IQ and Religion: What Research Shows

0 The Short Answer

Research finds a small negative average association between measured intelligence and religiosity in the populations that have been studied, and that finding is far weaker and far more conditional than the headlines it generates. The most cited source is a 2013 meta-analysis by Miron Zuckerman, Jordan Silberman and Judith Hall in Personality and Social Psychology Review, which pooled 63 studies and reported a modest inverse relationship.

Everything interesting about this topic lives in the qualifications, and there are many. The samples are overwhelmingly Western, largely American, and predominantly Christian, which means the finding describes a specific cultural context rather than a human universal. The measures of religiosity vary enormously between studies, and belief, attendance, denominational identity and fundamentalism are not the same variable. The effect size is small enough that the distributions overlap almost entirely. And the design is nearly always cross sectional, which leaves the direction of any causal arrow undetermined.

The rest of this page takes that finding apart carefully: what was actually measured, how big the effect is in terms you can picture, what explanations have been proposed and how well each holds up, where the evidence is weakest, and what the whole literature does and does not license anyone to say. If you want a single takeaway before the detail, it is this: an average difference of this size across groups tells you essentially nothing about any individual person, and anyone using it to do so has misunderstood the statistics rather than discovered something about human beings.

1 What the Meta-Analysis Actually Reported

Zuckerman, Silberman and Hall assembled studies spanning several decades that had measured both intelligence and some index of religiosity in the same sample, and pooled the results. Their headline finding was a negative correlation of modest size, and their paper is careful about the variability behind that average. Averages of this size rarely survive adjustment intact, and the well-being literature shows why: the positive association between ability and life satisfaction shrinks substantially once health, income and employment enter the model, because those conditions were carrying much of it.

Two details from the analysis matter more than the headline number. First, the strength of the association varied substantially by the type of sample. Studies of college students and adults tended to show stronger associations than studies of children and adolescents, a pattern the authors discussed at length and which admits several interpretations, including that the relevant variable is not ability at all but the educational and social environment that college attendance represents.

Second, the association varied by how religiosity was operationalized. This is not a footnote. A study measuring frequency of religious service attendance is measuring social participation, community embeddedness and family tradition as much as belief. A study measuring agreement with statements of doctrine is measuring something closer to conviction. A study measuring self reported religious identity is measuring how someone labels themselves on a survey. These produce different numbers because they are different constructs wearing one word.

Any honest summary of this literature has to begin by admitting that "religiosity" as a research variable is a collection of loosely related things, and that the reported effect depends on which of them a given study happened to measure.

2 How Big the Effect Is in Practical Terms

Correlations in the range reported here are easy to misread, so it helps to convert them into pictures rather than leaving them as decimals.

The practical translation is blunt. If you met a hundred devout believers and a hundred committed atheists and tested all of them, the two score distributions would sit almost on top of each other, with a slight average displacement you could not detect without measuring everyone. This is what a small effect looks like, and it is the single most important thing to hold onto when reading anything else about this subject.

3 The Measurement Problem on Both Sides

Before interpreting any association, it is worth asking whether the two things being correlated were measured well. Here, both sides have problems.

On the religiosity side, the problems are severe. Beyond the belief versus attendance versus identity distinction, there is social desirability: in strongly religious communities, people over report religious behavior on surveys, and in strongly secular ones, they under report it. There is also the question of what counts as religious at all, which varies across traditions in ways that Western designed instruments handle poorly. A survey item about weekly attendance at a place of worship maps onto some traditions cleanly and onto others barely at all.

On the intelligence side, the problems are subtler but real. Many studies in this literature did not use full professionally administered batteries. Some used vocabulary based proxies, some used short group administered tests, some used educational attainment as a stand in for ability. Vocabulary proxies are particularly awkward here, because vocabulary reflects reading habits and educational exposure, both of which vary with religious and cultural background for reasons that have nothing to do with reasoning capacity.

When two noisily measured variables are correlated, the observed relationship is attenuated relative to the true one, which pushes in the direction of underestimating. But when both measures also capture education and social background, the observed relationship is inflated by that shared contamination, which pushes the other way. The honest position is that the reported effect is somewhat uncertain in size, not just in interpretation.

4 The Explanations That Have Been Proposed

The authors of the meta-analysis offered several candidate explanations, and later researchers have added more. None is established, and they are not mutually exclusive.

• Cognitive style. The most discussed proposal holds that a more analytic and less intuitive processing style is associated both with higher performance on reasoning tasks and with greater skepticism toward supernatural claims. On this account, ability is not the operative variable; disposition toward reflection is, and ability correlates with it.
• Functional substitution. A second proposal is that religion supplies goods such as self regulation, community support, meaning and a sense of control, and that people with more educational and economic resources may obtain some of those goods through other routes. This makes the relationship a byproduct of circumstance rather than of cognition.
• Conformity and social context. A third is that ability correlates with willingness to depart from a majority position. In a religious society, that predicts less religiosity among high scorers. In a secular society, the prediction reverses, which is a testable implication that has not been examined nearly enough.
• Education as the real variable. A fourth possibility is that neither ability nor style is doing the work, and that the association is largely carried by years and type of education, which independently predicts both measured ability and secularization in many societies.

Notice how different these are in their implications. The first locates the effect in the individual's mind, the second in their material circumstances, the third in their society, and the fourth in an institution. The data available cannot cleanly distinguish among them, which is why the literature has produced far more speculation than resolution.

5 The Analytic Thinking Line of Research and Its Replication Trouble

The cognitive style explanation deserves its own section, because it generated the most famous experimental work in this area and because that work has since become a cautionary tale worth knowing. A neighboring literature has held up better: across seven studies, Stanovich and West found that susceptibility to anchoring, base rate neglect and the conjunction fallacy barely tracked measured cognitive ability, while belief bias in syllogisms and probability matching did.

A set of studies in the early 2010s reported that prompting people into an analytic mode of thinking, through various priming manipulations, reduced their expressed religious belief. The result was widely covered because it appeared to turn a correlation into a causal demonstration: shift the thinking style, shift the belief.

Subsequent replication attempts with larger samples did not reproduce the effect. This became one of many examples in psychology's replication reckoning, where a striking, heavily cited finding did not survive better powered testing. The related correlational work, linking performance on reflection tasks to lower supernatural belief, has held up better than the priming experiments, but correlational evidence returns us to the same interpretive problem we started with.

The lesson generalizes past this topic. When a small effect in a contested area is reported with an elegant causal story attached, the appropriate response is to wait for replication rather than to update your worldview. That is true here, and it is equally true of any study you encounter claiming to have found the mechanism behind a human difference.

6 Where This Question Came From

The research question has a history, and knowing it explains why the literature looks the way it does.

Studies comparing believers and non believers on mental tests date back to the early decades of psychometrics, when the field was young, ambitious and methodologically primitive by any modern standard. Many of those early studies used tiny convenience samples, most often college students at a single institution, with instruments that would not pass review today and with no attempt to control for the social background of participants. Some were conducted by researchers with visible commitments on the question, in both directions.

This matters because meta-analyses pool what exists, and what exists in older decades is thin. A pooled estimate that includes studies of forty undergraduates from 1960 alongside a properly sampled modern survey is averaging across enormous differences in quality. The better meta-analyses examine whether effects vary by study quality and publication year, and finding that they do is itself informative: it suggests part of what is being measured is the changing methods of a discipline rather than a stable fact about people.

There is also a publication bias question that applies with unusual force here. This is a topic where results are socially interesting in a way that null results are not, which is exactly the condition under which the published record drifts away from the underlying reality. A study finding no relationship is a study with no story, and stories get published.

7 Variation Within Religion Is Larger Than the Gap Between

Treating religiosity as a single dimension running from none to a lot obscures something important: the differences among religious people are far larger than the average difference between religious and non religious groups. The same asymmetry holds for cognitive results, which is why an IQ test with an adult reference group is read as a position in a distribution with a confidence interval, not as a label for the group a person belongs to.

Traditions differ enormously in what they ask of adherents intellectually. Some place scholarly study of complex texts at the center of religious life, with centuries of interpretive commentary that practitioners are expected to engage. Others emphasize direct experience, practice or community over doctrinal analysis. Within any single tradition, individual adherents range from those who have never examined a claim to those who have spent decades in rigorous theological argument. Folding all of that into one variable called religiosity, and then correlating it with a test score, discards nearly all of the meaningful structure.

Educational attainment complicates it further, and in ways that vary by community. Some religious communities have historically maintained exceptionally high rates of higher education, others have had educational access restricted by discrimination or poverty, and both patterns produce associations in survey data that have nothing to do with the psychology of belief. When a study reports an association between religiosity and measured ability in a sample drawn from a society with that history, it may be measuring the residue of who was permitted to go to school.

None of this makes the pooled finding meaningless. It does mean that the pooled finding is an average across radically heterogeneous groups, which is another reason it cannot be applied to any individual or any specific community without further evidence.

8 Why the Cultural Setting Changes Everything

Nearly all of this research was conducted in a handful of Western countries, most of it in the United States, with predominantly Christian participants. That is a serious limitation for a question about human psychology in general, and it interacts with the conformity explanation in a way that should make everyone cautious. Running the question across genuinely different societies would also mean solving the instrument problem, and the honest verdict on the nonverbal tests built for that job is that they are culture reduced rather than culture free, since format familiarity, schooling and prior test exposure all leave a mark. That is also why the norm group behind an accurate score has to be described rather than assumed, since a result only means something relative to the population the test was standardized on.

Consider two societies. In the first, religious participation is the cultural default, and departing from it carries social cost. In the second, secularity is the default and religious commitment is the departure. If ability correlates even weakly with willingness to hold a minority position, the same underlying psychology would produce a negative association in the first society and a positive one in the second. The observed correlation would be a fact about the society, not about religion or intelligence.

This is not a hypothetical objection. Levels of religiosity differ enormously between countries in ways that track history, state policy, economic development and institutional structure rather than anything cognitive. Any account of the individual level association that ignores that variation is describing one cultural moment and calling it a general law.

What would settle it is a large body of comparable research across genuinely diverse societies, using consistent measures. That body of research does not yet exist at the scale required, and until it does, generalizing beyond the studied populations is not supported.

9 The Direction of Causation Is Not Established

Almost every study in this literature measured both variables at one point in time in one group of people. That design can establish that two things travel together. It cannot establish which one moves first, or whether a third thing moves both. The same impasse turns up wherever a trait is correlated with ability: the Openness literature can point to reasoning making complex ideas rewarding, to curiosity accumulating knowledge, or to environments that reward both, and cross sectional data cannot choose between them.

Read the association in each possible direction and notice that all of them are plausible. Higher measured ability might lead to more questioning of inherited belief. Alternatively, growing up in a highly religious environment might be correlated with educational paths that produce lower measured scores, in which case the arrow runs backward through schooling. Or a third factor, such as family socioeconomic position or urban versus rural upbringing, might independently shape both religious participation and test performance without either influencing the other.

Longitudinal designs that follow the same people across years would help enormously, and there are far too few of them. Until that work exists, statements about intelligence "leading to" any religious position are interpretation dressed as finding, regardless of which direction the speaker prefers.

10 What This Research Cannot Be Used to Say

Because this is a topic where findings get weaponized in both directions, it is worth stating the boundaries explicitly rather than leaving them implied.

• It cannot rank individuals. With overlapping distributions and a small average difference, knowing a person's religious position tells you close to nothing about their score, and knowing their score tells you close to nothing about their beliefs.
• It cannot evaluate any belief. Whether a proposition is true is a separate question from the psychological characteristics of people who hold it. Treating a correlation as evidence about theology is a category error that no data can repair.
• It cannot justify condescension. The finding is often deployed as a rhetorical weapon, and doing so requires ignoring both the effect size and every qualification the researchers themselves attached.
• It cannot support policy or institutional decisions. Nothing about group level associations of this magnitude licenses treating individuals differently in any setting.

It is also worth naming the mirror image error. Some readers will want the effect to be exactly zero and will dismiss the entire literature as flawed. That is also a distortion. A small, contested, culturally bounded association is a real finding, and describing it accurately is more useful than pretending it does not exist.

11 Two Things That Are Constantly Conflated

Much of the confusion around this topic collapses if you keep two distinctions in view.

The first is group averages versus individual prediction. A reliable difference in group means and a useful predictor for an individual are entirely different animals. With enough participants, a tiny average difference becomes statistically detectable while remaining practically worthless for saying anything about one person. Nearly every misuse of this research consists of taking a fact of the first kind and asserting it as a fact of the second kind.

The second is religiosity as belief versus religiosity as belonging. Research that separates these two dimensions often finds them behaving differently, which makes sense: participating in a community you grew up in and holding specific metaphysical convictions are not the same act. Studies that lump them together produce averages of two different relationships and report the blend as though it were one thing.

Once those two distinctions are in place, most confident claims you encounter on this subject can be sorted quickly into careful and careless.

A third distinction deserves mention because it trips up otherwise careful readers: statistical significance is not effect size. With a sample of tens of thousands, an association of almost any magnitude becomes significant, meaning only that it is unlikely to be exactly zero. Headlines routinely report significance as though it were strength, and in large modern surveys that translation is close to meaningless. Ask for the correlation or the overlap, never for the p value.

A fourth is the difference between describing a population and explaining it. Even if every measurement problem discussed above were solved and the association held up perfectly, the finding would still be a description of a pattern with an unknown cause. Explanation requires the kind of longitudinal and cross cultural work that this field has barely begun, and confident causal stories in the absence of that work are exactly as speculative in one direction as in the other.

12 What Cognitive Tests Do and Do Not Assess

It helps to be concrete about what a cognitive battery is measuring, because that clarifies why it has nothing to say about belief.

A structured assessment samples performance across defined domains: verbal comprehension, fluid reasoning, visual spatial processing, working memory, processing speed and quantitative reasoning. Each is measured with tasks that have objectively correct answers, scored against a reference population, on a particular day under particular conditions. That is the entire scope of the instrument.

What it does not assess is enormous. It does not assess wisdom, judgment under uncertainty, moral reasoning, meaning making, or the capacity to sustain commitments. It does not assess whether someone's convictions are well founded. It does not assess character. A person can score in the top percentile and reason terribly about the questions that matter most to them, a mismatch that has its own research literature under the heading of rationality, and which sits uncomfortably beside any attempt to use test scores as a proxy for the quality of someone's beliefs.

Anyone who takes a cognitive assessment should understand that boundary clearly, and it applies just as much to what any IQ test measures as to the studies discussed here.

13 How to Read the Next Headline on This Topic

Coverage of this research is reliably worse than the research itself. Four questions will let you assess a claim quickly. Sociological correlates of IQ invite the same misuse everywhere; the delinquency research shows how a modest population correlation gets stretched into claims about individuals.

• What exactly was measured as religiosity? Attendance, belief, identity and fundamentalism give different answers. A headline that says "religious people" without specifying is already imprecise.
• Where were the participants from? A finding from American undergraduates is a finding about American undergraduates until demonstrated otherwise.
• How large is the effect, in overlap terms? Translate correlations into how much the distributions overlap. Most claims deflate immediately at this step.
• Is the design capable of supporting the causal language being used? Cross sectional data with causal verbs attached is the most common failure in coverage of this subject.

Those questions are not specific to this topic. They work on any reported association between a demographic characteristic and a psychological measure, which is a category of claim you will encounter constantly. A fifth question is worth carrying alongside them, which is whether the outcome in the headline was measured at all: the three papers behind the claim that generative AI lowers IQ recorded neural connectivity during essay writing, survey responses at a single point in time and self reported mental effort, and not one of them administered an intelligence test.

14 The Honest Summary

Research in mostly Western samples reports a small negative average association between measured intelligence and several indices of religiosity. The finding replicates across multiple studies well enough to take seriously as a description of those populations. Its size is modest, its measures are heterogeneous, its cultural generalizability is unestablished, its causal direction is unknown, and the proposed explanations range from cognitive style to educational access to social conformity without any of them being demonstrated.

That is genuinely all that can be said, and it is considerably less than what gets said. The gap between that paragraph and the confident claims made in both directions online is the reason this page exists.

If your interest in arriving here was your own cognitive profile rather than the sociology of belief, that is measurable directly, across separate domains with their own scores, rather than inferred from any demographic characteristic. A profile tells you where your measured strengths sit. It tells you nothing about what you should believe, and no test ever will.

15 Frequently Asked Questions

16 Sources Behind This Page

The relationship discussed here comes from published research, and the honest reading includes its limits. These are the primary sources behind the numbers on this page.

• Nisbett, R.E. et al. (2012). Intelligence: new findings and theoretical developments. American Psychologist, 67(2). The broad APA review of what moves measured intelligence and what does not.
• Plomin, R. & Deary, I.J. (2015). Genetics and intelligence differences: five special findings. Molecular Psychiatry. Open access. The standard modern review of what twin and DNA evidence does and does not show.
• Schmidt, F.L. & Hunter, J.E. (1998). The validity and utility of selection methods in personnel psychology. Psychological Bulletin, 124(2). The meta-analysis behind general ability as the strongest single predictor of job performance.
• Strenze, T. (2007). Intelligence and socioeconomic success: a meta-analytic review of longitudinal research. Intelligence, 35(5).
• Deary, I.J., Strand, S., Smith, P. & Fernandes, C. (2007). Intelligence and educational achievement. Intelligence, 35(1). A five year cohort of over 70,000 English schoolchildren.
• Kuncel, N.R., Hezlett, S.A. & Ones, D.S. (2004). Academic performance, career potential, creativity, and job performance: can one construct predict them all? Journal of Personality and Social Psychology, 86(1).
• Whalley, L.J. & Deary, I.J. (2001). Longitudinal cohort study of childhood IQ and survival up to age 76. BMJ, 322(7280). Open access.
• Batty, G.D. et al. (2007). IQ and mortality: a population based cohort study. International Journal of Epidemiology, 36(3), 538-553.
• Zagorsky, J.L. (2007). Do you have to be smart to be rich? The impact of IQ on wealth, income and financial distress. Intelligence, 35(5).
• Judge, T.A., Colbert, A.E. & Ilies, R. (2004). Intelligence and leadership: a quantitative review and test of theoretical propositions. Journal of Applied Psychology, 89(3).