MULTILEVEL MODELING OF DRUG USE TRAJECTORIES

点击翻译按钮获取中文摘要

This application proposes secondary data analyses of drug use and attitudes in three longitudinal samples: the Lexington Longitudinal Study, the National Longitudinal Survey of Youth (NLSY) Main sample, and the NLSY Child sample. Longitudinal analyses of substance use have frequently been hampered by the limitations of conventional methods. Statistical procedures are needed which can model change at the individual level, utilize data from all waves of measurement and include attriters, evaluate the effects of both fixed and time-varying covariates on developmental pathways for drug use, and model the influence of contextual factors. Drawing upon recently developed techniques, this study will apply three-stage mixed effects regression models to characterize and predict individual level variation in continuous drug outcomes in these secondary data. Generalized linear mixed model procedures are proposed to analyze series of discrete drug outcomes. Proposed analyses are distributed selectively across datasets, depending on the availability of appropriate measures. The analysis plan has six interdependent components. First, analysis will use unconditional random effects regression models to estimate and characterize individual differences in trajectories of drug use and drug-related attitudes over multiple waves of measurement. Second, analysis will use two-level mixed effects regression models to predict individual variation in drug use and drug-related attitudinal pathways as a function of both fixed and time- varying characteristics of persons, and to evaluate reciprocal relationships between rates of change in different drug outcomes. Third, analysis will use three-level mixed effects regression models to estimate and characterize individual level variation in relationships between predictors and drug outcomes as a function of multiple contextual settings (family, school, neighborhood, and peer network). Fourth, analysis will use mixed effects models to pool data across overlapping cohorts and evaluate the feasibility of modeling substance use trajectories over extended developmental sequences in this fashion. Fifth, exploratory analyses will evaluate the influence of random cross- classification of subjects into multiple contexts. Sixth, exploratory analyses will use latent variable growth models to model the effects of measurement error in drug use. The proposed study will contribute to the understanding of developmental pathways for drug use in youth and early adulthood. The study may also contribute to multiple policy dimensions in drug studies, including the design of prevention, intervention, and treatment strategies with greater recognition of individual differences in pathways of drug use over time; enhanced understanding of the consequences of individual level risk and protective factors on the dynamics of drug use; and the development of family and community level prevention and intervention programs and the integration of such programs with individual level approaches.