Does Religiosity Correlate With Better Social Outcomes?

The individual-level evidence is moderately strong and consistent in direction. Baier and Wright’s foundational meta-analysis (2001, Journal of Research in Crime and Delinquency) synthesized 60 studies producing 79 effect size estimates from research spanning 1962–1998. They report a mean effect of r = −0.12 with a standard deviation of 0.09, a median of −0.11, and about two-thirds of effects falling between −0.05 and −0.20. Their own summary is that religious beliefs and behaviors exert a moderate deterrent effect on criminal behavior. Not one of the 79 effect sizes was positive.

A later meta-analysis by Kelly, Polanin, Jang, and Johnson (2015, Criminal Justice Review 40(4):505–523) drew on 62 studies encompassing 193,656 adolescents across four decades and found somewhat larger effects, with bivariate correlations ranging from r = −0.16 to −0.22 across religious salience, church attendance, and various delinquency outcomes. Cheung and Yeung (2011, Children & Youth Services Review, 40 studies published 1995 to 2009, mean age 16.45) report a weak overall effect of Zr = 0.19. That figure is a Fisher’s z rather than a raw correlation, though at this magnitude the two are close, since Zr = 0.19 back-transforms to r = 0.188. It also pools constructive and destructive behaviour rather than measuring crime, and the authors report religious involvement relating more strongly to constructive behaviour than to destructive behaviour. The positive sign therefore does not read against the negative delinquency correlations above it.

All three of those meta-analyses assume the relationship is linear. Recent work says it is not. Drawing on six German surveys of more than 45,000 adolescents between 2013 and 2024, a 2026 paper in Justice Quarterly finds a U-shaped relationship between religiosity and violent offending: the least and the most religious adolescents both report more of it than those in between. For substance use the protective effect appears, as the older meta-analyses found, but it weakens at the highest levels of religiosity rather than continuing to climb. For property offences there is no association at all.

This does not overturn Baier and Wright. It is one country, one age band, and self-reported offending. But a curve and a straight line are different claims about the world, and a mean correlation of −0.12 cannot tell them apart. The honest position is that the direction is established and the shape is not.

The missing positive tail is the problem. Philosopher Eric Schwitzgebel took Baier and Wright’s own reported mean and standard deviation, generated 10,000 sets of 79 normally distributed effect sizes, and found that only 9 of them, 0.09%, came out entirely zero or negative. Either the true mean is more negative or the spread narrower than reported, or the distribution is not normal, or studies with positive effects were not published. This analysis is a post on his blog The Splintered Mind (11 April 2014), not a peer-reviewed paper, which is worth stating given that the finding it qualifies is a journal meta-analysis. The arithmetic is reproducible from the figures Baier and Wright publish.

Rodney Stark’s “moral communities” hypothesis (1982; elaborated 1996) argued that religion deters crime most effectively where collective religious commitment is high. Testing this in Provo, Utah versus Seattle, Washington, Stark found the church attendance–delinquency correlation was gamma = 0.45 in Provo versus gamma = 0.13 in Seattle. The two gammas come from the 1982 paper with Kent and Doyle; the 1996 paper tested the same idea on a national survey of high school seniors. The hypothesis has also not held up uniformly since. Using geocoded General Social Survey data, a 2012 test in the Journal for the Scientific Study of Religion found the individual-by-community interaction Stark predicts was not supported for any form of deviance examined, even while religiosity itself remained inversely related to several of them. Stansfield and Mowen (2019, Journal of Quantitative Criminology 35:493–516, N = 1,362 previously incarcerated people) found county-level religious adherence predicted lower recidivism while an individual’s own religious involvement did not, once context was accounted for. The same SVORI data has produced two other results worth putting beside it. Stansfield, Mowen, O’Connor and Boman (2017) found religious support did not significantly relate to reoffending at all, though it did predict desistance from substance use among those at low risk of recidivism. Mowen, Stansfield and Boman (2018) found neither baseline religious support nor change in it related to offending on its own, and only the interaction between the two did. One dataset, three papers, and the individual-level effect keeps failing to appear except in interactions.

At the societal level, the pattern flips. Gregory Paul’s cross-national analysis (2005, Journal of Religion and Society 7) of 17 prosperous democracies found the United States, the most religious wealthy nation, had among the highest homicide rates, while secular nations like Japan, Sweden, and Norway had the lowest. Paul’s method was scatter diagrams and bivariate correlations with no controls, and that is the whole of it. When Moore and Rees redid the exercise in 2010 (Evolutionary Psychology 8(4)) across the fifty states plus DC, using multiple regression, the support for Paul’s hypothesis came out minimal: education, income and race carried far more of the variance than religiosity did. Their verdict was that once the methodological problems are accounted for, Paul’s conclusions do not survive. Three statisticians asked to review the original for MSNBC concluded that the data was sound and the critics still had a point, which is roughly where it has stayed. Phil Zuckerman (2009, Sociology Compass) documented that more secular U.S. states like Vermont, Oregon, and New Hampshire have lower murder rates than highly religious states like Louisiana, Alabama, and Mississippi. Heaton (2006, Journal of Law and Economics 49(1):147–172) replicated the usual negative OLS result on county-level membership and crime data, then instrumented current religious participation with historic religiosity. The effect went negligible. He also found the arrow running the other way, from crime to religious participation, which is what the OLS estimate had been absorbing. Then he checked it a second way, comparing crime before and after Easter, and found no drop. Two methods, same answer.

The finding that religious Americans donate more to charity is among the most replicated in this literature. Using the Social Capital Community Benchmark Survey (~30,000 respondents, 2000), Arthur Brooks reported in Who Really Cares (2006) that religious people were 25 percentage points more likely to donate money, 91% against 66%, and gave about 3.5 times more, $2,210 against $642 for otherwise comparable households in 2000. Both figures are exact, and both need their definitions. Brooks counts as religious those attending a place of worship at least weekly, roughly 30% of the population, and as secular those who do not believe in a deity or attend once a year or less, roughly 20%. The comparison is between two tails and leaves out the half of Americans in between. The 3.5× also includes giving to the donor’s own congregation. Strip religious giving out and Brooks’s own figure for the remaining gap in donations to non-religious charities is $88 a year. Pew Research Center’s 2016 Religious Landscape Study found 65% of the highly religious donated money, time or goods to help the poor in the past week, against 41% of other adults, and 45% volunteered against 28%. “Highly religious” here means praying daily and attending services weekly, about 30% of US adults, measured against the other 70%. Pew’s own follow-up question locates the volunteering gap: 23% of the highly religious volunteered mainly through a church or other religious organisation, against 4% of everyone else. Subtract that and the shares volunteering through any other kind of institution are similar, 22% against 24%. The gap in the headline figure is church volunteering.

The picture changes substantially when you distinguish between religious and secular giving. The National Study of Congregations’ Economic Practices (Lake Institute on Faith and Giving, 2019, N = 1,231 congregations) reports that congregations spend nearly three-quarters of their budgets on salary and facilities combined, with 23% on buildings and utilities and 21% on missions and programs together. Within that 21%, NSCEP splits 11% for mission activity outside the congregation and 10% for specific program expenses, a category that includes worship activities during services. The study publishes no separate figure for direct charitable benevolence. With total giving to religion at $146.5 billion in 2024, and $151.6 billion in 2025, the roughly 72% going to salary and facilities is on the order of $105 billion a year, and the 11% for outward mission is on the order of $16 billion. How much of that reaches direct charitable services is not something NSCEP measures.

Putnam and Campbell’s American Grace (2010, Simon & Schuster) rests on the 2006 Faith Matters Survey, an hour-long telephone survey of roughly 3,100 American adults (ICPSR 36315, response rate 53%) with a panel re-interview six to nine months later. It confirmed the giving gap, then took apart the theological explanation for it. Zero of 25 theological beliefs correlated with the giving gap. The mechanism was entirely social: having “supercharged friends” at church. Dense congregational networks create peer pressure to give. Berry, Glaser, and Schildkraut (Everyday Democracy, University of Chicago Press, 2025, N = 1,000) found liberals were more likely to give to health, social service, and secular nonprofits, while conservatives’ giving advantage was driven almost entirely by congregational donations.

Tyler VanderWeele’s research group at the Harvard T.H. Chan School of Public Health has produced the most methodologically rigorous evidence for health benefits of religious attendance. Their landmark study (Li, Stampfer, Williams, and VanderWeele, 2016, JAMA Internal Medicine) followed 74,534 women in the Nurses’ Health Study from 1996 to 2012 and found that attending religious services more than once per week was associated with 33% lower all-cause mortality (HR = 0.67, 95% CI: 0.62–0.71). Cardiovascular mortality was 27% lower (HR = 0.73) and cancer mortality 21% lower (HR = 0.79). The study has the dose-response a reader should want: more than weekly gives 33%, weekly alone 26%, less than weekly 13%. It also reports the effect in absolute terms, which the relative figures do not convey. Across the sixteen-year follow-up, the most frequent attenders lived on average about five months longer.

VanderWeele, Li, Tsai, and Kawachi (2016, JAMA Psychiatry 73(8):845–851) followed 89,708 women from 1996 to 2010 and found weekly attendance associated with a roughly fivefold lower suicide rate, in the paper’s own words (HR = 0.16, 95% CI: 0.06–0.46). Read the base before the ratio: 36 suicides occurred across the whole cohort, which is what the confidence interval spanning 0.06 to 0.46 reflects. The result also splits sharply by denomination, HR = 0.05 for Catholics against 0.34 for Protestants, and the Protestant interval (0.10–1.10) includes no effect. The cohort is female nurses, predominantly white and Christian. Major meta-analyses confirm the direction. Yaden et al. (2022, Journal of Happiness Studies 23(8):4147–4163, 256 effect sizes, N = 666,085) found an overall correlation of r = 0.18 (95% CI 0.16–0.19) with life satisfaction. That headline pools two things that behave differently. Split out, religiosity gives r = 0.16 across 124 studies and 427,834 people; spirituality gives r = 0.30 across 34 studies and 24,447. The larger effect sits on about one-eighteenth of the sample, and it is the measure with the least to do with congregations. Poorolajal, Goudarzi, Gohari-Ensaf and Darvishi (2022, Journal of Research in Health Sciences 22(1):e00537, 63 studies, 8,053,697 participants) found an odds ratio of 0.31 for completed suicide (95% CI: 0.14–0.72). The confidence interval is wide, and the same paper finds much smaller associations for the more common outcomes: suicidal ideation OR = 0.83, suicide attempt OR = 0.84, and suicide plan OR = 0.93, which does not reach significance.

Three critical caveats temper these findings. First, Garssen, Visser and Pool (2021, International Journal for the Psychology of Religion 31(1):4–20) pooled 48 longitudinal studies across 59 independent samples and got r = 0.08, with a confidence interval of 0.06 to 0.10. The interval is narrow, so this is not a small effect that might be larger. It is a small effect measured precisely. They also separated eight different religion and spirituality predictors, and only two came out significant: participation in public religious activities (r = 0.08) and importance of religion (r = 0.09). The other six showed nothing. Their own reading is blunter than “small but significant”: they argue the effect is not substantial enough to be worth pursuing further, which drew published commentaries disputing exactly that. A 2024 panel study of 8,146 German adults found “almost no evidence” that changes in religiosity predicted future mental health changes.

Second, the “healthy-attender bias” is real: people who maintain routines (church, medication, exercise) may differ in personality, health behaviors, or social support in ways that drive health gains. Richard Sloan of Columbia (1999, Lancet; 2000, NEJM) argued many studies have weak methods and show only associations.

Robert Putnam’s Bowling Alone (2000) established that religious affiliation accounts for roughly half of all U.S. social capital. His follow-up with David Campbell, American Grace (2010, Faith Matters Survey, N = 3,108 with 1,909 re-interviewed), found religious Americans are 3–4 times more generous and substantially more civically engaged, 34% active in charitable organizations versus 15% for the non-religious.

General Social Survey data analysed by Valente and Okulicz-Kozaryn (2021, Review of Religious Research 63(3):343–379) shows a split: social religiosity (attendance, church membership) predicts more trust and less misanthropy, while individual religiosity (prayer, closeness to God, belief in God) predicts lower trust and more misanthropy. The two correlate and pull in opposite directions, which is why studies measuring only one of them disagree. The authors’ own gloss is that connecting with God appears to displace connecting with people. One limit on that, from the same authors on the same data: the pattern is not uniform across racial groups. Black and Latino weekly attenders show higher misanthropy, white weekly attenders lower. Communal religiosity builds trust; private belief without community does the opposite.

Whether religious prosociality extends beyond in-group boundaries is where evidence becomes most contested. Luke Galen (2012, Psychological Bulletin 138(5):876–906) argued that religious prosociality is largely a congruence fallacy reflecting stereotypes, self-report bias, and in-group favouritism. His sharpest methodological point is about the comparison group: studies routinely set high religiosity against “low religiosity”, meaning no denomination, no attendance, or religion rated unimportant, which lumps indifferent believers in with the actually nonreligious. His argument did not go unanswered. Myers and Saroglou published commentaries in the same journal, and Galen replied to both (138:918–923). Saroglou’s objection is that Galen moved from the known limits on the religion–prosociality link to denying the link exists.

Experimental evidence tends to support the parochialism critique. Hallin, Moche, Andersson and Västfjäll (2024, Judgment and Decision Making 19:e16, three studies, N = 1,719, in the USA, Sweden, Egypt and Lebanon) ran a dictator game in which players allocated money to a Christian, a Muslim and an atheist. When religious affiliation was visible, religious participants were more generous than non-religious ones. When it was hidden, that difference disappeared. All three groups gave more to their own, atheists included. The groups were not equally parochial, though. Muslims gave more to their ingroup than Christians or atheists gave to theirs, in the US sample and probably the Swedish one. Parochial generosity is the general pattern; its strength is not the same everywhere. Everett, Faber, and Cikara (2017, N = 621) found Christians showed in-group bias while atheists behaved impartially.

A partial reconciliation comes from Preston and Ritter (2013, Personality and Social Psychology Bulletin), who found that priming “religion” (institutional identity) enhanced in-group prosociality, while priming “God” (universal moral concept) enhanced out-group prosociality. The institutional and theological dimensions of religion may pull in opposite directions.

The cross-national data presents perhaps the most striking, and most misinterpreted, pattern in this literature. Phil Zuckerman’s Society without God (New York University Press, 2008, based on 149 in-depth interviews in Denmark and Sweden) documented that these countries enjoy among the lowest crime rates, lowest corruption, strongest economies and highest life satisfaction globally, while being about as irreligious as anywhere on earth. His figure of fewer than 25% believing in a personal God is exact, and it is narrower than it looks: it describes Danes and Swedes who identify as Christians, not the whole population. Alongside it he gives 10% believing in hell and 7% treating the Bible as the literal word of God. This page had also said 3–7% attend church weekly. About 3% of Danes attend regularly, which is well attested, but the 7% upper bound matches nothing on attendance and does match Zuckerman’s Bible figure, so the range looks assembled rather than measured.

Mapping Gallup World Poll religiosity data against the UN Human Development Index, Coyne and Roy (Rensselaer Polytechnic Institute, 2012) found a U.S. state-level correlation of r = −0.669, strongly negative. Countries where over 95% say religion is important cluster at the bottom of HDI rankings; countries where fewer than 20% say so cluster at the top. The United States sits anomalously among wealthy nations, with both high religiosity and relatively poor outcomes on several measures. Its homicide rate was 5.8 per 100,000 in 2023 on World Bank and UNODC figures, against about 1 for Denmark. US life expectancy reached 79.0 years in 2024, its highest recorded level, up from 78.4 in 2023; Sweden was about 83.3 in 2023 and 84.1 in 2024. The gap is real and it is narrowing: 4.1 years behind comparable countries in 2023, 3.7 in 2024. The Gini comparison of 0.39 against 0.28 is left as it was, because the number depends entirely on whether it is measured before or after taxes and transfers and this page does not say which.

The direction-of-causation question is addressed most comprehensively by Pippa Norris and Ronald Inglehart’s Sacred and Secular (Cambridge University Press, 2004/2011), drawing on World Values Survey data from 80+ societies. They proposed the existential security thesis: people raised in conditions of insecurity (economic instability, poor healthcare, absent social safety nets) develop stronger need for religion, while those raised in security do not. Solt (2014) provided temporal evidence: rising inequality in one year predicts rising religiosity the next, but changes in religiosity do not predict subsequent changes in inequality. The causal arrow runs from dysfunction to religion, not the reverse.

The country-level data is unambiguous. Stoet and Geary (2017, Intelligence 62:71–78) combined PISA (2000–2015) and TIMSS (2003–2011) scores with World Values Survey religiosity data across 50–79 countries and found correlations of r = −0.65 to −0.74 between national religiosity and science/math performance. Rachmatullah, Ha and Park (2019, South African Journal of Education 39(1), article 1611, 68 countries) found the strongest single pairing was between PISA science scores and the importance of God, r = −0.759. On composite scores the education–religiosity correlation was −0.793, and −0.747 among the 42 Christian-majority countries. Two of their results matter more than the headline. The correlation survives controlling for socioeconomic status, falling only to −0.701. But the religiosity–socioeconomic correlation collapses when education is controlled, from −0.532 to −0.142, which is not significant. Their path model therefore runs socioeconomic conditions to education to religiosity, with education as the mediator rather than wealth acting on belief directly.

At the individual level, however, the picture reverses or becomes negligible. The landmark meta-analysis by Zuckerman, Silberman, and Hall (2013, Personality and Social Psychology Review, 63 studies, ~70,647 participants) found intelligence–religiosity correlations ranging from r = −0.20 to −0.25. A later and larger meta-analytic update by Dürlinger and Pietschnig (2022, PLOS ONE 17(2):e0262699, 89 studies, 105 effect sizes, 201,457 participants) found a smaller association, r = −0.14, which they describe as small but robust. The direction holds across both; the magnitude roughly halves on the larger base. Either way these are small effects, explaining between 2% and 6% of variance.

More, within the United States, religiosity often predicts better academic outcomes at the individual level. A sibling study using the National Longitudinal Study of Adolescent Health (Horwitz, Domingue and Harris, 2020, Social Science Research 88–89:102426) compared siblings within the same family, which holds constant the family background that ordinary controls miss. The more religious sibling earned a higher high school GPA even with family fixed effects, and completed more years of education fourteen years later. The second result runs through the first: the attainment gap is there because of the GPA gap, not alongside it. Meisenberg, Rindermann, Patel and Woodley of Menie (2012, Temas em Psicologia 20(1):101–120), working from World Values Survey responses from 345,743 people in 96 countries, put it as a weak negative relationship at the individual level that is culturally amplified into far larger differences between countries. Their individual-level numbers are worth having in full. Education explains an average of 0.65% of the variance in religiosity within countries. The partial correlation is negative in 70 countries, positive in 25, and zero in one, and in sub-Saharan Africa it is more often positive than negative. So the country-level pattern this section is built on does not rest on a strong individual-level effect being aggregated. It rests on a very weak one, pointing different ways in different places. That is the ecological fallacy with the numbers attached.

Income inequality: Wilkinson and Pickett’s The Spirit Level (2009) documented that 11 categories of health and social problems are significantly worse in more unequal developed countries. Solt (2014) showed rising inequality predicts rising religiosity but not the reverse.

Welfare state strength: Gill and Lundsgaarde (2004, Rationality and Society) found a strong negative relationship between welfare spending and religious participation. Strong welfare → less need for religion → less religiosity; strong welfare → better outcomes.

Self-report bias: Hadaway, Marler, and Chaves (1993, American Sociological Review) found actual church attendance was roughly half of reported levels. Brenner (2011) confirmed this with time-diary data from 14 countries. The key independent variable in most U.S. studies is substantially inflated.

What remains uncertain is whether communal religious participation provides any benefits beyond what equivalent secular social engagement would provide, and whether the Scandinavian model of prosocial secularism is generalizable beyond its specific cultural and institutional context. The research community does not yet have definitive answers to either question, but the trajectory of evidence increasingly suggests that what religion provides is community, and community need not be religious.