Religion, Life Expectancy, and Disability-Free Life Expectancy Among Older Women and Men in the United States

Abstract

Objectives

Existing literature shows religion is associated with health and survival separately. We extend this literature by considering health and survival together using a multistate life table approach to estimate total, disability-free, and disabled life expectancy (LE), separately for women and men, for 2 disability measures, and by 2 indicators of religion.

Method

Data come from the Health and Retirement Study (1998–2014 waves). Predictors include importance of religion and attendance at religious services. The disability measures are defined by ADLs and IADLs. Models control for sociodemographic and health covariates.

Results

Attendance at religious services shows a strong and consistent association with life and health expectancy. Men and women who attend services at least once a week (compared with those who attend less frequently or never) have between 1.1 and 5.1 years longer total LE and between 1.0 and 4.3 years longer ADL disability-free LE. Findings for IADL disability are similar. Importance of religion is related to total and disabled LE (both ADL and IADL), but the differentials are smaller and less consistent. Controlling for sociodemographic and health factors does not explain these associations.

Discussion

By estimating total, disability-free, and disabled LE, we are able to quantify the advantage of religion for health. Results are consistent with previous studies that have focused on health and mortality separately.

Religion is perhaps one of the longest-standing considered social determinants of health, with studies of associations between religion and health dating back to the early 1900s (Durkheim, 1915; Hiltner, 1943). Religious institutions have played a role in public health throughout history, although their influence has not always been positive (E. L. Idler, 2014). This enduring interest is unsurprising given the understanding of religion’s role in shaping societal values and norms, and subsequently health-related behaviors (Strawbridge, Shema, Cohen, & Kaplan, 2001). In the United States, the overwhelming balance of this evidence has suggested that religion is salutary (Koenig, 2012; Krause, 2011; Larson, Swyers, & McCullough, 1998; Moreira-Almeida, 2013; Seybold & Hill, 2001). Much of this literature has focused on older adults who have higher morbidity and mortality and are therefore most likely to gain from religion’s health benefits (Chida, Steptoe, & Powell, 2009; Hill, Burdette, & Idler, 2011; Hill & Cobb, 2011; Lavretsky, 2010; Zimmer et al., 2016).

However, much of the evidence is based on studies wherein religious involvement is measured in terms of frequency of attendance or participation in religious services in a church or temple (Hummer, Ellison, Rogers, Moulton, & Romero, 2004). One concern with this measure of religiosity is the causal nature of the association (Sloan et al., 2000), since a positive association between attendance and health could mean that religiosity provides healthful benefits or it may reflect selection bias whereby healthier individuals are more physically able to attend services and participate in religious activities. Cross-sectional data makes it difficult to differentiate these pathways, though the association between religiosity and longer life is more convincing, due to the longitudinal nature of mortality studies, and has been shown to be a consistent robust connection (McCullough, Hoyt, Larson, Koenig, & Thoresen, 2000) with minimal bias due to reverse causation (E. Idler, Blevins, Kiser, & Hogue, 2017). Since McCullough’s meta-analysis, additional studies based on younger and older populations have strengthened the evidence (Dupre, Franzese, & Parrado, 2006; Hill, Angel, Ellison, & Angel, 2005; Musick, House, & Williams, 2004) by finding connections between religiosity and cause-specific mortality (cardiovascular disease and cancer) when adjusting models for a large range of lifestyle and risk factors (Li, Stampfer, Williams, & VanderWeele, 2016), and that religious and spiritual involvement had as strong or a stronger impact on mortality compared with other common health interventions, including fruit and vegetable consumption and statin therapy (Lucchetti, Lucchetti, & Koenig, 2011). As evidence continues to mount, studies are continually showing that a religiosity–longevity nexus is difficult to explain away.

Despite this evidence several questions remain. First, living longer does not necessarily mean living longer in good states of health (Jagger, 2006), and it is unclear whether religiosity is associated with both quantity and quality of life. The distinction is particularly important for those at older ages since life expectancy (LE) has been rising in the United States and around the world (Kinsella & Phillips, 2005), with ensuing debate as to whether these extra years result from health interventions that save lives but do little to improve health (Crimmins & Beltrán-Sánchez, 2011). One way of focusing attention on this is by considering both total and “healthy” LE.

Our focus here is therefore on both disability-free and total LE. Others have found a protective quality of religion on physical functioning and disability (E. L. Idler & Kasl, 1992, 1997a; Roff et al., 2006) but many early studies were based on cross-sectional data and those based on longitudinal data have produced mixed findings. Some have found a protective association of religion with disability (Hill, Burdette, Taylor, & Angel, 2016; E. L. Idler & Kasl, 1997b), while others have found either a contemporaneous association, but no association over time (Fitchett, Benjamins, Skarupski, & Mendes de Leon, 2013), or no association between religion and disability (Son & Wilson, 2011).

Second, disability, more so than mortality, may be subject to reverse causality, since functional limitations may act as a barrier for social activities such as religious service attendance (Kelley-Moore & Ferraro, 2001). In disability-free LE research, this can be addressed by considering results that are status-based, that is, conditioned upon baseline disability status, providing separate estimates for those not disabled versus disabled at baseline. Controlling for baseline health status can also help minimize the impact of reverse causality.

Third, religiosity is a complex social phenomenon encompassing different dimensions. We focus specifically on two measures, attendance at religious services and self-rated importance of religion. According to the comprehensive classification of survey items into domains of religiosity (E. L. Idler et al., 2003), attendance falls squarely into the domain of “public practice,” influencing health through adherence to risk-reducing behaviors, exposure to social networks and sources of support, and physiological responses to stressors. Self-rated importance of religion was not included in Idler’s classification but this item best falls into the domain of “religious intensity” or the degree to which one considers oneself to be religious, with the pathway to health being through psychological well-being and feelings of self-worth. While most studies have concentrated on participation alone, how religion associates with disability may be especially pertinent for religious intensity since it may be less prone to endogeneity.

Fourth, little research has carefully examined gender differences in the effects of religiosity on health. Gender is often considered a control variable, but infrequently are models examined separately for men and women. Findings from those studies that have looked at gender have been inconsistent. Using the Taiwanese National Health Interview Survey, Lin, Liang, and Chen (2017) find that religion has a salutary influence on a variety of health measures, including physical, psychological and general, but only among men. McFarland (2010) draws similar conclusions examining mental health outcomes. One of the reasons speculated for these findings is that, because men tend to have weaker support networks overall, they have more to gain from the social aspects associated with religious participation. Yet, how associations between religiosity and health vary by gender is far from settled, since others find no gender variation across physical functioning and other health outcomes (Headey, Hoehne, & Wagner, 2014; Hidajat, Zimmer, Saito, & Lin, 2013).

These various issues are addressed in the current article which examines the degree to which religion (measured by attendance and importance) is associated with quantity and quality of later life through the estimation of total, disability-free, and disabled LE for middle-aged and older adult men and women in the United States. By running separate models for men and women, we determine the gender variation in the associations. We examine two types of disability: activities of daily living (ADL) and instrumental activities of daily living (IADL). Further, to overcome potential reverse causality, we calculate both population and status-based estimates. Both approaches follow individuals prospectively to assess years of life lived with and without disability, but status-based estimates are made separately for those who are nondisabled and disabled at baseline. Finally, we control for several key sociodemographic and health variables that may influence the association between religion, disability, and mortality.

Method

Data Source

Data come from the Health and Retirement Study (HRS), an ongoing panel study of men and women aged 51 + in the United States, which began in 1992 with a cohort of preretirement-aged individuals born between 1931 and 1941. New cohorts were added in 1993 and 1998 to round out the sample over age 50, and additional cohorts are enrolled every 6 years (in 2004, 2010, etc.) to refresh the sample at the younger ages. For the cohorts used here, response rates range from 68% to 81% in the baseline wave and from 87% to 89% at each follow-up wave. HRS conducts about 20,000 interviews every 2 years using a combination of telephone and in-person interviews. The questionnaire covers a broad range of topics. Fact and date of death are verified through linkage to the National Death Index.

Disability Measures

We examine two measures of disability: ADL disability based on six activities: dressing, bathing, eating, walking across a room, getting in/out of bed, and using a toilet; and IADL disability based on five activities: preparing meals, shopping for personal items, using a telephone, taking medications and managing money. Participants are asked if they have any difficulty with the activity because of a physical, mental, emotional, or memory problem and to exclude difficulty that is expected to last less than 3 months. ADL(IADL) disability was defined as reporting difficulty performing at least one ADL(IADL).

Religion Measures

We use two measures of religion from the core HRS interview. The first is a question on self-rated importance of religion: “How important would you say religion is in your life; is it very important, somewhat important, or not too important?” The second is a question on religious service attendance: “About how often have you attended religious services during the past year?” We used a three-category indicator of religious attendance: at least once a week; less than once a week; never. The question on importance of religion was in every wave of HRS through 2012 while the question about attendance appeared intermittently between 1992 and 1995, but was not added as a regular question until 2004.

Control Variables

All models control for age and are stratified by sex. Demographic and socioeconomic control variables include race/ethnicity, a composite measure of coupleness and living arrangements, and education. To control for potential selectivity in baseline health, we included several measures of physical and mental health, including self-ratings of vision and hearing impairment, cognitive function, number of depressive symptoms (8-item CES-D), and number of chronic conditions (hypertension, diabetes, lung disease, heart disease, stroke, cancer, psychiatric problems, arthritis). Our measure of cognitive function combines self and proxy cognition measures following the coding scheme developed by Langa and Weir (Crimmins, Kim, Langa, & Weir, 2011). Supplementary Table 1 shows how the control variables were coded. All control variables are measured at the initial wave of the time series used in the analysis, as described later.

Study Samples

Because the two measures of religiosity were measured in different HRS waves, we have different (but overlapping) study samples for the analyses. The first comprises 8,524 men and 11,125 women aged 51+ residing in the community in 1998 who were followed for 16 years to 2014. The second sample comprises 8,014 men and 10,804 women who were aged 51+ and residing in the community in 2004 and followed for 10 years to 2014. The 1998 sample is used for the analysis of importance of religion, first measured in the 1998 wave, while the 2004 sample is used for the analysis of religious attendance, not available for the full sample until 2004. Both samples are restricted to individuals who had nonmissing data on the disability, religion and control measures (other than CES-D, as explained later); have a minimum of two observation points (including mortality); and who were living in the community in the “baseline” wave for the analysis (either 1998 or 2004). Individuals who moved into nursing homes or other long-term care facilities after the “baseline” wave are included. There is considerable overlap between the two samples; differences are that the 1998 sample includes individuals who had died or withdrawn between 1998 and 2004, whereas the 2004 sample includes individuals who entered the study after 1998 and/or who had reached their 51st birthday between 1998 and 2004.

Characteristics for the two samples are provided in Supplementary Table 1. The samples are similar in terms of sociodemographic characteristics and health status with mean age being 65, roughly two-thirds married or partnered, and disability prevalence of 16% for ADL disability and 13%–14% for IADL disability. With regard to religious affiliation, just over three-fifths of the sample are Protestant, one-quarter are Catholic, about 2% are Jewish, and between 6% and 8% report no religious affiliation. The 2004 sample is slightly more educated than the 1998 sample (46% vs 39% completed more than 12 years of schooling), reflecting the population trend of increasing education among older adults during this period.

Missing Data

The level of missing data for all measures used in the analysis is extremely small, less than 0.1% for most variables with the exception of CES-D, which is higher (3.8% in 1998 and 2.2% in 2004) since it is not administered to proxy respondents. As respondents with proxies tend to be less healthy than those who complete a self-interview and we did not want to exclude them, we included them by coding “missing” as a category on the CES-D. This resulted in 134 respondents in total being dropped from the 1998 sample, and 114 from the 2004 sample due to missing data on other variables.

Analytic Methods

Disability-free LE estimation begins with the calculation of the probability of transitioning across disability states and mortality. As illustrated in Figure 1, there are six possible transitions, including retention in nondisabled or disabled states, movements from nondisabled to disabled or vice versa, and movements from nondisabled and disabled to death.

Figure 1.

Depiction of possible transitions among disability states.

Using data from the 1998–2014 waves of HRS, we estimate total, disability-free, and disabled LE for persons aged 51 or older, by sex and two indicators of religion, controlling for sociodemographic and health factors, obtaining these expectancies at 10-year age increments from 55 through 95. We use the Stochastic Population Analysis for Complex Events (SPACE) software for estimating health expectancy (Cai et al., 2010). SPACE is a SAS-based program that estimates total and healthy LE using a multistate life table (MSLT) approach constructed from transition probabilities estimated by a multinomial logistic regression. SPACE provides estimates of standard errors associated with life and health expectancies based on a bootstrap method and allows for the inclusion of control variables in the models. All analyses are weighted and estimates of standard errors are adjusted to take into account HRS’s complex sample design. We use 300 bootstrap samples for each analysis to calculate standard errors and all analyses are stratified by sex. We report both population-based and status-based results (Saito, Robine, & Crimmins, 2014).

For analyses that examine importance of religion, we use nine waves of survey data (1998–2014) to estimate LE, disability-free LE, and disabled LE, and for analyses examining religious attendance, we use six waves of data (2004–2014). We report the total, disability-free, and disabled LE estimates in terms of remaining years and compare these across levels of religious attendance and importance, examining the statistical significance of differences through the 95% confidence intervals around estimates obtained from bootstrapping.

Validation

We compared our estimated LE by sex with published life table values for 2006, the midyear for our study period. Our estimates are very close to published LEs and not statistically different (19.8 estimated vs 19.7 published for women and 16.9 vs 17.0 for men). In addition, HRS mortality data has been examined extensively and shown to be complete (Weir, 2016).

Results

Distributions for Religion Measures

Men are less religious than women on both of the religion indicators that we examine. For attendance at religious services, 43% of women attended at least once a week during the past year compared with 32% of men, with men also being more likely to report having never attended (31% vs 23%). A large majority of both men and women consider religion to be very or somewhat important in their lives (82% for men and 92% for women), however women are more likely than men to view religion as very important (70% for women vs 51% for men) and less likely to view it as not too important (8% for women vs 18% for men).

Disability Transitions

Table 1 presents the number of transitions between disability states and death for the 1998 sample for ADLs and IADLs, by sex. The total number of transitions between two successive waves was nearly 63,000 for women and 45,500 for men, with the total transitions for IADL being slightly higher than for ADL due to a small number of missing transitions for ADLs.

Table 1.

Number of Transitions Between ADL and IADL Disability States and Death Over 16 Years of Follow-up (1998–2014), Among People Aged 51+ in 1998 (Unweighted)

Gender	State at time t	Disability type	State at time t + 1			Total
			Nondisabled	Disabled	Dead
Female	Nondisabled	ADL	41,125	5,637	1,823	48,585
		IADL	44,315	4,822	1,845	50,982
	Disabled	ADL	3,421	8,205	2,715	14,341
		IADL	2,456	6,821	2,697	11,974
	Total	ADL	44,546	13,842	4,538	62,926
		IADL	46,771	11,643	4,542	62,956
Male	Nondisabled	ADL	32,189	3,404	2,208	37,801
		IADL	33,121	3,265	2,321	38,707
	Disabled	ADL	1,928	3,839	1,871	7,638
		IADL	1,789	3,183	1,763	6,735
	Total	ADL	34,117	7,243	4,079	45,439
		IADL	34,910	6,448	4,084	45,442

Note. ADL = activities of daily living; IADL = instrumental activities of daily living.

Further examination shows stability in the transitions for both ADL and IADL disability with 65% of the transitions for women’s ADL disability (41,125 out of 62,926) starting and ending in a nondisabled state and 13% (8,205) starting and ending in a disabled state; for men, a similar percentage remain in the same state over time, but a higher percentage start and end nondisabled (71% vs 65%) and a lower percent start and end disabled (8% vs 13%). Transitions from disabled to nondisabled, indicative of ADL recovery, number about three-fifths those from nondisabled to disabled, reflecting an overall decline in function for men and women.

Patterns for IADL disability are similar with about 80% of transitions involving remaining in the same state (70%–73% nondisabled and 8%–11% disabled), the “rate” of recovery is just over half the rate of decline, and transitions for men indicate slightly lower levels of IADL disability than for women.

Although not shown here, the patterns for the 2004 sample (2004–2014) are similar to those for the 1998 sample, except there are fewer transitions overall and proportionately fewer transitions into death due to the shorter time period.

Estimates of Disability-Free and Disabled LE

We report estimates of total, disability-free, and disabled LEs at age 65, separately for men and women, by importance of religion (ADL: Table 2, IADL: Table 3) and religious service attendance (ADL: Table 4, IADL: Table 5). The estimates are from population-based models that include only the religion indicator (unadjusted) and that control for sociodemographic and health factors described previously (adjusted). (Supplementary Tables 2–5 present corresponding status-based estimates for the adjusted models.) The patterns for total, disability-free, and disabled LEs are similar for other ages. (The full matrix of expectancies by age may be obtained upon request.)

Table 2.

Estimated Total, ADL Disability-Free, and ADL Disabled Life Expectancy (LE) at Age 65, by Sex and Importance of Religion (1998–2014)

	Importance of religion			Differences between religious groups
	Very important	Somewhat important	Not too important	Very vs somewhat	Very vs not too important	Somewhat vs not too important
Gender and model	LE (SE)	LE (SE)	LE (SE)	Diff (95% CI)	Diff (95% CI)	Diff (95% CI)
Female
Unadjusted
Total LE	19.69 (0.19)	19.01 (0.26)	18.94 (0.59)	0.68 (0.16, 1.20)	0.75 (−0.42, 2.02)	0.07 (−1.17, 1.41)
Disability-free LE	14.13 (0.18)	14.22 (0.26)	14.38 (0.49)	−0.08 (−0.66, 0.44)	−0.25 (−1.20, 0.73)	−0.16 (−1.15, 0.93)
Disabled LE	5.56 (0.11)	4.79 (0.17)	4.56 (0.26)	0.77 (0.37, 1.09)	1.00 (0.45, 1.71)	0.23 (−0.47, 0.98)
Adjusteda
Total LE	20.33 (0.18)	19.11 (0.27)	18.52 (0.58)	1.22 (0.71, 1.73)	1.82 (0.76, 2.91)	0.60 (−0.64, 1.77)
Disability-free LE	14.82 (0.17)	14.16 (0.27)	13.85 (0.48)	0.66 (0.13, 1.17)	0.97 (−0.01, 1.79)	0.31 (−0.72, 1.32)
Disabled LE	5.51 (0.11)	4.95 (0.18)	4.66 (0.26)	0.56 (0.18, 0.88)	0.85 (0.24, 1.49)	0.29 (−0.37, 1.00)
Male
Unadjusted
Total LE	17.07 (0.19)	16.25 (0.30)	16.42 (0.31)	0.82 (0.11, 1.38)	0.65 (−0.06, 1.24)	−0.17 (−1.07, 0.57)
Disability-free LE	13.63 (0.18)	13.30 (0.31)	13.46 (0.33)	0.33 (−0.35, 1.00)	0.17 (−0.57, 0.75)	−0.16 (−1.08, 0.62)
Disabled LE	3.44 (0.09)	2.96 (0.12)	2.97 (0.15)	0.49 (0.20, 0.72)	0.48 (0.14, 0.81)	−0.01 (−0.32, 0.35)
Adjusteda
Total LE	17.59 (0.22)	16.53 (0.31)	16.70 (0.33)	1.06 (0.36, 1.69)	0.89 (0.08, 1.58)	−0.17 (−1.10, 0.73)
Disability-free LE	14.18 (0.22)	13.50 (0.32)	13.68 (0.34)	0.68 (−0.07, 1.35)	0.50 (−0.30, 1.17)	−0.17 (−0.94, 0.72)
Disabled LE	3.41 (0.09)	3.03 (0.12)	3.03 (0.17)	0.38 (0.15, 0.65)	0.39 (0.001, 0.75)	0.00 (−0.35, 0.33)

Note. ADL = activities of daily living; CI = confidence interval; LE = life expectancy.

^aAdjusted models control for race/ethnicity, coupleness status and household composition, education, cognitive status, depressive symptoms, vision impairment, hearing impairment, and number of chronic conditions.

Table 3.

Estimated Total, IADL Disability-Free, and IADL Disabled Life Expectancy at Age 65, by Sex and Importance of Religion (1998–2014)

	Importance of religion			Differences between religious groups
	Very important	Somewhat important	Not too important	Very vs somewhat	Very vs not too important	Somewhat vs not too important
Gender and model	LE (SE)	LE (SE)	LE (SE)	Diff (95% CI)	Diff (95% CI)	Diff (95% CI)
Female
Unadjusted
Total LE	19.67 (0.19)	18.99 (0.27)	18.96 (0.61)	0.67 (0.18, 1.28)	0.70 (−0.42, 2.04)	0.03 (−1.28, 1.40)
Disability-free LE	14.83 (0.19)	14.84 (0.27)	15.19 (0.57)	−0.01 (−0.61, 0.53)	−0.36 (−1.39, 0.84)	−0.35 (−1.56, 0.89)
Disabled LE	4.84 (0.12)	4.16 (0.18)	3.77 (0.24)	0.68 (0.31, 1.07)	1.07 (0.52, 1.55)	0.39 (−0.25, 0.90)
Adjusteda
Total LE	20.28 (0.19)	19.09 (0.27)	18.57 (0.61)	1.19 (0.71, 1.72)	1.71 (0.55, 2.87)	0.51 (−0.74. 1.77)
Disability-free LE	15.43 (0.18)	14.77 (0.28)	14.67 (0.56)	0.66 (0.12, 1.18)	0.76 (−0.30, 1.91)	0.10 (−1.04, 1.36)
Disabled LE	4.85 (0.12)	4.31 (0.19)	3.90 (0.23)	0.53 (0.17, 0.94)	0.95 (0.40, 1.40)	0.41 (−0.21, 0.94)
Male
Unadjusted
Total LE	17.06 (0.19)	16.19 (0.30)	16.56 (0.33)	0.87 (0.17, 1.48)	0.51 (−0.23, 1.14)	−0.36 (−1.31, 0.45)
Disability-free LE	13.90 (0.18)	13.43 (0.30)	14.04 (0.33)	0.47 (−0.22, 1.07)	−0.14 (−0.87, 0.47)	−0.61 (−1.55, 0.21)
Disabled LE	3.16 (0.10)	2.76 (0.17)	2.51 (0.13)	0.40 (0.16, 0.69)	0.64 (0.39, 0.95)	0.25 (−0.08, 0.54)
Adjusteda
Total LE	17.59 (0.23)	16.48 (0.30)	16.85 (0.31)	1.11 (0.40, 1.73)	0.74 (−0.07, 1.39)	−0.37 (−1.28, 0.42)
Disability-free LE	14.52 (0.21)	13.64 (0.30)	14.29 (0.33)	0.88 (0.12, 1.45)	0.23 (−0.56, 0.81)	−0.65 (−1.49, 0.17)
Disabled LE	3.07 (0.10)	2.84 (0.11)	2.56 (0.14)	0.23 (0.03, 0.56)	0.51 (0.25, 0.88)	0.28 (−0.12, 0.53)

Note. CI = confidence interval; IADL = instrumental activities of daily living; LE = life expectancy.

^aAdjusted models control for race/ethnicity, coupleness status and household composition, education, cognitive status, depressive symptoms, vision impairment, hearing impairment, and number of chronic conditions.

Table 4.

Estimated Total, ADL Disability-Free, and ADL Disabled Life Expectancy at Age 65, by Sex and Religious Service Attendance (2004–2014)

	Frequency of attendance at religious services			Differences between religious groups
	At least once per week (weekly+)	Less than once per week (<weekly)	Never	Weekly+ vs <weekly	Weekly+ vs never	<weekly vs never
Gender and model	LE (SE)	LE (SE)	LE (SE)	Diff (95% CI)	Diff (95% CI)	Diff (95% CI)
Female
Unadjusted
Total LE	21.34 (0.27)	19.19 (0.35)	16.23 (0.31)	2.14 (1.48, 2.91)	5.11 (4.48, 6.03)	2.97 (2.36, 3.79)
Disability-free LE	15.71 (0.24)	13.68 (0.29)	11.38 (0.30)	2.03 (1.33, 2.81)	4.33 (3.60, 5.10)	2.30 (1.66, 2.99)
Disabled LE	5.63 (0.18)	5.51 (0.19)	4.85 (0.18)	0.12 (−0.27, 0.65)	0.78 (0.40, 1.31)	0.66 (0.23, 1.12)
Adjusteda
Total LE	21.97 (0.31)	20.11 (0.40)	17.59 (0.35)	1.86 (1.05, 2.76)	4.38 (3.61, 5.26)	2.52 (1.69, 3.26)
Disability-free LE	16.34 (0.29)	14.66 (0.33)	12.71 (0.33)	1.68 (0.91, 2.56)	3.63 (2.82, 4.50)	1.95 (1.19, 2.64)
Disabled LE	5.63 (0.18)	5.45 (0.21)	4.88 (0.19)	0.18 (−0.27, 0.65)	0.75 (0.25, 1.21)	0.56 (0.13, 1.04)
Male
Unadjusted
Total LE	18.52 (0.27)	17.39 (0.32)	15.11 (0.30)	1.13 (0.54, 1.85)	3.41 (2.71, 4.20)	2.28 (1.42, 3.18)
Disability-free LE	14.92 (0.30)	13.93 (0.35)	11.82 (0.29)	0.99 (0.25, 1.84)	3.10 (2.29, 3.90)	2.11 (1.21, 3.04)
Disabled LE	3.60 (0.14)	3.46 (0.15)	3.29 (0.16)	0.14 (−0.23, 0.55)	0.30 (−0.06, 0.73)	0.17 (−0.13, 0.58)
Adjusteda
Total LE	19.01 (0.33)	18.35 (0.34)	16.38 (0.33)	0.66 (0.06, 1.43)	2.63 (1.77, 3.69)	1.97 (1.03, 2.93)
Disability-free LE	15.36 (0.37)	14.88 (0.37)	13.12 (0.32)	0.49 (−0.26, 1.35)	2.24 (1.33, 3.30)	1.76 (0.76, 2.70)
Disabled LE	3.65 (0.15)	3.47 (0.15)	3.26 (0.16)	0.17 (-0.21, 0.58)	0.39 (−0.03, 0.83)	0.21 (−0.11, 0.62)

Notes. ADL = activities of daily living; CI = confidence interval; LE = life expectancy.

^aAdjusted models control for race/ethnicity, coupleness status and household composition, education, cognitive status, depressive symptoms, vision impairment, hearing impairment, and number of chronic conditions.

Table 5.

Estimated Total, IADL Disability-Free, and IADL Disabled Life Expectancy at Age 65, by Sex and Religious Service Attendance (2004–2014)

	Frequency of attendance at religious services			Differences between religious groups
	At least once per week (weekly+)	Less than once per week (<weekly)	Never	Weekly+ vs <weekly	Weekly+ vs never	<weekly vs never
Gender and model	LE (SE)	LE (SE)	LE (SE)	Diff (95% CI)	Diff (95% CI)	Diff (95% CI)
Female
Unadjusted
Total LE	21.41 (0.27)	19.32 (0.35)	16.27 (0.29)	2.08 (1.35, 2.86)	5.14 (4.49, 6.04)	3.05 (2.36, 3.90)
Disability-free LE	16.41 (0.25)	14.43 (0.31)	11.79 (0.29)	1.98 (1.28, 2.64)	4.61 (3.88, 5.44)	2.64 (1.92, 3.48)
Disabled LE	5.00 (0.21)	4.90 (0.20)	4.48 (0.18)	0.10 (−0.35, 0.66)	0.52 (0.14, 1.05)	0.42 (0.07, 0.83)
Adjusteda
Total LE	21.99 (0.31)	20.17 (0.38)	17.49 (0.33)	1.81 (0.94, 2.67)	4.49 (3.70, 5.32)	2.68 (1.81, 3.55)
Disability-free LE	17.01 (0.31)	15.31 (0.34)	13.07 (0.32)	1.70 (0.88, 2.50)	3.95 (3.20, 4.87)	2.24 (1.51, 3.03)
Disabled LE	4.97 (0.20)	4.86 (0.20)	4.42 (0.19)	0.11 (−0.37, 0.65)	0.55 (0.08, 1.02)	0.44 (0.04, 0.81)
Male
Unadjusted
Total LE	18.62 (0.27)	17.50 (0.30)	15.09 (0.30)	1.12 (0.60, 1.75)	3.53 (2.86, 4.39)	2.41 (1.52, 3.34)
Disability-free LE	15.20 (0.28)	14.39 (0.28)	12.29 (0.27)	0.81 (0.31, 1.41)	2.90 (2.28, 3.80)	2.10 (1.30, 3.03)
Disabled LE	3.42 (0.16)	3.11 (0.12)	2.80 (0.13)	0.31 (−0.02, 0.70)	0.62 (0.23, 0.97)	0.31 (−0.003, 0.50)
Adjusteda
Total LE	18.96 (0.31)	18.34 (0.31)	16.45 (0.33)	0.62 (0.03, 1.23)	2.51 (1.76, 3.65)	1.89 (1.00, 2.89)
Disability-free LE	15.53 (0.33)	15.19 (0.30)	13.75 (0.31)	0.34 (−0.29, 0.91)	1.78 (0.99, 2.92)	1.44 (0.57, 2.47)
Disabled LE	3.43 (0.16)	3.15 (0.13)	2.70 (0.14)	0.28 (−0.10, 0.65)	0.73 (0.25, 1.14)	0.45 (0.12, 0.73)

Note. CI = confidence interval; IADL = instrumental activities of daily living; LE = life expectancy.

^aAdjusted models control for race/ethnicity, coupleness status and household composition, education, cognitive status, depressive symptoms, vision impairment, hearing impairment, and number of chronic conditions.

Importance of Religion

In unadjusted models, both women and men who view religion as very important have significantly longer total LE at age 65 than those who view religion as somewhat important (0.68 years longer for women and 0.82 years longer for men, p < .05; Table 2). Differences in total LE between the very and not too important groups are similar in magnitude (0.75 for women and 0.65 for men), but are not statistically significant, and there are no significant differences between the somewhat and not too important groups.

Most of the gain in LE for those who view religion as very important is in years ADL disabled as opposed to ADL disability-free. Women who consider religion as very important have an estimated 0.77 and 1.00 extra years of ADL disabled life compared with women who view religion as somewhat or not too important, respectively (p < .05). In contrast, there is little or no difference across religiosity groups in ADL disability-free LE for women. Similar to women, most of the gain in total LE for men is in ADL disabled LE, with significant differences between the very important versus somewhat or not too important groups (0.49 and 0.48 years, respectively, p < .05).

Controlling for sociodemographic and health factors has a somewhat surprising effect with differences in total LE increasing in the adjusted model, as did those in ADL disability-free LE (Table 2). For example, differences in total LE between women who consider religion to be very important compared with those viewing religion as somewhat important increased from 0.68 (unadjusted) to 1.22 years (adjusted) and the corresponding difference in ADL disability-free LE increased from −0.08 (unadjusted) to 0.66 years (adjusted). Similar increases were found in the differences in total and disability-free LE between women considering religion to be very important versus not too important between unadjusted and adjusted models. Controlling for sociodemographic and health factors reduced the differences for ADL disabled LE for women slightly (from 0.77 to 0.56 years for very vs somewhat and from 1.00 to 0.85 years for very vs not too important), though the differences remained statistically significant in the adjusted model. Differences for the somewhat versus not too important group also increased slightly in the adjusted model, though none reached statistical significance. Taken together, after adjusting for sociodemographic and health factors, gains in total LE for women who view religion as very important (vs somewhat or not too important) are fairly evenly divided between ADL disability-free and ADL disabled life, with the gain of 1.22 years (total LE) for the very versus somewhat important groups breaking down into 0.66 extra years disability-free and 0.56 years disabled.

The patterns in the adjusted versus unadjusted models for men are essentially identical to those for women, except that the differences across religious groups tend to be somewhat smaller for men than for women.

Adjusted results for those who begin in a nondisabled state and those who begin in a disabled state largely mirror those for the overall sample in terms of levels and significance of differences across religious importance groups (Supplementary Table 2). Both the nondisabled and disabled at baseline experience gains in disability-free and disabled LE, though, for women, more of the gain for those initially nondisabled is in disability-free years, whereas for those initially disabled, slightly more of the gain is in disabled years. For men, however, the gain in disability-free LE outweighs that in disabled LE slightly regardless of initial disability state.

The patterns for IADL disability are remarkably similar to those for ADL disability, in terms of the direction, magnitude, and statistical significance of the differences (Table 3). In the unadjusted models, most of the gain in total LE for those who view religion as very important (vs somewhat or not too important) is in IADL disabled LE, and this is true for both women and men. Also, as with ADL disability, differences in IADL disability-free LE increased in the adjusted compared with unadjusted models.

IADL status-based results were broadly similar to those from population-based models (Supplementary Table 3), except that, among those initially disabled, the gain in IADL disability-free life for women who view religion as very versus somewhat important and the gains in IADL disability-free and disabled LE for men were not statistically significant. Moreover, when comparing the very versus somewhat important groups for both women and men, more of the LE gain is in IADL disability-free years, but when comparing the very versus not too important groups, more of the gain tends to be in disabled years.

Religious Attendance

Differentials in LE are much larger and more consistently significant for attendance at religious services. Men who attend services at least once per week live 1.13 years longer on average than those who attend services less than once per week and 3.41 years longer than those who do not attend services at all (Table 4). For women, the differentials are even larger: a 2.14 year advantage for those who attend at least once a week versus less than once a week and a 5.11 year advantage over those who never attend. Differences between those who attend services less than once per week and those who never attend are also statistically significant.

In contrast to the findings for importance of religion, before adjustment for sociodemographic and health factors, most of the gain for those who attend services more frequently is in disability-free as opposed to disabled LE. For men, the gains in ADL disabled LE associated with more frequent attendance are not statistically significant. For women, the gains in both disability-free and disabled LE are statistically significant for most comparisons, but when considered in proportionate terms, a much larger share of the gain is in disability-free life.

Controlling for sociodemographic and health factors reduces the differentials across attendance groups some, in contrast to that observed for importance of religion. However, differences mostly remain statistically significant apart from the difference in ADL disability-free LE for men who view religion as very versus somewhat important.

The differences in total and ADL disability-free LE remain as pronounced in the models that stratify by initial disability state (Supplementary Table 4), providing additional evidence that the findings are not driven by reverse causality, that is, that religious attenders are initially more active and healthy than nonattenders. Here, the only exception is that the gain of 0.30 years in total LE for men who attend services weekly versus less than weekly is no longer statistically significant for those who were initially ADL disabled, whereas it is significant for those who were initially nondisabled and for all men combined.

As with importance of religion, differences for IADL LEs by religious attendance (Table 5 and Supplementary Table 5) are very similar to those for ADL, in terms of both their magnitude and statistical significance and the impact of controlling for sociodemographic and health factors. The only differences we observed relate to disabled LE for men—whereas none of the differences in ADL disabled LE were statistically significant, the differences in IADL disabled LE were slightly larger and statistically significant between some of the groups.

Discussion

The links between religion and both health and mortality have been the focus of extensive research over the past few decades. Mortality studies have by and large found a salutary effect of religion on longevity, though findings from studies focusing on health outcomes have been more mixed depending on the specific health outcome examined and whether the association is estimated concurrently or over time. In contrast to most studies that have examined health and mortality separately, we quantify the association between religion and disability and mortality jointly by using a multi-state life table approach to estimate total, disability-free, and disabled LEs, additionally stratifying analyses by gender to compare associations for men and women.

Of the two religion measures investigated, attendance at religious services had the stronger and more consistent association with total and disability-free LE with a “dose–response” effect across increasing frequency of attendance. The difference between those who attend regularly and those who never attend is substantial: 5 years for women and 3.4 years for men, with most of the gain in LE being in years of disability-free, as opposed to disabled life. These findings were observed for both ADL and IADL disability.

Status-based models accounting for initial disability state allowed us to assess whether significant associations between religion and LE were a result of reverse causality, that is, that good health leads to more frequent religious attendance. We found this not to be the case as associations found for the full sample held in status-based models with the same monotonic pattern of more frequent religious service attendance being associated with significantly higher total and disability-free LE in those initially disabled for both ADL and IADL disability. This pattern and order of magnitude is also largely evident among those who were initially nondisabled. These results, along with the broad range of baseline health factors that are included in the models, lend confidence that the findings are not due to reverse causality.

We also observed significant differences in LE by perceived importance of religion, however, they are smaller and less consistent than those for religious attendance. Unlike religious attendance, for importance of religion, the pattern was not monotonic; individuals viewing religion as very important had longer total and disabled LE than those for whom religion is somewhat or not too important, though differences between the latter two groups were not significant. Additionally, whereas most of the gain in total LE by religious attendance was in disability-free years, for importance of religion, more of the gain is in disabled years. The results suggest that the association between religiosity and LE differs by the dimension of religiosity examined. Taking part in religion appears potentially more important than the internal feeling or conviction for religion, which may confirm the impact of social support and network integration, often considered to be a mechanism through which religion influences health (Pirutinsky et al., 2011; Strawbridge et al., 2001).

We find some evidence of mediating effects of health and sociodemographic factors in the religion–disability association, although these effects operated differently for the two measures of religion. For religious attendance, controlling for these factors reduced the differences in total, disability-free, and disabled LE proportionately across religious groups, although the differences remained statistically significant. For importance of religion, controlling for these factors resulted in larger gains in both total and disability-free LE and slightly smaller gains in disabled LE for those who consider religion to be very important. Again, this finding suggests that the mechanisms underlying the association with health and LE may differ across dimensions of religiosity. The fact that differences across religion groups remain significant after adjusting for sociodemographic and health factors suggests that there are other factors underlying this association that we have not accounted for.

Gender comparisons are a key focus of our article, and although the direction and significance of associations were similar for women and men, differences in total, disability-free, and disabled LE across religion groups tended to be larger for women than men, mostly by a factor of 1.5 to 2. Furthermore, the mediating effects of sociodemographic and health factors appeared stronger for men than women in terms of reducing the differences in life and health expectancies, suggesting religion is more beneficial for women than men, though again this may reflect the influence of social support and network integration. Literature tends to show that older women generally have more and stronger network connections than do men and benefit more from these connections (Shumaker & Hill, 1991). Nevertheless, the gender differences are intriguing and suggest the need for further work to explore the different mechanisms at play.

Our results are consistent with previous research suggesting that religion is related to a number of health outcomes in older populations, but we advance the literature in several ways. Different dimensions of religiosity have rarely been examined and compared. Gender differences are often observed, though few studies examine men and women separately. Moreover, we have examined how religiosity associates with mortality and disability simultaneously, determining the association with both quantity and quality of life. Further strengths of our study are: the use of between 10 and 16 years of panel data from a nationally representative and large sample of older adults in the United States; controls for key covariates; and examining ADLs and IADLs separately which allows us to extend previous work by comparing the associations across these two disability indicators.

One limitation of the study is that our measures of disability are fairly basic, namely difficulty with any ADL(IADL). It is unclear whether patterns found will hold for more refined disability measures (e.g., that take severity into account) or for other health indicators. A second limitation is that our estimation of life and health expectancy at older ages is based on individuals who not only survive to those ages, but who also take part in a longitudinal survey. As a result, the sample on which our analysis is based is selective in terms of health and survival. This selectivity might attenuate the association between religion and health/mortality. Research focusing on the religion-health association in young and middle-aged adults would be valuable. Another potential limitation is that our measures of religion pertain to behaviors and beliefs at one point in time. Although we may not expect much shift in attendance patterns and importance of religion during later life, it is possible that religious practices and beliefs earlier in life have a lasting (and perhaps independent) effect on health in later life. In future work, we plan to incorporate measures of religious attendance and affiliation at different points over the life course.

Despite these limitations, the strength and consistency of the associations we observed between religious attendance and total and disability-free LE give us confidence in concluding that religious attendance is associated with increased quantity and quality of life. Importance of religion is also associated with increased quantity of life, though the findings regarding quality of life are more mixed. Further research is needed to understand the mechanisms underlying these associations and explore gender differences in more depth. Additionally, it would be beneficial to replicate this study in other settings with different religious traditions and practices to gain a better understanding of the generalizability of our findings.

Supplementary Material

Supplementary data are available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Funding

This work was supported by a grant from the John Templeton Foundation (57521), “Linking spirituality and religiosity to life and health expectancy: A global comparative study.”

Author Contributions

M. B. Ofstedal helped plan the study, prepared the data, conducted descriptive data analysis, and had the lead role in writing and revising the article. C.-T. Chiu helped plan the study, conducted the life expectancy analysis, and helped write/revise the article. C. Jagger helped plan the study and write/revise the article. Y. Saito helped plan the study, conduct data analysis, and write the article. Z. Zimmer helped plan the study and had a substantial role in writing/revising the article.

Conflict of Interest

None reported.