Statistical Analyses

We examined sample characteristics by age and used chi-squared tests to assess age differences in baseline demographic and socioeconomic characteristics. Means and standard deviations for time-varying covariates (i.e., depressive symptoms, the psychosocial resources, and the health factors) were examined by age and for each wave of data collection, but bivariate correlations among all continuous variables were only assessed at baseline. Our multivariate analyses modeled change in depressive systems over time using linear mixed models with a random intercept and random slope for time. Linear mixed models are appropriate for modeling repeated observations across waves that are nested within individuals. Time was modeled as years since baseline (i.e., Wave 1 = 0 years, Wave 2 = 4 years, and Wave 3 = 8 years) and models were built sequentially.

Model 1 includes time, age, and the sociodemographic characteristics. Model 2 adds the three psychosocial resources. We then separately tested several interactions between age and time (i.e., Model 3), hope (Model 4), purpose in life (Model 5), and religiosity (Model 6). Preliminary analyses examined the interaction between time and each psychosocial resource. None of these interactions were statistically significant; therefore, we focused the rest of our analyses on interactions between age and time and age and each psychosocial resource. A significant interaction between age and time would suggest that the rate of change in depressive symptoms differs by age; a significant interaction between age and a psychosocial resource would indicate that there are age differences in the effects of the resource on average levels of depressive symptoms. We evaluated significant interactions in more detail by estimating and graphing the simple slopes for depressive symptoms within each age group and examining age-stratified models. The stratified models added the health factors to test whether they accounted for linear change in depressive symptoms and the effects of the psychosocial resources. Stata 15 was used for all analyses and the significance level was set to p < .05.