The findings from our analyses are contingent on the validity of the assumption that design changes in the NHSDA/NSDUH do not influence prevalences differentially by age or other demographic characteristics. That is, while we expect methodological variation to effect prevalence estimates, we assume that those effects do not differ markedly across demographic groups. This assumption was tested by comparing effect sizes in each redesigned survey to those from the preceding year. Significant differences would be manifest as interactions between demographic variable and survey-year. In theory, significant differences in these tests could correspond to true secular trends, but in practice, true trends would be difficult to observe over a single year. Hence, we tested for interactions between survey year and age, sex, or race/ethnicity for the following pairs of surveys, which correspond to major methodological changes: 1994 vs. 1993, 1999 vs, 1998, and 2002 vs. 2001. In only one case was a significant interaction detected: In the 1994 survey, compared with the 1993 survey, the odds for binge drinking among 12–14 year olds increased significantly whereas those for other age categories did not (Interaction Wald-χ2=17.0, p=0.002). The implications of such an effect are discussed below.
