Estimating the effect of intervention compliance on long-term outcome trajectories: Application of the latent adherence growth curve model in a cluster-randomized trial of the good behavior game.

The effectiveness of universal, school-based interventions in preventing the emergence, maintenance, or escalation of behavioral difficulties in children and adolescents is supported by a growing body of research. Nevertheless, this understanding is somewhat limited to “what works” assessments, leaving aside fundamental questions on relevant underlying mechanisms that make an intervention effective under specific circumstances, for some groups, or even at certain times. Using data from the cluster randomized controlled trial of the Good Behavior Game in England, involving 3,084 children in 77 schools, we showcase the latent adherence growth curve model, a multilevel structural equation modeling approach that takes account simultaneously of intervention compliance, subgroup moderator effects, change over time, and higher-level data structures. This statistical framework allowed a more nuanced view of the effects of the Good Behavior Game on the long-term developmental trajectories of disruptive behavior among male and female students. Our approach enabled us to ascertain the effect of child- and school-level covariates on intervention compliance and change in disruptive behavior over time, as well as the effects of both compliance and noncompliance on outcomes. In doing so, we found evidence of positive effects (reduction of disruptive behavior) on female compliers, but also iatrogenic effects among male noncompliers (increased disruptive behavior). We discuss the implications of our approach and findings for further research and practice.