The NESARC sample and its associated sampling weights were originally constructed in order to provide nationally representative estimates. Nevertheless, the sample included individuals from all 50 states with sample sizes ranging from 69 in Vermont to 3932 in California (median sample size across states, 490 persons). The use of nationally representative data to address associations at the state-level can be accomplished through multilevel regression (Gelman, 2007; Lax and Phillips, 2009; Park et al., 2004) where potential lack of representativeness of samples within states is accounted for by controlling for individual-level covariates in the model. To lessen concerns regarding the representativeness (or lack thereof) of NESARC data at the state-level, an investigation was conducted to compare the demographic representation of the sample at the state level. Correlations and Bland-Altman plots (Bland and Altman, 1986, 1999) were used to compare weighted and unweighted state-level NESARC demographic variables with the ACS state demographics. Since correlations between NESARC demographic variables and the ACS were higher using completely unweighted state estimates rather than the weighted estimates, we concluded that the NESARC was representative at the state level, and we performed the state- and multi-level NESARC analyses without the sampling weights.
