Does Childhood Religiosity Delay Death?

An aspect of adulthood religiosity found to significantly predict childhood reports of religious importance in fixed-effects twin models and included in our main statistical analysis was religious importance (ranging from “not at all” to “very important”). The pattern of twin differences suggests that relatively less religious adults tend to report childhood religion as more central to their family than do the most highly religious adults. Several other correlated variables were also significant in fixed-effects twin models when religious importance was excluded; these include how important spirituality is for the respondent and how spiritual they believe they are (both ranging from “not at all” to “very important), and the frequency of relying on religion for comfort or in their daily activities (both coded from “never” to “often”). For the sake of parsimony, we do not retain these final three variables in our analysis, because results were unchanged with their inclusion.

Additional analyses also incorporated a measure of respondent’s current region of residence (Midwest, South, Northeast, and West), as childhood census region was unfortunately not available in the MIDUS data. Results were unchanged with the inclusion of this variable, so we excluded it from final analyses.

Results were substantively similar using listwise deletion to handle missing data.

Since the MIDUS data did not contain information about a respondent’s childhood religious denomination, additional analyses included a dichotomous measure of belonging to a conservative religious denomination in adulthood (coded using the RELTRAD coding scheme; see Steensland et al. 2000) to test whether those belonging to stricter religious groups at some point in the life course had a different mortality risk. Those reporting belonging to conservative religious traditions in adulthood were not at an elevated mortality risk compared to those belonging to non-conservative religious traditions. Main results were also unchanged with the inclusion of this variable, so it was ultimately excluded from final analyses.

Subsequent analyses considered a progressive modeling strategy, starting with the covariates from Model 3 of Table 2 and then adding in covariates in the same fashion as we did for Model 4 (health) and Model 5 (health behaviors). Results were substantively similar, so we present only the results from the fully adjusted model in Table 3.