Good Faith and Bad Health: Self-Assessed Religiosity and Self-Assessed Health of Women and Men in Europe

To mention but a few: economic growth (Barro and McCleary 2003; Durlauf et al. 2012), trust (Johansson-Stenman et al. 2009; Berggren and Bjørnskov 2011), the size of the shadow economy (Schneider et al. 2015), criminality (Brauer et al. 2013), tolerance (Berggren et al. 2019), the work ethic (van Hoorn and Maseland 2013) and entrepreneurship (Wiseman and Young 2014).

For examples of the diversity of the literature, see, e.g. Ellison and Levin (1998), Idler et al. (2003), Ellison and Hummer (2010), Hill and Cobb (2011) and Zimmer et al. (2019). Baker et al. (2018) identify a complex pattern between religious/secular identity and health outcomes in the United States. They find that “atheists report significantly fewer problems with physical health, mental health and pain than both nonaffiliated theists and some religious individuals” and that “[a]theism is the religious or secular identity with the healthiest outcomes on the basic psychiatric symptoms analyzed” (p. 53). Thus, while many affiliated theists benefit health-wise from their religiosity, atheists likewise seem to benefit from their (secular) identity, and sometimes even more so.

A key concern is reverse causality—e.g., a negative relationship between religiosity and health could reflect sick people seeking religious comfort and consolation (Ferraro and Kelley-Moore 2000).

We use the terms “self-assessed” religiosity and “self-assessed” health, since these are indicators from surveys based on people’s own assessments of how religious and how healthy they are. “Subjective” or “self-rated” are synonyms to “self-assessed”, used in other studies.

Nicholson et al. (2010: 233) point out that “… as this was a cross-sectional study, the potential for reverse causality is a major problem. … [R]everse causality is an important alternative explanation of our findings. These analyses could equally well be argued to show that poor health reduced attendance at religious services.”.

This study forms part of an emerging body of longitudinal studies on religion and health that overcome some of the methodological challenges of cross-sectional studies but which predominantly use U.S. data (VanderWeele 2017b). See, e.g., VanderWeele (2017c) and Saad et al. (2019).

The data from these rounds are stacked. The first round does not include information on parental-birth country.

We also have data for Turkey and Israel, but to keep the sample more homogeneous, we do not include them in the main analysis. We have conducted all tests with them included, and there is no qualitative change in our findings (available upon request).

The individuals in our sample are similar to the general population in the countries in which they were born on observables, including self-reported health (Ljunge 2016).

The importance of the mother in transmitting cultural traits like religiosity has been shown to hold for social trust as well (Ljunge 2014a). Note that we cannot, due to a lack of data, differentiate between mothers who migrated to their present country of residence early and those who did so later in life.

By using this sample with immigrant mothers we also relate to a literature on migrant health in Europe (Huijts and Kraaykamp 2012; Rechel et al. 2013), which we complement by explicitly focusing on the role of religiosity.

In a robustness test, we code the replies from 1 (“very bad”) to 5 (“very good”) and conduct a linear regression analysis (see the Sect. 4.2).

In a robustness test, we also account for individuals’ overall life satisfaction in several ways, in order to capture cultural differences in attitudes to life, which could contribute to comparability of the subjective-health data.

The EVS/WVS variables are averaged for two reasons: Long-time averages are arguably better measures of persistent levels, and using many waves greatly increases the number of countries available.

We use the dataset compiled by Samanni et al. (2010). Using Gini data from The Standardized World Income Inequality Database yields similar results.

The IQ has been validated by Lynn and Meisenberg (2010) as a measure of cognitive ability.

For more information about this method, see pp. 2–3 of the supplementary material.

There are more studies showing that religion is transmitted in families (Bisin and Verdier 2000; Bengtson et al. 2013; Jacob and Kalter 2013; van de Pol and van Tubergen 2014).

The standardized effect is computed as the estimate for religiosity times the standard deviation of religiosity as a fraction of the standard deviation of health.

The significant estimate on the tertiary education of the father is not a robust finding.

The correlation between self-assessed health and the objective health measure life expectancy in the mothers’ birth countries is 0.76, indicating that both measures capture an objective dimension of health.

There are also indications that cognitive skills are transmitted from parents to children (Coneus et al. 2012).

For example, Deaton (2011) remarks that time preference, captured by patience here, could affect both religiosity and health.

In addition, we have investigated whether effects differ depending on whether countries of residence are Catholic or not, and we do not find any statistically significant difference. Hence, it seems as if is religiosity as such, rather than belonging to a particular denomination, that is important.

Russia and Germany are the two largest ancestral groups. Scandinavians could be particularly secular, while Turkish ancestry could be particular since most are Muslims and Europe is overwhelmingly Christian. The Baltics are excluded since most immigrants have Russian ancestry.

There is not much variation in the kind of religion that dominate the mothers’ birth countries: Since they all stem from Europe, almost all are from nations characterized by Christianity. When checking, through sample restrictions, whether Protestant, Catholic or Orthodox Christianity display differential effects, we do not find any evidence that they do—it is the strength of the religious commitment or belief that matters.

This exercise also addresses concerns that people do not reply honestly to surveys about religiosity in very religious countries out of fear. Since our results hold when excluding the most religious countries, we consider it likely that the results are not driven by such behavior.

The result is similar when running the ordinary least squares model of column (3), Table 2, separately for women and men: strongly significant and positive estimate for women and close to zero and insignificant for men. We also tested the same model using the entire sample of the ESS (185,000 women and 159,000 men), and again found a positive estimate for how religiosity is related to bad health for women and a close-to-zero estimate for men. Results are available on request.

Due to the smaller samples when separating the genders, the first-stage results are estimated less precisely, which in turn makes it challenging to perform additional robustness checks in the gender subsamples. The point estimate on religious degree becomes higher for women and so does the standard error, resulting in a wide confidence interval for the true effect. The point estimate for men is close to zero with a standard error like the baseline specification, consistent with a zero true effect.

The theoretical idea, and the results, are also broadly consistent with studies that indicate that individuals with strong religious commitments are less likely than secular individuals to hold egalitarian gender-role attitudes (Diehl et al. 2009) and that more religious societies are less gender equal (Klingorová and Havlíček 2015; Schnabel 2016). Furthermore, it is certainly conceivable that factors that previous studies have shown can explain how religiosity affects health negatively could affect women more strongly than men, but we cannot test them with our data. For example, a negative effect of religiosity on health could result from religious people being less concerned with life on earth, including physical exercise (Feinstein et al. 2010; Kortt and Dollery 2014), being more “fatalistic” and not dealing with negative developments (Azaiza et al. 2010), having been subjected to bad treatment as children, e.g., corporal punishment, genital mutilation, sexual assault or no vaccination (Grogan-Kaylor and Otis 2007; Jegede 2007; Terry 2008; Gershoff 2010; Cappa et al. 2013) or being more mentally strained by internal or external conflicts (Exline 2002; Ellison and Lee 2010; Hill and Cobb 2011).

Still, it bears noting that for generalizability to make sense, one needs to have obtained an interpretable estimate, and we believe we have obtained an estimate that we can interpret as an effect of religiousness on health. This contrasts with much of the literature, as it does not address reverse causality, and for that reason, correlations obtained without an ability to rule out reverse causality cannot readily be generalized in a meaningful way. Hence, it can arguably be claimed that we are in a better position to offer generalizability even with an inclusion of mothers from another European country.