It is documented that the elderly are more religious and that religiosity is associated with better health & lower mortality; contrary to wide‐held beliefs, religiosity decreases with greater expected proximity to death

Aging, Proximity to Death, and Religiosity. Marie Lechler  Uwe Sunde. Population and Development Review, August 24 2020. https://doi.org/10.1111/padr.12358

Rolf Degen's take: https://twitter.com/DegenRolf/status/1344583676466769921

Abstract: Considerable evidence has documented that the elderly are more religious and that religiosity is associated with better health and lower mortality. Yet, little is known about the reverse role of life expectancy or proximity to death, as opposed to age, for religiosity. This paper provides evidence for the distinct role of expected remaining life years for the importance of religion in individuals’ lives. We combine individual survey response data for more than 300,000 individuals from 95 countries over the period 1994–2014 with information from period life tables. Contrary to wide‐held beliefs, religiosity decreases with greater expected proximity to death. The findings have important implications regarding the consequences of population aging for religiosity and associated outcomes.

Robustness

The result of a decline in religiosity with greater proximity to death is robust to a variety of robustness checks. In particular, a potential concern for the validity and robustness of the results is multicollinearity due to the systematic correlation between remaining years of life and age. In order to explore the robustness of the results with respect to this concern, we computed variance inflation factors for the estimates for remaining years of life and age obtained on the full sample. The findings do not reveal any evidence for excessive multicollinearity.12

Second, the result is not driven by the particular measure of religiosity in terms of subjective importance of religion in life. The pattern also robustly emerges when using alternative measures of religiosity. This is illustrated by the results for alternative measures related to the subjective strength of religious beliefs, importance of God, or a composite measure of responses regarding the importance of religion, the belief in God, and the importance of God in life by ways of a principal component.13

The result is also robust to the use of alternative estimation methods. In particular, the results robustly emerge when applying interval regression methods that account for the interval censoring of responses as consequence of the coarse response scale.14

The result regarding a decline in religiosity as the expected remaining years of life decrease also robustly emerges for countries with different levels of socioeconomic development. In particular, the respective coefficient estimates from an extended specification that allows for different coefficients for each bin of remaining life years for the groups of OECD and non‐OECD countries reveal very similar patterns.15 This is reassuring in light of the different levels of religiosity in rich and less developed countries.16

The main results are also robust to the inclusion of separate controls for birth cohorts. The joint identification of age, period, and cohort effects is possible due to the estimation of flexible, semiparametric specifications for five‐year bins of age and cohort. Most importantly, the effect of remaining years of life on religiosity is identified from the variation in remaining life expectancy across country‐age‐sex cells.17 Besides a decline in religiosity for greater proximity to death and an increase in religiosity with age, the results document a decline in religiosity for later birth cohorts that is consistent with an increasing secularization among younger cohorts.18

Another method to identify cohort and period effects separately from the age pattern in religiosity is the inclusion of control variables that incorporate differences across cohorts in a nonlinear way (Heckman and Robb 1985).19 We apply this methodology building on the hypothesis that personality and beliefs are formed during the critical period of adolescence, when individuals are particularly susceptible to environmental conditions (Arnett 2000). In view of the fact that democracies typically grant freedom of religion whereas religious practices are more regulated in less‐democratic environments, we use the exposure to democracy during life as a cohort proxy variable for religiosity. This measure builds on existing work that has pointed out the role of the institutional environment for preferences for democracy (Fuchs‐Schündeln and Schündeln, 2015). The estimation results for these extended specifications confirm the decline in religiosity the lower the expected number of remaining years of life as well as the increase in religiosity with age.20

Finally, the result of a decline in religiosity with greater proximity to death consistently emerges also in other data sets. To explore the robustness of the main results, we replicated the analysis using the Gallup World Poll, which contains comparable information about the importance of religion and demographic characteristics as the World Value Survey. The results that emerge from this exercise are qualitatively and quantitatively almost identical to those obtained before.21

Religiosity, religious service attendance, and health

The analysis so far has focused on the role of proximity to death, as opposed to age, for religiosity. A potential explanation for the result of declining religiosity with greater proximity to death is a decline in the participation in religious activities, in particular in the attendance of religious services at the end of life. Indeed, the findings for alternative outcomes also reveal a similar pattern for the attendance of religious services as dependent variable.22 If health‐related limitations imply reduced attendance, as suggested by evidence for older age by Ainlay, Singleton, and Swigert (1992), and if lower attendance is associated with lower subjective religiosity, either due to lower awareness or as consequence of less frequent and intense social interactions with other individuals during religious services, this might explain the empirical results as consequence of health deterioration at the end of life.

The robustness of the main results to the inclusion of subjective health status as a control variable should already account for a health confound to some extent. In order to explore the distinct predictions of a health effect working through attendance and a genuine effect of proximity to death, we also estimated more extensive specifications that allow for an interaction effect of health and remaining years of life.

The results for attendance as dependent variable show that better health indeed increases attendance of religious activities, while closer proximity to death reduces attendance.23 A negative interaction between health and remaining years of life indeed indicates that better health partly compensates for the decline in religiosity closer to death, but the effect is too small to eliminate (or even reverse) the main result that religiosity decreases with greater expected proximity to death. When considering the importance of religion as dependent variable, health is positively related to religiosity, whereas the interaction effect turns out to be insignificant. The positive health effect is indeed consistent with the previous literature. However, the main finding of a negative effect on religiosity of greater proximity to death remains unaffected by this extension.

Another, more direct, way to explore the possibility that the results for religiosity are driven by attendance of religious services is to control for attendance when conducting the main estimates. The respective estimation results reveal indeed that respondents state a greater importance of religion in their lives when they attend religious services more often.24 The qualitative results regarding the decline of religiosity along with a decline in expected remaining years of life remain unaffected, however. Although the coefficient estimates are reduced to about half the size compared to the estimates reported in the main results, the patterns of religiosity with remaining years of life and age remain robust and significant.

Heterogeneity by sex, religious affiliation, and development

Heterogeneity by sex has been found to be a key feature of religiosity in the existing literature, with women being more religious than men.25 This pattern also emerges in the estimation results of this paper, which reveal a significantly higher level of importance of religion in life for women than for men. A question that emerges in the context of the previous results is therefore whether there is not only a sex difference in average religiousness, but also in the life cycle patterns of religiosity. To account for differences in the age pattern, we estimated extended specifications of the empirical model (2a) that allow for sex‐specific age effects. The results of these estimates reveal similar age patterns for women and men, although the age profile is slightly more pronounced for women.26 To explore whether the role of remaining life years for religiousness varies systematically by sex, we also estimated extended specifications that allow for a sex‐specific pattern of remaining lifetime. Again, the results reveal a similar pattern as for the baseline.27 If anything, the decline in religiousness is even more pronounced for women, but the sex differences are not significant.

Religions differ in many dimensions, including behavioral norms, beliefs about afterlife, and concepts of salvation. This likely maps into the motives for religious participation and the role of proximity to death or age for religiosity. In order to investigate this issue and explore possible heterogeneity in the role of expected remaining life years for religiosity, we replicated the analysis separately for individuals reporting different religious affiliations. The main result regarding a decline in religiosity in association with a smaller number of expected remaining years of life holds for respondents that report to be of Christian, Muslim, or Buddhist faith.28 Together, these religious denominations cover more than 60% of the sample. The pattern is not significant (and even opposite in slope) for respondents that report to be of Hindu or Jewish faith. Together, these respondents only make up less than 4% of the sample, however. The pattern is qualitatively similar as in the full sample, although somewhat tilted to the positive and not significant, for individuals reporting to be member of no religion, who make up for around 17 percent of the sample.29

Even within Christianity, the beliefs about afterlife as well as social norms differ substantially across denominations. For instance, recent work by Becker and Woessmann (2018) has pointed out that religious beliefs and social norms are a possible explanation for a higher propensity of suicide among Protestants compared to Catholics. Replicating the analysis for Catholics and Protestants indeed reveals differences in the gradient for remaining life years among the two denominations. In particular, the gradient is more pronounced for Protestants than for Catholics, which suggests a more pronounced erosion of religious beliefs.30

To further explore heterogeneity in the nexus between remaining life years and religiousness with respect to the overall level of economic development, we replicated the analysis with a linear specification of the effect of expected remaining life years while allowing for country‐specific slope parameters.31

The coefficient estimates are positive for the vast majority of countries in the sample, with smaller coefficient estimates for countries with higher levels of per capita income, as illustrated in Figure 4. In the estimation sample, religiosity and economic development exhibit a strong negative correlation across countries, but there is no significant relationship between average expected remaining years among survey respondents and economic development across countries. This implies that the pattern of heterogeneity is not due to sample composition. Instead, together with the result that religiosity is predicted to be lowest among a young population in environments with a short life expectancy, this finding implies that population aging in terms of increased life expectancy is expected to be associated with a stronger increase in religiosity in less developed countries, including many African countries. This provides important insights regarding the consequences of demographic aging for developing countries given that religiosity has been associated with faster economic development at the aggregate level and with better health and greater resilience at the individual level. In particular, the consequences of aging are expected to be stronger among less developed countries. At the same time, the results indicate that the effects of demographic change on religiosity and related outcomes might be limited among some of the more developed countries.