Session: New Methods to Investigate Statistical Mediation: Time to Event Models and Methods to Improve Confidence Limit Estimation and Power (Society for Prevention Research 21st Annual Meeting)

 This symposium will cover several critical issues pertaining to statistical mediation and its applications to prevention research.  The use and research of mediation analysis has steadily increased over the years because of its utility to prevention research.  Identifying mediators of interventions allows researchers to develop more efficient and more successful interventions.  Three posters discuss the use of survival analysis in the context of mediation for prevention researchers. The first poster examines the power to test mediation using censored continuous and discrete time survival data.  The second introduces Discrete Time Survival Mediation (DTSM) and discusses the utility of this method in examining event occurrence in prevention research.  The third survival analysis poster provides an empirical example of DTSM using data from a study examining substance use among children.  The fourth poster examines the bias and stability of various mediation effect sizes and provides recommendations for prevention researchers on which effect size to use under what circumstances.  The fifth poster describes how confounder bias affects tests for mediation, describes several methods for assessing confounding in single mediator models, and applies the methods to prevention research data.  The sixth poster describes how statistical power is affected with the addition of a second mediator at various sample sizes describing how adding mediators is a method to increase the statistical power of a study.  The seventh poster reviews two methods for planned missingness for a mediator in single mediator models; the Auxiliary Variable Model and the Latent Indicator Model.  This poster discusses the bias of each method and discusses how the use of planned missingness in mediation studies can help extend resources for prevention researchers and increase statistical power to detect effects. The final poster describes a Monte Carlo method and computer program for mediated effect confidence interval estimation that is more accurate that existing normal theory methods.  Together these posters address a number of important issues related to mediation and provide valuable information for assisting prevention researchers seeking to understand how an intervention achieved effects on an outcome variable.