We weighted the data to adjust for differential probabilities of sample selection. We conducted logistic regressions for the multivariate analyses; these regressions took appropriate account of the complex sample design. For interpretation of trends over time, we ran logistic regressions with linear and quadratic terms for year.