Date of Award

Spring 4-2019

Degree Name

Doctor of Philosophy (PhD)



Second Advisor


Third Advisor



In behavioral science research, many outcomes of interest can be influenced by interpersonal relationships (Gyarmathy & Neaigus, 2007). To assess such outcomes, data can be collected using dyads. Each dyad has two elements, an actor, who responds to a stimulus and a partner, who can potentially influence the actor (Kenny, Kashy, & Cook, 2006). One popular model for analyzing dyadic data is the Actor Partner Interdependence model (APIM). In this study, we proposed a variable selection method applied to a probit Bayesian Hierarchical Generalized Linear Model (Bayesian HGLM) to fit the APIM to dyadic data. The proposed method used stochastic search technology to identify key predictors of the Bayesian HGLM for APIM. It included a component for selecting interactions; selecting only interactions with both main effects also included. The proposed method was evaluated in two different forms, with simulated data and with real data. When we evaluated the method using simulated data, we examined its performance on 5 different simulated scenarios with varying associated predictors and two different sample sizes: a large sample size (2000 dyads) and a small sample size. And when we evaluated the method using real data, we used baseline data from an evaluation of the program Its Your Game-Family (IYG F). The baseline data set had the complete information of 61 dyads. Across the 5 scenarios, the proposed variable selection method selected the correct variables over 85% of the simulated data sets in either sample size. And using the real data, the proposed variable selection method selected one construct out of 6 to be associated with the binary outcome. Thus, using the real data, we concluded that the construct of teenage Sex Communication Self-Efficacy Relational explains the outcome Sexual initiation, and the effects are equal across dyad members (teenager-parent). In conclusion, in this study, we implemented the first variable selection procedure specifically to analyze dyadic data, based on stochastic search technology. The selection procedure can be applied in any research study that involves dyadic data from the APIM model with a binary outcome and a set of continuous covariates.