Darin Erickson, PhD

The models we use tend to be complex, mostly to overcome data and study design constraints. First, correlation exists in almost all of our datasets, and I use a number of statistical models to accommodate this correlation including random intercept models, correlated residual models, random coefficient/growth curve models, general estimating equations, and time series, to name a few. Second, recent work has involved the use of tailored or SMART designs. These study designs complement traditional RCT designs and incorporate multiple randomizations to efficiently test factorial experiments. Finally, I employ person-centered mixture models to measure and examine unobserved heterogeneity. Latent class analysis allows us to measure constructs that are categorical in nature but not directly observable. Examples of constructs I have modeled using LCA include alcohol control policies (for states as well as college campuses), substance use, alcohol enforcement practices by law enforcement agencies, food shopping, physical activity, and use of sexually explicit media. I also use mixture techniques longitudinally, including using them to cluster longitudinal trajectories (e.g., tobacco and alcohol use during adolescence and young adulthood) and to examine movements between latent classes over time (i.e., latent transition analysis) and correlates of these movements.