A full multivariable model including all the covariates was then estimated such that the latent hazard function with its 12 binary time-specific event indicators was regressed on the set of time-invariant covariates (all factor loadings fixed to 1.0 to reflect proportional covariate effects across time), and any covariate that violated the proportionality assumption had freely estimated regression paths on the binary time-specific event indicators to allow for time-varying effects. The results of this model allowed interpretation for each individual covariate effect on alcohol onset after adjusting for all other covariates.
