Date of Award

8-2025

Degree Name

Doctor of Philosophy

Department

Psychology

First Advisor

Cynthia Pietras, Ph.D.

Second Advisor

Anthony DeFulio, Ph.D.

Third Advisor

Hugo Curiel, Ph.D.

Fourth Advisor

Ya Zhang, Ph.D.

Keywords

Behavior analysis, hierarchical linear models, single-case designs, statistics, visual analysis

Abstract

Behavior analysts often use single-case experimental designs to evaluate behavioral interventions. Data are typically analyzed through the visual analysis of graphs. Visual analysis has several limitations, however. Visual analysis has only moderate inter-rater reliability and small effects can be difficult to detect, especially if data are variable. Hierarchal Linear Models (HLMs) have been proposed as method for quantifying results of single-case designs. HLMs are a type of statistical model that can evaluate the effectiveness of treatment with nested data, whereby sessions are nested within treatment conditions. An advantage of HLMs over other statistical tests is that they do not require averaging across participants. Prior research found that HLMs could detect small effects in variable data and identified a relationship between a subject variable and behavioral outcomes. The goal of the present study was to expand on this research by comparing visual analysis and results of HLMs to single-case experimental design graphs (ABA reversal designs) when data varied systematically in effect size, trend, autocorrelation, and variability, and when overall effects from sets of graphs were evaluated. Graphs were computer generated so that programmed effect sizes were known. Board Certified Behavior Analysts (n = 18) served as participants and were asked to evaluate effect size, trend, and variability of graph sets containing one, four, and sixteen graphs. Participants’ ratings were compared to each other and to the results of an HLM. Reliability of visual analysis was moderate and comparable to prior research. HLMs had smaller overall error rates than visual analysis, especially when autocorrelation and trend were present. HLMs performed equally well to visual analysis when single graphs were presented, but better when multiple graphs were present. These findings suggest that HLMs may be useful aids to visual analysis of reversal-designs graphs when evaluations involve multiple graphs, and when data show significant trend or autocorrelation. Alternative aids to visual analysis are recommended when only a single graph is analyzed.

Access Setting

Dissertation-Open Access

Download

Share

COinS