Did 'Big Gods' build civilisation? The moralizing gods debate and the retracted Nature paper

Did belief in all-seeing, moralizing gods help humans build large societies of strangers? Or did "Big Gods" appear only after societies were already big? This question produced one of the most public disputes in the science of religion. It includes field experiments in eight and then fifteen societies, family-tree analyses of Pacific cultures, and a 2019 Nature paper on the Seshat database that was retracted in 2021. This post explains what each study found, what the retraction did and did not show, and why both sides now talk about supernatural punishment rather than high gods.

Part 5 of How did religion develop?. Previous: The Axial Age and the one God. Next: From Jesus to the Schism.

The hypothesis

Ara Norenzayan's Big Gods (2013, bibliographic) and a 2016 target article with colleagues set out a cultural-evolution account. Religious beliefs first arose as by-products of ordinary cognition (Part 1). Over the last 10–12 millennia, cultural variants with "increasingly potent, moralizing, supernatural agents, credible displays of faith, and other psychologically active elements conducive to social solidarity" were selected because they raised cooperation with co-religionist strangers and helped groups win competition. The authors call large-scale cooperation and prosocial religion "importantly linked and mutually energizing" (Norenzayan et al. 2016).

That last phrase matters. The thesis as stated allows causation in both directions. A finding that complexity came first does not refute it outright, but it does refute the strong popular version: that moralizing gods were a precondition for large societies.

The field games

Purzycki and colleagues (2016, Nature) ran an anonymous, cheatable coin game with 591 people in eight communities: in Vanuatu, Fiji, Brazil, Mauritius, Siberia and among the Hadza of Tanzania. Each player privately assigned 30 coins between two cups using a die roll. Honest play gives about half to each, so a skew measures favouritism. The abstract reports that "the higher participants rated their moralistic gods as punitive and knowledgeable about human thoughts and actions, the more coins they allocated to geographically distant co-religionist strangers" (Purzycki et al. 2016). According to a secondary summary, average allocations stayed below an even split even at the highest ratings of punishment, though the uncertainty interval included it. The study is correlational, and the lead author warned against reading it as showing that religious people are nicer. The design also had no condition for people of other religions.

A 15-society extension with 2,228 participants found that higher ratings of gods as monitoring and punishing predicted less local favouritism and more sharing with distant co-religionists (Lang et al. 2019). Then a preregistered reanalysis across the same 15 cultures separated punishment from moral concern, measured by asking people what angers or pleases their god. Its conclusion: "gods' moral concerns do not play a direct, cross-culturally reliable role in motivating cooperative behaviour" (Bendixen et al. 2023). Watching and punishing gods seem to matter more than gods who care about morality as such. None of these studies manipulates belief. All are correlations within sites.

The Pacific family trees

Watts and colleagues (2015) used 96 Austronesian cultures and Bayesian models of trait evolution on 4,000 language trees. Their abstract reports that "broad supernatural punishment drives political complexity, whereas MHGs follow political complexity" (Watts et al. 2015). MHGs are moralizing high gods. Broad supernatural punishment can come from any agent or process, including ancestors. The caveats are real. Only 6 cultures had moralizing high gods. None of the directional Bayes factors exceeds about 5 against independence, and against the unconstrained model they are all below 2, which is weak evidence for direction.

A global study of 583 societies in the Ethnographic Atlas found moralizing high gods more likely in poorer environments and in politically complex societies with herding, with neighbours' beliefs the strongest predictor, which points to diffusion (Botero et al. 2014). Its "91% accuracy" is an out-of-sample AUC of 0.91, not 91% of cases classified correctly.

The Seshat paper and its retraction

In 2019 Whitehouse and colleagues used Seshat, a databank of 414 societies across 30 regions and 10,000 years, to argue that moralizing gods "follow—rather than precede—large increases in social complexity", appearing only after "megasocieties" of about a million people (Whitehouse et al. 2019). It was a striking reversal of the Big Gods story.

Within weeks a group including several Big Gods proponents posted a critique. Its abstract states that "61% of Seshat data points on moralizing gods are missing values", that the authors had recoded those as absences, and that "when we confine the analysis only to the extant data or use various standard imputation methods, the reported finding is reversed" (Beheim et al. 2019 preprint). The critics' code shows that only one of 30 regions had a known absence of moralizing gods before a presence, and they dispute even that one (reanalysis repository). Nature published the critique in 2021 (Beheim et al. 2021), and the authors retracted the paper on 7 July 2021 (retraction note).

The retraction was not a concession on the conclusion. As reported by Retraction Watch, the authors accepted that they should have labelled many of those data points as absent or inferred absent rather than unknown, while maintaining that their original conclusions remained strongly supported, and they pointed readers to new papers. Brian Beheim told Retraction Watch the objection "wasn't so much their conclusion, but that they weren't transparent about the amount of missing data involved" (Retraction Watch). The two framings differ: the authors say the error was a label, the critics that genuine unknowns were treated as absences.

Historians attacked the data as well. Re-coding religion variables for 12 polities in China's Middle Yellow River valley with expert codes from the Database of Religious History, they found Seshat agreed with the experts in "only under a third of cases" (Slingerland et al. 2020).

The "retake" and its critics

The authors' revised analysis appeared in Religion, Brain & Behavior 13(2), published online in June 2022 and printed in the 2023 volume. Using an expanded codebook and a more direct causal test, it concludes "that the largest increases in social complexity do indeed precede Big Gods in world history and that Big Gods did not contribute to the evolution of sociopolitical complexity" (Whitehouse et al. 2022/2023). A companion paper found that "intergroup warfare, supported by resource availability" played the largest role in the rise of both complexity and moralizing religion (Turchin et al. 2022). Commentators in the same issue, including Purzycki, Bendixen and Lightner, argue that the emergence and drivers of moralistic supernatural punishment "remain unresolved" (commentary).

A further problem cuts across both camps. Moralizing gods first appear in Seshat together with writing, so any such beliefs in societies without writing are structurally invisible. That biases any Seshat test towards "complexity first".

Small societies have moralizing gods too

The most important reframing comes from Lightner, Bendixen and Purzycki (2023, Evolution and Human Behavior). Their title states their conclusion: moralistic supernatural punishment is probably not associated with social complexity. The standard cross-cultural "high god" variable requires a god who created or directs the universe, which has "led researchers to underestimate the presence of moralizing gods by systematically producing false negatives", more often in simpler societies. The likely result is "a spurious positive association", and "ethnographic evidence supports the prevalence of moralizing gods among small-scale societies" (Lightner et al. 2023).

Through the lenses

Cultural evolution and quantitative history

What survives is methodological. With Seshat's coverage, the answer to "which came first?" depends almost entirely on how missing data are treated, and neither side's original causal claim is strongly identified from the 2019 data. The action has moved from high gods to supernatural punishment by any agent, which looks old and widespread. The open datasets make this checkable: D-PLACE, Pulotu, Seshat's Equinox release and both sides' analysis code are on GitHub, under licences that allow non-commercial reuse with attribution.

History

The historians' critique is about expertise as much as statistics. Coding a religion's moral content from tertiary sources, and with one expert or none, produces numbers that look precise and are not. Rüpke (2022) adds that religion stabilizes dissent as often as unity, which a single "moralizing gods: present/absent" column cannot capture.

Catholic theology

The Big Gods literature asks whether belief in watchful gods is useful for cooperation. Christian theology makes a different claim: that God exists and is good. The Catholic Church distinguishes "theological faith" from "belief in the other religions", which it describes as "that sum of experience and thought that constitutes the human treasury of wisdom and religious aspiration" (Dominus Iesus §7). On that distinction, a functional story about belief is an account of the human side, not a verdict on the divine one. Where the theological lens finds something congenial: Lightner and colleagues' finding that moral, punitive supernatural beings are widespread among small-scale peoples fits Nostra Aetate's observation of a perception "among various peoples" of a hidden power over human events. Where it strains: the evidence favours watching, punishing beings of many kinds, not one God, and the notes are clear that it supports nothing like a primordial monotheism (Part 7).

Orthodox theology

Hart's classical theism (Part 1) offers a pointed reply to the whole framing. If God is the ground of being rather than a being among others, then a "Big God" who monitors behaviour and punishes cheats is a being among beings, not the God of classical theism as Hart describes it. Hart treats popular pictures of God as a cosmic agent who fills explanatory gaps as a strawman. The notes record this as an inference, and the obvious reply applies: the god in a field participant's answers is often exactly a watcher and punisher. Both points can be true. The science studies the god of practice, and the theologian defends the God of doctrine.

What we can say with confidence

• The 2019 Nature paper was retracted in 2021 after a critique of how 61% missing data were coded.
• Its authors' revised analysis, in Religion, Brain & Behavior 13(2), still concludes that complexity came before Big Gods. Critics dispute that.
• In field games, rating one's god as watching and punishing goes with more fairness to distant co-religionists. A god's moral concerns as such do not reliably predict it.
• Supernatural punishment from some agent is common in small-scale societies.

What remains open

• Whether moralizing gods came before, after or alongside large societies; the data cannot yet say.
• Whether belief causes cooperation; all the game evidence is correlational.
• How much of the global pattern reflects diffusion from a few complex societies.

Sources

• Beheim, B. et al. (2019). PsyArXiv preprint. DOI; (2021). Nature 595:E29–E34. DOI; reanalysis code
• Bendixen, T. et al. (2023). Evolutionary Human Sciences 5:e18. DOI
• Botero, C. A. et al. (2014). PNAS 111(47). PMC4250141
• CDF (2000). Dominus Iesus. vatican.va
• Hart, D. B. (2013). The Experience of God. Secondary summary
• Lang, M. et al. (2019). Proc. R. Soc. B 286:20190202. Record
• Lightner, A. D., Bendixen, T. & Purzycki, B. G. (2023). Evolution and Human Behavior 44(6):555–565. DOI
• Norenzayan, A. (2013). Big Gods. Princeton UP (bibliographic).
• Norenzayan, A. et al. (2016). Behavioral and Brain Sciences 39:e1. Europe PMC
• Purzycki, B. G. et al. (2016). Nature 530:327–330. Europe PMC; lead author's caution via Oxford
• Purzycki, B. G., Bendixen, T. & Lightner, A. D. (2022). Commentary, RBB 13(2):207–214. DOI
• Retraction Watch (2021). Critique topples Nature paper on belief in gods. Link
• Rüpke, J. (2022). Big Gods and Big Rituals. Journal of Cognitive Historiography 6. DOI
• Second Vatican Council (1965). Nostra Aetate. vatican.va
• Slingerland, E. et al. (2020). Journal of Cognitive Historiography 5. PDF
• Turchin, P. et al. (2022). RBB 13(2):167–194. DOI
• Watts, J. et al. (2015). Proc. R. Soc. B 282:20142556. PMC4375858
• Whitehouse, H. et al. (2019). Nature 568:226–229 (retracted). Europe PMC; retraction note DOI
• Whitehouse, H. et al. (2022/2023). A review and "retake". RBB 13(2):124–166. DOI
• Datasets: D-PLACE Ethnographic Atlas, Pulotu, Seshat Equinox, original Seshat analysis code

The notes behind this post: 05_big_gods_debate.md.

Part 5 of How did religion develop?. Previous: The Axial Age and the one God. Next: From Jesus to the Schism: what historians can establish about early Christianity.

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