Functional analyses have identified negative-reinforcement contingencies (such as the termination of academic tasks or self-care routines) as the most common contingencies maintaining severe problem behavior (such as aggression and self-injury) exhibited by individuals with intellectual and developmental disabilities (Hanley et al., 2003; Iwata, Pace, et al., 1994b). Individuals displaying escape-maintained problem behavior are likely to experience fewer opportunities to learn new skills because parents, teachers, and other caregivers may avoid these evocative teaching situations (Carr et al., 1991). However, after identifying that escape or avoidance of tasks serves as a reinforcer for severe problem behavior, function-based interventions may ameliorate these behavioral concerns by disrupting the contingency between problem behavior and task termination (Geiger et al., 2010).

Differential reinforcement of alternative behavior (DRA) procedures, particularly those targeting compliance, have strong empirical support as effective, function-based interventions for this class of problem behavior (Payne & Dozier, 2013). DRA for escape-maintained problem behavior typically includes (1) arranging escape extinction, in which an implementer continues to issue instructions despite problem behavior; and (2) differentially reinforcing an appropriate response (e.g., compliance with instructions) with a break from demands. In many applications, this break may be enhanced by including other preferred events or activities as potential sources of positive reinforcement (e.g., Drifke et al., 2017; Lalli et al., 1999; Piazza et al., 1996; Zangrillo et al., 2016).

DRA is typically initiated using a continuous reinforcement schedule (i.e., each compliant response results in a break period) to ensure a strong contingency between compliance and reinforcer delivery, but over time, this schedule should be thinned to produce a more socially acceptable intervention model for caregivers and teachers (Tiger & Hanley, 2021). That is, consumers should eventually be able to complete entire tasks such as getting dressed, brushing their teeth, or completing an academic assignment before accessing a break. Schedule thinning involves progressively increasing the amount of work an individual must complete before contacting reinforcement. Although the literature includes successful examples schedule thinning, the thinning process may result in increased problem behavior, even when interventions seemingly eliminated these behaviors under dense schedules (Hagopian et al., 1998).

Schedule thinning involves introducing and progressively increasing the duration of delays between the onset of an instructional period (i.e., the delivery of a first instruction) and the delivery of reinforcement (i.e., a break). The more tasks that are required, the longer the temporal delay between the onset of instruction and its eventual offset. Problem behavior may increase during schedule thinning because the reinforcing efficacy of a stimulus after a delay is diminished relative to its efficacy when delivered immediately (Madden & Johnson, 2010). Thus, it may be the case that the subjective value of the delayed reinforcer is insufficient to continue to serve as a reinforcer for compliance. Further, as work requirements increase, the presentation of repeated demands would further establish escape as a reinforcer and would be more likely to evoke the problem behavior that historically produced escape, especially when compliance is reinforced on a lean schedule. For example, Smith et al. (1995) showed that escape-maintained problem behavior occurred more often with shorter interinstruction intervals relative to longer interinstruction intervals. Finally, the leaning of reinforcement density for compliance may set the stage for a resurgence of problem behavior (Kimball et al., 2023). That is, reducing the overall density of reinforcement for compliance may result in the reoccurrence of other behavior (e.g., problematic behavior) that have historically resulted in that same reinforcer.

There are at least two variations of schedule thinning described in the literature. One variation involves maintaining a constant reinforcement duration while work requirements increase progressively. For example, in Lalli et al. (1995), an individual was initially required to complete one work task prior to experimenters honoring a break request by terminating instruction for 30 s. The experimenters progressively increased the work requirements from 1 to 16 tasks, but still delivered only a 30-s break at the completion of that chain. In another variation of schedule thinning, the duration of delivered reinforcement increases incrementally, commensurate with increases in work requirements. For instance, in Piazza et al. (1996), DRA work requirements increased from an FR-1 to an FR-20 schedule while the duration of reinforcement increased from 30 to 270 s. Arranging progressively longer reinforcer durations may decrease the likelihood of problem behavior by increasing reinforcement magnitude (such that the subjective value is retained at a higher level over longer periods) and by decreasing the density of instruction (i.e., by increasing the time between each instructional period).

Despite these arguments in favor of increasing reinforcement durations during schedule thinning, it is also possible that programming for incrementing reinforcer durations would confer limited therapeutic benefit. In that case, the extended reinforcement durations would limit the number of therapeutic or instructional trials delivered during a session and potentially slow the schedule-thinning process. Although both models of schedule thinning have been described in the literature, no study to date has directly compared these procedures with escape-maintained problem behavior. Therefore, the purpose of the current study was to compare the efficacy and efficiency of schedule thinning when reinforcement durations were either held constant or increased incrementally with three individuals who engaged in escape-maintained problem behavior.

Method

Participants and Setting

Larry was a 17-year-old, white male diagnosed with autism spectrum disorder (ASD) who was referred for the assessment and treatment of aggressive and self-injurious behavior (SIB). He communicated primarily by pointing to images using an application on his iPad. Ivan was an 8-year-old, Black male diagnosed with ASD who was referred for the assessment and treatment of aggressive and destructive behavior. Ivan communicated with a limited repertoire of single words or short phrases. Dan was a 5-year-old, white male diagnosed with Smith-Magenis syndrome who was referred for the assessment and treatment of aggressive, destructive, and self-injurious behavior. Dan communicated using several modified signs.

Each participant attended a clinic specializing in the assessment and treatment of severe behavior disorders for 3-hr appointments, 5 days a week. Each appointment was divided into 10-min assessment or treatment sessions; this typically yielded 12–15 sessions per appointment. All sessions were conducted in therapy rooms (approximately 3 × 3 m) equipped with a one-way observation mirror. Session therapists were graduate students in a behavior analysis program who were either completing practicum hours or held their BCBA certification. Each session was conducted or directly observed by the first author.

Measurement and Interobserver Agreement

Observers recorded instances of aggressive (hitting, kicking, pinching, scratching, pushing, and throwing objects at an experimenter), self-injurious (head hitting with hand or objects and head banging against the walls of the therapy room), and destructive behavior (hitting, throwing, or ripping objects with hands) from behind an observation mirror during each 10-min session on laptop computers using InstantData. Observers recorded the frequency of instructions provided by the experimenter (e.g., “Put the toys away”) and participant compliance with the instruction, defined as completing the instruction within 5 s of a vocal or model prompt. These measures are reported as a percentage of instructions resulting in compliance.

To assess interobserver agreement (IOA), a second observer simultaneously, but independently, collected data during 66%, 31%, and 47% of sessions, overall, for Larry, Ivan, and Dan, respectively. The distribution of IOA sessions across phases is presented in Table 1. Observers’ scoring records from each session were divided into 10-s intervals and were compared on an interval-by-interval basis using the proportional-agreement method. Intervals in exact agreement received a score of 1. Intervals not in exact agreement received a score by dividing the smaller number of scored responses by the larger number of responses. The sum of all interval scores were then divided by the total number of intervals, and the resulting quotient was converted into a percentage. These calculations yielded mean agreement scores of 99% for aggression (range: 94%–100%), 99% for self-injury (range: 80%–100%), 93% for instructions (range: 75%–100%), and 93% for compliance (range: 62%–100%) for Larry; 98% for aggression (range: 89%–100%), 99% agreement for destruction (range: 89%–100%), 92% for instructions (range: 45%–100%), and 93% for compliance (range: 49%–100%) for Ivan; and 99% for aggression (range: 82%–100%), 98% for self-injury (range: 85%–100%) and 96% for destruction (range: 89%–100%), 94% (range: 79%–100%) for instructions, and 99% for compliance (range: 89%–100%) for Dan.

Table 1 Percentage of sessions with IOA data in each phase
Full size table

Procedures

Preference Assessment

Experimenters conducted the Reinforcer Assessment for Individuals with Severe Disabilities (RAISD; Fisher et al., 1996) with participants’ caregivers to nominate preferred leisure items. Experimenters then conducted paired-stimulus preference assessments (Fisher et al., 1992) with nominated items to identify the two highest preferred items to deliver as reinforcers during the DRA conditions.

Functional Analysis

The second author began the functional assessment process by conducting an open-ended, semi-structured interview with caregivers. This interview covered (1) family demographic information; (2) medical, educational, and therapeutic histories; (3) current developmental presentation; and (4) environmental accompaniments of problem behavior. The latter portion of the interview was used to rule in and rule out functional analysis test conditions and to identify relevant evaluative contexts within each test condition (e.g., what types of demands and disruption of what materials commonly evoked problem behavior).

Experimenters then conducted a functional analysis of problem behavior, based upon the procedures of Iwata, Dorsey, et al. (1994a) with each participant. Each functional analysis included (1) escape sessions, in which a therapist presented task instructions using graduated prompting and terminated those instructions for 30-s following the occurrence of problem behavior and (2) toy-play control sessions in which no instructions were delivered during the session. Functional analyses also included tests for behavioral sensitivity to attention (Ivan and Dan only) and access to tangible items as reinforcers, but, most relevant to the current investigation, all functional analyses identified that target problem behavior was maintained by negative reinforcement in the form of escape from instruction. Results of these functional analyses are depicted in Fig. 1.

Fig. 1
figure 1

Functional analysis of problem behavior. Note. Results of a functional analysis for Larry (top), Ivan (middle), and Dan (bottom panels). Each assessment identified that escape from demands served as a reinforcer for participants’ problem behavior

Full size image

Initial Treatment Evaluation

Experimenters first established a baseline of problem behavior. Similar to the escape sessions of the functional analysis, the experimenter presented instructionsFootnote 1 to put toys (Larry) or laundry (Ivan and Dan) in a basket using a graduated prompting procedure. That is, the experimenter would deliver a vocal instruction (e.g., “Put the toy away”) and wait 5 s. If the task was not initiated within 5 s, the experimenter would then repeat the instruction and present a model of the target response. If the task was not initiated within another 5 s, the experimenter would repeat the instruction, provide physical guidance, and restart the prompting sequence. Compliance following a vocal or model prompt resulted in praise and the presentation of the next instruction. Problem behavior at any time resulted in the termination of instructions for 30 s. Task related materials remained present in the therapy room (e.g., toys or clothes spread across the floor) during these break periods.

Following baseline, experimenters implemented DRA sessions, which were similar to baseline except experimenters continued their prompting sequence following problem behavior (i.e., arranged escape extinction) and instead ceased delivery instructions during a 30-s break with access to preferred leisure activities (manipulable toys for Larry; an iPad for Ivan and Dan) and conversational attention after each instance of compliance with a vocal or model prompt. Baseline and DRA conditions were compared in an ABA reversal design for Larry and Ivan and an ABAB reversal design for Dan.

Schedule Thinning Comparison

After demonstrating reductions in problem behavior to be a result of DRA, experimenters compared two variations of schedule thinning. Instructional procedures were identical to DRA sessions in both variations. During schedule-thinning sessions with fixed-reinforcement durations, the work requirements to produce an enriched break increased by one instance (e.g., FR 2, 3, 4, 5) following every three consecutive sessions in which problem behavior was below 10% of its baseline level. Within each session, problem behavior during the work period resulted in a restart of their work requirements (e.g., if the participant complied with four instructions and then engaged in problem behavior, they were required to complete an additional five instructions without problem behavior to satisfy the FR-5 schedule requirement). As the FR schedule increased, the enriched break duration in these sessions remained at 30 s.

Schedule thinning sessions with incrementing-reinforcement durations were similar except that the enriched break durations increased by 5 s at each step. For example, when the FR schedule increased to 2, 3, 4, and 5 responses, the reinforcement duration earned increased to 35, 40, 45, and 50 s, respectively. Each step, and its associated reinforcer duration, are listed in Table 2.

Table 2 Reinforcement durations (s) during the task-chaining comparison
Full size table

Fixed- and incrementing-reinforcement sessions were counterbalanced in pairs of sessions with the order determined randomly, conforming to a multielement experimental design. Each condition advanced through fading steps independently until one condition reached a predetermined terminal criterion. Such criteria were selected in consultation with participants’ caregivers (an FR-30 schedule for Larry; and an FR-20 schedule for Ivan and Dan). To promote discrimination between conditions, tasks were completed using uniquely colored materials. That is, Larry was to place manipulable toys in a red bucket during fixed-reinforcement sessions and a green bucket during incrementing-reinforcement sessions. Ivan and Dan were to place laundry in a blue basket during fixed-reinforcement sessions and a white basket during incrementing-reinforcement sessions. In addition, the experimenter affixed a corresponding, colored sheet of paper to the therapy room wall. After one condition met terminal criterion, sessions of this condition were conducted in isolation (i.e., an independent verification phase; Barlow & Hayes, 1979) to ensure observed effects were not a result of carry-over effects between conditions.

Procedural Integrity

During the schedule-thinning comparison, there were two primary independent variables: (1) the number of tasks to be completed prior to reinforcement and (2) the duration of reinforcement delivered. To ensure these independent variables were implemented by therapists as programmed, we sampled 20% of sessions for each participant from the schedule thinning comparison using a random number generator. This random sampling drew 31 and 33 fixed and incrementing reinforcement sessions for Larry, 12 fixed and 10 incrementing reinforcement sessions for Ivan, and 6 fixed and 13 incrementing reinforcement sessions for Dan for a total of 103 sampled sessions.

We identified the programmed FR schedule for the sampled sessions from session logs, and then analyzed the raw data streams from these sessions. These data streams involved a record of each frequency key scored as well as the time stamp at which that key was scored. From these data we identified each instructional episode within a session. We defined an instructional episode beginning with the first instance of an instruction scored and ending with a time interval between an instance of compliance and at least 25 s before the next instance of an instruction (the latter defining the onset of the next instructional episode). Within each instructional episode, we then counted the number of consecutive instances of compliance (i.e., those not interrupted by an instance of problem behavior). We subtracted the number of instances of compliance from the programmed FR schedule and divided the difference by the programmed FR schedule. We converted the absolute value of that difference into a percentage and subtracted that from 100% to arrive at a procedural integrity score for instruction delivery. As an example of this calculation, if therapists required six instances of compliance before delivering a break, but the programmed FR schedule required five instances, the difference of 6 (obtained) – 5 (programmed) = 1 error. Then, 1(error)/5(programmed) = 20% error or 80% integrity. We omitted any instructional episodes within a session that were incomplete (e.g., those that involved presenting instructions within the last 25 s of session termination). These calculations yielded integrity scores of 98% (range: 90%–100%), 98% (range: 93%–100%), and 98% (range: 94%–100%) of fixed reinforcement sessions and scores of 99% (range: 93%–100%), 98% (range: 93%–100%), and 100% (range: 99%–100%) for incrementing reinforcement sessions for Larry, Ivan, and Dan, respectively. We did not explicitly score the duration of breaks, so we were unable to check therapist integrity of break durations.

Results

The results of Larry’s, Ivan’s, and Dan’s evaluations are depicted in Figs. 2, 3 and 4, respectively. Each participant’s data are shown across two panels with the occurrence of problem behavior in their top panel and compliance with instructions on their lower panel. The initial DRA evaluations are shown in the first three phases for each participant (four for Dan). Larry (Fig. 2) engaged in a mean of 3.2 instances of problem behavior per min (top panel) and complied with 48% of instructions (bottom panel) during the initial baseline. Problem behavior decreased to 0.1 instances per min during the last five sessions of DRA (a 97% reduction from baseline), and again increased to a mean of 6.7 per min during a return to baseline. Likewise, compliance increased to average 87% during DRA and decreased back to 82% in the return to baseline.

Fig. 2
figure 2

Treatment evaluation: Larry. Note: Problem behavior is displayed in the top panel and compliance in the bottom panel. Following an initial exposure to and replication of the effects of DRA, schedule thinning with fixed and incrementing reinforcement durations was compared in a multielement design

Full size image
Fig. 3
figure 3

Treatment evaluation: Ivan. Note: Problem behavior is displayed in the top panel and compliance in the bottom panel. Following an initial exposure to and replication of the effects of DRA, schedule thinning with fixed and incrementing reinforcement durations was compared in a multielement design

Full size image
Fig. 4
figure 4

Treatment evaluation: Dan. Note: Problem behavior is displayed in the top panel and compliance in the bottom panel. Following an initial exposure to and replication of the effects of DRA, schedule thinning with fixed and incrementing reinforcement durations was compared in a multielement design

Full size image

Ivan (Fig. 3) engaged in a mean of 1.3 instances of problem behavior during baseline, which then decreased to near zero rates (approximately a 99% reduction) during DRA and returned to a mean of 4.7 per min in a return to baseline. Ivan’s compliance changed from 40% in baseline, to 100% during DRA, and decreased to 17% during the reversal to baseline. Dan (Fig. 4) engaged in a mean of 1.6 instances of problem behavior per min during baseline, which reduced to zero levels during DRA. Problem behavior then resumed at 1.3 instances per min in a return to baseline and again decreased to near zero levels (M < 0.1; a 98% reduction from BL) in a return to DRA. Dan complied with 88% and 98% of instructions during DRA conditions and 11% and 39% of instructions during the two baseline periods. Thus, the initial stages of DRA were shown to result in reduced problem behavior and increased compliance for all three participants.

The multielement, schedule-thinning comparisons are shown in the final phase of each panel in Figs. 2, 3 and 4. Fixed-reinforcement sessions are denoted by open circles and incrementing-reinforcement sessions as filled circles. In addition, the operating FR requirements in each condition are indicated by horizontal lines running atop the phase with ticks indicating the points at which each schedule incremented. Across the evaluation, Larry engaged in higher rates of problem behavior during fixed reinforcement sessions (M = 0.6 per min) than incrementing reinforcement sessions (M = 0.4 per min). These differential rates of problem behavior affected the efficiency of the two schedule-thinning procedures. Larry met the terminal goal (an FR-30 schedule) after 153 sessions with incrementing reinforcement. However, he only reached the FR-19 step in the same amount of exposure to schedule thinning with fixed-reinforcement durations.

Ivan also engaged in higher rates of problem behavior during fixed-duration sessions (M = 0.4 per min) than when reinforcement durations increased during schedule thinning steps (M = 0.1 per min). After 84 sessions, Ivan reached the terminal goal of an FR-20 schedule during incrementing-reinforcement sessions relative to the FR-13 schedule during fixed-reinforcement sessions. Similar to Larry and Ivan, Dan engaged in near zero levels of problem behavior across both schedule thinning procedures initially. However, problem behavior emerged during the FR-11 step for fixed sessions and the FR-13 step for incrementing-reinforcement sessions. Due to these elevated levels of problem behavior, this comparison was terminated for Dan, and we adopted an alternative treatment approach (data not shown). Neither schedule-thinning procedure reached the terminal goal of 20 tasks but thinning progressed further given incrementing-reinforcement and occasioned overall lower levels of problem behavior (M = 0.5 per min) than did fixed-duration reinforcement (M = 0.8 per min).

To highlight the impact of small differences in the rate of problem behavior across many sessions, the cumulative instances of problem behavior for each participant are depicted in Fig. 5. Across this evaluation, Larry engaged in 377 fewer instances of self-injury (a 39% reduction) in incrementing-reinforcement relative to fixed-reinforcement conditions. Ivan engaged in 252 fewer instances of aggression (a 69% reduction) and Dan engaged in 170 fewer instances of aggression (a 45% reduction) in incrementing-reinforcement conditions relative to fixed-reinforcement conditions. Problem behavior rates were similar across conditions at the early stages of the comparison when reinforcement durations were most similar and became more disparate at as the differences in reinforcer magnitude became larger in the later stages of the evaluation.

Fig. 5
figure 5

Cumulative problem behavior during schedule thinning. Note: Depicts the cumulative instances of problem behavior each participant exhibited during schedule thinning with fixed (open circles) and incrementing (filled circles) reinforcement periods. The initiation of each schedule thinning step is shown along the top of each panel. As the disparity in reinforcement duration became greater, differences in the occurrence of problem behavior became progressively more apparent

Full size image

To highlight differences in the speed of meeting mastery criterion (three consecutive sessions at a given FR-step with a 90% reduction in problem behavior relative to baseline levels) for both fixed- and incrementing-reinforcement procedures, the number of sessions required for Larry and Dan to meet criterion at each FR-step requirement are depicted in Fig. 6,Footnote 2 with incrementing-reinforcement sessions shown in the top row and fixed-reinforcement sessions shown in the bottom row. Sessions to mastery were nearly identical up to the FR 9 and FR 8 schedule step for Larry and Dan, respectively. However, after that point the number of sessions required to meet mastery progressively decreased (Larry, top left panel) or remained consistently low (Ivan, top right panel) during incrementing-reinforcement sessions. By contrast, the number of sessions required to meet the same criteria appeared to increase progressively as the work requirements increased during fixed-reinforcement sessions (bottom left and right panels for Larry and Ivan, respectively). Thus, not only was arranging incrementing reinforcement a more efficient procedure, but the gains in efficiency appeared to increase exponentially as work requirements increased.

Fig. 6
figure 6

Sessions required to meet sequential FR-schedule advancement requirements. Note: Depicts the number of sessions to meet mastery criterion to advance to the next FR-schedule during schedule thinning with incrementing-reinforcement durations (top rows of panels) or fixed-reinforcement durations (bottom rows of panels) for Larry (left column) and Ivan (right column)

Full size image

The greatest potential cost of incrementing reinforcement durations seems to come from a loss of instructional time as longer periods of reinforcement will result in fewer instructional episodes per session. To assess this cost, we first calculated the mean number of instructions issued during the last three sessions of the highest achieved FR schedule during fixed-reinforcement conditions. For example, Larry received a mean of 88 (range: 84–92) instructions per session during the FR-19 step of the fixed-reinforcement condition. We then compared that mean to the equivalent FR step of the incrementing-reinforcement conditions. For instance, Larry received a mean of 51 (range: 40–57) instructions per session during the FR-19 step of incrementing-reinforcement conditions. We then divided mean of the incrementing conditions by the mean of the fixed-reinforcement conditions and converted that quotient into a percentage. Larry received 42% fewer instructions during incrementing reinforcement sessions relative to the fixed-reinforcement sessions at the equivalent FR-19 schedule. Ivan received 22% fewer instructions per incrementing-reinforcement session relative to fixed-reinforcement sessions at FR-13 schedule. Dan received 64% fewer instructions during incrementing-reinforcement sessions relative to fixed-reinforcement sessions at the FR-11 schedule. However, these data are flawed indicators of instructional efficiency.

Our procedures required compliant responses to occur consecutively without the occurrence of problem behavior to satisfy our reinforcement schedule. Said another way, when a participant engaged in problem behavior, their work schedule reset and compliance with additional instructions was needed to produce reinforcement. Thus, sessions with higher levels of problem behavior would involve more frequent instruction delivery and more opportunities for compliance relative to an identical session without problem behavior. For instance, Dan received a mean of 90 instructions per session and engaged in a mean of 30 instances of problem behavior during his last three FR-11 sessions with fixed reinforcement, relative to 33 instructions per session and zero instances of problem behavior during his last three incrementing-reinforcement sessions. Trying to evaluate instruction or compliance data as a means of instructional efficiency is confounded by our problem behavior omission requirements. Future research could eliminate the DRO contingency to gain a clearer measure of differential instruction delivery across these two schedule thinning procedural variations.

We believe that a better indicator of therapeutic efficiency, for the current data set, was the speed with which participants met their treatment goals.Footnote 3 The differential occurrence of problem behavior affected the success of each procedure at attaining the terminal goal of schedule thinning. Both Larry and Ivan met their terminal thinning goals (FR 30 and FR 20) during incrementing-reinforcement conditions after 153 and 84 sessions, respectively, but failed to do so after an equal exposure to fixed-reinforcement conditions, which reached the FR 19 and FR 13 steps, respectively. Neither procedure resulted in meeting the terminal goals for Dan but incrementing-reinforcement sessions advanced further before we decided to terminate this comparison. Thus, across participants incrementing-reinforcement durations resulted in a more effective treatment model that optimized instructional time. We did not continue to conduct fixed-reinforcement sessions after incrementing-reinforcement sessions met their terminal criteria. Provided more time, fixed-reinforcement schedule thinning may have achieved similar goals. However, the amount of time required to achieve these goals would have been substantively more than required by incrementing-reinforcement durations during fading.

Discussion

The current study compared two variations of schedule thinning following DRA with three individuals who engaged in escape-maintained problem behavior. Incrementally increasing the duration of reinforcement delivered commensurate with each increase in work requirement resulted in decreased instances of problem behavior relative to thinning in which reinforcement duration was held constant. Increasing reinforcement duration also decreased the amount of instructional time within each session, so it is important to compare the benefits gained relative to the cost of this manipulation. Based purely on rates of problem behavior, differences between incrementing and fixed-reinforcement procedures may seem marginal. That is, incrementing-reinforcement sessions averaged 0.2, 0.3, and 0.3 instances of problem behavior per min less than fixed-reinforcement sessions for Larry, Ivan, and Dan, respectively. However, given the amount of exposure to thinning procedures, these accumulated small differences were substantial (a reduction of 377, 252, and 170 fewer instances of aggression for Larry, Ivan, and Dan, respectively). Thus, the impact of delivering incrementally larger durations of reinforcement was associated with a notable decrease in aggression over time.

We evaluated changes in the speed of meeting criterion over time for both fixed- and incrementing-reinforcement conditions (data depicted in Fig. 6). Sessions to criterion were nearly identical initially, but the number of sessions required to meet the same criteria increased as the work requirements increased during fixed-reinforcement sessions. However, the number of sessions to meet criterion remained low for incrementing sessions, even as the work requirements progressed. Thus, not only was arranging incrementing reinforcement a more efficient procedure, but the gains in efficiency appeared to increase exponentially as work requirements increased.

Increasing the duration of reinforcement may affect reductions in escape-maintained problem behavior either by increasing the magnitude of delivered reinforcement (Fulton et al., 2019; Trosclair-Lasserre et al., 2008) or by diminishing aversive aspects of instruction delivery by providing longer breaks between instructional episodes. Such a distinction is not possible in the current data set due to the use of duration-based reinforcement procedures. That is, by definition, expanding break periods decreases the frequency of instructional episodes. If one wanted to separate those influences in future research, they could potentially do so by providing access to positive reinforcers following compliance for which increases in magnitude are not measured in units of time. For instance, one could provide consumable snacks as a positive reinforcer for compliance and increase the magnitude of reinforcement in terms of the size or number of snack items while holding the consumption duration constant during both fixed and incrementing-reinforcement conditions.

The differences between incrementing- and fixed-reinforcement conditions may be conceptualized as changes in the unit price of reinforcement (Roane et al., 2007). Unit price refers to the ratio of the cost of a commodity (i.e., the amount of work required to produce that commodity) relative the amount of the commodity earned (i.e., the duration of reinforcement). Thus, at the onset of DRA in the current study, 30 s of reinforcement (the commodity) had a unit price of 1. In the fixed-reinforcement condition, when the price (schedule) was increased to an FR 2, the unit price for the same commodity doubled (2) and then tripled (3) when the work requirements were increased to an FR 3, and so on. The unit price also increased during incrementing-reinforcement arrangements, but to a lesser extent. For example, the transition from an FR 1 to an FR 2 and FR 3 resulted in a unit price changing from 1 to 1.7 to 2.3, respectively.Footnote 4 Thus, increases in unit price were slowed by incremental increases in reinforcement duration.

The experimenters chose not to hold unit price constant in the current investigation for practical rather than experimental reasons. As an example, maintaining a constant unit price in Larry’s case would have involved providing 30 s of reinforcement for each completed task. With his terminal requirement of placing 30 toys into a bucket, maintaining a constant unit price would require a 15-min reinforcement period. Such lengthy reinforcement periods may not be practical in common application. The terminal reinforcement duration at the FR-30 step for Larry was approximately 3 min in this study. A more formal evaluation of social validity is warranted but balancing instructional time with earned break durations may be influential in the adoption of such procedures in homes and schools.

The magnitude of reinforcement increments in this study were selected fairly arbitrarily with a goal of maintaining socially acceptable break periods. Thus, the overall successful treatment outcomes may be fortuitous, particularly given the relatively small number of participants included. There may be individuals, like Dan in the current study, for whom small reinforcement increments are insufficient at maintaining compliance as work requirements increase; larger increments may be necessary. Thus, we recommend the results of this evaluation be interpreted not that reinforcement durations should be incremented by 5 s at each step, but rather that there are therapeutic benefits to incrementing-reinforcement as schedule thinning progresses. Identifying the sufficient and necessary increases per step remains an area for future investigation.

It is also plausible that incrementing reinforcement per se was not responsible for the differences in responding across conditions, but rather maintaining a sufficient, critical magnitude of reinforcement was necessary to maintain compliance under longer work schedules. That is, Larry and Ivan were each earning 175 s and 125 s of reinforcement at the FR-30 and FR-20 work requirements, respectively. Delivering reinforcement at those durations at the onset (i.e., the FR-1 schedule) and maintaining that fixed duration during schedule thinning could have yielded similar outcomes to our incrementing reinforcement procedures. Future research should conduct this comparison evaluating both its efficacy and efficiency relative to incrementing reinforcement procedures.

Fixed-reinforcement schedule thinning may also have been more effective if evaluated in isolation. The multielement alternation between sessions delivering a larger magnitude of reinforcement for compliance (incrementing reinforcement sessions) could promote lower levels of reinforcement earning and higher levels of problem behavior during sessions delivering a smaller duration of reinforcement (fixed reinforcement sessions) due to behavioral contrast (Reynolds, 1961; Weatherly et al., 1997). This potential confound could be resolved using other experimental designs but doing so would introduce additional confounds such as sequence effects in reversal designs and individual differences in across-participant comparison designs.

The success of incrementing- (and fixed-) reinforcement procedures may also have been bolstered by the inclusion of “enhanced” breaks (i.e., those in which positive reinforcers are delivered) relative to breaks in which task demands are removed, but no additional reinforcers are delivered. There are likely several therapeutic benefits to including positive reinforcers during breaks. First, the delivery of additional reinforcement may increase the likelihood that a desirable behavior such as compliance may be differentially strengthened relative to problem behavior, which may or may not historically have produced positive reinforcers when parents, teachers, and other caregivers have withdrawn their instruction. Said another way, the inclusion of positive reinforcers could bias responding towards the more desirable response option (Lalli et al., 1999; Slocum & Vollmer, 2015). Second, the inclusion of positive reinforcers within “demand” contexts may decrease the value of escaping from those contexts. Several studies have shown that the inclusion of positive reinforcers (either contingently or noncontingently) during work periods can reduce the occurrence of escape-maintained problem behavior (Horner et al., 1997; Ingvarsson et al., 2008, 2009; Lomas Mevers et al., 2014). Third, a handful of studies have shown that when provided a choice between receiving a break or receiving access to positive reinforcers, a plurality of participants with escape-maintained problem behavior have preferred positive reinforcers (e.g., DeLeon et al., 2001; Fisher et al., 2005; Kodak et al., 2007; Lalli et al., 1999). Thus, our interpretation of these data is that breaks should be enhanced by positive reinforcers when treating escape-maintained problem behavior to increase treatment efficacy, minimize aversive aspects of instruction, and to provide the most socially acceptable treatment options available.

There are a few limitations to this study that should be noted and addressed in follow-up studies. Although we analyzed our implementation of scheduled work requirements to ensure procedural integrity, we did not formally measure the duration of break periods during our fading comparison. Doing so would have required scoring the onset and offset of break periods based upon the presentation and removal of instructional materials by the therapist. This formal measurement of procedural integrity would have bolstered our findings. The capturing of break durations would also have permitted us to remove reinforcement periods from our calculations of problem behavior rates. That is, the rate of problem behavior across fixed and incrementing reinforcement conditions may have been more similar when comparing only instructional periods rather than across overall sessions.

Although we demonstrated that escape from the presented tasks served as a negative reinforcer via our functional analyses, we did not include a formal process for including and excluding tasks prior to our functional analysis (e.g., Call et al., 2009). We included tasks that were nominated by caregivers as evocative events preceding problem behavior. Thus, these tasks were socially accurate approximations of what those participants would experience in their typical environments. However, if those nominated tasks failed to evoke problem behavior during the functional analysis and subsequent baseline evaluation, additional formal assessments would have been warranted. We also did not evaluate the generality of these outcomes to other tasks within participants. Participants had historically completed each of these tasks at some level prior to their involvement in this study; it is not clear if similar outcomes would have been seen with more novel, acquisition tasks.

Although conceptual questions about the causes of the differences between these procedures remain, we believe there are practical benefits of incrementing reinforcement durations. We believe that offering additional “compensation” is a reasonable accommodation as we increase the amount of work that we ask of our clients, and in doing so, we create an environment characterized by greater access to sources of positive reinforcement. We encourage practitioners to adopt incrementing-reinforcement procedures when schedule thinning with their clients while researchers continue to parse apart the mechanisms responsible for these benefits.