Association for Behavior Analysis International

Application of basic statistical measures (e.g., t-tests, ANOVA) to single-subject designs have been a source of conflict in Behavior Analysis because, in part, these tests aggregate behavioral variability across subjects and time, eliminating much of the data that behavior analysts find important. Multilevel modeling (MLM) is a statistical technique that addresses these concerns and is commonly used when data are naturally clustered (e.g., student clusters in classrooms, which are also clustered in various schools across a district). With MLM, the value of a statistical parameter for a specific case depends on the levels of the each cluster for that case. A single subject can serve as a cluster of data and, therefore, MLM can provide subject-by-subject predictions. In a single-subject or small-n design, statistical comparisons based on the IVs of interest are enhanced when the models have already accounted for intrasubject variability. In theoretical modeling of behavior, subject-by-subject model parameters can be obtained while simultaneously accounting for group-level patterns in the data. This tutorial will demonstrate using MLM to analyze experimental data from a single subject design and also to conduct subject level model fitting. The analyses will be conducted in R, a popular, free software package for statistical analyses.

Researchers, research-practitioners, students

Dr. DeHart received his B.A. and Ph.D. from Utah State University under the mentorship of Dr. Amy Odum. In July of 2017, he began his current position as a post-doctoral fellow with Dr. Warren Bickel at the Fralin Biomedical Research Institute at VTC. Dr. DeHart’s primary research interests include the behavioral economics of addiction and other health behaviors including cigarette smoking and obesity as well as the application of advanced statistical methods to behavioral data. His early research investigated novel methods of reducing impulsive choice using framing and financial education and his dissertation applied structural equation modeling to better understand the effects of delay length and outcome magnitude on delay discounting. His current research interests are twofold. First, he is interested in measuring the abuse liability of different risky products including tobacco cigarettes and e-cigarettes and how demand for those products can be changed using public-health narratives. Second, he is interested in understanding the relationship of delay discounting to various health behaviors. In this line, he has applied advanced statistical methods including structural equation modeling, machine learning algorithms, and mixed-effects modeling. Dr. DeHart’s work has been recognized by various popular media outlets including the Wall Street Journal and he currently serves on the editorial board for the Journal of the Experimental Analysis of Behavior.

Jonathan E. Friedel is a research psychologist in the Bioanalytics Branch at the National Institute for Occupational Safety and Health. As part of the Organizational and Behavioral Research Team, he works on several grant funded projects focused on worker safety in laboratory workers, distracted driving, and data analytics for organizations using behavior based safety. He is currently the primary investigator for a grant funded project designed to use behavioral economics to quantify the factors that affect safety-related decision making in small businesses. He obtained his PhD in experimental psychology from Utah State University where he focused on delay discounting and behavioral economics. He obtained a MS in Behavior Analysis from University of North Texas.