What religions do — the function theories weighed against cross-cultural evidence
Asking "what is religion for?" the way the classic theories did — each crowning one function as the function — produces a contest nobody wins. The cross-cultural record makes better sense if the theories are sorted by which question they actually answer:
1. Why do religious representations exist at all? The by-product theories win this layer. Boyer and Barrett's cognitive program explains the supply side: human minds over-detect agency, reason effortlessly about invisible minds, and preferentially remember minimally counterintuitive concepts — so god-like representations arise everywhere without any function required, the way moths find flames without flames being "for" moths. The vault's own motif work supports the universality this predicts while also flagging its limits: flood myths recur on every inhabited continent, but their distribution is non-uniform (sparse in flood-prone Africa — great-flood), so "same minds, same stories" cannot be the whole account; population history and transmission chains do real work that pure cognition does not explain.
2. Why do some representations get institutionalized, scaled, and kept? The group-function theories win this layer — as selection pressures, not origins. Here the evidence is of four kinds:
- Cohesion (Durkheim): collective ritual demonstrably bonds groups; the lament cults this vault has dated — Tammuz wailing in Jerusalem (Ezekiel 8:14, dying-rising-god), Osiris rites, Adonia festivals — are precisely organized collective emotion, persisting for millennia, which is what Durkheim's "society worshipping itself" predicts institutions of collective effervescence should do.
- Cooperation-enforcement (Norenzayan's Big Gods): moralizing, punishing, omniscient gods correlate with large-scale anonymous cooperation across societies, and experimentally with reduced cheating. But the causal arrow is genuinely contested: the flagship historical test claiming complexity precedes moralizing gods (Whitehouse et al. 2019) was retracted in 2021, and the reanalysis dispute remains open. The correlation survives; the causal story is unresolved (filed below).
- Legitimation: the most under-sung and best-dated function in the record. The earliest large religious corpora this vault has touched are royal instruments: the Pyramid Texts exist to make a king's afterlife succeed and his successor's rule cosmically grounded (dying-rising-god — Osiris–Horus), the Sumerian King List has kingship "lowered from heaven," and Zhou China runs legitimacy through Heaven's mandate. Wherever early states appear, religion is doing sovereignty work, with class-1/2 evidence.
- Identity-marking and costly signaling (Sosis): religious communes with costlier requirements outlived secular communes in the 19th-century data (Sosis & Bressler 2003, Cross-Cultural Research 37); boundary-marking practices (dietary law, circumcision, distinctive dress) recur exactly where group boundaries are under pressure — post-exilic Judah being the canonical dated case.
- Anxiety-reduction (Malinowski) answers a third question — not why religion exists or persists, but when it intensifies: Trobriand islanders ritualized dangerous open-sea fishing, not safe lagoon fishing (Malinowski, Magic, Science and Religion, 1948); modern data echo it (religiosity rises after earthquakes and in high-existential-risk populations — e.g. Bentzen, Economic Journal 2019 on earthquake districts). Situational, real, but parasitic on representations the other layers supply.
3. The layered model is itself supported by the transmission evidence. The motif verdicts show that what religions carry is governed by ordinary cultural transmission — literary borrowing along scribal corridors (great-flood: contact), faithful inheritance down language families (sky-father: descent), and independent reinvention under shared pressures (dying-rising-god: convergence strands). Crucially, the sky-father case shows traditions inherit vocabulary far more faithfully than importance — Zeus stayed central while his exact cognate Dyaus became a fossil. Something other than inheritance decides which gods matter, and the function layer (what work the god does for the community now) is the obvious candidate. Function is the selection environment; cognition is the variation generator; transmission is the mechanism. That is the integrated claim.
So: what are religions mainly used for in culture? On the dated evidence — binding groups, enforcing trust beyond kin, legitimating power, and marking who belongs, with comfort under uncertainty as the situational intensifier — all running on representations the human mind produces for free.
Simple explanation
Human brains come with quirks: we see agents everywhere (a noise at night feels like a someone), we find stories about invisible persons easy to remember and retell. So ideas about gods and spirits pop up everywhere, the way weeds grow in every garden — nobody plants them, they just suit the soil. That's why religion exists.
But why does it stay, and grow huge temples and priesthoods? Because groups that used these ideas got real things done: a village that swears by the same god trusts each other at the market; a king crowned by heaven gets obeyed without an army on every corner; a community that fasts and sacrifices together knows who's truly committed, because faking it is expensive; and when the sea is dangerous or the harvest might fail, the rituals keep people calm enough to act. Societies kept the god-ideas that did this work, the way gardeners keep the weeds that turn out to be herbs.
And religions spread the same three ways words do: handed down from ancestors (Zeus and the Vedic sky-father are literally the same name, said two ways), borrowed from neighbors (the Bible's flood story machinery comes from older Babylonian tellings), or reinvented independently (people who farm all watch crops die and return, and tell god-stories about it). None of this says the gods are real or unreal — it explains the human side: why the stories rhyme and why the institutions last.
Gaps found
| Gap | What's fuzzy | Open question filed |
|---|---|---|
| Causal order of big gods and big societies | "Gods enforce cooperation" vs. "big societies invent enforcing gods" — flagship dataset retracted; I cannot state the arrow | Q8 |
| Costly-signaling evidence base | The commune/kibbutz studies are narrow (19th-c. USA, Israel); does the survival effect generalize? | Q9 |
| Why inherited gods rise or fall | "Function selects which gods matter" is my claim, but Zeus-vs-Dyaus is one data point; needs systematic test | Q10 |
| Convergence's engine | I say "same minds + same problems reinvent similar stories," but flood myths are sparse in Africa — so what exactly triggers reinvention where it happens? | Q5 (distribution data) |
| "Function" language smuggles teleology | Saying religion is "for" cohesion blurs origin and current use — the essay separates layers, but the vocabulary keeps collapsing them; needs sharper terms when claims are formalized | Q2 (register, narrowed not closed) |
| Secular substitution limits | If Scandinavian-style societies replace religious cooperation machinery, under what conditions does substitution succeed or fail? | Q9 (adjacent; tracked there) |
The strongest honest case against this synthesis: functionalism is a hindsight machine. Any institution that persists can be assigned a "function" after the fact — persistence is the only evidence, and the explanation is unfalsifiable. The one serious attempt to make the cooperation-function claim historically testable at scale collapsed: Whitehouse et al. 2019 was retracted over how missing data were coded, and the underlying databases (Seshat, ethnographic atlases) encode the very theoretical assumptions they are meant to test. Meanwhile the cleanest dated facts in this vault are transmission facts, not function facts: we can prove the flood story was borrowed (contact) and the sky-father inherited (descent), but we cannot prove either was kept because it bonded groups — that part is inference dressed as observation. Add that the most secular societies on earth are among the most cooperative and least anxious, and the function theories face a dilemma: if religion's functions are fully substitutable by courts, insurance, and football, then "what religion is for" explains nothing distinctive about religion at all — it was never for anything; it is a by-product that institutions opportunistically rode, and the by-product theory alone, without the functional superstructure, is the parsimonious model. On this view my "layered synthesis" is a diplomatic compromise that grants every theory a participation trophy precisely because none of them can be tested hard enough to lose.
The layered model's reply — substitutability shows function is real but multiply realizable; the retraction shows fragile measurement, not absent effect; and convergent reinvention of similar institutions is itself evidence of selection — is fair, but the steelman correctly identifies where the synthesis is softest: the selection step is the least directly evidenced link in the chain.
- Causal-order resolution against cooperation-function: corrected Seshat-class analyses robustly showing moralizing-god beliefs consistently follow state-scale complexity by centuries → demotes Big Gods from selection pressure to legitimation tool; confidence in the layered model's cooperation leg drops to low, legitimation leg rises.
- Costly-signaling replication failure: systematic failure of the commune-longevity effect outside the original datasets → identity-marking layer loses its best quantitative support, drops to qualitative-historical status.
- A documented zero-function tradition: a well-described society maintaining a costly, long-lived religious system with demonstrably no cohesion, trust, legitimation, identity, or anxiety payoff → directly falsifies the selection claim (by-product-only model wins that case).
- Strong secular-substitution evidence: longitudinal evidence that secularizing societies suffer no transitional deficit in trust, cooperation, or mental health as religious infrastructure exits → suggests the "work" religion does was never load-bearing, raising the by-product-only model's standing.
- Priming-literature collapse: the religious-priming experimental base (already shaken by replication critiques) fully failing meta-analytic scrutiny → removes the experimental leg of cooperation-enforcement, leaving it correlational only; confidence drops a tier.