Analyses were conducted using SAS software version 9.2 (SAS Institute, 2008). General Estimating Equations (GEEs) were used to account for the clustering of participants within states and to allow models to include state-level variables. GEE models are crucial to employ in this case in order to avoid underestimation of errors associated with state-level variables and natural clustering of data. While NESARC sampling weights have been constructed to provide nationally representative sample estimates, they were not utilized in these analyses for two reasons. First, sampling weights must be analyzed using SUDAAN software, which does not currently support calculation of GEE models (Research Triangle Institute, 2004). Second, since the analyses were concerned with relative effect sizes, and not making population-level estimates of the number of people affected, the sampling weights were not necessary to employ (Korn and Graubard, 1991).
