Prevention programs like the Drug Resistance Strategies (DRS) Intervention are designed to work by affecting intermediate variables like refusal confidence, positive expectancies, and substance use norms, which in turn reduce substance use. Often,
programs like DRS are cluster-randomized to classrooms, which allows for causal inference of program effects on outcomes. However, establishing support for the hypothesized mediating mechanisms is difficult because individuals cannot also be randomly assigned to levels of the mediators. Modern causal inference methods are being developed to improve identification of effective mediators (see, e.g., Coffman & Zhong, 2012; Pearl, 2001; Robins & Greenland, 1992). These methods, however, have been primarily applied when there is randomization at the individual level rather than the cluster level. When the program is cluster-randomized, this clustering must be taken into account when using causal inference methods like inverse propensity score weighting. This talk introduces steps for conducting causal mediation analysis for cluster-randomized trials by examining the effect of the DRS Intervention on alcohol use through alcohol refusal confidence, positive expectancies of alcohol use, and peer alcohol use norms.
- Coffman Presentation
