Major Data Analysis Issues in Drug Abuse Prevention Research

Abstract

This chapter has provided an overview of some of the data analysis issues confronting prevention researchers. Several decisions made prior to the study clarify the data analysis, including randomization of units to conditions, refinement of measures of relevant constructs, linking of program components to potential mediators, and strategies to reduce confounding of program effect estimates due to attrition. The mixed design with one between- and one within-subjects factor is complicated in actual prevention research when there are multiple levels of possible analyses and when there are missing data. New techniques based on hierarchical models adequately model differential growth among participants, the nesting of participants within schools, and inclusion of partially missing data. Such analyses now reflect state-of-the-art prevention evaluation. Conditional and unconditional models can be used to estimate program effects when there is incomplete randomization. The analysis of mediating variables which was described is likely to increase understanding of the mechanisms of program effects. Finally, future developments in statistical analyses of drug prevention studies were considered. In the discussion of the new statistical advances, it is important to keep in mind that the usefulness of statistical methods rests on the quality of the data collected, the research design, and on the truth of the substantive theory.