An analysis of survey data to determine significant risk factors associated with adolescent marijuana use through utilization of sample weighting methods

摘要

This investigation seeks to identify factors associated with adolescent marijuana use in the 30 days prior to survey response collection in the 2012 National Survey on Drug Use and Health (NSDUH). Both inverse probability weighted and unweighted backwards elimination multivariate logistic regression modeling techniques were used to determine these factors. Final models compared the magnitude of the difference between odds ratios, the selection of final variables, the statistical significance of selected variables, and the overall fit of the models to determine whether or not we believed a weighted model was more appropriate for this type of complex sampling survey data. ududOur analysis showed that age, tendency towards risky behavior, importance of religious beliefs, academic grades, cigarette use, and alcohol consumption were significant predictors of marijuana use. In addition, the odds of marijuana use in those who smoke cigarettes and consume alcohol are much higher than the odds in those who do not partake in either.ududThe public health significance of this study is that the results can be used to help public health officials understand the risk factors that affect an adolescent’s decision to use marijuana. This insight would allow them to collaborate with policy makers to more accurately identify at risk teens and allow for avoidance, earlier detection, and treatment strategies.ududThe assumptions of logistic regression were met, but few model diagnostics were available for the weighted model due to the lack of appropriate statistical diagnostics in the Stata statistical software. However, based on our results, we believe the weighted model, which incorporates the complex sampling methods used in the data collection, is more sufficient for our data. Although the available diagnostics revealed similar results for both models, we saw notable differences in the odds ratios for race and academic grades, which leads us to believe that weights are a necessary component of the model.