We used an individual differences approach to test three interrelated cognitive hypotheses: 1) the autistic spectrum is inversely related to belief in God, 2) mentalizing mediates this relation, and 3) mentalizing mediates the long-known, but theoretically ambiguous, gender gap wherein men show reduced religious belief than women. Study 1 compared religious belief in a sample of adolescents with clinical diagnoses of autism with a neuro-typical sample matched on relevant socio-demographic characteristics. Studies 2–4 replicated and extended our findings to three distinct non-clinical samples that measured autism as a continuous variable rather than as a clinical diagnosis. These studies also enabled formal mediation analyses, in which we tested our hypotheses with multiple mediation bootstrapping based on 9999 resamples [27], an analytic technique for simultaneously testing multiple potential mediators (thus avoiding the inferential pitfalls of sequential analyses). In all analyses where gender was included, it was coded (female = 0, male = 1), and all other variables were standardized. Studies 2–4 also controlled for additional socio-demographic and psychological variables related to autism, mentalizing, and religiosity that addressed several alternative explanations, which we evaluate in the General Discussion.

For Studies 2–4, approval was obtained from the Behavioral Research Ethics Board (BREB) at the University of British Columbia, and all participants provided written informed consent (for participants who completed the study via the internet, consent was provided by clicking a designated button online; this was approved by the UBC BREB). For Study 1, the parent was given a letter of information and signed a written consent form and the child was given an assent form to read and sign. Approval for the study was obtained from the University of Western Ontario Research Ethics Board.

In a logistic regression model with autism diagnosis and IQ predicting belief in God, autistic participants were only 11% as likely as neuro-typical controls to strongly endorse God, OR = .11, 95% CI = .01, .96, Wald = 3.98, p = .046, and IQ was unrelated to belief, OR = 1.01, 95% CI = .96, 1.06, Wald = .22, p = .64.

This study allowed comparison of individuals with autism diagnoses with a matched control group, but the small sample size rendered statistical mediation testing unfeasible. As an alternative, we entered IQ and parental ratings of adolescent mentalizing tendencies (the two were uncorrelated r = .15, p = .50) as independent predictors of belief in God. In this logistic regression model, only mentalizing was a significant predictor: for each standard deviation decrease in mentalizing, participants were only 21% as likely to strongly endorse God, OR = .21, 95% CI = .06, .73, Wald = 6.07, p = .01, and again IQ was not a significant predictor, OR = 1.03, 95% CI = .98, 1.08, Wald = .96, p = .33.

Study 2 utilized a Canadian student sample and measured belief in God, a standard self-report measure of the autism spectrum, and both mentalizing (The Empathy Quotient) and systemizing [28] as potential mediators. Empathy has been used extensively to detect individual differences in adult mentalizing tendencies, including perspective taking, interest in others' beliefs and desires, and understanding emotions. It is inversely correlated with autism and with being male [8], [28]. Systemizing assesses individual differences in abilities and interests concerning physical and rule-based systems. Because systemizing is positively correlated with autism and with being male [28], but is typically orthogonal to mentalizing, it was included as a second potential mediator. Neither mentalizing nor systemizing has religious content or share conceptual resemblance with the belief in God measure. We tested our two primary hypotheses with autism and gender (respectively) predicting belief in God, and mentalizing (Empathy) and systemizing as potential mediators.

First, we tested our autism-related hypothesis (Fig. 1A). As hypothesized, higher autism scores predicted lower belief in God, β = −.13, p = .02 (controlling for gender). Critically, as predicted, mentalizing significantly mediated this relationship, b_{MentalizingIndirect} = −.07, 95% CI = −.14, −.01. Systemizing was not a significant mediator, b_{SystemizingIndirect} = .00, 95% CI = −.007, .009.

We next examined the effect of gender (Fig. 1B). Consistent with past research, there was a trend for men to report weaker belief in God than women (controlling for the autism spectrum), b = −.21, p = .08 (this gender gap was significant at p = .05 when autism was not controlled). As hypothesized, there was a significant indirect effect, such that men were lower in mentalizing, and in turn lower mentalizing predicted lower belief in God, b_{MentalizingIndirect} = .04, 95% CI = .003, .103; there was no significant indirect effect for systemizing, b_{SystemizingIndirect} = −.001, 95% CI = −.086, .084, even though gender more strongly predicted systemizing (b = .71) than mentalizing (b = −.21).

Study 3 replicated and extended findings from Study 2 to a broad national sample of American adults, controlled for a range of covariates of the autism spectrum and religious belief, and used an alternative measure of belief in and a personal relationship with God. We again tested mentalizing (The Empathy Quotient) and systemizing as mediators. In addition, we measured and tested two additional potential mediators. Two of the five basic facets of personality [29], Conscientiousness and Agreeableness, are reliably related to religious belief and involvement in both previous research [30], [31] and also in the current sample. Therefore it is plausible that they could also explain links between autism or gender and belief in God.

In a logistic regression model with autism spectrum predicting belief in a personal God, controlling for gender, age, educational attainment, income, and frequency of religious attendance, for each standard deviation increase in autism scores, participants were only 80% as likely to strongly endorse a personal God, OR = .80, 95% CI = .66, .97, Wald = 5.25, p = .02. Mentalizing emerged again as a significant mediator of this relationship, b_{MentalizingIndirect} = −.20, 95% CI = −.37, −.04, while systemizing did not, b_{SystemizingIndirect} = −.007, 95% CI = −.03, .008 (Fig. 2A). Furthermore, neither of the two personality measures were significant mediators, b_{ConscientiousnessIndirect} = −.02, 95% CI = −.07, .02, b_{AgreeablenessIndirect} = −.10, 95% CI = −.20, .004.

In a separate logistic regression model, men were only 60% as likely to report strong belief in a personal God as women, controlling for the autism spectrum, age, income, educational attainment, and frequency of religious attendance, OR = .60, 95% CI = .41, .89, Wald = 6.55, p = .01. Replicating the pattern from Study 2, there was a significant indirect effect such that men were lower in mentalizing, and in turn lower mentalizing scores predicted lower belief in a personal God, b_{MentalizingIndirect} = −.15, 95% CI = −.28, −.06 (Fig. 2B); there was no significant indirect effect for any other potential mediator, b_{SystemizingIndirect} = −.007, 95% CI = −.03, .008, b_{ConscientiousnessIndirect} = −.02, 95% CI = −.07, .02, b_{AgreeablenessIndirect} = −.10, 95% CI = −.20, .004. It was noteworthy that systemizing again failed as a mediator of the gender effect, despite the fact that, once again, gender more strongly predicted systemizing (b = .62) than mentalizing (b = −.33) (Fig. 2B).

Finally, in a comprehensive logistic regression model predicting high belief in a personal God, including all predictors (autism and gender), covariates (age, education, income, religious attendance), and potential mediating variables (mentalizing, systemizing, Agreeableness, and Conscientiousness), mentalizing emerged as a specific, independent, and robust predictor of belief in a personal God, OR = 1.41, Wald = 6.01, p = .01 (Table 1). Older age and religious attendance also independently predicted belief in God; Agreeableness and systemizing were statistically marginal predictors (Table 1).

Study 4 again replicated our findings in another national, broad sample of American adults, and controlled for four different covariates (age, education, frequency of religious attendance, and a new 3-item measure of interest in math, science, and engineering, or IMSE). In addition to the Empathy Quotient, we added a second distinct measure of mentalizing-the performance-based Reading the Mind in the Eyes or Mindreading task [32].

In a logistic regression model, the autistic spectrum inversely predicted belief in God, controlling for gender and the four covariates: for each standard deviation increase in autism scores, participants were only 66% as likely to strongly endorse God, OR = .66, 95% CI = .53, .84, Wald = 12.11, p = .001. The two distinct mentalizing measures significantly and independently mediated this relationship: b_{EmpathyIndirect} = −.25, 95% CI = −.41, −.10, and b_{MindreadingIndirect} = −.07, 95% CI = −.14, −.02 (Fig. 3A).

A gender gap was again observed: men were 47% as likely as women to strongly endorse God, OR = .47, 95% CI = .30, .75, Wald = 10.07, p = .002, controlling for all four covariates and the autism spectrum. As hypothesized, both mentalizing measures independently mediated this gender effect: b_{EmpathyIndirect} = .27, 95% CI = .12, .45, and b_{MindreadingIndirect} = .11, 95% CI = .02, .25 (Fig. 3B). In a comprehensive logistic regression model, the two mentalizing measures emerged as significant predictors. Lower education and religious attendance also independently predicted belief in God (Table 2).

We found new evidence for an inverse link between the autism spectrum and belief in God that was explained by mentalizing, as predicted by cognitive theories of religion [10]–[13]. Mentalizing also explained the widely observed [24] gender gap in religious belief. Our findings should be interpreted with some caution; although the results held controlling for several key socio-demographic characteristics, further research conducted in other cultural contexts should assess generalizability of findings [33]. Most of the measures were self-report (or observer-report in Study 1), which are known to have their limitations. Moreover, the correlational natures of the observations are another limitation and preclude definite causal inferences without further experimental research. Nevertheless, results were robust to various methodological checks, including different sampling strategies, alternative measures of autism, mentalizing, and religious belief, and the inclusion of several theoretically relevant control variables addressing several alternative accounts.

Specifically, one alternative is that high levels of autism cause adjustment difficulties in social situations, leading to lower levels of religious attendance, which in turn reduce religious belief. Contrary to this prediction, the effect of autism on belief in God remained significant after controlling for religious attendance (Studies 3–4), and disappeared only after controlling for mentalizing. This demonstrates that the effect of autism on belief exists even after removing the considerable overlap between belief in God and religious attendance. Relatedly, the relationship between the autism spectrum and belief cannot be solely a by-product of the more challenging social circumstances of autistic individuals, as identical patterns emerged when autism was measured as a continuous variable in a non-clinical sample of university students sharing similar social circumstances (Study 2).

A second possible alternative is reverse-causation: that religious involvement somehow causes higher levels of mentalizing, which in turn predict low scores on the autism spectrum. One causal path for this alternative is that belief in God encourages greater social involvement in religious groups and activities, which in turn increases mentalizing tendencies and decreases the likelihood of being on the autism spectrum. This interpretation did not receive support in two studies, because holding constant frequency of religious attendance did not eliminate the effect of mentalizing on belief in God. Moreover, it fails to account for the gender findings (belief in God cannot cause gender), whereas the mentalizing hypothesis parsimoniously explains both the autism and gender effects. Another reverse-causation pathway is that religious involvement leads to greater levels of mentally-simulated social engagement with supernatural agents believed to have elaborate mental states, which in turn encourages more mentalizing, and lower autism scores. Future research could test this hypothesis, but we note that this alternative pathway is compatible with the hypothesis that mentalizing deficits constrain religious belief.

Third, it is possible that the autism spectrum is associated with interest in math, science, and engineering (IMSE), which in turn reduces religious belief. However, Study 4 statistically controlled for IMSE, which did not independently predict belief in God (Table 2). Similarly, systemizing, a variable closely linked to IMSE, failed as a mediator (Studies 2–3). Fourth, the link between autism and low belief in God was not explained by general intelligence: autism remained a significant predictor of low belief in God even after statistically controlling for IQ (Study 1), and education (Studies 3–4), which is typically correlated with IQ. Fifth, the two basic personality dimensions that are most reliably predictive of religiosity, Agreeableness and Conscientiousness [30], similarly failed as mediators.

Cognitive approaches to religion emphasize that a reliably developing social cognitive mechanism-mentalizing or theory of mind-is a key foundation that supports the intuitive understanding of God or gods. Present findings bolster this hypothesis, and further demonstrate that mentalizing deficits undermine not only intuitive understanding of God, but belief as well. Furthermore, these findings suggest one reason why, despite rich sociocultural diversity, key aspects of religion reoccur across history and cultures. Additionally, the robust gender gap in religious belief has been recognized for decades, although its origins continue to be vigorously debated [24]–[26]. Our findings contribute to this debate by providing an important and previously overlooked psychological explanation for the overrepresentation of men among disbelievers.

Finally, we emphasize that our data do not suggest that religious disbelief solely arises through mentalizing deficits; multiple psychological and socio-cultural pathways likely lead to a complex and over-determined phenomenon such as disbelief in God or gods. Therefore, mentalizing deficits are one pathway among several to disbelief. Analytic cognitive processing that suppresses or overrides the intuitions that make theism cognitively compelling [34] and exposure to secular cultural contexts lacking cues that one should believe in God or gods [35] also likely promote religious disbelief. In other words, the present results suggest that disbelief can result from mentalizing deficits, but it can also arise from multiple other sources, holding constant mentalizing tendencies.

A complete scientific account of religious belief and disbelief therefore requires consideration of not only cognitive underpinnings such as mentalizing and other core cognitive biases such as dualistic intuitions and teleological or purpose-driven thinking [12], [36]. Equally important in explaining their cultural prevalence, supernatural agent beliefs­once cognitively available-can be co-opted for motivational and social functions, because of both their palliative effects on existential anxieties [1] and their facilitative effects on cooperation in large, anonymous groups in a cultural evolutionary process [37], [38]. Finally, the prevalence and content of supernatural agent beliefs, although constrained by core social cognitive capacities, respond to and fluctuate with socio-demographic conditions across time and cultures [39]. Within this broader theoretical landscape, these studies present new evidence for a social cognitive mechanism underlying one source of individual differences in religious belief.