target responding returned to elevated levels for all participants during the return to baseline. During reinforcement schedule thinning, Tommy and Emmitt reached the terminal multiple schedule of reinforcement more quickly in the S^D and S^Δ condition relative to the S^D-only condition; however, these differences were minimal. Tommy met mastery in the S^D and S^Δ condition in two fewer sessions than in the S^D-only condition, and Emmitt met mastery in the S^D and S^Δ condition in one fewer session than in the S^D-only condition. In contrast, Riley reached the terminal multiple schedule of reinforcement more quickly in the S^D-only condition relative to the S^D and S^Δ condition. Again, this difference was minimal. Riley met mastery in the S^D-only condition in four fewer sessions than in the S^D and S^Δ condition. Riley’s progression though reinforcement schedule thinning was slower than for either of the other participants. His responding was low and variable. Due to minimal responding, a single incorrect response dramatically lowered his percentage correct alternative responding across conditions. This improved considerably with the use of contingency-specifying statements that began in both conditions at Session 60.

Figure 1.

Target response rate (top panel) and percentage of correct alternative responses (bottom panel) for Tommy during Experiment 1. The lower horizontal dotted line represents an 80% reduction for rate of target responses. The upper horizontal dotted line represents the 80% criterion for percentage of correct alternative responses. X/Y indicates the duration (in seconds) of the reinforcement and extinction components, respectively.

Figure 2.

Target response rate (top panel) and percentage of correct alternative responses (bottom panel) for Emmitt during Experiment 1. The lower horizontal dotted line represents an 80% reduction for rate of target responses. The upper horizontal dotted line represents the 80% criterion for percentage of correct alternative responses. X/Y indicates the duration (in seconds) of the reinforcement and extinction components, respectively.

Figure 3.

Target response rate (top panel) and percentage of correct alternative responses (bottom panel) for Riley during Experiment 1. The lower horizontal dotted line represents an 80% reduction for rate of target responses. The upper horizontal dotted line represents the 80% criterion for percentage of correct alternative responses. X/Y indicates the duration (in seconds) of the reinforcement and extinction components, respectively.

Across all three participants, progression through reinforcement schedule thinning was hampered to a greater extent by variable and low levels of correct alternative responding than it was for elevated rates of target responding. Target responding during most sessions of reinforcement schedule thinning remained below the 80% reduction criterion; however, correct alternative responding often failed to meet its 80% criterion. Variable and low levels of correct alternative responding were most pronounced during the middle of the reinforcement-schedule-thinning progression, suggesting that this part of the progression was critical for learning the “meaning” of the schedule-correlated stimuli.

One somewhat unexpected finding is that all three participants engaged in fewer target responses in the S^D and S^Δ condition relative to the S^D-only condition. Tommy did not engage in the target response during reinforcement schedule thinning in the S^D and S^Δ condition, Emmitt engaged in twice as many target responses in the S^D-only condition than in the S^D and S^Δ condition, and Riley engaged in 3.5 times more target responses in the S^D-only condition than in the S^D and S^Δ condition.

When comparing correct and incorrect alternative response rates during reinforcement schedule thinning, there were no consistent differences between the two multiple-schedule arrangements (see Table 1). Tommy and Emmitt had slightly higher discrimination indices during the S^D-only condition compared to the S^D and S^Δ condition. However, Riley had a slightly higher discrimination index during the S^D and S^Δ condition. Additional data on correct and incorrect alternative response rates during reinforcement schedule thinning for each participant are located in Supporting Information.

Table 1.

Alternative Response Rates and Discrimination Indices During Experiment 1

Child	Correct Alternative Responses per Minute		Incorrect Alternative Responses per Minute		Discrimination Index
	SD & SΔ	SD Only	SD & SΔ	SD Only	SD & SΔ	SD Only
Tommy	1.45	1.64	.41	.31	.78	.84
Emmitt	1.45	1.53	.68	.42	.68	.79
Riley	1.47	1.44	.44	.48	.77	.75

Note. The discrimination index here represents the proportion of correct alternative responses divided by the total number of correct and incorrect alternative responses.

Performance across the two conditions was highly similar within and across participants, suggesting that our use of a multielement design may have facilitated responding across the two conditions for all three participants, despite our use of distinct schedule-correlated stimuli in each condition. Unfortunately, other within-subject experimental designs (e.g., reversal) are likely to present other issues (e.g., history effects) that may carry their own interpretative difficulties.

Experiment 2

Basic research has generally shown periods of extinction and stimuli associated with extinction to be aversive. For example, alternating periods of reinforcement and extinction can produce aggression and emotional responding (Azrin et al., 1966), and the aversive properties of extinction can transfer to stimuli correlated with extinction (i.e., the S^Δ; Perone, 2003; Terrace, 1971). Terrace (1971) showed that when given the opportunity, pigeons allocated responding toward a key that terminated an S^Δ. Similarly, Case et al. (1985) found that pigeons allocated responding toward an observing key that produced an S^D rather than an observing key that produced an S^Δ. It is important to note that in these studies, the presence or absence of schedule-correlated stimuli did not affect the schedule of reinforcement in effect; rather, it affected only the presence or absence of the schedule-correlated stimuli. In general, pigeons tend to respond away from observing keys that produce stimuli correlated with extinction and away from observing keys that produce stimuli correlated with the leaner of two reinforcement schedules (Perone, 2003). Finally, presenting an S^Δ contingent on a response can function as punishment (Jwaideh & Mulvaney, 1976), and neutral stimuli may develop aversive properties through systematic pairings with extinction (Azrin, 1961; Perone, 2003).

Distinct from applied studies on the efficacy of various multiple-schedule arrangements, applied researchers have also studied participants’ preferences for schedule-correlated stimuli by allowing participants to choose across different multiple-schedule arrangements. For example, Tiger et al. (2006) compared compound schedules that included (a) an S^D and an S^Δ, (b) an S^D only, and (c) neither an S^D nor an S^Δ (i.e., a mixed schedule of reinforcement). Participant preferences amongst the multiple-schedule arrangements were correlated with the relative number of errors made under each arrangement. That is, participants who made more errors (i.e., responding during the S^Δ) preferred the S^D-only arrangement, while participants who made fewer errors preferred the S^D and S^Δ arrangement.

Tiger et al. (2008) also compared multiple schedules that did and did not use a stimulus to signal extinction. In their study, all four participants preferred the S^D and S^Δ arrangement relative to the S^D-only arrangement. This is likely due to the fact that the S^D and S^Δ arrangement was more effective at teaching the discrimination than the S^D-only arrangement. The authors indicated that the signal of the extinction component was more salient in the S^D and S^Δ arrangement. In addition, participants made fewer errors during the extinction component in the S^D and S^Δ arrangement relative to the S^D-only arrangement.

As noted in Experiment 1, one limitation of previous applied research in this area is that researchers often programmed the same stimulus to serve as the S^D across conditions that do and do not signal extinction. We corrected this issue by using three distinct schedule-correlated stimuli (i.e., one stimulus for each of the two S^Ds and a third for the S^Δ) to help minimize potential carryover effects that may result from programming common discriminative stimuli across experimental conditions.

Although not a limitation per se, applied researchers have relied on the use of concurrent-chains procedures to assess preference for different multiple-schedule arrangements. In a concurrent-chains procedure, selection amongst multiple, concurrently available stimuli in the initial link (e.g., choosing between two rooms, each associated with a distinct condition) initiates a terminal link. In this line of research, the terminal link is a brief session with a distinct arrangement of schedule-correlated stimuli. These procedures have proven helpful in assessing preference for different multiple-schedule arrangements. One downside of this approach, however, is that the concurrent-chains procedure does not allow for the detection of momentary changes in preference, as participants are unable to make more than one initial-link choice per trial. Measuring momentary changes in preference could be important, as it may allow for a more dynamic assessment of choice.

Basic research on choice has suggested ways to address some of the aforementioned limitations of the applied literature in this area. For example, basic studies have assessed preference for different multiple-schedule arrangements using a concurrent-operants arrangement, rather than a concurrent-chains procedure, to allow for momentary changes in preference (e.g., Jwaideh & Mulvaney, 1976; Mulvaney et al., 1974). In the studies by Jwaideh and Mulvaney (1976) and Mulvaney et al. (1974), observing keys were located on either side of a key that produced food on an intermittent schedule of reinforcement. All keys were illuminated yellow at the start of the session. Pecking one observing key (e.g., left) produced the S^D (e.g., green) or produced no change (i.e., no S^Δ), depending on the schedule component in effect (i.e., reinforcement or extinction, respectively). The other observing key (e.g., right) produced both the S^D (e.g., green) and the S^Δ (e.g., red), again depending on the schedule component in effect. The contingencies associated with the observing keys varied across conditions, but the underlying schedules of reinforcement were unchanged by pecks at the observing keys, thus providing an ongoing examination of preference for various multiple-schedule arrangements. Across conditions, pigeons generally responded away from the observing key that produced the S^Δ.

Perone and Baron (1980) failed to replicate this study with human participants. A group of men pulled plungers to earn money on an alternating schedule of reinforcement consisting of (a) extinction or (b) a high-effort response requirement. When given observing keys similar to those in the animal studies, participants responded towards the observing key that produced the S^Δ, unlike responding in the animal research. Perone (2019) hypothesized that one reason for the differing results is that the humans had a relatively lower response-effort requirement during the extinction component, which may have functioned like a break from the component with the high-effort response requirement. Perone (2019) corrected this in a later experiment in which not responding (i.e., not clicking a button on a computer before it disappeared) during either the reinforcement or extinction component caused participants to lose money. With this arrangement, the extinction component no longer functioned as a reinforcer, and participants allocated responding toward the S^D, similar to the animal research.

Experiment 2 aims to borrow procedures from the basic literature and to correct some of the methodological limitations in the applied literature to assess the preference for different multiple-schedule arrangements in children with autism.

Method

Whereas Experiment 1 evaluated participants’ ability to acquire the successive discrimination of a multiple schedule under each of two multiple-schedule arrangements, the purpose of Experiment 2 was to assess preference for concurrently available multiple-schedule arrangements by examining allocation of responding across two observing responses when those observing responses produced different multiple-schedule arrangements.

We added two similarly colored index cards that differed in color from all other programmed stimuli to serve as the materials for observing responses for Experiment 2. Therapists placed these index cards on opposite sides of the table, in the middle of which were the materials for the target and the alternative response. This arrangement allowed for a left and a right observing response. Touching one of these index cards constituted an observing response. Observing responses produced a 30-s observing interval during which time, schedule-correlated stimuli associated with that observing response (i.e., left or right) could appear or change (e.g., a colored light might turn on or off). Whether an observing response produced or changed schedule-correlated stimuli depended on which component of the compound schedule was currently in effect and which observing response occurred. For example, the S^Δ would not appear when the extinction component was in effect during a condition that programmed only the S^D for that specific observing response. If, however, the participant emitted the other observing response during the same extinction component, the S^Δ would appear if the newly emitted observing response was programmed to present both the S^D and S^Δ (or was programmed to present only the S^Δ). A second therapist located behind the one-way observation window controlled the light bulb in the session room and thus controlled the schedule-correlated stimuli.

When components of the reinforcement schedule changed within a 30-s observing interval, the therapist presented or removed the schedule-correlated stimuli as needed in accordance with the condition currently associated with that observing response. For example, if the participant engaged in an observing response using the card that signaled the S^D only but did so when the extinction component was in effect, nothing happened (i.e., the light remained off). However, the S^D appeared when the reinforcement component began. If, during a 30-s observing window for the same observing response, the reinforcement component ended and the extinction component began, the S^D turned off. Similar events happened for other observing responses. For example, if the participant engaged in an observing response using the card that signaled the S^D and S^Δ, the S^D replaced the S^Δ when the extinction component ended and the reinforcement component began.

Thus, a mixed schedule of reinforcement, devoid of any schedule-correlated stimuli, ran continuously in the background of each session. Observing responses were not required to obtain reinforcement; however, they could facilitate efficient responding by producing a stimulus indicating which component of the reinforcement schedule was currently in effect (i.e., reinforcement or extinction). At any point in the session, participants could emit target and alternative responses, the latter of which alone produced reinforcement and only during the reinforcement component.

We used the same schedule-correlated stimuli in Experiment 2 as we programmed for each participant in Experiment 1 so that schedule-correlated stimuli did not overlap across multiple-schedule arrangements. Each component of the compound schedule lasted approximately 30 s (i.e., variable-time [VT] 30-s schedule; range, 20–40 s). As in Experiment 1, we used a quasirandom sequence to order components within each session. That is, components of the same type could occur consecutively with no more than two occurring back to back. Thus, the longest possible combined reinforcement or extinction component was 80 s. For all reinforcement components lasting over 20 s, the participant had the opportunity to contact reinforcement multiple times.

Response Measurement

Data collection was identical to Experiment 1, except observers also recorded left and right observing responses. Location was always relative to the data collector’s point of view. That is, a left observing response remained a left observing response even if the participant moved to the opposite side of the table. A left observing response entailed touching the index card to the left of the target and alternative responses. A right observing response entailed touching the index card to the right of the target and alternative responses. All sessions lasted 5 min. We conducted sessions consecutively with approximately one minute between each. We conducted research with Tommy for two hours per day and conducted approximately 8–10 sessions per day. The study took place over approximately 2 weeks. We conducted research with Emmitt and Riley for 30 min per day and conducted approximately 4 sessions per day. Emmitt completed Experiment 2 in 12 weeks, and Riley completed Experiment 2 in 8 weeks.

Interobserver Agreement

We collected IOA using the same procedures from Experiment 1 and did so for 26% of Tommy’s sessions. Coefficients averaged 97% (range, 90%–100%) for correct alternative responses, 98% (range, 83%–100%) for incorrect alternative responses, 100% for target responses, 95% (range, 80%–100%) for left observing responses, and 95% (range, 80%–100%) for right observing responses. We collected IOA on 20% of Riley’s sessions. Coefficients averaged 99% (range, 93%–100%) for correct alternative responses, 98% (range, 93%–100%) for incorrect alternative responses, 100% (range, 97%–100%) for target responses, 93% (range, 90%–100%) for left observing responses, and 99% (range, 93%–100%) for right observing responses. Due to technical difficulties, we were unable to access Emmitt’s timestamped primary data from Experiment 2. Therefore, we calculated IOA for Emmitt using the total-agreement method for each session with IOA by dividing the smaller value recorded by either observer by the larger value recorded by either observer and by then multiplying by 100. We collected IOA on 20% of Emmitt’s sessions. Coefficients averaged 100% for correct alternative responses, 100% for incorrect alternative responses, 100% for target responses, 94% (range, 67%–100%) for left observing responses, and 91% (range, 83%–100%) for right observing responses.

Procedures

Pretraining (Not Displayed).

The purpose of this condition was to teach the observing response. Pretraining occurred with each S^D from the two multiple-schedule arrangements from Experiment 1 and began with the light off and extinction arranged for target and alternative responding. Once the participant made an observing response, the second therapist from behind the observation window presented one of the two S^Ds (depending on pretraining condition) for 30 s, during which time the therapist reinforced the alternative response on an FR 1 schedule. In pretraining, both observing responses produced the same S^D, but the S^Ds differed across pretraining conditions. Additional observing responses were required to produce the S^D again, as well as the reinforcement contingency for the alternative response. We conducted a minimum of six pretraining sessions (i.e., three with each S^D) and until the participant engaged in eight observing responses for two consecutive sessions in each pretraining condition.

S^D and S^Δ vs. S^D Only.

Following pre-training, we examined preference for the S^D and S^Δ arrangement relative to that for the S^D-only arrangement to evaluate potentially aversive properties of the S^Δ. Touching one observing card (e.g., right) produced the S^D during the reinforcement component and produced no stimulus change during the extinction component. Touching the other observing card (e.g., left) produced the S^D during the reinforcement component and produced the S^Δ during the extinction component. We reversed the contingencies associated with the two observing cards after the participant emitted consistent observing responses to one of the two observing cards for three consecutive sessions, such that a change in response allocation would suggest a preference for one of the two multiple-schedule arrangements and not a side or location bias. Prior to each session, the therapist conducted a forced-choice of each observing response so that the participant experienced the contingencies associated with reinforcement and extinction for both observing responses.

Because Riley demonstrated an initial bias toward the left observing response during pretraining and at the start of this condition, beginning with Session 2, his therapist alternated the location of Riley’s chair at the table such that sometimes the “left” observing response was located to Riley’s left but to Riley’s right during other sessions. However, this arrangement led to a decrease in the number of observing responses, and Riley made numerous incorrect alternative responses. Therefore, his therapist implemented response restriction for the observing card during non-observing intervals beginning with Session 7. That is, Riley had to make an observing response to either observing card before accessing the alternative response materials. The therapist delivered the communication card as soon as Riley made an observing response, regardless of whether the reinforcement or extinction component was currently in place. Riley’s levels of observing responses increased following this change.

S^D Only vs. S^Δ Only.

Following the previous comparison (S^D and S^Δ vs. S^D only), we next compared the S^D-only arrangement with an S^Δ-only arrangement. Since all participants either preferred the S^D and S^Δ arrangement over the S^D-only arrangement or had no preference, we wanted to compare the S^D-only arrangement with an S^Δ-only arrangement to evaluate whether the S^Δ was aversive (i.e., participants would respond away from it) when both arrangements used only one discrete stimulus. Similar to the condition above, we reversed the contingencies of the observing responses to rule out side- or location-biased responding.

S^Δ Only vs. Mixed Schedule (Riley Only).

We compared an S^Δ-only arrangement with a mixed schedule for Riley. Prior to this comparison, Riley responded consistently toward a specific observing card, regardless of whether we placed the card to the right or to the left of Riley. Despite this apparent location bias, Riley’s percentage of correct alternative responses remained consistently high, regardless of the multiple-schedule arrangement in place for his preferred observing response. This suggested strongly to us that Riley’s responding was under the discriminative control of the various schedule-correlated stimuli but that he had an overarching preference for the location of the observing response. For this reason, we correlated his preferred observing response (i.e., left) with a mixed schedule of reinforcement and his nonpreferred observing response (i.e., right) with the S^Δ-only arrangement. Observing responses to his preferred observing card now resulted in the alternative response materials (i.e., the communication card) but no schedule-correlated stimuli. This arrangement pitted Riley’s preference for a specific location of the observing response against his preference for the information provided by the S^Δ. As with the other conditions, we later reversed the contingencies associated with the two observing responses.

Results and Discussion

Figure 4 displays the results of Experiment 2 for Tommy. Across the first three phases, Tommy responded toward the observing card that produced the S^D and S^Δ over the observing card that produced the S^D only. Across the next three phases, we compared the S^D-only condition with the S^Δ-only condition. Tommy responded toward the observing card that produced the S^D only rather than the observing card that produced the S^Δ only. Tommy’s preference hierarchy in order of most to least preferred was S^D and S^Δ, followed by S^D only, followed then by S^Δ only. Throughout Experiment 2, Tommy’s percentage of correct alternative responses averaged 90%.

Figure 4.

Proportion of right observing responses (top panel) and rate of observing responses (bottom panel) during Experiment 2 for Tommy. “Left” indicates responses to the left observing card. “Right” indicates responses to the right observing card. The dotted line represents indifference.

Figure 5 displays the results of Experiment 2 for Emmitt, whose response pattern differed markedly from Tommy’s pattern of responding. Emmitt showed no clear preference for any multiple-schedule arrangement, and this apparent indifference persisted across comparisons that included only one schedule-correlated stimulus associated with each observing response (i.e., the S^D-only and S^Δ-only arrangement). Throughout Experiment 2, Emmitt’s percentage of correct alternative responses averaged 100%.

Figure 5.

Proportion of right observing responses (top panel) and rate of observing responses (bottom panel) during Experiment 2 for Emmitt. “Left” indicates responses to the left observing card. “Right” indicates responses to the right observing card. The dotted line represents indifference.

Figure 6 displays the results of Experiment 2 for Riley, who demonstrated a persistent bias for the observing response located on the left side of the table. Because this location bias did not negatively impact his high percentage of correct alternative responses nor did it interfere with his high rate of obtained reinforcement across multiple-schedule arrangements, we compared the S^Δ-only arrangement with a mixed schedule of reinforcement in his final two phases. Riley responded toward the observing card associated with the S^Δ-only arrangement even when that observing card was located to his right, suggesting that Riley preferred the information provided by the S^Δ over his overarching preference for the left observing response. It is reasonable to assume that Riley would show a similar preference for other multiple-schedule arrangements (e.g., S^D and S^Δ, S^D only) when compared to a mixed schedule of reinforcement. Throughout Experiment 2, Emmitt’s percentage of correct alternative responses averaged 75%.

Figure 6.

Proportion of right observing responses (top panel) and rate of observing responses (bottom panel) during Experiment 2 for Riley. “Left” indicates responses to the left observing card. “Right” indicates responses to the right observing card. RR stands for response restriction of the alternative response materials during non-observing intervals. The dotted line represents indifference.

General Discussion

Although the results of Experiment 1 indicated that the two multiple-schedule arrangements (i.e., S^D and S^Δ, S^D only) were similarly effective in producing a successive discrimination with low rates of target responding, Experiment 2 showed three highly disparate patterns of preference across the three participants. Tommy demonstrated a preference for the S^D and S^Δ arrangement over the S^D-only arrangement and a preference for the S^D-only arrangement over the S^Δ-only arrangement. Thus, Tommy’s preference hierarchy is partially consistent with the hypothesis that the S^Δ was somewhat aversive, as he consistently responded away from the observing response that produced the S^Δ only when pitted against the S^D only. However, Tommy preferred the S^D and S^Δ arrangement over the S^D-only arrangement. One interpretation of this pattern of responding involves the information provided by the absence of the schedule-correlated stimuli across the multiple-schedule arrangements. In the S^D-only and S^Δ-only arrangements, termination of the schedule-correlated stimuli occurred when (a) the extinction component or the reinforcement component began, respectively, or (b) the 30-s observing interval ended. However, in the S^D and S^Δ arrangement, termination of the schedule-correlated stimuli occurred only when the 30-s observing interval ended. Thus, the S^Δ, when used in conjunction with the S^D, provided information as to when extinction was in effect during the S^D and S^Δ arrangement, and its absence in this condition indicated that another observing response was required. This dual function of the S^Δ in the S^D and S^Δ arrangement, but not in either of the other two arrangements, may have made the S^D and S^Δ arrangement more preferred.

Riley’s results offer additional insight into this hypothesis. We implemented response restriction of the communication card during non-observing intervals for Riley due to his initial low rate of observing responses. Thus, the S^Δ for Riley did not provide additional information because we signaled the end of the observing interval by removing the communication card. Riley’s indifference across the multiple-schedule arrangements may have been due to the fact that all arrangements provided equal amounts of information about which schedule component was in effect, unlike in Tommy’s evaluation. Researchers looking to evaluate this possibility may consider using an additional stimulus (e.g., a timer) to indicate the termination of the observing interval, signaled independently from the absence of the schedule-correlated stimuli.

Emmitt demonstrated a lack of preference for all three of the multiple-schedule arrangements. It is important to note that Emmitt’s percentages of correct alternative responses remained near 100% for all sessions during Experiment 2. In addition, at the end of the study, the therapist asked Emmitt several questions regarding the study (e.g., What do the observing cards do?, “What does the blue light mean?”). Emmitt described the contingencies of each observing card correctly. That is, Emmitt stated that one observing card produced the light that meant “Go” so that he could exchange the other card and play with his toys. He also correctly stated that the other observing card produced a different light that meant “Stop,” and he could not exchange the other card to play with his toys. These two factors suggest that Emmitt’s lack of preference in Experiment 2 was not the result of failing to understanding the contingencies and instead suggest that his indifferent responding was due to indifference for the various multiple-schedule arrangements.

Riley demonstrated a strong bias toward the observing response located on the left side of the room, regardless if that card was located to the right or to left of Riley’s position. However, like Emmitt, Riley consistently engaged in high percentages of correct alternative responses. When we compared the S^Δ-only arrangement with a mixed schedule, Riley’s responding shifted away from the left observing response and to the right observing response associated with the S^Δ-only arrangement. This pattern reversed when we switched the contingencies associated with the observing responses, suggesting that some information about the schedule components currently in effect was better than no information from his preferred (left) observing response.

At the very least, these results suggest that preference for multiple-schedule arrangements is complex and that efficacy (Experiment 1) can be a poor predictor of preference (Experiment 2). Furthermore, preference for one multiple-schedule arrangement may not be a good predictor of preference for other multiple-schedule arrangements. Tommy preferred the S^D and S^Δ arrangement over the S^D-only arrangement, suggesting that the S^Δ was not sufficiently aversive to predict he would strongly favor the S^D-only arrangement over the S^Δ-only arrangement. However, Tommy showed clear preference for the S^D-only arrangement over the S^Δ-only arrangement.

Our results provide little support to suggest the S^Δ was aversive. Terrace (1971) demonstrated that pigeons responded to terminate an S^Δ, and Perone (2003) showed that pigeons responded away from a key that signaled the S^Δ. Our results are similar in that Tommy responded away from an observing response that signaled the S^Δ only when it was compared with an observing response that signaled only the S^D. However, both Emmitt and Riley responded equally towards the S^Δ-only arrangement as they did towards the S^D-only arrangement. Emmitt’s responding to the right observing card stabilized at 50% even though he could appropriately state the contingencies for each observing response, and Riley preferred the S^Δ-only arrangement over a mixed schedule of reinforcement. Continued research is needed to more fully parse out the factors that affect preference for different multiple-schedule arrangements and to do so with a larger sample of participants. Additionally, it may be interesting to examine multiple-schedule arrangements that explicitly do or do not program a common S^Δ. Also, future research should include participants who engage in destructive behavior as it is important to determine the extent to which translational findings such as these generalize to more applied contexts.

It should be noted that early work by Terrace (1966) is particularly relevant for the present discussion. Terrace found that behavioral contrast, peak shift, and general activity in the presence of the S^Δ were greatly modulated by the subject’s history with that stimulus. Pigeons exposed to an S^Δ early on in training and then trained to emit minimal errors in the presence of the S^Δ using errorless-teaching procedures did not show an aversion to the S^Δ as did other pigeons not trained with these procedures. Therefore, the nature in which the S^Δ is introduced (and the errors that occur in its presence) may modulate whether an S^Δ becomes aversive. We did not use errorless-teaching procedures in Experiment 1, and all three participants in that study emitted numerous errors in the presence of the S^Δ and in the absence of the S^D when learning the successive discrimination. It remains unclear what role the generally consistent histories of performance across participants in Experiment 1 played in the highly disparate findings of Experiment 2.

The present study expands upon the research conducted by Tiger et al. (2006), Tiger et al. (2008), and Landa and Hanley (2016) by using unique schedule-correlated stimuli across multiple-schedule arrangements similar to the study conducted by Pizarro et. al (2020). Our results were similar to those of Pizarro et. al (2020) in that there were no large differences between the efficacy of the two multiple-schedule arrangements. Our Experiment 2 translated the basic approach used by Jwaideh and Mulvaney (1976) to measure preference by pitting different multiple-schedule arrangements against one another and allowing for repeated measures of choice. Future research may adopt similar procedures for addressing other questions in this general line of inquiry.