Session: Three M's in Prevention Research: Mediation, Multilevel Modeling, and Missing Data (Society for Prevention Research 24th Annual Meeting)

Statistical and methodological innovation in prevention research has allowed researchers to ask more interesting and complex questions. Many prevention researchers use mediation analyses to understand the causal chain of relations between three or more variables. Multilevel data are also present in a large proportion of prevention intervention studies, with participants nested within a variety of geographical or organizational groups. It is also possible to combine these approaches to examine the causal process at multiple levels using multilevel mediation. In any analysis, appropriately dealing with unplanned missing data can be difficult for researchers. However, it is also possible to use modern missing data methods to enhance research and leverage resources by using a planned missing data design. The three papers in this symposium examine issues related to several combinations of these “three Ms” in prevention research. The papers help researchers address specific issues in implementing prevention research studies: How should I best design my study?  How should I best analyze my data? Recommendations and practical application for applied researchers are explored.

The first paper, “Using Planned Missing Data to Leverage Resources in a Single Mediator Model,” considers a research scenario where researchers interested in assessing an indirect or mediated effect must choose to either collect a large sample of data with an inexpensive measure or a smaller sample of data with a more expensive measure. To address issues of limited resources in study design, a design with purposeful missing data may utilize information from a large sample of the inexpensive mediator(s) and a small subsample of the expensive measure. This study makes recommendations as to when and how researchers might be able to leverage missing data to their benefit.

The second paper, “A Comparison of Bootstrap Confidence Intervals for Multilevel Mediation,” considers the use of bootstrapping to create appropriate confidence intervals when using the most common multilevel model for prevention research: the 2-1-1 mediation model. The paper compares seven approaches to creating confidence intervals for the 2-1-1 mediation model.

The third paper, “Missing data strategies for real world multilevel data: What should I do when the recommended methods fail?,” explores options for addressing missing data in multilevel modeling when the recommended strategies of maximum likelihood estimation or multiple imputation cannot be implemented.  Currently, few to no guidelines exist for researchers faced with choosing which suboptimal approach to use for missing data analysis. The results of this study will help researchers decide how to proceed when the recommended methods for missing data handling fail.