Estimating Mediation Effects in Prevention Studies
Project InformationPrincipal Investigator: Matt Fritz
Funding Agency: NIH-NIDA
Award Date: July 30, 2015
Project URL: N/A
For more information please contact Matt Fritz at firstname.lastname@example.org.
While prevention interventions are designed to change specific behaviors, interventions rarely change these behaviors directly. Instead, interventions targets intermediate variables, such as the knowledge, attitudes, and intentions, which are related to the behaviors. Changing these intermediate variables, known as mediators, in turn causes changes in the behavior.
Mediation analysis is one statistical method used to investigate these mechanisms of change. This method improves interventions and tests theories by providing information about the mediating processes through which interventions achieve effects. As interventions become more sophisticated, it is essential that optimal statistical mediation methods continue to be developed and disseminated for prevention research designs.
The overall objective of this project is to develop and evaluate new methodologies for investigating how drug prevention programs achieve their effects. It also aims to apply these methodologies to answer substantive questions in prevention research and to disseminate this information to methodologists and substantive researchers.
The topics examined in this project address critical issues in the types of data collected and designs used in current and future prevention research. Each overlapping aim has a methodological goal and a substantive goal. The specific aims are:
1. To investigate the precision and accuracy of confidence limit estimation and significance testing of traditional mediation methods, as well as new advances in mediation methods using causal mediation and Bayesian approaches. The substantive aim will be to review the prior information available in prevention studies to increase precision and accuracy of mediated effect estimation and to investigate the veracity of assumptions required for causal mediation analysis.
2. To provide practical information regarding the influence of both measurement error and confounder bias on mediation analysis. The substantive aim is to develop solutions for the types of measurement error and confounder bias that could compromise mediation analysis in prevention research.
3. To develop and evaluate mediation methods for longitudinal data. The substantive aim will focus on the timing and types of longitudinal theoretical change in prevention research.
4. To develop methodologies to identify mediated effects across subgroups. The substantive aim will focus on identifying types of moderator effects and how moderation analysis could be improved using Bayesian and causal mediation perspectives.
5. To apply mediation methods developed above to existing prevention research data thereby evaluating the practical utility of the methods.
6. To disseminate state-of-the-art mediation methods to prevention researchers and others interested in mediation analysis. The overall goal is to translate and evaluate methods in technical publications so that they can be more easily and accurately applied by researchers.