All analyses were carried out in SAS version 9.0, using the “surveylogistic” and “surveyfreq” procedures, in conjunction with survey design variables, including weights, to account for the complex sample design of the NSDUH in variance estimation. Time-trend analyses of prevalence within methodologically contiguous NHSDA/NSDUH segments were conducted, in which binge drinking prevalence was modeled as a function of survey year, relative to the first year in the segment, assuming a log-linear trend (i.e., a single odds ratio describing change per-year). For the 1994 survey, which utilized a split-sample design, we analyzed only the larger “B” survey, which was methodologically identical to the 1995–1998 surveys. Time-trend analyses of odds ratios across the entire NSDUH series were conducted by combining all survey years into a single logistic regression analysis, predicting binge drinking from age category, survey year, and their interaction. The interaction effect is the odds ratio describing the change in the main effect odds ratio over time, relative to the first time point (1979), and corresponds to the slope of a plot of log(OR) versus survey year, where a constant trend
