Introduction

Token economies are a frequently implemented behavior management strategy and represent a common application of behavior analysis to behaviors of social concern (e.g., noncompliance, class disruptions; Boerke and Reitman 2011). As described by Ayllon and Azrin (1968), token economies deliver conditioned reinforcers (e.g., tokens) contingent on a target behavior to increase the frequency of that behavior. After a predetermined number of conditioned reinforcers are earned, these reinforcers can be exchanged for a backup reinforcer (e.g., time with computer). For example, one token might be delivered to a student after completion of a work activity. Completion of ten work activities results in the individual having earned ten tokens which can be traded in for 10-min access to a preferred video. Despite their long-standing use, research on token economies has decreased in recent years (Matson and Boisjoli 2009). Thus, limited research exists to assist practitioners with identifying the conditions under which optimal performance will occur within a token economy.

In practice, the design of the token economy may be one of the most challenging aspects of implementing conditioned reinforcement programs for clinicians and educators. Token economies are comprised of numerous interdependent components, including the token production schedule (schedule by which responses produce tokens) and token exchange schedule (schedule by which tokens are exchanged for backup reinforcers; Hackenberg 2009), in addition to being influenced by dimensions of reinforcement (e.g., quality, magnitude) of the backup reinforcer for which the tokens are exchanged. Several studies have provided evidence for the importance of these considerations (Hupp et al. 2002; Field et al. 2004). For example, Field and colleagues demonstrated that shortening the latency to delivery of the backup reinforcer after earning the predetermined number of conditioned reinforcers decreased problem behavior and increased more appropriate behavior for three children admitted to a residential care facility. Intuitively, adjusting the immediacy of reinforcer delivery, along with the quality, rate, and magnitude of reinforcer delivery will affect the efficacy of token economy programs. However, few studies have evaluated differential effects of these dimensions of reinforcement. Furthermore, although several studies have demonstrated benefits of various dimensions of conditioned and backup reinforcers (Field et al.; Hupp et al.), no study has shown a method for establishing which dimensions of reinforcement are most relevant when developing token economies.

When determining the dimensions of reinforcement that may affect the success of token economy arrangements, the matching law may provide a conceptual framework to begin studying these relations. The matching law (Herrnstein 1961) predicts response allocation across multiple response options that produce different schedules of reinforcement within a concurrent schedules arrangement. For example, Herrnstein showed greater response allocation to a response option that was associated with the relatively higher rate of reinforcement according to a concurrent VI–VI schedules arrangement. As predicted by the matching law, many studies have shown that dimensions of reinforcement can be manipulated to result in a shift in responding from problem to appropriate behavior (Falcomata et al. 2010; Friman and Poling 1995; Gardner et al. 2009; Piazza et al. 2002). For example, Gardner and colleagues showed lower levels of escape-maintained problem behaviors during a work period with higher-quality attention compared to a work period with lower-quality attention.

Neef et al. (1992) conducted one of the first translational studies in which the effect of various dimensions of reinforcement was systematically evaluated. Specifically, they examined the effect of quality and rate of reinforcement on response allocation. They hypothesized that multiple dimensions of reinforcement may affect human choice responding. Thus, they evaluated response allocation for a choice option that was associated with a high-quality reinforcer (nickels) delivered at a low rate of reinforcement (high quality/low rate) against a choice option associated with a low-quality reinforcer (program money) delivered at a high rate of reinforcement (low quality/high rate). Results for three participants showed greater response allocation to the choice associated with the higher-quality reinforcer, despite the lower rate of reinforcement it produced. Other studies (Mace et al. 1994) have shown similar findings.

Neef et al. (1994) extended these findings by evaluating the effect of other dimensions of reinforcement on human choice responding. During this study, preference for rate of reinforcement, quality of reinforcement, immediacy of reinforcement, and response effort were evaluated within a concurrent schedules design. For each participant, six choice arrangements were presented; in each condition, two dimensions of reinforcement were manipulated, while other dimensions were held constant. Results for all six participants showed clear preferences among the dimensions of reinforcement, and these preferences differed across participants. This study was influential in that it provided a basis for understanding the aspects of behavioral treatment programs that could affect problem behavior (e.g., Falcomata et al. 2010). For example, Neef et al. hypothesized that Subject 5 in their study, who demonstrated a preference for immediacy of reinforcement, would benefit from a treatment that produced immediate reinforcement contingent on the alternative behavior.

Neef and Lutz (2001) extended this study by using the results of the preference assessment for dimensions of reinforcement to inform behavioral treatment of two children engaging in disruptive behavior in the classroom setting. Similar to previous studies, preference for dimensions of reinforcement was individualized. During a second, follow-up experiment, these authors designed a treatment program based on the results of the preference assessment and compared it to a baseline condition within a reversal design. Results showed lower levels of disruptive behavior when the treatment program was based on the results of the preference assessment. It is unknown whether similar results would be obtained when compared with a treatment program that involved a reinforcement arrangement that was mismatched to the results of the preference assessment.

Since the Neef et al. (1992, 1994) and Mace et al. (1994) studies, few studies have evaluated the application of preference assessments for dimensions for reinforcement and their potential application to behavioral treatment programs. Thus, the purpose for the current study was to first replicate the Neef et al. (1994) study to identify more- and less-preferred dimensions of reinforcement for students engaging in problem behaviors in classroom settings (Experiment 1). Following the preference assessment for dimensions of reinforcement, Experiment 2 applied this information to develop treatment arrangements that were matched and mismatched to the results of the preference assessment to treat problem behaviors maintained by negative reinforcement in the form of escape from demands.

Experiment 1

Method

Participants

Three children admitted to a specialized psychiatric unit providing multidisciplinary treatment for children diagnosed with intellectual and developmental disabilities participated in this study. All three participants were admitted to a partial hospitalization program providing psychiatric and behavior analytic treatment for problem behavior (for a detailed description of the program, see Gabriels et al. 2012). Caregiver and teacher report of tasks correlated with the occurrence of problem behavior-guided selection of each participant’s academic tasks. All three participants’ school teams either had a token economy in place already or expressed their willingness to implement one at school.

Mathias was a 6-year-old boy diagnosed with autism spectrum disorder and unspecified anxiety disorder. He communicated using full and complex sentences and had average cognitive abilities as measured by the Wechsler Intelligence Scale for Children, Fifth Edition (WISC-V). He was admitted to the partial hospitalization program due to concerns regarding aggression, property destruction, and noncompliance when presented with academic tasks at school. Mathias’ target problem behavior during this study was noncompliance. Academic tasks for Mathias were single-digit addition problems (e.g., 3 + 2 = 5).

Gregory was a 9-year-old boy diagnosed with autism spectrum disorder, generalized anxiety disorder, and unspecified disruptive, impulsive control, and conduct disorder. He communicated using full and complex sentences and had average cognitive abilities as measured by the WISC-V. Gregory was admitted due to concerns regarding aggression, property destruction, and inappropriate vocalizations (e.g., cussing) that he displayed at school. Each of these problem behaviors were targeted during the current study. Academic tasks for Gregory were single-digit multiplication and handwriting. Handwriting tasks consisted of Gregory writing words spoken by an experimenter.

Nathaniel was a 14-year-old young man diagnosed with unspecified anxiety disorder and autism spectrum disorder. He communicated using full and complex sentences and had average cognitive abilities as measured by the WISC-V. Nathaniel was admitted due to concerns regarding homicidal ideation, aggression, and inappropriate vocalizations that he displayed at school. Problem behaviors targeted during the current study were inappropriate vocalizations. The academic task used during Nathaniel’s analysis was handwriting. Handwriting tasks consisted of Nathaniel writing sentences based on a story starter (e.g., a story about going to school).

Setting and Materials

Sessions were conducted in classrooms on the psychiatric unit. Classrooms were approximately 4.5 m × 6.1 m. Each classroom was equipped with six desks, one large table, and leisure activities. Academic work was presented on 20.32-cm × 25.4-cm pieces of paper.

Response Definitions Observation System and Interobserver Agreement

Dependent variables for Experiment 1 task completion, time allocation, and selection.

Task Completion Task completion was defined as completion of an adult request. Mathias was presented with one math problem at a time. Each time he completed a math problem, task completion was recorded. Task completion was a permanent product, and the frequency of tasks completed was divided by the total session time to produce responses per minute.

Time Allocation Time allocation was defined as the amount of time spent on a side of the therapy room when it was divided in two during the quality assessment. Thus, time allocation was coded as a duration measure. Duration of time allocation was measured using a timer. To calculate percentage of time allocation, we divided the amount of time spent on either side of the therapy room by the total session duration and then multiplied by 100.

Selection A selection response was defined as vocalizing or gesturing to a response option. Thus, selection was coded as frequency of occurrence. Selections during Experiment 1 were used to determine preference for dimensions of reinforcement. To determine preference for dimensions of reinforcement, total selections of each dimension were divided by three (i.e., the total number of opportunities to select that dimension). The two dimensions of reinforcement that were selected most frequently were defined as more preferred, and the two dimensions of reinforcement selected least frequently were defined as less preferred.

Data were collected using data sheets by observers who stood across the room to collect data for each dependent variable. Total interobserver agreement (IOA) was used for problem behavior and task completion the current study. Total IOA was calculated by dividing the lower number of target behaviors scored by one observer (either primary or reliability) by the higher number of target behaviors scored by the second observer for Gregory and Nathaniel. For Mathias, IOA for noncompliance was calculated by dividing the lower duration of noncompliance scored by one observer by the higher duration of noncompliance scored by the second observer.

During the functional analysis, IOA was calculated for 20% of sessions for Mathias, 23% of sessions for Gregory, and 100% of sessions for Nathaniel. IOA was 100% for all three participants. During the quality assessment, IOA was calculated on 100% of sessions and was 100%. During the preference assessment for dimensions of reinforcement, IOA was calculated on 100% of choices for Mathias and Gregory and 50% of choices for Nathaniel. IOA for all three participants was 100%.

Experimental Design

Each participant’s functional analysis of problem behavior was conducted within a multielement design to evaluate the reinforcer(s) maintaining problem behavior. The quality assessment was conducted within a concurrent schedules design to identify relatively higher- and lower-preferred stimuli. Experiment 1 was conducted within a concurrent chains arrangement to evaluate participant preference for each dimension of reinforcement. That is, response selection produced an opportunity to comply with demands to access the programmed reinforcer.

Experimental Procedures

Functional Analysis A functional analysis was conducted for all three participants based on procedures outlined by Iwata, Dorsey, Slifer, Bauman, and Richman (1982/1994). The results of each participant’s functional analysis suggested his problem behavior was maintained by negative reinforcement in the form of escape from academic tasks.

Quality Assessment Following the functional analysis, a preference assessment was conducted to identify higher- and lower-quality stimuli to be used during Experiments 1 and 2. Toys were shown to be highly preferred based on the results of a free-operant preference assessment. The classroom was evenly divided, and similar toys were placed on each side. The experimenter stood on one side of the room and directed the participant to the middle of the room. The experimenter, who had limited interactions with the participants prior to this assessment, described to the participant that one side of the room was where he could play with the experimenter (interactive play) and the other side of the room was where he could play alone. The experimenter then instructed the participant to select where he wanted to play and told them that they could change sides at any time. Midway through the session, the experimenter reminded the participant that they could move to the other side if he wanted. Sessions of the quality assessment lasted 5 min and were conducted three times for each participant prior to the start of Experiment 1.

Experiment 1During Experiment 1, the experimenter (who conducted the quality assessment as well as all subsequent sessions) escorted the participant into the classroom and directed him to sit at a table. The experimenter then presented the choice to the participant. Each choice was contrived to evaluate two dimensions of reinforcement, while the other two dimensions of reinforcement being held constant. Each choice was presented once. That is, every option was presented with every other option once. Meeting the requirements of a differential reinforcement of alternative behavior (DRA) schedule of reinforcement produced token delivery. That is, engaging with the task for a predetermined period resulted in token delivery. The therapist used a timer to measure task engagement and engaged in token delivery when the DRA schedule requirements were met. The DRA schedule varied depending on the choice. Please see Table 1 for a description of each choice. Session length for each choice varied depending on the condition. Varied session length occurred because of choice-specific components, like delay to reinforcement. However, each academic demand was always removed after the participant earned five tokens. Problem behavior resulted in session time pausing until the participant began working for 3 consecutive seconds. A choice would have been terminated if 5 consecutive minutes elapsed without task engagement. Participants could not change options after one was made. After the reinforcement period ended, the next choice was introduced.

Table 1 Description of dimension of reinforcement during the preference assessment for dimensions of reinforcement
Full size table

The rate of token delivery was determined based on the average latency to problem behavior during escape sessions of the functional analysis for each participant. The average latency to problem behavior for Mathias was 20 s, 50 s for Gregory, and 90 s for Nathaniel. To increase the likelihood, the participant would contact reinforcement, the schedule of token delivery was shortened such that tokens were delivered after completing work for 15 s for Mathias, 45 s for Gregory, and 60 s for Nathaniel when the high rate schedule of reinforcement was in effect. The low rate schedule of token delivery was working for 45 s (Mathias), 90 s (Gregory), and 120 s (Nathaniel). The higher-quality reinforcer for each participant was play with an adult, and the lower-quality reinforcer was play alone. Immediate reinforcement was when the participant accessed reinforcement immediately after earning five tokens; delayed reinforcement was when the participant needed to wait 6 min before reinforcer delivery after earning five tokens. High-magnitude reinforcement was when the participant accessed 6 min of reinforcement with the higher/lower-preferred stimulus; low-magnitude reinforcement was when the participant accessed 2 min of reinforcement with the higher/lower-preferred stimulus.

Rate versus Quality During the first-choice arrangement, participant preference for rate of token delivery and quality of reinforcement was evaluated. The experimenter indicated that the participant could either earn tokens at a high rate that could be exchanged for 2-min access to a low-quality reinforcer (HR/LQ) or earn tokens at a low rate to play with the high-quality reinforcer for 2 min (LQ/HQ). For example, in Mathias’ case, the experimenter said, “You can earn tokens every 15 s to get a 2-min playtime by yourself, or you can get tokens every 45 s to get a 2-min playtime with me. Do you want to work less to play by yourself or more to play with me?” The same amount of playtime (i.e., magnitude of reinforcement) was delivered immediately after earning all five tokens (i.e., immediacy of reinforcement). The participant independently completed work following their choice. Academic assistance was available from the experimenter during this choice and all subsequent choices too.

Quality versus Immediacy The second choice evaluated participant preference for quality of reinforcement or immediacy of reinforcement. The experimenter indicated that the participant could earn tokens to gain access to a low-quality reinforcer for 2 min immediately after earning all tokens (LQ/Imm) or to gain access a to high-quality stimulus after waiting 6 min (HQ/Del) for the reinforcer to be available for 2 min. Tokens were delivered at a high rate and the backup reinforcer was available for 2 min for both response options. An example description of this choice was, “Mathias, you can either earn tokens every 15 s to play by yourself right after you earn five tokens or you can earn tokens every 15 s to play with me after waiting for 6 min. Would you rather work to play by yourself right away or work to play with me after waiting?”.

Rate versus Immediacy The third choice evaluated participant preference for rate of token delivery and immediacy of reinforcement. The experimenter indicated that the participant could earn tokens at a high rate to gain 2-min access to a high-quality reinforcer 6 min after finishing work (HR/Del) or earn tokens at a low rate to gain 2-min access to a high-quality reinforcer immediately (LR/Imm) after finishing work. An example description of this choice for Mathias was, “You can earn tokens for working every 15 s to play with me after waiting for 6 min or you can earn tokens for working every 45 s to play with me right after you finish work. Would you rather earn tokens more quickly and then wait to play with me or earn tokens more slowly and play with me right after finishing work?”

Quality versus Magnitude The fourth choice evaluated participant preference for quality of reinforcement and magnitude of reinforcement. The experimenter indicated that the participant could earn tokens at a high rate for working to gain 2-min access to a high-quality reinforcer (HQ/LM) immediately after earning five tokens or earn tokens at a high rate for working to gain access to a low-preferred reinforcer (LQ/HM) for 6 min immediately after earning five tokens. An example description of this choice for Mathias was, “You can earn tokens every 15 s to earn a 2-min playtime with me or earn tokens every 15 s to earn a 6-min playtime by yourself. Would you rather work for a smaller playtime with me or work for a bigger playtime by yourself?”

Magnitude versus Immediacy The fifth choice evaluated participant preference for magnitude of reinforcement and immediacy of reinforcement. The experimenter indicated that the participant could work to earn tokens at a high rate to gain 6-min access to a high-quality reinforcer 6 min after finishing work (HM/Del) or earn tokens at a high rate to earn access to 2-min access to a high-quality reinforcer immediately after finishing work (LM/Imm). An example description of this choice for Mathias was, “You can earn tokens every 15 s to earn a 6-min playtime after waiting for 6 min or you can earn tokens every 15 s to earn a 2-min playtime right when you are done with work. Would you rather work for a big playtime after waiting or work for a small playtime right when you are done with work?”

Rate versus Magnitude The final choice evaluated preference for rate of token delivery and magnitude of reinforcement. The experimenter indicated that the participant could earn tokens at a high rate to gain access to a high-quality reinforcer for 2 min immediately (HR/LM) after completing work or earn tokens at a low rate to gain access to a high-quality reinforcer for 6 min immediately (LR/HM) after completing work. An example description of this choice arrangement for Mathias was, “You can earn tokens every 15 s to play with me for 2 min after finishing work or earn tokens every 45 s to play with me for 6 min after finishing work. Would you rather earn tokens more quickly for a small playtime or earn tokens more slowly for a bigger playtime?”

Results and Discussion

Each participant’s quality assessment (please see Table 2) to establish conditions of higher- and lower-quality stimulus presentation showed play with tangibles and experimenter was the higher-quality stimulus. Each participant allocated 100% of choice responding to the side of the room with the tangibles and the experimenter.

Table 2 Percentage of allocation during the quality assessment
Full size table

Data for participant selections during the preference assessment for dimensions of reinforcement are displayed in Table 3. Mathias’ preferred dimensions of reinforcement were a high rate of token delivery (selection = 66%) and immediate reinforcement (selection = 66%). Quality (selection = 33%) and magnitude of reinforcement (selection = 0%) were less preferred. Gregory’s preferred dimensions of reinforcement were rate (selection = 100%) of token delivery and magnitude of reinforcement (selection = 66%). Quality (selection = 0%) and immediacy of reinforcement (selection = 33%) were less preferred. Nathaniel’s preferred dimensions of reinforcement were rate of token delivery (selection = 66%) and magnitude of reinforcement (selection = 100%). Immediacy (selection = 33%) and quality of reinforcement (selection = 0%) were less preferred. Mathias did not respond to one choice arrangement (rate versus immediacy of reinforcement) due to noncompliance with adult instructions. Problem behavior did not occur during the other choice arrangements for Mathias and did not occur at all for Gregory and Nathaniel.

Table 3 Preference assessment for dimensions of reinforcement
Full size table

Similar to Neef et al. (1994), preference for dimensions of reinforcement became clear during the brief preference assessment for dimensions of reinforcement implemented during Experiment 1. Results from this preference assessment for dimensions of reinforcement can likely be translated to a treatment context (see also Falcomata et al. 2010). Much research has shown that matching a treatment arrangement to hypothesized preference for dimensions of reinforcement would result in socially relevant treatment outcomes (Field et al. 2004; Hupp et al. 2002; Neef and Lutz 2001). The purpose of Experiment 2 was to extend this study by validating the results of the preference assessment conducted in Experiment 1 by showing that token economy arrangements matched to preference produced decreases in problem behavior and increases in task completion than less-preferred token economy arrangements.

Experiment 2

Participant, Setting, and Materials

The same three participants also participated in Experiment 2. The materials and setting remained the same.

Response Definitions, Observation System, and Interobserver Agreement

Dependent variables for Experiment 2 were aggression and disruption (Gregory), noncompliance (Mathias), inappropriate vocalizations (Gregory and Nathaniel), task completion (all participants), time allocation (all participants), and selection (all participants).

Problem Behaviors Aggression was defined as Gregory’s hand or leg forcefully coming into contact with an experimenter’s body. Disruption was defined as Gregory pushing away, scribbling on, or tearing work materials. Inappropriate vocalizations (Nathaniel only) were defined as using obscene language or making negative statements about an activity. Aggression, disruption, and inappropriate vocalizations were recorded as frequency of occurrence and represented as responses per minute. To calculate rate of problem behavior, we divided the frequency of problem behaviors by the session duration in minutes. Noncompliance (Mathias only) was defined as failure to comply with an adult request within 5 s. For example, if Mathias was directed to solve 1 + 1, and he crawled underneath the table for 5 s, noncompliance would be counted until he began completing the requested problem. Noncompliance was recorded as duration of occurrence and was represented as percentage of session. To calculate noncompliance, we divided the total number of seconds in which Mathias engaged in noncompliance by the session duration multiplied by 100.

Task CompletionTask completion was defined as completion of an adult request. For example, Mathias was presented with one math problem at a time. Completion of each math problem resulted in one task completed. Task completion was a permanent product, and the frequency of tasks completed was divided by the total session time to produce responses per minute. Session time included only time spent with a demand in place. Reinforcement time was excluded.

Total IOA was calculated in the same way as in Experiment 1. IOA calculated on the treatment evaluation was collected on 25, 38, and 60% of sessions for Mathias, Gregory, Nathaniel, respectively. IOA for Mathias averaged 98.3% (range 96.7–100%). IOA for Gregory was 100%. IOA for Nathaniel was 66.8% (range 33–100%). After Nathaniel’s session in which IOA was 33%, both observers clarified the response definitions and IOA for the next session was 100%.

Experimental Design

Experiment 2 was conducted within an ABAB (A = more-preferred token economy arrangement; B = less-preferred token economy arrangement) reversal design for Mathias and Gregory and an ABA reversal design for Nathaniel to evaluate changes in problem behavior (or noncompliance in Mathias’ case) and task completion when each token economy arrangement was in effect.

Procedures

Each participant was directed to a table in the classroom. The same academic tasks used during Experiment 1 were then presented, and the participants were again required to earn five tokens to access the backup reinforcer. At the beginning of each session, directions for the relevant condition were described to the participant. As a reminder, the two dimensions of reinforcement that were selected most frequently during the preference assessment for dimensions of reinforcement were defined as more preferred and the two dimensions of reinforcement selected least frequently were defined as less preferred. Mathias’ more-preferred token economy arrangement was to work for 15 s to earn 2-min access to a low-quality reinforcer immediately after finishing work. Mathias’ less-preferred token economy arrangement was to work for 45 s to earn 6-min access to a high-preferred reinforcer after waiting for 6 min following completion of the work activity. Gregory’s more-preferred token economy arrangement was to work for 45 s to earn 6-min access to a low-quality reinforcer after waiting for 6 min. His less-preferred token economy arrangement was to work for 90 s to earn 2-min access to a high-quality reinforcer immediately after finishing work. Nathaniel’s more-preferred token economy arrangement was to work for 60 s to earn 6-min access to a low-quality reinforcer after waiting for 6 min. His less-preferred token economy arrangement was to work for 120 s to earn 2-min access to a high-quality reinforcer immediately after finishing the work activity.

Sessions of each condition continued until clear visual separation between problem behavior occurring during the more- and less-preferred arrangements was observed. Session length varied but either ended at the conclusion of the 2 or 6-min reinforcement period or following 5 min of no work completion. Like Experiment 1, experimenter used a timer to measure task engagement and deliver tokens per the relevant schedule of reinforcement. During sessions of both the more-and less-preferred token economy arrangements, problem behavior always produced negative reinforcement. That is, escape extinction was not implemented, and the participants could choose not to complete work at any time. The experimenter waited until the participant either began to work before delivering tokens or until 5 min of no task completion occurred. No prompting strategies were used to increase task completion.

Results and Discussion

Results for all three participants showed higher rates of problem behavior under the less-preferred token economy arrangement as compared to the more-preferred arrangement. Differences in task completion were less evident between participants. Mathias (Fig. 1, top panel) engaged in zero or near zero levels of noncompliance during the more-preferred token economy arrangement (M = 1.7%; range, 0–6.7%). He completed an average of 1.35 RPM (range, 1.2–1.4 RPM) math problems when working under the more-preferred token economy arrangement. He engaged in relatively higher levels of noncompliance during the less-preferred token economy arrangement. Noncompliance occurred during an average of 94.1% (range, 76.7–100%) of each session. Lower rates of task completion occurred too (M = 0.45 RPM; range, 0–1.4 RPM).

Fig. 1
figure 1

Problem behavior and task completion data for Mathias (top panel), Gregory (middle panel), and Nathaniel (bottom panel). R, rate of reinforcement; I, immediacy of reinforcement; Q, quality of reinforcement; and M, magnitude of reinforcement

Full size image

Gregory (Fig. 1, middle panel)) did not engage in problem behavior during the first implementation of the more-preferred token economy arrangement. Task completion occurred at stable levels (M = 1.7 RPM; range, 1.4–2.0 RPM). In contrast, problem behavior increased to an average of 0.64 RPM (range, 0.36–0.92 RPM) during sessions of the less-preferred token economy arrangement. Task completion remained relatively stable during these sessions though (M = 0.73 RPM, range, 0.53–0.93 RPM). A return to the more-preferred and less-preferred token economy arrangements produced similar results for both problem behavior (more preferred M = 0.03 RPM; less preferred M = 0.5 RPM) and task completion (more preferred M = 1.4 RPM; range, 0.8–2.0 RPM; less preferred M = 1.67 RPM; range, 0.67–2.67 RPM).

Nathaniel (Fig. 1, bottom panel) did not engage in problem behavior during the first and second implementations of the more-preferred token economy arrangement. Task completion occurred at an average rate of 1.8 RPM (range, 1.4–2.0 RPM). Problem behavior occurred at a higher rate during the only implementation of the less-preferred token economy arrangement (M = 1.8 RPM; range, 1.4–2.0 RPM). Task completion was exhibited at approximately equivalent rates to the more-preferred token economy arrangement (M = 1.7 RPM).

Results for all three participants validated the results of the preference assessment for dimensions of reinforcement conducted during Experiment 1. Mathias, Gregory, and Nathaniel each showed greater rates of problem behavior during the lower-preferred arrangement. Mathias showed corresponding decreases in task completion during the lower-preferred arrangement. Gregory and Nathaniel did not show this pattern and completed approximately equivalent amounts of work during both the more- and less-preferred arrangements. These data indicate that the preference assessment for dimensions of reinforcement was useful for informing the design of an effective behavioral treatment to decrease escape-maintained problem behaviors, as each participant’s problem behavior responded well to the higher-preferred treatment arrangement. As was shown by Neef et al. (1992, 1994), combining preferred dimensions of reinforcement shifted response allocation to task completion from problem behavior for one participant. Important to note, too, is that the results of one participant’s preference assessment would be unlikely to positively influence the behavior of another participant. Thus, incorporating individual preference into treatment seems to be highly important.

General Discussion

The current two-experiment study sought to implement and validate a preference assessment for dimensions of reinforcement to inform token economy arrangements. Clear preferences for dimensions of reinforcement were identified for each of three participants (see Table 3). The dimensions of reinforcement were then combined to form a more- and less-preferred token economy arrangement. Results for all three participants showed higher levels of problem behavior during the less-preferred arrangement. Mathias engaged in lower rates of task completion during the less-preferred arrangement compared to the more-preferred arrangement. Gregory and Nathaniel completed tasks at approximately the same rates regardless of the arrangement.

As was suggested by Neef et al. (1994), the results of this preference assessment can be used to inform a treatment arrangement designed to decrease problem behavior and increase an alternative behavior. For the current study, all three participants showed differentiation for rates of problem behaviors between the more- and less-preferred token economy arrangements. Mathias showed differences in his task completion data. Mathias’ task completion data covaried with increases (and decreases) in problem behavior, which would be predicted because problem behavior (noncompliance) was based on the absence of task completion. In contrast, Gregory and Nathaniel’s rates of task completion remained stable throughout the study. Previous studies have also failed to show covariation between problem behavior and task completion with task completion persisting even as problem behavior increases (Rispoli et al. 2013; Slocum and Vollmer 2015). Rispoli and colleagues delivered response-independent access to tangibles or escape and found that differences in problem behavior emerged, but task completion was relatively stable across conditions for some of the participants. It is plausible that in some cases, task completion may be maintained by different sources of reinforcement than problem behavior (Wacker et al. 2011). For example, Wacker et al. (2011) showed the persistence of task completion well before problem behaviors failed to show resurgence. It could be that task completion is maintained by even subtle adult attention, whereas problem behavior is maintained by escape from demands (Fewell et al. 2016). Future research might continue investigating this to understand the sources of reinforcement maintaining both appropriate and problem behaviors.

The matching law (Herrnstein 1961) helps predict choice allocation within concurrent schedules arrangements. Previous research has shown that delivering reinforcement at a higher rate for one response alternative produces higher response allocation to that alternative when an equivalent response is concurrently available but produces reinforcement at a lower rate. Exceptions to the matching law occur when other dimensions of reinforcement are introduced to the concurrent schedule arrangement. For example, Neef et al. (1992) showed higher response allocation to a response that produced a higher-quality stimulus at a lower rate than a lower-quality stimulus at a higher rate. Neef and colleagues went on to extend these analyses by evaluating the influence of multiple dimensions of reinforcement on choice allocation (Neef et al. 1994).

Surely, though, these data need to be interpreted within the context they were evaluated. The high-quality stimulus evaluated during the quality assessment only assessed preference for one experimenter’s attention and a limited array of leisure activities. At school, adults with other histories of reinforcement and other types of leisure activities would be available and could affect the likelihood preference assessment results will result in positive treatment outcomes across settings. One example of this effect was documented by Lalli et al. (1999) for children engaging in problem behavior maintained by negative reinforcement. These researchers showed decreased rates of problem behavior when negative reinforcement (escape from a demand) was paired with positive reinforcement (access to edibles) contingent on task completion compared to when negative reinforcement alone was delivered. This could perhaps provide additional importance for taking into account individual preference when creating token economies so that an enriched break is being delivered following token collection.

One limitation for the current study was that the preference assessment and treatment evaluation were relatively brief. Participants experienced each choice once. Changes in allocation could have occurred following a second or third exposure to each choice. An example of the brevity of the treatment evaluation occurred with Nathaniel when one session was conducted during the first implementation of the more-preferred token economy arrangement. This was done in an attempt to implement a time-efficient analysis that could be used in applied settings. However, this produces at least two effects. First, due to the brevity of the analysis, it remains unclear how this treatment approach would operate in a classroom environment. It is also not reasonable to expect parents or caregivers to deliver reinforcers every 15 s, such as in Mathias’ case. Thus, future research should evaluate whether schedule thinning within the relevant dimensions of reinforcement will maintain positive treatment effects. Second, given diagnostic characteristics of disorders like attention deficit hyperactivity disorder, which are characterized by impulsivity, it is unclear how stable preferences might be for these types of children. Future research could investigate stability of preference for dimensions of reinforcement.

A second limitation of the current study was that rate of token delivery and response effort (i.e., amount of work that needed to be completed prior to token delivery) were confounded. A higher rate of token delivery produced a condition of overall lower response effort because less time was spent in a task demand relative to the lower rate of token delivery arrangement. This may have affected choice responding for all participants, as problem behavior was maintained by negative reinforcement and rate of reinforcement was one of the preferred dimensions for each participant. Future research extending this study should attempt to control for this confound, perhaps by standardizing time spent with a demand in place and requiring more tokens to be earned in the high rate of token delivery choice arrangement. This change would also isolate the effects of the preference during the token exchange process as opposed to evaluating preferences for both the token production and also token exchange processes during a token economy.

In conclusion, the current study replicated the results presented by Neef et al. (1994) and extended them to show how a preference assessment for dimensions of reinforcement can be used to determine treatment conditions to maintain appropriate responding during behavioral treatment. The brief preference assessment for dimensions of reinforcement was completed relatively quickly and helped produce effective treatment arrangements. We hope that these data will assist practitioners developing token economies in their classrooms, and also encourage additional research into effective token economy arrangements in the future.