Livert D, Rindskopf D, Saxe L and Stirratt M
Multivariate Behavioral Research, 36(2): 155-184, 2001
With the increasing prevalence of community-based health and social interventions, multilevel research models for analyzing the efficacy of these interventions are becoming more popular. This article discusses the appropriateness of multilevel modeling, focusing on method and study design rather than on study results. To elucidate their point, the authors use the example of Fighting Back, a program designed to test the feasibility of reducing substance abuse through coordination and expansion of community efforts. The authors examine general analytical difficulties inherent in evaluating community-based drug abuse prevention programs. One prominent example of these difficulties is the pairing of multiple controls with a treatment community. Without multiple, well-paired controls against which to analyze treatment communities, certain effects cannot be analyzed appropriately. The authors also discuss randomization, weighting and issues of external validity. As the authors conclude, study design is crucial. No analytical technique can remedy fundamentally flawed designs. This point is even more important when it comes to the ultimate evaluation issue of generalizing results: only in well designed, randomized studies is sound extrapolation possible.
Associated Grant ID: 037368