linear and quadratic effects of age were included as time-varying covariates, and sex and self-reported race were included as timeinvariant covariates. Random effects of the intercept, linear and quadratic slopes of age, and all time-varying predictors were examined, with non-significant random effects constrained to zero to yield a parsimonious final analytic model. The alcohol consumption variable was rescaled (divided by 100) for the purpose of model convergence. Parameters were estimated using maximum likelihood estimation with robust standard errors in Mplus, which is robust to nonnormality of variables and non-independence of observations (Muthén & Muthén, 1998–2012).
