Detecting Selection Bias in Community Disseminations of Universal Family-Based Prevention Programs

The goals of the present study were to demonstrate a method for examining selection bias in large-scale implementations of community-based family skills programs, and to explore the nature of selection bias in one such implementation. We used evaluation data from a statewide dissemination of a popular substance abuse prevention program (N programs = 42; N youth = 294). The program’s evaluation measures were designed to match publicly available data on risk and protective factor scales collected in the state’s schools, which enabled us to construct a comparison sample of non-participants (N = 20,608). We then examined the risk status of adolescents in both groups to determine whether risk and protective factor scores were related to the probability of program participation. Participation was predicted by both demographics and risk and protective factor scores. Among families with younger adolescents, program attendance was associated with lower risk; among families with older adolescents, participation was associated with both higher risk (on parental management skills) and lower risk (on substance use). Selection effects must be identified and corrected for in order to calculate valid estimates of program benefits, but in community-based disseminations, the necessary supplemental comparison sample is difficult to obtain. The evaluation design and analytic approach described here can be used in program evaluations of real-world, “bottom-up” dissemination efforts to identify who attends a program, which in turn can help to inform recruitment strategies, to pinpoint possible selection influences on measured program outcomes, and to refine estimates of program costs and benefits.