variable, which distinguished respondents who were in the age ranges 18–29, 30–44, 45–59, and ≥60 y at interview. The sex-by-cohort interaction was examined to determine if the well-known gender difference in drug use has become smaller in recent years in some or all countries studied. Sociodemographic correlates of lifetime use at the time of interview were examined using logistic regression analysis [38]. Sociodemographic variables (e.g., family income, relationship status) were coded as of the time of interview, not as of the time of first drug use, which means that the associations examined might reflect influences of previous drug use on sociodemographic characteristics. It is consequently illegitimate to interpret the associations in temporal terms. Instead, the associations provide only cross-sectional descriptive information. Standard errors of cumulative incidence estimates, survival coefficients, and logistic regression coefficients were estimated using the Taylor series linearisation method [39] implemented in the SUDAAN software system [40,41]. Survival coefficients, logistic regression coefficients, and their 95% confidence intervals were exponentiated and are reported in the form of odds ratios (ORs) for ease of interpretation. Multivariate significance tests were made with Wald χ2 tests using design-based coefficient variance-covariance matrices. Significance tests were consistently evaluated at the 0.05 level with two-sided
