The proportional hazards assumption for the GZTS scales was evaluated by testing the interaction of time by trait. The interactions were not significant for most scales, supporting the proportionality assumption, except for General Activity and the estimated Conscientiousness. These analyses suggested that the risk of death was not proportional across different levels of General Activity and Conscientiousness, a problem usually addressed by stratifying the non-proportional predictor (32, 46). So, in addition to models that used the continuous scores, we tested models in which General Activity and Conscientiousness were trichotomized, using cut-offs at 1 SD below and above the mean. Because women were enrolled in the study about 20 years later, additional analyses compared the pattern of associations for women and men. To statistically test for sex differences in mortality risk, we added a sex by trait interaction term in a second step of the Cox’s regression model. Additional analyses examined personality predictors of mortality when limiting the sample to deaths due to cardiovascular disease, and when individuals assessed within five years from the outcome were excluded.
