In developmental science, a common and essential goal of research is to understand the relations that exist between constructs of interest. These constructs are not always directly observable and, in some cases, refer to abstract or theoretical factors. In those instances, the constructs end up being inferred from the results of measures or self-reported assessments taken across one or more items. The goal of this dissertation was to demonstrate the use of regression modeling methods for analyzing the relation between latent variables composed of multi-item measures.
This dissertation examined regression methods for multi-item measures in three separate studies. The first was an empirical study using a latent variable modeling framework to examine the roles of playfulness, stress, and coping. The second study used the data from the first study to compare four different approaches for estimating regression relationships with multi-item measures. The third study used a statistical simulation to demonstrate how data characteristics (i.e., sample size, effect size, and reliability) differentially impact the accuracy of different methods. Together, this dissertation examined the practical and theoretical considerations in regression analysis using latent variables. The findings from this dissertation have implications for developmental scientists and provide empirically grounded guidance for the selection of regression estimation methods to use for assessing latent variable relations.