Common Statistical Methods for Primary and Secondary Analysis in Substance Abuse Research

Abstract

This chapter presents statistical methods and issues commonly encountered in the design and analysis of substance abuse research studies. It begins with a general discussion contrasting primary and secondary data analysis, followed by an overview of study design from the perspective of the conduct of primary research, including hypothesis and planned analysis specification, sampling schemes, and power analysis. Next, descriptions of study characteristics are included from the perspective of secondary analysis, paying particular attention to characteristics that need to be considered when determining appropriate analytic methods and interpreting results. Statistical methods reviewed include: various types of regression (linear regression, logistic regression, survival analysis), related topics, such as moderators and mediators, as well as multilevel models (for longitudinal or clustered observations), and latent variable modeling techniques, including structural equation modeling, latent class analysis, latent transition analysis, and growth mixture modeling. Finally, overviews of four major special topics particularly important when using secondary data are provided, which include: multiplicity of hypotheses, combining data and results from multiple studies, missing data, and propensity scores. Where helpful, concepts and methods are illustrated using practical examples.