Why are humans religious? What the cognitive science of religion found, and what failed to replicate

Why are humans religious? For about thirty years the cognitive science of religion (CSR) has tried to answer that with experiments: is belief in gods a by-product of how our minds detect agents and look for purpose, or an adaptation that helped groups cooperate? Some of the field's best-known findings have since failed large replications. This post sorts what held up from what did not, and then asks what the results mean through scientific, philosophical, Catholic and Orthodox lenses.

Part 1 of How did religion develop?. Next: The first signs of religion.

Two big ideas: by-product or adaptation

The by-product view, described as the most popular among CSR researchers (Pyysiäinen & Hauser 2010, cited via a secondary summary), holds that religious beliefs are outputs of mental tools that evolved for other jobs. The usual list is:

• a hair-trigger tendency to detect agents (Guthrie 1993, bibliographic);
• theory of mind, our habit of reading intentions into others;
• a memory advantage for ideas that are only slightly counterintuitive, such as a person who can walk through walls (Boyer 2001, bibliographic);
• a bias towards explaining nature by purpose;
• an intuitive sense that minds and bodies are separate (Bloom 2004, bibliographic).

The adaptationist view holds that religion, or at least some religious packages, helped groups survive by binding members together. Costly rituals signal commitment, and belief in watchful gods may curb cheating. Most theorists now combine the two. Atran and Henrich (2010) and Norenzayan and colleagues (2016) treat by-product cognition as the raw material, with cultural (not genetic) evolution selecting the variants that held groups together. This "dual inheritance" synthesis is now the field's dominant framework.

What failed to replicate

Analytic thinking and disbelief

Gervais and Norenzayan (2012, Science) reported four experiments in which prompting analytic thinking, including showing people Rodin's The Thinker, lowered self-reported religious belief. A preregistered direct replication with about 950 participants found nothing (d = 0.07, in the wrong direction, 95% CI −0.12 to 0.25; Sanchez et al. 2017). A commentary notes that it replicated only the Thinker study. The finding failed again in the Social Sciences Replication Project, which retested 21 Nature and Science experiments and replicated 13 (Camerer et al. 2018). A 13-country correlational test with 3,461 people found that the link between analytic style and disbelief appeared robustly only in pooled analyses and in three individual countries (Gervais et al. 2018).

Mentalizing, autism and belief in God

Norenzayan, Gervais and Trzesniewski (2012, PLoS ONE) proposed that weaker mentalizing, the capacity to read other minds, makes belief in a personal God harder. Their first study compared 12 autistic and 13 neurotypical adolescents, and the effects in their larger samples were modest. A five-study replication with more than 67,000 participants, with Gervais among its co-authors, found that autism-spectrum and empathy scores added essentially nothing once demographics were included. In the Dutch sample of 65,561 the added variance was below 0.1% (Maij et al. 2017). The paper's title states its result: mentalizing skills do not differentiate believers from non-believers, but credibility-enhancing displays do.

One discrepancy is worth flagging. A 2022 review by van Elk describes Maij et al. as having replicated a small but robust mentalizing effect (van Elk 2022). That conflicts with the 2017 paper's own abstract and statistics. This post follows the primary paper.

Priming God and generosity

Shariff and Norenzayan (2007) reported that unconsciously priming God concepts made people more generous in an anonymous economic game. A meta-analysis of 92 studies found a bias-corrected effect on prosocial behaviour of about g = 0.18, present among religious participants only (Shariff et al. 2016). A sceptical reanalysis using a different bias correction put the prosocial estimate at or below zero, and concluded that only a registered replication could settle it (van Elk et al. 2015). A preregistered replication with 650 participants found no effect of implicit primes (Gomes & McCullough 2015). A Bayesian follow-up supported the null for implicit priming even among religious people, but found a small, reliable effect of explicit religious reminders among religious participants only (Billingsley et al. 2018).

Fear and hyperactive agency detection

If belief in spirits comes from a threat-sensitive "agency detector", people should see more phantom agents when they feel in danger. The direct tests have been weak or null. A 2024 virtual-reality study of 199 people did not find more false agent detection under expected danger, and its authors say this casts doubt on the threat-triggered version of the idea (Szymanek et al. 2024; the notes could not verify the full citation). An earlier six-experiment test exists (Maij, van Schie & van Elk 2019), but the notes did not read its results, so this post makes no claim about what it found. Correlational studies do link stronger supernatural belief to a perceptual bias towards seeing agents, though in a US sample of 797 people agency attribution, autism and empathy measures together explained only about 3% of the variance in belief (Maij et al. 2017).

What held up

Purpose-based thinking. Children readily explain nature by purpose: clouds are "for" raining. Kelemen (2004) argued that by about age five the label "intuitive theist" has some explanatory value. The most striking evidence is from adults: under time pressure, physical scientists at top US universities endorsed more unwarranted purpose-based explanations than they did at leisure (Kelemen, Rottman & Seston 2013). They still endorsed fewer than comparison groups overall. Kelemen has said she does not regard the tendency as innate or hard-wired.

Memorable, slightly strange ideas. Cross-cultural recall experiments in France, Gabon and Nepal found better recall for concepts that break one category expectation (Boyer & Ramble 2001). But the definition of "minimally counterintuitive" has shifted between authors. Purzycki and Willard argue that the theory has serious conceptual and empirical problems (Purzycki & Willard), and a Rutgers dissertation describes the literature as mired in empirical contradictions. This finding stands, but loosely.

Dualism, with a twist. Early studies found children reasoning as if minds survive death. A 24-country study of more than 10,000 people, still under review when van Elk summarised it, found mental states judged more likely to continue after death than bodily ones. But the most common answer for both was that they stop, which fits an "intuitive materialism" better than universal dualism (van Elk 2022).

What predicts who believes: watching what people do

The strongest, most consistent predictor of belief in these studies is not a cognitive bias but credibility-enhancing displays (CREDs): seeing parents and others act on their faith in ways that would be costly if they did not mean it. Henrich (2009) modelled why learners would trust displays over words. In Maij et al.'s data, CREDs explained about 10% of the variance in belief, against about 3% or less for the mentalizing measures. Van Elk's 2022 review calls parental CREDs the strongest predictor of supernatural belief. The caveat is that most of this evidence is retrospective self-report, and the notes found no experimental or longitudinal study establishing causation in childhood.

Twin studies point the same way. In one sample of adult male twins, heritability of religiousness was 12% for remembered adolescent religiousness but 44% for current adult religiousness, while the shared family environment dropped from 56% to 18% (Koenig et al. 2005). The family shapes religion in youth, and adults drift towards a temperamental set point. Heritability says nothing about which religion people hold. That is learned.

Are children born believers? Barrett (2012, Born Believers) argues that children's default intuitions make them ready for belief in gods. Banerjee and Bloom (2013) reply that spontaneous theism without cultural input is, as far as anyone knows, non-existent. A study of 5- and 6-year-olds supports the role of testimony: children from churchgoing or parochial school backgrounds judged miracle-story characters real, and secular children judged them fictional (Corriveau, Chen & Harris 2015). The defensible middle position is that children are born receptive, and culture supplies the content.

Ritual and cohesion. Field studies are suggestive but small. At a Spanish fire-walk, heart rates of 38 participants synchronised between walkers and related spectators (Konvalinka et al. 2011). At a Hindu festival in Mauritius, 86 men were studied, and those who performed or watched the painful kavadi ritual gave more to the temple, though participants were not randomly assigned (Xygalatas et al. 2013). In historical data on 83 nineteenth-century American communes, costly requirements went with longer survival for religious communes but not secular ones (Sosis & Bressler 2003).

Through the lenses

Cognitive and evolutionary science

The field has corrected itself, mostly from inside. The researchers who ran the original experiments co-authored many of the replications. What fell was the causal-experimental layer: priming, threat-induced agent detection, and individual differences in mentalizing. What survives is descriptive: humans find agentive and purpose-based explanations easy, and they learn their particular faith from people whose lives show they mean it. The field's centre of gravity has moved from "cognitive biases make religion natural" to "cultural learning does most of the work, and biases shape which ideas spread".

Philosophy: does a cause of belief debunk it?

Philosophers on both sides largely agree that explaining how a belief arose does not, on its own, show it is false. Inferring that would be the genetic fallacy (Jong & Visala 2014). The live debate is narrower: whether the mechanisms that produce religious belief would produce it whether or not it were true. Wilkins and Griffiths (2013) argue that evolution links fitness to truth for commonsense and scientific beliefs but not religious ones (the notes have their argument at summary level only). Theist philosophers, including Murray (2009) and Plantinga (2011), reply that this assumes what it needs to prove (both bibliographic). One observation from the notes, offered as an inference: since testimony and CREDs now carry more weight than agency detection, the question shifts from evolutionary debunking towards the epistemology of trusting what others show and tell us.

Catholic theology

The Catechism states that religious expression is "so universal that one may well call man a religious being" (CCC §28), and Vatican I taught that God can be known "by the natural light of human reason" (Dei Filius). CSR's finding that religious thinking comes easily and early fits this almost too neatly. Several leading CSR researchers are themselves believers. Justin Barrett, then at Fuller Theological Seminary, argues with Kelly James Clark that a natural, non-inferential tendency to believe in God can be treated as innocent until proven guilty, in the tradition of Thomas Reid (Clark & Barrett 2011; abstract level). Where it strains: CSR also explains the specific contents of early belief (spirits, ancestors, hidden agents) as by-products. A critic of Clark and Barrett points out that the natural beliefs CSR finds are not the creeds of the world religions (PhilArchive). That objection cuts both ways: it weakens support for specific doctrines from natural cognition, and it weakens attempts to debunk them, since creeds come from testimony and tradition.

Orthodox theology

Alexander Schmemann's picture of the human being as homo adorans, the creature defined by worship (For the Life of the World, bibliographic), is a structural cousin of CCC §28: religion is universal because humans are made for it. David Bentley Hart adds a philosophical move that matters here. On classical theism, God is not one agent among others but the ground of all being (The Experience of God, 2013, via a secondary summary). If so, an account of why humans detect hidden agents does not bear directly on the God of classical theism. The notes flag the obvious reply: the God of ordinary believers often looks very like the agent CSR describes. Both sides of that exchange need their own sources before anyone leans on it.

What we can say with confidence

• Agentive and purpose-based explanations come easily to children and adults, across cultures.
• Several flagship CSR experiments failed large, preregistered replications: analytic-thinking primes, implicit God primes, and the mentalizing–belief link.
• Who believes is predicted far better by cultural input, especially credible displays of faith, than by any measured cognitive bias.
• A causal account of belief does not by itself show the belief false. Philosophers on both sides largely accept this.

What remains open

• Whether threat-driven agency detection plays any real role.
• Whether children's intuitions amount to "belief" before culture shapes them.
• Whether CREDs cause belief; the evidence is mostly retrospective and correlational.
• Whether the mechanisms CSR describes are reliable in the religious domain. That is a philosophical question that the experiments cannot settle.

Sources

• Atran, S. & Henrich, J. (2010). The evolution of religion. Biological Theory 5(1). PDF
• Banerjee, K. & Bloom, P. (2013). Would Tarzan believe in God? Trends in Cognitive Sciences 17(1). Summary
• Barrett, J. L. (2012). Born Believers. Free Press. Publishers Weekly
• Billingsley, J., Gomes, C. M. & McCullough, M. E. (2018). Royal Society Open Science 5(8). PMC6124061
• Bloom, P. (2004). Descartes' Baby. Basic Books (bibliographic).
• Boyer, P. (2001). Religion Explained. Basic Books (bibliographic).
• Boyer, P. & Ramble, C. (2001). Cognitive Science 25. Record
• Camerer, C. F. et al. (2018). Nature Human Behaviour 2:637–644. FLoRA record
• Catechism of the Catholic Church §28. vatican.va
• Clark, K. J. & Barrett, J. L. (2011). Journal of the American Academy of Religion 79(3). Record
• Corriveau, K. H., Chen, E. E. & Harris, P. L. (2015). Cognitive Science 39(2). PDF
• First Vatican Council, Dei Filius (1870). papalencyclicals.net
• Gervais, W. M. & Norenzayan, A. (2012). Science 336:493–496. doi:10.1126/science.1215647
• Gervais, W. M. et al. (2018). Judgment and Decision Making 13(3). JDM
• Gomes, C. M. & McCullough, M. E. (2015). JEP: General 144(6). PDF
• Guthrie, S. (1993). Faces in the Clouds. Oxford UP (bibliographic).
• Hart, D. B. (2013). The Experience of God. Secondary summary
• Henrich, J. (2009). Evolution and Human Behavior 30:244–260. PDF
• Jong, J. & Visala, A. (2014). International Journal for Philosophy of Religion 76(3). Record
• Kelemen, D. (2004). Psychological Science 15(5). PDF
• Kelemen, D., Rottman, J. & Seston, R. (2013). JEP: General 142(4). PubMed
• Koenig, L. B. et al. (2005). Journal of Personality 73(2). Record
• Konvalinka, I. et al. (2011). PNAS 108(20). PMC3100954
• Maij, D. L. R. et al. (2017). PLoS ONE 12(8):e0182764. PMC5568287
• Maij, D. L. R., van Schie, H. T. & van Elk, M. (2019). Religion, Brain & Behavior 9(1). Record
• Murray, M. (2009) and Plantinga, A. (2011), in and after Schloss & Murray (eds.), The Believing Primate. OUP (bibliographic)
• Norenzayan, A., Gervais, W. M. & Trzesniewski, K. H. (2012). PLoS ONE 7(5). PLoS
• Norenzayan, A. et al. (2016). Behavioral and Brain Sciences 39:e1. PDF
• Purzycki, B. G. & Willard, A. K. MCI theory: a critical discussion. Religion, Brain & Behavior. PDF
• Pyysiäinen, I. & Hauser, M. (2010). Trends in Cognitive Sciences 14(3), via Wikipedia
• Sanchez, C. et al. (2017). PLoS ONE 12(2). PMC5325262
• Schmemann, A. For the Life of the World (bibliographic).
• Shariff, A. F. & Norenzayan, A. (2007). Psychological Science 18(9). FLoRA
• Shariff, A. F. et al. (2016). Personality and Social Psychology Review 20(1), as summarised in van Elk et al. 2015.
• Sosis, R. & Bressler, E. R. (2003). Cross-Cultural Research 37(2). SAGE
• Szymanek et al. (2024). Journal of Psychology of Religion and Spirituality. PDF
• van Elk, M. et al. (2015). Frontiers in Psychology 6:1365. PMC4569810
• van Elk, M. (2022). Frontiers in Psychology 13:949131. PMC9714272
• Wilkins, J. & Griffiths, P. (2013). Evolutionary debunking arguments in three domains. PhilSci-Archive
• Xygalatas, D. et al. (2013). Psychological Science 24(8). Aarhus news

The notes behind this post: 01_cognitive_evolutionary.md.

Part 1 of How did religion develop?. Next: The first signs of religion: burials, cave art and Göbekli Tepe.

Produced with AI assistance (Claude) from the cited notes in the religion-origins project.