generation were created to facilitate interpretation. These reparamaterized models had the same number of parameters as the models with the interaction terms. Once final models with significant interactions were determined, tests of proportional hazards, i.e. the assumption that the risk associated with different levels of a variable remain proportional over time, were computed using Schoenfeld residuals (Grambsch & Therneau, 1994). When the assumption was violated for generation or sex (e.g., generation X and baby boomer generation hazards differed, or male and female hazards differed) the violation was resolved by creating interactions with time so that hazards were proportional within each risk period. Violations of the proportional hazards that were made by covariates were not corrected, though. Instead, the average association with the outcome across time was used. This is because covariate violations did not affect the interpretation of the associations of sex or generation with help-seeking and treatment (Allison, 1995; 2010). Standard errors were adjusted for the non-independence of observations within families using the Huber-White robust variance estimator. All analyses were performed using Stata Statistical Software Release 15 (StataCorp, 2017).
