RANDOM EFFECT MODELS FOR SUBSTANCE ABUSE RESEARCH

点击翻译按钮获取中文摘要

DESCRIPTION: (Applicant's Abstract) This is for secondary data analysis that will apply and test a state-of-the-art multilevel statistical approach to research on the prevention of substance abuse. The analytical technique to be tested, the random-effect model (REM) allows simultaneous processing of information from different levels (e.g., student, class, school, etc.). Conventional approaches with single-level analyses and conventional multilevel models contain shortcomings in specifying substantive models or in statistical testing of hypotheses, which in turn may lead to unreliable results and erroneous conclusions. The REM provides a more appropriate mechanism to represent the complexity of the multilevel data and a more adequate model specification to test the impact of social contexts. The ability to test contextual effects makes the application of REM a very promising alternative for data analyses in substance abuse research. Having the flexibility to model random group effects, the REM also offers a more powerful alternative to examine the impact of social context on substance use. The extension of this approach to substance abuse research may help us understand more about the impact of prevention programs. In this proposed project, hierarchical linear models will be developed and applied to data from a substance abuse prevention study to identify important contextual and individual effects on substance use. Results obtained from the REM will be compared to that of other conventional multilevel approaches. In addition, the issues of robustness to violations of normal distribution and sensitivity to the number of groups of the REM will be investigated. Other potential methodological concerns, such as dichotomized outcome, latent factor, and survival analysis, will also be examined. Findings from this project should provide guidelines for appropriate REM application, a better understanding of the contribution of environmental contexts to effects of prevention programs, and more valid estimation of program effects on substance abuse.