Self-Control Training: A Scoping Review

Abstract

We conducted a scoping review of the behavior analytic self-control training (SCT) literature. To identify included articles, we searched key terms in six databases for articles published between 1988 and 2021. We included empirical articles that used a behavioral approach to self-control training with human participants for whom increasing self-control choice was a clinically significant goal and measured self-control and impulsive choice as dependent variables. Twenty-five experiments from 24 articles with a total of 79 participants were included in the review. This review aims to summarize the characteristics of SCT procedures and outcomes, provide recommendations for future research directions, and offer practical suggestions to clinicians incorporating SCT into practice. We examined similarities across studies regarding the independent variables manipulated in SCT, dependent variables measured, metrics of successful interventions, and assessment of generalization and maintenance of self-control choice. Twenty-one experiments arranged concurrent self-control- and impulsive-choice options with positive reinforcement, and four experiments arranged self-control training with negative-reinforcement contingencies. Variations of SCT included progressively increasing delays, intervening activities, signaled delays, antecedent rules, and commitment responses. Providing an intervening activity during the delay was largely successful at increasing self-control choice. Maintenance and generalization of increased self-control choice were assessed in two and three experiments, respectively. Future research should focus on improving the generality of SCT procedures in clinical settings by increasing terminal delays, fading out intervening activities, including probabilistic outcomes, and combining appetitive and aversive outcomes.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40617-023-00885-y.

As humans, we are always making choices to engage in one or more responses from a near-infinite number of available options. For example, at the end of the workday, we can choose to go to dinner, go to the gym, watch TV, or engage in any number of other activities. Often, our choices are influenced by the immediacy of the outcomes: some choices are followed directly by their consequences, whereas others have delayed outcomes. When presented with the opportunity to choose between a smaller reinforcer delivered immediately and a larger reinforcer delivered after a delay, the choice for the smaller immediate reinforcer is referred to as an impulsive choice,1 and the choice for the larger delayed reinforcer is referred to as a self-control choice (Rachlin & Green, 1972). Although the choice for the smaller-sooner reinforcer is not always problematic (e.g., taking a rest day vs. exercising every single day), making exclusively impulsive choices can be maladaptive.

Impulsive choice is often studied using delay-discounting tasks: participants choose between a smaller-sooner reinforcer and a larger-later reinforcer, the delay to which is varied across trials. At short delays, the larger-later reinforcer is often selected, but with increasing delays to its delivery, choice often shifts to the smaller-sooner reinforcer. This discounting pattern is observed across species (see Vanderveldt et al., 2016; and Rung & Madden, 2018, for cross-species reviews), but the present review focuses only on self-control and impulsive choice with humans. Delay sensitivity varies; some people will continue to choose the larger-later reinforcer as the delay duration increases, but others favor the smaller immediate reinforcer at relatively shorter delays. This heightened sensitivity to increased delay is associated with negative behavioral outcomes, including substance misuse and the likelihood of relapse (Coffey et al., 2013; Stanger et al., 2012). A tendency toward impulsive choice is also associated with risky sexual behavior (Celio et al., 2016), alcohol use disorder (Jones et al., 2015), cigarette use (Bickel et al., 1999), gambling (Vitaro et al., 1999), and obesity (Epstein et al., 2010).

Although neurotypical people often make self-control choices rather than impulsive choices in laboratory delay-discounting tasks (e.g., Logue et al., 1986; Schweitzer & Sulzer-Azaroff, 1988), a tendency toward impulsive choice is commonly observed in specific populations, including individuals with developmental delays (Dixon et al., 1998), attention deficit hyperactivity disorder (ADHD; Binder et al., 2000), and traumatic brain injuries (TBI; Dixon & Tibbetts, 2009). Choosing the impulsive rather than the self-control option can be problematic when delays to reinforcement are inevitable or when the choice for the impulsive option has detrimental outcomes. For example, a child in a classroom may raise their hand to request attention when the teacher is unavailable. The self-control choice would be to wait until the teacher is available, but some children (e.g., children diagnosed with ADHD) may be more likely to make an impulsive choice, such as engaging in disruptive behavior to obtain attention sooner. As such, there is a body of literature on self-control and impulsive choice with human populations across a variety of diagnoses. Of particular interest are behavioral interventions targeted at increasing self-control choice (thereby decreasing impulsive choice) and lengthening the tolerated delay to the larger reinforcer while maintaining choice for the self-control option.

Investigations influencing self-control versus impulsive choice in clinical populations have appeared in previous behavior-analytic literature, starting with Ragotsky et al. (1988), in which three participants with severe intellectual disability were offered choices between one piece of cereal delivered immediately and three pieces of cereal delivered at increasing delays. Results were reminiscent of the curves obtained in delay-discounting research with nonhuman animals and typically developing humans: choice for the larger reinforcer dominated when the delay to receipt was brief, but preference switched to the smaller-sooner option at increasing delays to larger-reinforcer delivery (Ragotsky et al., 1988). However, once a short delay was introduced for the smaller reinforcer, choice shifted toward the larger reinforcer delivered after a relatively longer delay. A similar procedure shifted choice allocation from impulsive to self-control choice in preschool children with a history of impulsive choice-making (Schweitzer & Sulzer-Azaroff, 1988). These early results suggested that modifying schedule parameters could produce clinically meaningful decreases in impulsive choice and sparked further investigations of the mechanisms behind shifting choice allocation, giving rise to behavioral interventions collectively known as self-control training (SCT).

SCT provides participants choices between an impulsive option, such as receiving one preferred edible reinforcer immediately, and a self-control option, such as receiving five edible reinforcers after a 60-s delay. One or more independent variables (i.e., training components) are applied to increase self-control choices for individuals who previously engaged in impulsive choices. Frequently, the primary dependent variable is response allocation across the impulsive and self-control choice options, but other dependent variables are often included, such as the delay to reinforcer delivery or engagement in an intervening activity during the delay.

Since the publication of Ragotsky et al. (1988) and Schweitzer and Sulzer-Azaroff (1988), numerous experimental attempts to shift choice toward the self-control option in clinical populations have been published. Two articles (Falligant, 2020; McKeel & Dixon, 2014) have described some of the common training procedures seen in the literature. Overall, the SCT literature is somewhat disparate in terms of procedures, baseline assessments, manipulated variables, and reporting of outcomes. A scoping review would contribute to the literature by identifying patterns and variations in experimental and training components, synthesizing outcomes, and informing recommendations for future research and practice. Thus, the objectives of the present review are to (1) gather and organize literature to which practitioners and researchers can refer; (2) summarize and integrate procedural approaches across the reviewed literature; (3) summarize and integrate outcomes across the reviewed literature; (4) suggest future research based on identified gaps; and (5) discuss the social significance of procedures and outcomes, including meaningfulness of behavior change, maintenance and generalization, and social acceptability of procedures. We were interested in examining commonalities between SCT articles published between 1988 and 2021 in terms of the employed independent variables and their outcomes, metrics of successful interventions, and generalization and maintenance of self-control choice.

Method

Search Procedure

We conducted this review according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR; Tricco et al., 2018). To identify publications that met inclusion criteria for this scoping review, we entered the search string ((impulsivity OR impulsiveness) OR ("self-control") OR ("self control") OR ("delay exposure") OR ("delay tolerance") OR ("impulsive choice") OR ("self control choice") OR ("self-control choice") OR ("temporal discounting") OR (“delay discounting”) OR ("delayed reinforcement") OR ("reinforcer delay")) AND (children OR adolescents OR adults) AND (training or intervention or procedure) AND (reinforcement OR reinforcing OR reinforcer OR punisher OR punishing OR punishment OR aversive) OR ("self control training" OR "self-control training")) into six databases: APA PsycInfo, Education Resources Information Center, Education Research Complete, Scopus, Pubmed, and Web of Science (search completed in July 2021). Our search protocol was not preregistered. Following the initial search, duplicates were removed within and across database search results. After the first author removed all duplicates, 1,961 items remained. The first and second authors then read the abstracts of all remaining articles to screen for inclusion or exclusion. A hand search of the reference lists of the articles was then conducted by the third author to identify additional articles that met the inclusion criteria. The third author obtained and reviewed the abstracts from the articles cited in the references sections of the included articles.

Secondary Article Screening and Scoring

To calculate interrater agreement of the screening procedure, the second author independently screened the abstracts of 651 of the 1,961 experiments (33.20%) previously screened by the first author to identify articles that met the inclusion criteria. Two authors completed an independent, secondary scoring of the included articles for 32%–48% of the experiments for each scoring category. Following individual screening or scoring, all discrepancies were discussed by all authors and resolved based on consensus prior to reporting results. The process for resolving disagreements involved revisiting the operational definitions for each category with all authors present to reduce the likelihood of coder bias. All results presented in this review are based on resolved disagreements.

Inclusion Criteria

We included experiments published between 1988 and June 2021 that met the following criteria: (1) manipulated a nonpharmacological independent variable; (2) evaluated response allocation across two or more mutually exclusive and experimenter-defined choice options that varied by the delay to access and some other reinforcer or response dimension (i.e., quality, quantity, duration of access, response effort, or programmed density of reinforcement); (3) systematically implemented SCT as an intervention (i.e., experiments in which impulsive choice was only assessed were excluded); (4) conducted SCT with human participants for whom increasing self-control choice was a clinically significant goal; and (5) participants experienced the contingencies associated with each choice (i.e., studies that employed hypothetical choice options were excluded). The earliest date in the search range was 1988 to identify publications following Schweitzer and Sulzer-Azaroff (1988) and Ragotsky et al. (1988).

Article Scoring

All experiments were scored and reviewed independently. The first author scored all included experiments for variables in the following categories.

Article and Participant Characteristics

We identified the authors, title, journal, and publication year for each experiment. In each included article, we recorded the number of participants and scored each participant’s reported characteristics. For each participant, we recorded age, gender, race/ethnicity, socioeconomic status, and diagnoses. If gender was not explicitly reported (e.g., “Participant X was a 5-year-old boy”), it was scored based on pronouns included in the method, results, or discussion when available (e.g., “He did not engage in challenging behavior”). If any of the above characteristics were not included in the article, they were scored as “not reported.” We also recorded the settings in which each participant experienced SCT, coded as home, clinic, school, work, or other.

Procedural Variables

We coded the type of preassessment, dependent variables, and whether there were assessments of maintenance and generalization. We created categories of common independent and dependent variables identified in the SCT literature. The SCT conditions in each experiment were scored based on these categories, which included (1) progressive delay; (2) intervening activity; (3) signaled delay; (4) antecedent rule; and (5) commitment response. We also scored manipulations of reinforcer or response parameters when they were part of the SCT training procedure (e.g., adjustments in task difficulty). For experiments that employed intervening activities, we recorded the type of activity. Many of the identified studies evaluated more than one SCT condition in isolation or as part of a treatment package (i.e., multiple SCT conditions were implemented simultaneously), which was noted in the scoring. Experiments were categorized as SCT with positive reinforcers or negative reinforcers to organize the discussion of results.

Task Characteristics

Reinforcer Dimensions

The SCT experiments that arranged positive reinforcement all manipulated delay to reinforcement (i.e., selection of the impulsive option led to immediate or faster delivery of the reinforcer compared to a selection of the self-control option) and at least one additional reinforcer dimension. We identified four dimensions that were manipulated in the included articles: magnitude (e.g., one snack item immediately vs. five snack items after a delay), duration (e.g., 1 min of access to a preferred item immediately vs. 5 min of access after a delay), quality (e.g., access to a highly-preferred item immediately vs. a low-preference item after a delay), and rate (e.g., access to reinforcers on a lean interval schedule of reinforcement immediately vs. a dense interval schedule of reinforcement after a delay). For each participant, the reinforcer dimensions were scored according to the dimension (or multiple dimensions) manipulated in addition to delay.

Task Arrangement

Experiments that employed negative-reinforcement contingencies arranged escape from a task as the reinforcer. Delay was manipulated in all experiments with negative-reinforcement arrangements, but rather than adjusting the delay to reinforcer delivery as is common in the arrangements with positive reinforcement, it was the delay to the task that varied in these experiments. We scored the parameters of the task that were manipulated in addition to delay in order to alter the value of the reinforcer. Magnitude was scored if the self-control and impulsive choices were differentiated based on the number of tasks required to access reinforcement (e.g., one task now vs. three tasks later). Difficulty was scored if relatively more difficult tasks were required for the impulsive choice and easier tasks were arranged for the self-control choice. Aversiveness was scored when highly aversive tasks were associated with the impulsive choice and less aversive tasks were associated with the self-control choice; in these cases, the aversiveness level of each included task was evaluated for each participant prior to SCT.

Outcomes

As a measure of efficacy, we scored participant-level outcomes of each SCT condition in relation to each dependent variable. Scorers inspected the results section, conducted visual analysis of the graphed data for each experiment, and recorded a binary (successful/unsuccessful) decision regarding whether the intervention altered each dependent variable for each participant based on the stated goals of the intervention (e.g., increased response allocation to the self-control choice, increased task engagement, decreased challenging behavior relative to baseline). Interventions were deemed successful if (1) visual analysis of the graphed data indicated a systematic change in behavior following the intervention in accordance with intervention goals; and (2) the experimental design permitted confident attribution of behavior change to the intervention. Interventions were deemed unsuccessful if the data indicated (1) no systematic change in behavior following the intervention based on visual analysis of the graphed data; (2) behavior change in an undesired direction (e.g., shifts in choice toward the impulsive option, decreased task engagement, increased challenging behavior relative to baseline); or (3) behavior change that could not reasonably be attributed to the intervention due to the design of the experiment.

Results

The initial search yielded 2,613 items. After removing duplicates, 1,961 items underwent initial screening. From the abstract screening, 22 articles were identified as meeting inclusion criteria, and 2 additional articles were identified by the third author by searching the reference lists of the identified articles. One article contained two experiments (Perrin & Neef, 2012); thus, 24 articles with 25 experiments were included in the review (see Fig. 1). To organize the discussion of results, a distinction was drawn between SCT experiments whose procedures used positive reinforcement (21 experiments) and those that used negative reinforcement (four experiments).

Fig. 1.

Search results. Note. Figure 1 displays a flowchart of the number of studies identified in each step of the search

Article Characteristics

Objective 1 of this review was to gather and organize literature to which researchers and practitioners can refer. Table 1 displays a summary of article characteristics, including the distribution of publications across journals, the number and percentage of studies published in 5-year intervals between 1988 and 2021 (June), and the number of participants included in each experiment.

Table 1.

Article characteristics

	n	%
Journal
Journal of Applied Behavior Analysis	20	83.33%
Journal of Experimental Analysis of Behavior	1	4.17%
European Journal of Behavior Analysis	1	4.17%
Journal of Developmental and Physical Disabilities	1	4.17%
The Behavior Analyst Today	1	4.17%
Year Published
1988–1991*	1	4.17%
1992–1996	0	0%
1997–2001	6	25%
2002–2006	5	20.83%
2007–2011	3	12.50%
2012–2016	6	25%
2017–2021	3	12.50%
Number of Participants**
1	3	12%
2	3	12%
3	13	52%
4	2	8%
5	1	4%
6 or more	3	12%

Table 1 displays the distribution of article characteristics for all reviewed articles, including the number and percentage of articles published in common behavior analytic journals, range of year published, and number of participants in each experiment

*The first range of publication years is a 3-year range because SCT research started with Schweitzer and Sulzer-Azaroff in 1988. All subsequent publication ranges are 5-year blocks

**The number of participants is calculated out of 25 included experiments instead of 24 included studies. Perrin and Neef (2012) included three participants in Experiment 1 and four participants in Experiment 2

Journals

Twenty included articles (83.33%) were published in the Journal of Applied Behavior Analysis. One of each of the remaining included articles was published in the Journal of Experimental Analysis of Behavior, European Journal of Behavior Analysis, Journal of Developmental and Physical Disabilities, and The Behavior Analyst Today (see Table 1).

Participant and Setting Characteristics

Participant ages ranged from 3 to 50 years (see Table 2 for summary data and Appendix A in the supplemental materials for a participant-by-participant table of participant characteristics). Of the 79 total participants across the 25 included experiments, 55 participants were children (69.62%), and 24 were adults (30.38%). The most prevalent participant diagnoses across participants were autism spectrum disorder or “developmental disability” (22 participants; 27.85%) or autism spectrum disorder and intellectual disability (4 participants; 5.06%), intellectual disability (17 participants; 21.52%), ADHD (10 participants; 12.66%), and traumatic brain injury (7 participants; 8.86%). Fourteen participants (17.72%) were labeled “typically developing” or were reported as having no diagnoses and were referred for behavioral concerns such as hyperactivity or impulsivity. Forty-five participants (56.96%) were male and 20 (25.31%) were female. Gender was not reported for 14 (17.72%) participants.

Table 2.

Participant and setting characteristics

	n	%
Age
Children	55	69.62%
Adults	24	30.38%
Diagnosis
ASD/DD	22	27.85%
ASD+ID	4	5.06%
ID	17	21.52%
ADHD	10	12.66%
TBI	7	8.86%
TD/None	14	17.72%
Unavailable	1	1.27%
Other	4	5.06%
Gender
Male	45	56.96%
Female	20	25.32%
Not Reported	14	17.72%
Race/Ethnicity
Black/African American	6	7.59%
Not Reported	73	92.41%
Setting
School	39	49.37%
Clinic	20	25.32%
Home	14	17.72%
Work	3	3.80%
Not Reported	3	3.80%

Table 2 displays participant age, diagnosis, reported gender, and reported race/ethnicity and the setting in which SCT took place. ASD/DD = autism spectrum disorder or developmental disability, ID = intellectual disability, ADHD = attention deficit hyperactivity disorder, TBI = traumatic brain injury and TD = typically developing. The category of other includes “emotional disturbance” and the educational designation of “other health impairment.” In some cases, additional diagnoses were reported by the authors in addition to the scored categories (e.g., traumatic brain injury and schizophrenia, ASD and seizure disorder)

The most common setting in which SCT occurred was school (39 participants; 49.37%), followed by clinic (20 participants; 25.32%), home (14 participants; 17.72%), and work (3 participants; 3.80%); see Table 2. The setting could not be determined for three participants (3.80%); in these cases, the authors reported that sessions occurred either at school or home, but the settings were not reported on an individual-participant basis (Gokey et al., 2013).

The reported race and ethnicity for 6 of 79 participants (7.59%) was reported as Black or African American. Race and ethnicity were not reported for the remaining 73 of 79 participants (92.41%). Socioeconomic status was not reported for any participants.

Experiments Using Positive Reinforcement

In the experiments that employed positive reinforcement, reinforcers were presented contingent on choosing the self-control option and completing the required delay or more immediately following the selection of the impulsive-choice option. This type of procedure allows for the direct manipulation of reinforcer parameters: the reinforcer that follows a selection of the self-control alternative can be made larger, more valuable, or accessible for a longer time compared to the reinforcer associated with the impulsive option. In contrast, in procedures that employ negative reinforcement (e.g., escape from a task), it is difficult to directly manipulate reinforcer parameters, such as magnitude. Instead, they are often indirectly manipulated through differentiating task parameters, such as difficulty level. Such examples of procedures employing negative reinforcement are discussed below, in the section labeled Experiments Using Negative Reinforcers. The present section will discuss the 21 experiments that employed choices between positive reinforcers.

Procedural Characteristics

Objective 2 of this article was to summarize and integrate procedural approaches across the reviewed literature. Table 3 displays a summary of procedural characteristics of the included studies, including preassessments, dependent variables, postassessments (i.e., maintenance or generalization), and activity type (for the 17 experiments that used intervening activities during the delay to reinforcement).

Table 3.

Procedural characteristics scored by experiment

	SR+ (N = 21)		SR- (N = 4)
Preassessments
Waiting Assessment	6	28.57%	0	0%
Naturalistic Baseline	8	38.10%	0	0%
Choice Baseline	19	90.48%	4	100%
Magnitude Assessment	9	42.86%	2	50%
Maintenance and Generalization Assessments
Maintenance	2	9.52%	0	0%
Generalization	3	14.29%	0	0%
Dependent Variables	-
Response allocation	21	100.0%	4	100%
Task engagement	1	4.76%	0	0%
Challenging behavior	2	9.52%	0	0%
Terminal delay	12	57.14%	0	0%
Intervening Activities (n = 17)
Academic task	7	41.18%	-	-
Motor task	7	41.18%	-	-
Toy interaction	2	11.76%	-	-
Mediating verbal behavior	1	5.88%	-	-

Table 3 is a summary of the number of included experiments that conducted each type of preassessment and postassessment, measured each category of dependent variables, and included each type of intervening activity (if applicable). Note that intervening activity percentages were calculated out of 17, which represent the 17 out of 21 positive-reinforcement experiments that implemented an intervening activity as part of SCT

Preassessments

Various assessments of self-control and impulsive choice were conducted as measures of baseline or pretraining. Eight experiments (38.10%) included a naturalistic baseline during which participants were instructed to engage in a task for as long as they could without programmed reinforcement for task engagement. The dependent variable for naturalistic assessments was duration of task engagement, usually reported in seconds. Six experiments (28.57%) included a waiting assessment, during which participants were instructed to wait as long as they could before accessing reinforcement, and the dependent variable was duration of wait time in seconds. Nineteen experiments (90.48%) used a choice baseline, in which participants were asked to select between a smaller or lower-quality reinforcer delivered immediately (i.e., impulsive choice) or a larger or higher-quality reinforcer delivered after a delay (i.e., self-control choice), and the dependent variable was response allocation across choice options. Finally, nine (42.86%) experiments employed a magnitude assessment that involved the participant choosing between a smaller or a larger reinforcer, both delivered immediately, with response allocation across the options as the dependent variable. The purpose of magnitude assessments was to determine whether choice was sensitive to magnitude manipulations by confirming that baseline choice was allocated more often toward the larger reinforcer when there was no delay to either reinforcer delivery. In one case, the preassessments were used to identify the task parameters that were subsequently arranged on an individual-participant basis for the SCT intervention (Neef et al., 2001). Neef et al. (2001) conducted their initial assessments to determine the parameters of reinforcement to which participants’ response allocation was most sensitive. The reinforcer dimension was then selected for each participant based on individual assessment results.

Reinforcer Dimensions

There were 67 participants in the included experiments that used positive reinforcement. One participant did not experience SCT as an intervention (Schweitzer & Sulzer-Azaroff, 1988); thus, the following percentages were calculated out of 66 total participants. See Table 4 for a summary of reinforcer dimensions across participants. Delay to reinforcement was manipulated for all. Reinforcer magnitude was manipulated for 45 participants (68.18%); duration was the next most common, manipulated for 13 participants (19.70%); quality for 3 participants (4.55%); and rate for 1 participant (1.52%). Reinforcer dimension could not be ascertained for three participants (4.55%): all three were exposed to the manipulation of either magnitude or duration, but dimension was not reported on an individual-participant basis (Dunkel-Jackson et al., 2016). One participant (1.52%) was exposed to a magnitude manipulation with a food reinforcer during some sessions and the manipulation of the duration of access to television in others (Vollmer et al., 1999).

Table 4.

Task characteristics

	SR+ (N = 66)		SR- (N = 9)
Reinforcer Dimensions
Magnitude	45	68.18%	-	-
Duration	13	19.70%	-	-
Quality	3	4.55%	-	-
Rate	1	1.52%	-	-
Undetermined	3	4.55%	-	-
Magnitude and Duration*	1	1.52%	-	-
Task Arrangement
Magnitude	-	-	4	44.44%
Difficulty	-	-	1	11.11%
Aversiveness	-	-	3	33.33%
Magnitude + Difficulty**	-	-	1	11.11%

The top portion of Table 4 displays the percentage of participants who were exposed to manipulations of each reinforcer dimension for SCT experiments that arranged positive reinforcement. The bottom portion displays the percentage of participants who were exposed to manipulations of magnitude, difficulty, and aversiveness of tasks for experiments that implemented negative reinforcement

*One participant (“Dale”) was exposed to a magnitude manipulation with a food reinforcer during some sessions and the manipulation of the duration of access to television in others

**In one experiment, Perrin and Neef (2012) combined manipulations of magnitude and difficulty for one participant

SCT Conditions

Progressive delay was scored when the delay to the self-control option was systematically increased across trials following a predetermined rule. Intervening activity was scored when participants had the opportunity to engage in an additional activity that was made available or instructed during the delay to the reinforcer in the self-control choice option. Signaled delay was scored when an extra stimulus (i.e., not the original instruction) was presented specifically to signal the presence or duration of the delay to the reinforcer following a self-control choice. Antecedent rule was scored when the experimenter stated a vocal or written rule about the contingencies before the intervention began.

Table 5 depicts the number of applications coded for each SCT condition across experiments using positive reinforcement. An application was defined as the number of times an SCT condition was implemented.2 For studies that implemented positive-reinforcement procedures, there were a total of 66 participants across experiments and 87 total applications (i.e., some participants were exposed to more than one SCT condition). The most prevalent training arrangement was an SCT condition including progressive delay + intervening activity as a training package with 37 applications (42.53%), followed by progressive delay alone with 20 applications (22.99%), and intervening activity alone with 11 applications (12.64%).

Table 5.

SCT condition applications for procedures using positive reinforcers

	PD	IA	SD	PD + IA	PD + AR	PD + SD	Applications (participants)
Benedick & Dixon (2009)	3	-	-	-	3	-	6 (3)
Binder et al. (2000)	-	-	-	3	-	-	3 (3)
Bloh (2010)	1	-	-	2	-	-	3 (3)
Dixon et al. (1998)	-	-	-	3	-	-	3 (3)
Dixon & Holcomb (2000)	-	-	-	6	-	-	6 (6)
Dixon & Cummings (2001)	-	-	-	3	-	-	3 (3)
Dixon and Falcomata (2004)	-	-	-	1	-	-	1 (1)
Dixon, Horner, Guercio (2003a)	-	-	-	1	-	-	1 (1)
Dixon & Tibbetts (2009)	-	-	-	3	-	-	3 (3)
Dixon et al., (2003b)	3	-	-	3	-	-	6 (3)
Dunkel-Jackson et al. (2016)	-	-	-	3	-	-	3 (3)
Falcomata & Dixon (2004)	-	-	-	2	-	-	2 (2)
Gokey et al. (2013)	-	-	-	3	-	-	3 (3)
Juanico et al. (2016)	-	5	-	-	-	-	5 (5)
Neef et al. (2001)	-	3	-	-	-	-	3 (3)
Newquist et al. (2012)	-	3	3	-	-	-	6 (3)
Passage et al. (2012)	-	-	-	1	-	-	1 (1)
Schweitzer & Sulzer-Azaroff (1988)	5	-	-	-	-	-	5 (5*)
Staubitz et al. (2020)	6	-	-	3	6	1	16 (6)
Vessells et al. (2018)	2	-	-	-	-	4	6 (4)
Vollmer et al. (1999)	-	-	1	-	-	1	2 (2)
Total applications	20	11	4	37	9	6	87 (66)
% of applications	22.99	12.64	4.60	42.53	10.34	6.90

The number of applications of each SCT component or package in each experiment with positive reinforcement. Some participants experienced more than one SCT condition. Thus, the total number of applications (in bold) may exceed the total number of participants (in italic) as displayed in the right-most column. The percentage of total applications for each SCT condition is displayed in the bottom row. PD = progressive delay; IA = intervening activity; SD = signaled delay; AR = antecedent rule

*There were six participants in Schweitzer & Sulzer-Azaroff (1988), but only five participated in SCT

We also identified the type of intervening activity (academic tasks, motor tasks, toy interaction, or mediating verbal behavior) when applicable (See Table 3). Of the 17 experiments that arranged positive reinforcement and included an intervening activity, 7 experiments (41.18%) presented academic tasks, 7 experiments (41.18%) presented motor tasks, 1 experiment (5.88%) instructed participants to engage in mediating verbal behavior, and 2 experiments (11.76%) were toy interaction.

Dependent Variables

Dependent variables were coded as response allocation, terminal delay, task engagement, and challenging behavior (see Table 3). All studies measured response allocation as a dependent variable, which was a requirement for inclusion in this review. Response allocation was defined as the distribution of choice across experimenter-arranged self-control choices and impulsive choices, usually reported as percentages or proportions. Twelve experiments (57.14%) included terminal delay as an additional dependent variable, which was scored when experimenters measured the duration of the achieved delay to reinforcement following the self-control choice. Terminal delay included total time engaged in an intervening activity, if applicable (e.g., an intervening activity was present during the delay and the length of the delay was the primary dependent measure). One experiment (4.76%) reported on task engagement, defined as experimenter measurement of task completion, on-task behavior, or accuracy of task completion (e.g., number of math problems completed, proportional duration of on-task behavior, or proportional correctness of completed math problems) during an intervening activity. Two experiments (9.52%) measured challenging behavior as a dependent variable when experimenters defined and measured behavioral excesses targeted for reduction (e.g., aggression).

Outcomes of SCT with Positive Reinforcement

Proximal Outcomes by Dependent Variable

Objective 3 of this article was to summarize and integrate outcomes across the reviewed literature. Each SCT condition was assessed in terms of its success in altering the four dependent variables described above. Figure 2 displays successful and unsuccessful applications of SCT conditions on the dependent variables. The results of the SCT conditions are reported in the following sections, organized by dependent variable (see Appendix B in Supplemental Materials for results by participant).

Fig. 2.

Applications of SCT conditions in experiments using positive reinforcement. Note. Figure 2 displays successful and unsuccessful applications of SCT components on each of four dependent variables (i.e., response allocation, terminal delay, task engagement, challenging behavior) for all experiments that used positive reinforcement. Data coders used visual inspection and recorded whether there was a demonstrated change in the DV and coded each as successful or unsuccessful. An application was defined as an SCT component (or package) evaluated with a participant. In some studies, participants experienced more than one component/package, and each was scored as an application. PD = progressive delay; IA = intervening activity; SD = signaled delay; AR = antecedent rule

Response Allocation

Progressive delay + intervening activity was successful at shifting response allocation toward the self-control option for 35 of 37 total applications (94.59%; see Fig. 2, upper-left panel, PD+IA; Binder et al., 2000; Bloh, 2010; Dixon et al., 1998; Dixon & Holcomb, 2000; Dixon & Cummings, 2001; Dixon & Falcomata, 2004; Dixon et al., 2003a; Dixon & Tibbetts, 2009; Dixon et al., 2003b; Dunkel-Jackson et al., 2016; Falcomata & Dixon, 2004; Gokey et al., 2013; Passage et al., 2012; Staubitz et al., 2020). However, intervening activity alone (without progressive delay) appeared likewise effective; all 11 applications (100%) of intervening activity alone resulted in a shift toward the self-control option (Fig. 2, upper-left panel, IA; Juanico et al., 2016; Neef et al., 2001; Newquist et al., 2012). Within these studies, there were no apparent differences in effectiveness across the various types of intervening activities.

Progressive delay alone was less effective, with successful shifts in response allocation in only 12 of 20 applications (60%; Fig. 2, upper-left panel, PD; Benedick & Dixon, 2009; Bloh, 2010; Schweitzer & Sulzer-Azaroff, 1998; Staubitz et al., 2020; Vessells et al., 2018). The combination of progressive delay and antecedent rule did not drastically improve the effectiveness of progressive delay, resulting in successful changes to response allocation in six of nine applications (66.67%; Fig. 2, upper-left panel, PD+AR; Benedick & Dixon, 2009; Staubitz et al., 2020). In contrast, adding a signaled delay appeared to make progressive delay more effective, with five of six applications (83.33%) resulting in increased choice for the self-control option relative to baseline (Fig. 2, upper-left panel, PD+SD; Staubitz et al., 2020; Vessells et al., 2018; Vollmer et al., 1999).

Signaling the delay to reinforcement was the least effective intervention for shifting response allocation, resulting in increased choice for the self-control option in only one of four applications (25%) (see Fig. 2, upper-left panel, SD; Newquist et al., 2012; Vollmer et al., 1999).

Terminal Delay

Combined progressive delay + intervening activity successfully increased terminal delay in 28 of 31 applications (90.32%; Fig. 2, upper-right panel, PD+IA, Binder et al., 2000; Bloh, 2010; Dixon et al., 1998; Dixon & Holcomb, 2000; Dixon & Cummings, 2001; Dixon et al., 2003a; Dixon & Falcomata, 2004; Dixon & Tibbetts, 2009; Dixon et al., 2003b; Dunkel-Jackson et al., 2016; Falcomata & Dixon, 2004; Gokey et al., 2013, Passage et al., 2012; Staubitz et al., 2020), whereas progressive delay alone was successful in 4 of 10 applications (40%; Fig. 2, upper-right panel, PD; Benedick & Dixon, 2009; Bloh, 2010; Dixon et al., 2003b; Schweitzer & Sulzer-Azaroff, 1988; Staubitz et al., 2020; Vessells et al., 2018). SCT with progressive delay + signaled delay was evaluated with one participant and was found to be unsuccessful in increasing terminal delay (Staubitz et al., 2020; Fig. 2, upper-right panel, PD+SD); however, this result should be interpreted with caution due to the small sample size and lack of replication across subjects. There were six applications of progressive delay + antecedent rule, two of which successfully increased terminal delay (33.33%; Fig. 2, upper-right panel, PD+AR, Benedick & Dixon, 2009; Staubitz et al., 2020).

Task Engagement and Challenging Behavior

Task engagement was measured for five applications of intervening activity and was found to increase self-control choice in 100% of applications (Juanico et al., 2016; Fig. 2, lower-left panel, IA). Two studies measured challenging behavior for six total participants (Dixon & Cummings, 2001; Gokey et al., 2013). Challenging behavior decreased for three of six participants (50%; Fig. 2, lower-right panel). All three participants in the study by Dixon and Cummings (2001) displayed a decrease in challenging behavior (i.e., self-injury, aggression, and slamming on the floor) when SCT included an intervening activity (relative to baseline and SCT conditions without an intervening activity). Results from Gokey et al. (2013), who measured negative vocalizations, suggest that intervening activity had no effect on challenging behavior; it should be noted that challenging behavior was minimal for all three participants at baseline and during all subsequent experimental phases (i.e., challenging behavior was already low in baseline). One methodological difference between these two studies is that Gokey et al. (2013) withheld the reinforcer associated with the self-control choice if negative vocalizations occurred during the delay. As an alternative, Dixon and Cummings (2001) delivered the self-control choice independent of challenging behavior. Participants were exposed to two conditions to compare levels of aggression and self-injury in conditions with and without an intervening activity. Together, these results might suggest that intervening activities may reduce challenging behavior that is already occurring and may be unlikely to induce or increase challenging behavior from low levels.

Terminal Outcomes

Each experiment was scored for the inclusion of maintenance and generalization assessments (Table 3). If a maintenance assessment was conducted on any dependent variable, the scorer recorded whether results were maintained and at what time point the maintenance data were collected. If a generalization assessment was conducted, the scorer recorded the parameters across which generalization was assessed and whether or not self-control choice generalized.

Maintenance

Maintenance was assessed in only 2 of 21 experiments with positive reinforcement (9.52%); both reported on maintenance of changes in response allocation (Benedick & Dixon, 2009; Passage et al., 2012) and one also reporting maintenance of achieved terminal delay (Passage et al., 2012). Benedick and Dixon (2009) shifted response allocation toward the self-control option for all three participants using progressive-delay training combined with an antecedent rule. They conducted a 6-month follow-up test to evaluate the maintenance of self-control choice with an antecedent rule and found that increases in self-control choice maintained for all three participants. In addition, Passage et al. (2012) successfully increased response allocation toward the self-control option at the terminal-delay goal of 420 s using a combination of progressive-delay training and an intervening activity (completing workbook pages). A maintenance probe conducted 41 days after treatment cessation using procedures identical to the choice baseline revealed that self-control choice maintained at the terminal-delay goal for the single participant in the study.

Generalization

Generalization of self-control responses was assessed in 3 of the 21 experiments that arranged positive reinforcement (14.29%). Passage et al. (2012) used progressive delays combined with an intervening activity (a matching task) to increase choice for the self-control option at increasing delays for one participant, which was found to generalize across activities when the intervening activity was changed to a novel sorting task.

Bloh (2010) increased response allocation toward the self-control choice at terminal-delay goals for three participants, using progressive-delay training for one participant and a combination of progressive-delay training and an intervening activity (completing a matching task) for two participants. To assess generalization across reinforcers and settings, participants were offered choices between novel reinforcers (a shorter duration of gameplay immediately or a longer duration of gameplay after the terminal delay) in a setting frequented by the participants but not used during training. Choice for the self-control option at the terminal delay was found to generalize across reinforcers and settings for all three participants.

In a similar procedure, Dunkel-Jackson et al. (2016) successfully increased choice for the self-control option at the terminal-delay goal for all three participants. The experimenters conducted probe trials in which the procedure was identical to the naturalistic baseline but with a novel intervening activity during the delay and found that choice for the self-control option and successful completion of the terminal-delay duration generalized to the novel task.

Experiments Using Negative Reinforcement

Procedural Characteristics

Four experiments from three publications implemented SCT with negative reinforcement, in which participants chose between completing tasks that varied in delay and one other parameter, with escape from a task as the negative reinforcer (Perrin & Neef, 2012; Porter & Sy, 2020; Lerman et al., 2006). This experimental arrangement results in a reversal of the choice options: the more immediate option becomes the self-control choice (e.g., completing a smaller, less difficult, or less aversive task now) and the delayed option becomes the impulsive choice (e.g., completing a larger, more difficult, or more aversive task later). Because the reinforcer earned was typically escape from the task, the value of the reinforcer (i.e., escape) could not be manipulated by increasing the duration of access to it, as trial lengths were usually held constant. Instead, the reinforcer value was indirectly manipulated by altering task parameters in addition to the delay: magnitude (e.g., number of math problems that had to be completed); difficulty or effort, associated with a slower response rate and a higher rate of errors relative to a different version of the same task; and aversiveness, typically determined through baseline assessment of choice between different types of tasks (i.e., the tasks chosen least and/or last were considered most aversive). The outcomes from the reviewed SCT literature using negative reinforcement contingencies are summarized and integrated in the following sections (Objective 2).

Preassessments

None of the four experiments (0%) with negative reinforcement included a waiting assessment or a naturalistic assessment (see Table 3). All four experiments (100%) conducted a choice baseline. Two of the four experiments (50%) conducted a magnitude assessment. Porter and Sy (2020) also conducted an aversiveness verification assessment to identify high- and low-aversive tasks.

Task Arrangement

There were 13 applications of SCT with negative reinforcement across nine participants.3 Tasks were arranged to differentiate the relative value of escape between impulsive and self-control choices (see Table 4). The magnitude of the required task was manipulated for four participants (44.44%) and difficulty for one participant (11.11%). Experimenters varied task aversiveness for three participants (33.33%; Porter & Sy, 2020). In one experiment, Perrin and Neef (2012) combined manipulations of magnitude and difficulty for one participant (11.11%).

SCT Conditions

We used the same independent variables to categorize SCT conditions in experiments with positive and negative reinforcement, with the exception of progressive delay and commitment response. For experiments using negative reinforcement, we used the broader term altered delay to describe an SCT condition that varied delays to reinforcement. Altered delay referred to systematic changes in the delay to reinforcement but was not limited to progressively increasing delays. All experiments altered the delays to both response options to manipulate the relative value of escape from the tasks but did not always do so in a progressively increasing fashion. For example, Perrin and Neef (2012) increased or decreased the delay to both tasks by 10–30 s based on each participant’s responding throughout the phases. We also scored other task manipulations beyond delay that were part of the independent variable (e.g., altered delay + manipulating task aversiveness as part of the training component; Porter & Sy, 2020).

Commitment response as an independent variable was only seen in experiments that used negative reinforcement. When a commitment response was offered, the participant had the opportunity to commit to the self-control choice in advance of the choice point. Selection of the commitment response rendered the impulsive-choice option unavailable. If the participant did not opt in when the commitment response was available, the choice opportunity occurred as usual. See Table 6 for a summary of applications of SCT conditions across each experiment that arranged negative reinforcement.

Table 6.

SCT condition and task arrangement applications for procedures using negative reinforcement

	Altered Delay + Magnitude	Altered Delay + Difficulty	Altered Delay + Difficulty + Magnitude	Altered Delay + Aversiveness	Altered Delay + Magnitude + CR	Altered Delay + Difficulty + CR	Applications (participants)
Perrin & Neef Exp 1 (2012)	2	2	1	-	-	-	5 (4)
Perrin & Neef Exp 2 (2012)	-	-	-	-	1	2	3 (3*)
Porter & Sy (2020)	-	-	-	3	-	-	3 (3)
Lerman et al. (2006)	2	-	-	-	-	-	2 (2)
Total applications	4	2	1	3	1	2	13 (9*)
% of applications	33.33	16.67	8.33	25.00	8.33	16.67

Table 6 shows the number of applications of each SCT condition + task arrangement manipulation for experiments that employed negative reinforcement. Some participants experienced more than one SCT component + task arrangement combination. The total number of applications of SCT for each experiment is in bold and the total number of participants in italic in the right-most column. The percentage of total applications of each SCT condition and task arrangement combination is displayed in the bottom row

*The three participants in Experiment 2 of Perrin & Neef (2012) also participated in Experiment 1

Outcomes of SCT with Negative Reinforcement by Independent Variable

For all experiments that used negative reinforcement, response allocation was the only dependent variable measured. All four experiments implemented some form of altered delay. There were 13 total applications of SCT across four experiments (Lerman et al., 2006; Perrin & Neef, 2012, Exp. 1 & 2; Porter & Sy, 2020). There were 10 applications of altered delay plus manipulation of a task parameter, and three applications of altered delay + commitment response plus manipulation of a task parameter (Fig. 3; see Appendix C in Supplemental Materials for results by participant). To achieve Objective 3 of the current review, the outcomes across the reviewed studies with negative-reinforcement contingencies are discussed by independent variable in the following sections.

Fig. 3.

Applications of SCT conditions in experiments using negative reinforcement. Note. Figure 3 displays successful and unsuccessful applications of SCT conditions on response allocation to the self-control choice for all experiments that used negative reinforcement. Data coders used visual inspection and recorded whether there was a demonstrated change in the DV and coded each as successful or unsuccessful. An application was defined as an SCT component (or package) evaluated with a participant. In some studies, participants experienced more than one component/package, and each was scored as an application. AD = adjusted delay; mag = magnitude manipulation; diff = difficulty manipulation; CR = commitment response

Altered Delay

Altering the delay in addition to manipulating task magnitude was effective for four of four applications (100%; Fig. 3, AD+Mag). Altering the delay in addition to manipulating the difficulty of the task across choice options was effective for one of two applications (50%). For the participant for whom difficulty manipulations were ineffective, manipulating both difficulty and magnitude of the task shifted response allocation (Perrin & Neef, 2012, Experiment 1; Fig. 3, AD+Diff+Mag). Manipulating the aversiveness of the task in addition to the delay was effective for three of three participants (100%) in the study by Porter and Sy (2020; Fig. 3, AD+Aversiveness).

Commitment Response

Perrin and Neef (2012, Experiment 2) added a commitment response to the altered delay plus task parameter manipulation, which was effective for all three participants (100%). One participant was exposed to a magnitude manipulation (Fig. 3, AD+CR+Mag), and two participants were exposed to a difficulty manipulation (Fig. 3, AD+CR+Diff).

Terminal Outcomes

None of the four experiments that arranged negative reinforcement assessed maintenance or generalization (Table 3).

Discussion

The purpose of this review was to identify the components of SCT that shift impulsive choice to self-control choice with humans for whom increasing self-control choice is a socially significant goal. We identified and reviewed 25 experiments between 1988 and 2021 that conducted SCT with humans who engaged in impulsive choices during baseline. Twenty-one of the 25 experiments (84%) arranged SCT with positive-reinforcement contingencies, and 4 experiments (16%) arranged negative-reinforcement contingencies. Within the reviewed studies, there are examples of variations of SCT that employed progressively increasing delays, intervening activities, signaled delays, antecedent rules, and commitment responses, resulting in a shift in response allocation from impulsive to self-control choice across a variety of participant populations and settings. Although additional research is needed to refine best-practice recommendations for SCT, we offer some general considerations and recommendations for clinicians based on our synthesis of the reviewed studies.

Clinical Implications

Training Components

An SCT component that frequently resulted in increased self-control choice was providing an intervening activity during the delay in isolation or as part of a package. In isolation, intervening activities increased self-control choice (i.e., response allocation) in all cases for which it was used. When implemented as a package with progressive delay, intervening activity was effective for 94.59% of applications. It should be noted that progressive delay alone was effective for only 60% of applications. These findings suggest that the opportunity to engage in an intervening activity may be a stronger impetus for making self-control choices than exposure to progressive delays.

Based on the reviewed literature, intervening activity alone appears to be a good first-line approach when implementing SCT for the purpose of shifting response allocation. Intervening activity alone and intervening activity + progressive delay were similarly effective when response allocation was the dependent variable. In general, it may be less resource-intensive to provide only an intervening activity than it is to implement an intervening activity plus progressive-delay training. Thus, for the sake of practicality, we recommend clinicians consider starting with intervening activity alone.

That being said, when the goal is to increase the terminal delay to the delivery of the reinforcer associated with the self-control option (or to increase terminal delay and shift response allocation), we recommend clinicians consider implementing intervening activity + progressive delay as a package. Gradually exposing individuals to longer delays while including the intervening activity was most effective in the reviewed studies for increasing the duration of the terminal delay while also promoting self-control choice (e.g., Gokey et al., 2013; Falcomata & Dixon, 2004). Although one might expect work-related tasks (e.g., Gokey et al., 2013; Dunkel-Jackson et al., 2016) to be a less effective intervening activity than interaction with preferred leisure items (Newquist et al., 2012), both types of activities appeared to be similarly effective. Clinicians may then choose an intervening activity that is contextually appropriate. For example, if the clinical goal is to increase self-control choice and task engagement, practitioners may consider choosing an intervening activity of engaging with a goal-related task (e.g., academic tasks).

Humans are always behaving and will therefore engage in some behavior (covert or overt, instructed or self-selected) during delays. For example, when sitting in a waiting room at the doctor’s office, one is not “doing nothing.” Instead, one may reach for their phone, read a magazine, or think about plans for later. Descriptive research on delay discounting and self-control commonly includes observation of individuals engaging in other “distracting” activities during the wait period that may be facilitative. For example, in a series of experiments assessing delay tolerance with children using the “marshmallow test,” researchers reported that some children independently engaged in verbal mediation or an incompatible behavior (e.g., covering their eyes or averting their gaze, sitting on their hands, verbally stating the contingency; Mischel, 2014). Engaging in covert or overt activities during the delay is associated with increased self-control choice and tolerance of longer delays.

Variations of SCT may differ in their practicality for practitioners. Providing choices for intervening activities may further promote self-control choice and increase delays to more socially significant durations. In other areas of research, having the opportunity to choose an activity prior to a work period has been shown to increase task engagement (Tasky et al., 2008; Wilson et al., 2006). A few of the reviewed studies incorporated additional choice-making opportunities during SCT (Dixon & Falcomata, 2004; Passage et al., 2012; Dixon & Tibbetts, 2009).

Although SCT with an intervening activity was almost uniformly effective in the reviewed studies, there may be some practical limitations to including an intervening activity. For example, many of the studies included adult-led activities, such as providing physical prompts for motor tasks (Dixon et al., 2003b). When conducting SCT in naturalistic settings, it may be difficult to insert adult-led intervening activities in some settings (e.g., classroom), because frequent attention from the implementor is required to continue instructing new intervening activities and set delay values. It may be beneficial to teach individuals to independently select an activity that can be completed independently (e.g., Juanico et al., 2016; Pelios et al., 2003). One way to teach independent engagement in a series of activities is by teaching a child to use an activity schedule (e.g., Reinert et al., 2020), which can increase task engagement and independent initiation of, and transitions between, activities (see Koyama & Wang, 2011, for a review). Intervening activities may also be combined with demand fading to increase task engagement and work completion to target academic goals.

Preassessments

There was substantial variability in the types of preassessments used in the reviewed articles (e.g., naturalistic baselines, waiting assessments or choice baselines, magnitude assessments). In some cases, the purpose of the preassessment was to evaluate whether participants engaged in impulsive choice prior to the interventions or to establish baseline data. Another possible use for individualized preassessments is to determine sensitivity to various dimensions of reinforcement or response prior to conducting SCT. This approach may be more effective than arbitrarily selecting a reinforcer or response dimension and can help inform individualized SCT (e.g., Neef et al., 2001; Falcomata et al., 20104). We recommend conducting individualized assessments to confirm participant sensitivity, as participants may demonstrate individual differences in sensitivity to manipulations of reinforcement parameters. For example, some individuals may prefer a shorter duration of access to reinforcement such that the overall work session is shortened; likewise, there may be idiosyncratic preferences for distributed versus accumulated reinforcers (see Ward-Horner et al., 2017) or attention versus other reinforcers (Harper et al., 2021).

Future Directions

Objective 4 of this review was to suggest future research based on gaps identified in the existing SCT literature. To augment the reviewed SCT literature, we suggest five areas of future research.

SCT Conditions

Some SCT conditions occurred with enough frequency in the reviewed literature to yield preliminary conclusions regarding outcomes (e.g., intervening activities and intervening activities in combination with progressive delay). However, some of the SCT conditions in the literature only appeared within packages, which limits our analysis of their individual outcomes. For example, implementation of an antecedent rule only appeared in a package with progressive delay (Benedick & Dixon, 2009; Staubitz et al., 2020). In addition to understanding how SCT conditions affect self-control choice in isolation, future research should also explore combinations of SCT conditions that have not yet been evaluated in the literature. For example, intervening activity + signaled delay, antecedent rule + signaled delay, and any combination of three or more SCT conditions have not yet been evaluated in an SCT package. Some SCT conditions may be more or less likely to appear in package interventions due to relative ease of implementation or resources required. Future research could conduct component analyses to identify the active variables responsible for increases in self-control choice (Ward-Horner & Sturmey, 2010).

Challenging Behavior

Although only two studies included challenging behavior as a dependent variable (Dixon & Cummings, 2001; Gokey et al., 2013), results suggested that less challenging behavior occurred in the intervening activity condition (Dixon & Cummings, 2001), or the intervening activity condition did not evoke challenging behavior where there had previously been little or none (Gokey et al., 2013). When working with individuals who engage in challenging behavior during delays, it may be beneficial to include an intervening activity that is incompatible with challenging behavior. It is perhaps surprising that experiments that implemented negative reinforcement with aversive tasks did not directly measure challenging behavior during the SCT phases, despite reporting a history of challenging behavior or displaying challenging behavior during preassessments (Porter & Sy, 2020; Lerman et al., 2006). However, if challenging behavior is escape-maintained and increasing self-control choice is a socially significant goal, then practitioners may consider setting up SCT with negative reinforcement.

The existing literature on delay tolerance training (see Brown et al., 2021, for a comparative analysis) may provide additional strategies for increasing self-control choice when challenging behavior occurs during delays. Nonbehavior analytic research has also demonstrated that working in groups during delays may increase delay tolerance (Koomen et al., 2020). Only one of the studies included in this review implemented SCT with groups of participants (Dixon & Holcomb, 2000). As the authors discuss, implementing SCT in group settings may have important implications for increasing cooperative behavior and may be more appropriate for large-group settings such as group homes or classrooms. Future research may investigate the effects of group settings on individual self-control choice, as well as group contingencies and gamification, (e.g., incorporating SCT into the Good Behavior Game; Barrish et al., 1969; see Joslyn et al., 2020, for a practical guide on the Good Behavior Game).

Socially Significant Outcomes

Objective 5 of this review was to discuss the social significance of the outcomes from the SCT literature. Three critical components of evaluating the social significance of training procedures and outcomes include: (1) meaningful behavior change; (2) maintenance and generalization of treatment effects; and (3) the social validity and acceptability of the procedures. Many of the reviewed studies demonstrated effective applications of SCT in that one or more training components resulted in an increase in self-control responding. However, presumably one desirable aspect of SCT would be to produce self-control choices for individuals in naturalistic situations. Aside from the 24-h delay achieved by participants in Neef et al. (2001), the achieved delays following SCT across all of the other reviewed experiments were less than 1 h. The next-longest achieved delay was 51 min 42 s (Dixon & Tibbetts, 2009). Both within and outside of the settings used in the SCT literature, people encounter delays longer than 1 hr or 1 day, and there is little experimental evidence to suggest that increased tolerance to relatively short delays in specific situations is likely to generalize to longer delays in more naturalistic circumstances. For example, tolerating a 400-s delay with the aid of a programmed intervening activity and a nearby teacher or clinician likely does not translate to choice for and toleration of significant naturalistic delays of days, weeks, or months. Although increases from seconds in baseline to minutes after intervention are a starting point, future research should investigate methods for generalizing self-control choice to longer delays.

Another critical feature of social significance is the extent to which behavior change generalizes across settings, people, and environmental stimuli and maintains across time (Baer et al., 1968; Stokes & Baer, 1977). We found it somewhat discouraging that only 3 of the 25 reviewed experiments included an analysis of generalization (Bloh, 2010; Dunkel-Jackson et al., 2016; Passage et al., 2012). Generalization was not assessed in any of the reviewed SCT experiments that arranged negative-reinforcement contingencies. Additional research is needed to assess the generalizability of self-control choice-making with positive and negative-reinforcement contingencies.

Two of the 25 reviewed experiments conducted a formal maintenance assessment (Benedick & Dixon, 2009; Passage et al., 2012), and the maximum amount of time that was assessed and achieved among these experiments was 6 months (Benedick & Dixon, 2009). With only two formal maintenance assessments, the degree to which self-control choice maintains in human populations remains largely unknown. Future research should assess the maintenance of self-control choice more than 6 months after treatment cessation to determine the persistence of treatment outcomes over time.5 In addition, none of the reviewed experiments that arranged negative reinforcement contingencies conducted maintenance assessments. Maintenance assessments following SCT with negative reinforcement will be necessary to determine how likely it is that increases in self-control choice controlled by negative reinforcement will maintain. Further extensions of SCT research should also investigate relapse (e.g., resurgence, renewal, reinstatement) of impulsive choice and strategies to reduce the likelihood of relapse.

We did not identify any formal assessments of social validity in the reviewed studies. However, some practical considerations suggest SCT procedures may need to be adapted to increase their social acceptability. In the reviewed literature, intervening activities were frequently effective as SCT components. However, the activities used in these studies may not always be available or appropriate in naturalistic settings. For example, it may be appropriate for a child who attends a religious service with their family to engage in a quiet activity like coloring, but such an intervening activity is often inappropriate for an older child or an adult. One reviewed experiment (Gokey et al., 2013) used an intervening activity to achieve the terminal delay and then faded the activity while maintaining successful completion of the delay following a choice for the self-control option. These results suggest that an intervening activity can be used to initially increase self-control choice and successful waiting, and then be faded when the ultimate goal is for a participant to wait quietly during the delay with no intervening activity. However, additional evaluations would be required to determine the extent to which SCT procedures are acceptable to direct and indirect consumers. In particular, the social validity of interventions may partially depend on individual participant characteristics, such as age, gender, race/ethnicity, and socioeconomic status. Research and subsequent application should take into account the cultural needs and preferences of consumers. As with all research with human participants, future research should measure and report the aforementioned participant characteristics to inform culturally sensitive SCT and behavior-analytic practice in general (see Jones et al., 2020, for a discussion). In sum, researchers interested in SCT should strongly consider evaluating social validity (e.g., training acceptability, achieving meaningful wait times) and the generalization and maintenance of the training outcomes.

Furthermore, behavior analysis and related fields need to continue contending with the question of how interventions aimed at affecting client choices interface with values of self-determination and autonomy (Bannerman et al., 1990). Peterson et al. (2021) provide a behavioral conceptualization of self-determination in which they propose that building skills in self-control, choice, and self-management is critical for individuals to maximize their self-determination.6 Future researchers should consider how SCT can be implemented in a noncoercive manner such that opportunities for choice are opened rather than closed for clients and participants.

Combining Consequences

The choices humans make every day are not always controlled by positive reinforcement; rather, choices can be controlled by escape or avoidance. Four of the 25 (16%) reviewed experiments used SCT with negative reinforcement. Impulsive and self-control choices controlled by negative reinforcement are understudied compared to positive reinforcement in the current literature, and future studies that arrange SCT with negative reinforcement are warranted. In addition, future research could program positive and negative reinforcement into concurrently available choices, which may be more analogous to what an individual experiences in naturalistic environments. For example, an individual may have a choice between exercising and doing another activity, such as watching TV or attending a social event. By choosing to forgo exercise (i.e., impulsive choice), the immediate reinforcer is avoidance of a non-preferred activity. However, by choosing to engage in exercise (i.e., the self-control choice), the larger reinforcers include improved long-term physical and mental health outcomes, which are additive and delayed by years across the lifespan, if they occur at all.

Probabilistic Delayed Outcomes

Many of the choices we make are between reinforcers and punishers that differ in terms of delay (rather than two reinforcers delivered after different delays), and some of these choices are also gambles: sometimes there is no guarantee that the delayed consequence will come at all, be it reinforcing or punishing. Consider the example of smoking, which produces immediate and fleeting reinforcement (reinforcing physiological effects, escape from withdrawal symptoms, and social reinforcement, among others) but can result in probabilistic delayed punishers in terms of lung cancer and other smoking-associated negative health outcomes. One could argue that choosing to smoke is the impulsive option whereas refraining from smoking is the self-control option, as it produces a larger-later reinforcer in terms of avoidance of negative health outcomes. However, lung cancer is a probabilistic outcome of smoking, because only 10%–15% of smokers will be diagnosed with lung cancer (Burns, 2000; Crispo et al., 2004). SCT may be a first step in promoting self-control choice among clinical populations, but its utility has mostly been demonstrated in clinical and school settings with guaranteed outcomes that are often not reflective of real-world contingencies. Further research is warranted involving SCT with consequences that are both probabilistic and delayed.

Review Limitations

There were several limitations to the present review. First, we did not evaluate the experimental rigor of the reviewed research studies. All of the reviewed studies involved an experimental design, but we did not establish standards for inclusion or temper the weight of outcomes for studies based on the strength of those designs (such as characterizing the degree of control) or the inclusion of other elements related to quality, such as reporting treatment fidelity or interobserver agreement. A future review of this literature could establish stricter standards for inclusion and employ standardized evaluations of rigor for single-subject and group designs. Second, as a scoping review, our overall aim was to catalog the characteristics of the extant literature and identify gaps for future research. Thus, any patterns we highlight about the efficacy of different SCT methods are tentative in nature. If behavior analysts continue publishing research on SCT, researchers should consider employing quantitative meta-analysis techniques to lead to more concrete conclusions on the efficacy and effect sizes of various training techniques. Related to this, we did not complete a review of unpublished literature (i.e., “grey literature,” e.g., unpublished dissertations); thus, there may be findings not represented within this review. Future reviews should consider including evidence from grey literature to avoid conclusions due to publication bias (see Hopewell et al., 2007, for a discussion). Third, the replicability of our search and screening procedures is unknown. We encourage researchers conducting scoping reviews of behavioral analytic research to preregister their search and scoring protocols and take measures to increase the replicability of their procedures.

From the existing literature, there is emerging evidence of the utility of SCT. The reviewed studies demonstrate numerous examples of implementing SCT with outcomes including increases in self-control choice, improved tolerance of delays to reinforcement (up to 24 h), increases in task engagement, and decreases in challenging behavior. For clinicians, if the goal is simply to increase self-control choice in clinical and education settings at relatively short delays, there appear to be at least a few viable training components (i.e., intervening activities, progressive delay). However, it is likely the case that parents and caretakers hope to see general and overall increases in self-control choice, as impulsive choice is likely to occur in more than one particular setting, especially in clinical populations. Thus, the majority of our recommendations for future research are to increase the generalizability of SCT procedures to “real-life” situations: lengthening terminal delays, fading out intervening activities when appropriate, including probabilistic outcomes, and combining appetitive and aversive outcomes. Continued systematic research that arranges SCT procedures to mimic naturalistic settings will address gaps in the literature regarding the generalizability of SCT outcomes and will be critical for eventually distilling best-practice recommendations. Behavior analysts should continue exploring how SCT and other procedures aimed at affecting choice can be implemented in noncoercive ways and support the self-determination of our clients (Peterson et al., 2021) and avoid compromising individuals’ rights to make decisions for themselves, regardless of whether their selections would be labeled by others as “self-control” or “impulsive” (Bannerman et al., 1990).

Author Contributions

The first, second, and third authors contributed equally to the article search and abstract screening. The first author did the primary scoring for included experiments, drafted the first version of this article, and drafted the method and discussion sections of the current article. The second author scored interrater agreement, drafted the introduction, and drafted the discussion. The second, third, and fourth authors drafted the results section. All four authors contributed to editing and revising all sections of the article.

Funding

None

Data Availability

Data and materials are available upon reasonable request to the corresponding author.

Declarations

Ethical Approval and Consent to Participate

N/A

Consent for publication

N/A

Conflicts of Interest/Competing Interests

The authors declare no conflicts of interest.