Religion and Mental Health: Is the Relationship Causal?

Perhaps the best studied outcome concerning the relationship between religious participation and mental health concerns depression. In a comprehensive review of the literature, Koenig et al. (2023) discuss more than 20 rigorous longitudinal studies, with control for baseline depression, with the vast majority indicating a protective association between religious participation and subsequent depression, mostly with data from the USA, but also from the Netherlands, Australia, Canada, the UK, and Chile, along with one large study with participants from 10 European countries (Croezen et al., 2015). A recent systematic review (Balboni et al., 2022) of the religion and health literature between 2000 and 2022 likewise indicated that of 32 longitudinal studies of the relationships between religion/spirituality and depression, with control for baseline outcomes and with a sample size of at least 1000, 77% of the studies evaluating religious service attendance found notable evidence for a protective association with depression. A meta-analysis of longitudinal studies with control for baseline confounding and baseline depression (Garssen et al., 2021; VanderWeele, 2021a) indicated that service attendance was associated with 33% reduction in the odds of depression incidence (OR = 0.67; 95% CI, 0.58–0.81).

There is also considerable evidence relating religious participation and suicide. Koenig et al. (2023) report a number of rigorous longitudinal studies indicating a protective association between religious participation and suicide or suicide behavior and ideation. In the aforementioned systematic review of the literature (Balboni et al., 2022) of longitudinal studies with control for baseline outcome measures and sample sizes of at least 1000, three out of four studies indicated that religious service attendance was associated with lower suicide. While there are a limited number of large rigorous studies, the studies that have been carried out provide strong evidence and indicate large effects. For example, VanderWeele et al. (2016) report weekly religious service attendance associated with an 84% reduction in the risk of completed suicide (HR = 0.16; 95% CI, 0.06–0.46) over 16 years.

In contrast with the associations relating religious participation to depression and to suicide, the evidence concerning anxiety is more mixed. In the systematic review of Balboni et al. (2022), anxiety is not among the outcomes for which the evidence was rated as strong, with only half of the studies suggesting evidence for a protective association. In addition, some of the largest longitudinal studies of religious participation and anxiety indicate either little evidence (Li et al., 2016) or fairly small effect sizes (Chen et al., 2020). It may be the case that religious participation decreases anxiety and provides peace for some individuals or in some contexts but increases anxiety for or in others.

The evidence is relatively stronger once again for the relationship between religious participation and substance use. Koenig et al. report (2023) report numerous rigorous longitudinal studies indicating a protective association between religious participation and lower smoking, alcohol consumption, and drug use. In their systematic review of studies on religion/spirituality and health, Balboni et al. (2022) report that of 38 longitudinal studies with control for baseline outcomes and sample sizes of at least 1000, the vast majority of these indicate protective associations with smoking, alcohol abuse, and drug use.

There is likewise considerable evidence for moderately strong longitudinal associations between religious participation and various aspects of mental well-being including happiness, life satisfaction, meaning, purpose, and numerous other health and well-being outcomes also (Balboni et al., 2022; Chen et al., 2020; Koenig et al., 2023). Much of the longitudinal research on religion and well-being concerns life satisfaction, and Balboni et al. (2022) report that of 10 longitudinal studies with control for baseline outcomes and sample sizes of at least 1000 focused on life satisfaction, 7 of these (70%) indicate a beneficial association. Less longitudinal research has been carried out with regard to meaning and purpose, but the existing longitudinal research does also suggest a beneficial association (Chen et al., 2019, 2020; Krause & Hayward, 2012).

The evidence for associations between religious participation and suicide, depression, and substance use is especially strong. A summary of key studies is found in Table 1. Further evidence synthesis through quantitative meta-analysis restricted to more rigorous longitudinal studies with control for baseline outcome is still needed as this has thus far effectively only been carried out for depression (and for mortality for physical health) (Chida et al., 2009; Garssen et al., 2021; VanderWeele, 2021a). However, even at present, the various studies, systematic reviews, and meta-analyses employing longitudinal designs, with large sample sizes, extensive confounding control, and control also for baseline outcomes (VanderWeele, 2021b) provides some, albeit not definitive, evidence that the relationship between religious participation and these mental health outcomes is causal.

Further evidence comes from sensitivity analysis concerning the robustness of these associations to potential unmeasured confounding. Sensitivity analysis for unmeasured confounding (VanderWeele & Ding, 2017; VanderWeele, 2021a) of the meta-analysis concerning religious service attendance and depression (Garssen et al., 2021) was carried out. This work indicates that to explain away the longitudinal association between religious service attendance and depression, an unmeasured confounder that was associated with greater service attendance and with lower depression across studies by risk ratios of 1.74 each, above and beyond the measured covariates, could suffice, but weaker joint confounder associations could not. To shift the lower confidence interval of the estimate to the null, an unmeasured confounder that was associated with greater service attendance and with lower depression across studies by risk ratios of 1.46 each could suffice, but weaker joint confounder associations could not. Similar but yet stronger robustness to potential unmeasured confounding also pertains to the associations between religious service attendance and suicide. VanderWeele et al. (2016) report that for an unmeasured confounder to explain away the HR estimate of 0.16 (95% CI, 0.06–0.46), an unmeasured confounder associated with both increased service attendance and decreased suicide incidence by risk ratios of 12-fold each, above and beyond the measured confounders, could suffice, but weaker joint confounder associations could not. To bring the estimate’s upper confidence limit to the null, an unmeasured confounder associated with both increased service attendance and decreased suicide incidence by risk ratios of 3.7-fold each, above and beyond the measured confounders, could suffice, but weaker joint confounder associations could not.

Yet further evidence for causality for these relationships comes from quasi-experimental study designs within the economics literature. With regard to depression, quasi-experimental designs, using within-school variation in adolescents’ peers’ religiosity as an instrumental variable, support a causal relationship between religious participation and lower rates of depression (Fruehwirth et al., 2019). Likewise with regard to death by suicide, poisoning, and alcoholic liver disease (sometimes called “deaths of despair”), quasi-experimental designs, using the repeal of “blue laws” that restrict business activity on Sundays as an instrumental variable, support evidence for a protective causal effect of service attendance on the incidence of these “deaths of despair” (Giles et al., 2023). Such quasi-experimental and instrumental variable designs are themselves subject to assumptions to establish causation, but a different set of assumptions than the observational longitudinal research. That the results seem to be relatively consistent across these different designs and assumptions strengthens evidence yet further (VanderWeele, 2021b).