The effect of population heterogeneity on statistical power in the design and evaluation of interventions

Abstract

Do interventions work and for whom? For this article, we examined the influence of population heterogeneity on power in designing and evaluating interventions. On the basis of Monte Carlo simulations in Study 1, we demonstrated that the power to detect the overall intervention effect is lower for a mixture of two subpopulations than for a homogeneous population with the same average effect size. We also examined how obtaining covariate information for the outcome and risk status (i.e., latent subclass membership) affects power. For Study 2, we moved from the simulation to assess power in the design and evaluation of an intervention for antisocial children. We illustrated the importance of considering population heterogeneity because interventions have different impacts on different ‘‘classes’’ of participants. We also illustrated that power increases by including covariates that are correlated with risk status. The results provide guidelines for the design and evaluation of interventions.