Growth curve models for zero-inflated count data: an application to sexual risk behavior
In: SAGE Research Methods. Cases
For this case study, we will focus on our quantitative analysis of a longitudinal dataset consisting of 598 emerging adult gay and bisexual men looking at how romantic relationship beliefs are associated with changes in condomless anal sex. Our aim in this case is to discuss the application of longitudinal modeling techniques for count data. Several considerations arise when analyzing longitudinal data for count-dependent variables. Typically, count-dependent variables are modeled using applications of Poisson regression. However, many times, behavioral outcomes exhibit greater variability than what would be predicted by the Poisson distribution (i.e., there are an excess number of "zeroes" in the data.) Thus, this case aims to describe the application of an applied generalized linear growth mixture model technique using Poisson regression to an analysis of sexual risk behavior among a cohort of young sexual minority men. We first discuss the background relating to sexual risk behavior. We then detail the process, challenges, and lessons learned in terms of performing zero-inflated Poisson growth models.