The NESARC multilevel analyses used hierarchical logistic regression (SAS 9.2 Proc GLIMMIX) of individual marijuana outcomes (n=34,520 for marijuana use and abuse/dependence, n=1453 for abuse/dependence among users) regressed on individual- and state-level covariates, including state-level medical marijuana law. The hierarchical logistic regression included a random intercept for state to account for possible correlation of individuals within state not explained by state-level covariates. In addition, a random effect for the primary sampling units (nested within states) was included to account for the complex clustered NESARC sampling design.
