Paul (male, age 17) was diagnosed with autism spectrum disorder (ASD), bipolar disorder, and schizophrenia. Paul engaged in vocal conversation with staff and peers and was understood by familiar listeners. He resided in treatment centers since age 13 years from which he was most recently discharged due to high rates of aggressive behavior and frequent elopement. At the time of the study, Paul was enrolled in an alternative day-treatment program for students with an instructional individualized education plan who could not be adequately supported within their resident school district’s special education services. Results from a parent-completed Adaptive Behavior Assessment System (2nd ed.; Gray & Carter, 2013) indicated composite and subscale scores at or below the 0.1 percentile level, suggesting Paul would benefit from training and support with adaptive behavior across multiple skill areas. At the time of the study, Paul was also receiving language and cognitive supportive programming guided by the PEAK Relational Training System (e.g., PEAK-Equivalence; Dixon, 2015).

To meet this vocational goal while remaining consistent with his comprehensive programming, and to evaluate the use of EBI establish declarative knowledge about employment settings, we adapted a program from the PEAK Relational Training System Equivalence Module (PEAK-E; Dixon, 2015)—Program 10-F: Informational Resources. The adapted program includes references to specific stimuli (A, B, and C) that vary across each employment setting and provide a better fit for his vocational programming goals. The restated goal was:

When taught to respond to a photo of a sample employee (A) by selecting the correct option from an array of eight corresponding job titles (B) (A–B), and to match the sample job title (B) with the correct option from an array of eight textual description of the job responsibilities (C) (B–C), the participant is able to respond by selecting the textual description of the job responsibilities (C) when provided with an array of eight sample employee photos (A) (A–C). The participant is also able to respond by identifying the correct photo of the sample employee (A) when provided with an array of eight textual descriptions of job responsibilities (C) (C–A).

In addition to adapting the program stimuli to meet Paul’s vocational goals, we added the C–A test trial type that is necessary for demonstration of stimulus equivalence. It is not only important to identify the responsibilities of people on the jobsite (A–C) but also to identify people who can perform a specific responsibility on the jobsite (C–A), which is reflected in the adapted program. The A stimuli used in programming were unfamiliar to simulate learning new information at the initial conditions of employment.

Figure 1 shows the stimuli used across the three employment settings. We selected these sites as they were common vocational placements for students in the program. Each jobsite contained a total of eight equivalence classes, including 24 total stimuli associated (8 classes x 3 class members) with each setting and 32 target relations (8 classes x 4 relations). The intervention was presented via PowerPoint slideshows using a Dell Inspiron 3576 laptop computer with built-in webcam and internal microphone using the procedures described by Belisle et al. (2021a) for creating computerized training content. The webcam and microphone were used only for online sessions which occurred on three occasions due to COVID-19 precautions and dangerous road conditions. Otherwise, programming occurred directly at the school program in a room separate from other students to minimize distractions.

Stimuli Used in the Present Study across the Three Employment Settings, including the Person/Employee (A), the Job Title (B), and the Job Responsibility (C)

The two primary dependent variables were the percentage correct responding for trial types where correct responding was directly trained [matched photos and employee names (A) to corresponding job titles (B) (A–B) and job titles (B) to corresponding job responsibilities (C) (B–C)] and tested equivalence trial types [employees (A) to corresponding job responsibilities (C) (A–C), and job responsibilities (C) to corresponding employees (A) (C–A)]. Trials were delivered in eight-trial blocks corresponding to the eight relations contained in each block. Each stimulus–stimulus pair was presented on one single occasion within each block and the location of each stimulus within the block was randomized. Only a single relation type was targeted within a single block. All sessions were videotaped. We evaluated point-by-point agreement (Belisle et al., 2021a, b) by comparing the scores of two independent observers for each trial across 33% of all trial in the present study. The independent rater recorded scores for each trial from video recordings of the training sessions. The percent agreement was determined by dividing the number of agreements by the total number of trials, multiplied by 100. Arithmetic mean IOA was 90% (range: 87%–93%). We also evaluated implementation fidelity using a PEAK implementation fidelity checklist (Belisle et al., 2016; see supplemental material).

We used a nonconcurrent multiple baseline across employment settings to assess the efficacy of EBI in establishing the directly trained relations (A–B and B–C) and tested relations (A–C and C–A). Baseline sessions were introduced to determine Paul’s initial levels of correct responding. During the baseline phase, reinforcement was never provided for correct responses and prompting was never delivered contingent on incorrect responses. The responses were simply recorded, and the implementer progressed to the next trial. The total number of trial blocks in each baseline phase for each jobsite were staggered. In addition, stability was required in the baseline phase before progressing to the first training phase. Training for each jobsite occurred nonconcurrently, such that the second job was introduced after the first was mastered (all relation types), and the third job was introduced after the second was mastered (all relation types).

Training occurred in two successive training phases (A–B training followed by B–C training). The first training phase for each job site targeted A–B relations. The slideshow trial presentation was programmed such that when Paul selected the correct B stimulus from the array, he was immediately provided visual (i.e., computer screen presented a visual that said, “Correct!”) feedback and the implementer provided verbal praise. If a response was incorrect, the screen did not progress, no visual feedback was provided, and the implementer provided feedback that the answer was incorrect (e.g., “What’s your second choice?”). Paul was provided with the opportunity for three incorrect answers, after which the implementer provided the correct answer and asked Paul to repeat the correct answer prior to selecting the correct response that initiated the visual feedback sequence.

The skill was considered mastered once participants achieved three consecutive trial blocks with at least 90% mean accuracy and a stable or increasing trend. In some phases, additional trials were conducted after mastery when the research team had not yet met to collectively determine if mastery was achieved. Once the mastery criteria were met, Paul progressed to the B–C training phase. This phase was identical to the A–B training phase except that the stimulus presentation was consistent with the B–C trial type. Again, once the mastery criteria were met, Paul progressed to a testing phase that included the derived relations (A–B and B–C). Testing of these relations was identical to the baseline phase and trial block types alternated between A–C and C–A testing. We developed an experimental design integrity checklist (see supplemental materials) based on this basic arrangement. The mean fidelity of the experimental design was 94%, with a range of 91% to 100%. Notable exceptions were additional training trials after the mastery criterion for two phases (overtraining) and failure to meet the mastery criterion in one of the training phases (B–C hospital) before progressing to a new phase.

Consistent with the applied nature of this study, during the B–C training phase in the hospital setting, there was a 21-day break in training due to Paul’s illnesses, behavior, and transportation challenges. To compensate for this training break, we provided booster training sessions interspersed with test probes prior to the final testing phase.