LAUSR

Evaluations of prevention programs, such as the PAX Good Behavior Game (PAX), often have multiple outcome variables (e.g., emotional, behavioral, and relationship problems). These are often reported for multiple time points (e.g., pre- and post-intervention) where data are multilevel (e.g., students nested in schools). In this paper, we present both variable-oriented and person-oriented statistical approaches, to evaluate an intervention program with multilevel, longitudinal multivariate outcomes. Using data from the Manitoba PAX Study, we show how these two approaches provide us with different information that can be complementary. Data analyses with the variable-oriented approach (multilevel linear regression model) provided us with overall PAX program effects for each outcome variable; the person-oriented approach (latent transition analysis) allowed us to explore the transition of multiple outcomes across multiple time points and how the intervention program affects this transition differently for students with different risk profiles. We also used both approaches to examine how gender and socio-economic status related to the program effects. The implications of these results and the use of both types of approaches for program evaluation are discussed.