CoChat

How many atheists are there?

Open access
Gervais, Will M, Najle, Maxine Belén · 2017 · PsyArXiv (OSF Preprints)

The DOI matches its Crossref record; 25 of the 26 references it cites were found in a registry or an index; none of the 40 sources its claims cite is checked for support yet.

Abstract

One crucible for theories of religion is their ability to predict and explain patterns of belief and disbelief. Yet, religious nonbelief is often heavily stigmatized, potentially leading many atheists to refrain from outing themselves even in anonymous polls. We used the unmatched count technique and Bayesian estimation to indirectly estimate atheist prevalence in two nationally representative samples of 2000 U.S. adults apiece. Widely-cited telephone polls (e.g., Gallup, Pew) suggest USA atheist prevalence of only 3-11%. In contrast, our most credible indirect estimate is 26% (albeit with considerable estimate and method uncertainty). Our data and model predict that atheist prevalence exceeds 11% with greater than .99 probability, and exceeds 20% with roughly .8 probability. Prevalence estimates of 11% were even less credible than estimates of 40%, and all intermediate estimates were more credible. Some popular theoretical approaches to religious cognition may require heavy revision to accommodate actual levels of religious disbelief.

Overview

CoChat Summary

TL;DR

Using the unmatched count technique and Bayesian estimation in two nationally representative U.S. online samples (N = 4,000), Gervais and Najle estimate that about 26% of American adults may be atheists, far above telephone-poll self-reports of 3–11%. The authors argue the gap suggests substantial underreporting due to stigma, though they caution the indirect method has mixed validity evidence and wide uncertainty.

Key findings

  • The pooled indirect atheism prevalence estimate was 26% (97% HPDI [13%, 39%] in text; Table 3 reports 26 [14, 40]), versus a self-report estimate of 17% [14%, 21%] (Table 3: 17 [14, 20]); the posterior probability that indirect estimates exceed self-reports was .93.
  • Sample I yielded an indirect estimate of 32% [11%, 54%]; Sample II yielded 20% [6%, 35%].
  • The model indicated greater than .99 probability that atheist prevalence exceeds 11%, roughly .8 probability that it exceeds 20%, and that 26% was about 3 times as credible as 17% and more than 20 times as credible as 11%.
  • Self-reported atheism was 21% among Millennials and 14% among Baby Boomers, but indirect estimates were both around 26%; the probability that Millennials exceeded Baby Boomers dropped from greater than 99% on self-reports to about 48% on indirect measures.

Methods

  • Two preregistered nationally representative online samples of 2,000 U.S. adults each (total N = 4,000), contracted from YouGov.
  • Used the unmatched count technique: Sample I included baseline, target (“I believe in God” as not true), and self-report conditions; Sample II included baseline, atheist (“I do not believe in God”), and math-impossibility (“I do not believe that 2 + 2 is less than 13”) conditions.
  • Bayesian estimation with a hierarchical/multilevel pooled model; priors were centered on Gallup’s 11% self-report estimate with SD = 1; estimates are summarized with point estimates and 97% HPDIs, with demographic contrasts for gender, education, politics, and age.

Limitations

  • Paper-stated: The authors caution that estimates have “considerable estimate and method uncertainty” and that 26% should not be uncritically accepted; the pooled 97% HPDI is wide (text: [13%, 39%]; Table 3: [14, 40]) (text-evident).
  • Paper-stated: Validity evidence was mixed. In Sample II, the math-impossibility condition yielded a “bizarre” 34% prevalence, which the authors call their “most damning result” and say may reflect innumeracy, incomprehension, inattentiveness/jesting, sampling error, or a genuine flaw in the unmatched count technique; Sample I lacked religious demographics for validity checks.
  • Paper-stated: The U.S. data should not be used as a global baseline; underreporting likely varies with cultural norms, and precise global estimates are not currently possible. The preregistered analysis plan also varied slightly from the adopted one (BEST vs. McElreath), though inferences did not appreciably change.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.31234/osf.io/edzda_v1

Cites

25 of 25 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Posterior atheist prevalence estimates from Samples I and II. Yaxis represents the relative credibility with which different parameter values could have plausibly generated the observed data. Values higher on the y-axis represent more plausible parameter estimates; values lower on the y-axis represent less plausible parameter estimates.
Figure 1 Posterior atheist prevalence estimates from Samples I and II. Yaxis represents the relative credibility with which different parameter values could have plausibly generated the observed data. Values higher on the y-axis represent more plausible parameter estimates; values lower on the y-axis represent less plausible parameter estimates.CC-BY · From the PDF
Figure 2 Posterior atheist prevalence estimate from pooled model. Y-axis represents the relative credibility with which different parameter values could have plausibly generated the observed data. Values higher on the yaxis represent more plausible parameter estimates; values lower on the yaxis represent less plausible parameter estimates.
Figure 2 Posterior atheist prevalence estimate from pooled model. Y-axis represents the relative credibility with which different parameter values could have plausibly generated the observed data. Values higher on the yaxis represent more plausible parameter estimates; values lower on the yaxis represent less plausible parameter estimates.CC-BY · From the PDF
Image 3
CC-BY · From the PDF
Image 4
CC-BY · From the PDF
Image 5
CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 0 Not valid · 30 Not checked · 59 Own results and interpretations

50 more claims

Create a free account to see all 89 claims and to verify the 15 references still open.

See all claims — free account
  • It may state two separate claims as one. 13
  • Probably not a substantial claim. 25

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 25 references

Research works printed but not linked: 4

  1. Kruschke, J. K. (2010). Doing bayesian data analysis: A tutorial introduction with r: Academic Press.
  2. McElreath, R. (2016). Statistical rethinking: A bayesian course with examples in r and stan (Vol. 122): CRC Press.
  3. Stark, R., & Bainbridge, W. S. (1985). The future of religion: Secularization, revival, and cult formation: Univ of California Press. How many atheists are there? 26
  4. Wilson, E. O. (1999). Consilience: The unity of knowledge (Vol. 31): Vintage.

Metadata

Authors
Gervais, W. M., & Najle, M. B.
Affiliations
University of Kentucky (US)
Venue
PsyArXiv (OSF Preprints)
Published
2017-01-23
OpenAlex
W4409673081
Cited by
15
References
26
Citations by year
2024: 14 · 2021: 1
Type
posted content
Language
en
Open access
Locations
PsyArXiv (OSF Preprints) (submittedVersion, cc-by)
Funders
John Templeton Foundation (48275)
Topics
  • Social Sciences › Sociology and Political Science › Religion and Society Interactions
  • Social Sciences › Health › Religion, Spirituality, and Psychology
  • Social Sciences › Arts and Humanities › Philosophy › Study and Philosophy of Religion
Keywords
atheists, non-belief, indirect estimation, Bayesian estimation, religious stigma, self-disclosure, religious cognition, faith and unbelief

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-21, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-21 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: How many atheists are there? https://app.cochat.ai/papers/10.31234/osf.io/edzda_v1
The paper
Gervais, W. M., & Najle, M. B. (2017). How many atheists are there? PsyArXiv (OSF Preprints). https://doi.org/10.31234/osf.io/edzda_v1

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.