1. Introduction
Scholars in the cognitive and evolutionary science of religion (CESR) have attempted to identify the cognitive foundation of religious (dis)belief. In the course of these attempts, they have proposed various hypotheses about atheism (e.g. Norenzayan & Gervais 2013). For example, the mind-blind hypothesis suggests that people who have difficulty representing supernatural religious agents (e.g. gods and spirits) as having human-like mental states tend to be atheists because they do not perceive these agents as affecting their lives and cannot have imaginary interactions with them. The ‘inCREDulous’ hypothesis, based on the cultural learning theory, assumes that people remain indifferent to the existence of supernatural agents unless they are exposed to cultural stimuli, particularly credibility-enhancing displays (CREDs). According to Henrich (2009), CREDs are ‘displays (often actions) that indicate a model’s degree of commitment to, or belief in, verbally expressed representations’ (p. 258). Indeed, CREDs play a role in the transmission of various types of cultural information, such as food preferences, opinions, and altruistic behavior (e.g. Kraft-Todd et al. 2018). Similarly, witnessing religious CREDs, such as frequent prayers and religious activities, can lead to people adopting the model’s religious beliefs. Thus, the inCREDulous hypothesis predicts that those less exposed to religious CREDs are less likely to believe in supernatural agents.
The analytic atheism hypothesis explains atheism in terms of thinking style. Previous research (e.g., Byrd 2025; Evans & Stanovich 2013) has identified two relatively distinct thinking styles: the intuitive and the analytic thinking styles. The intuitive thinking style relies on initial intuition and is less likely to engage in conscious deliberation. In contrast, the analytic thinking style shows the opposite pattern. Because belief in supernatural agents often involves intuitive rather than analytic thinking, this hypothesis predicts that people whose dominant thinking style is analytic will more likely reject religious beliefs.
Although other hypotheses have been proposed (e.g., the apatheism hypothesis), the current study tested the mind-blind, inCREDulous, and analytic atheism hypotheses. In particular, we focused on the analytic atheism hypothesis because the cultural dependence of the association between analytic thinking and religious disbelief has been discussed (e.g., Gervais et al. 2018; Stagnaro et al. 2019; Weiss et al. 2021). The accumulation of data from diverse samples is required. In the following section, we first describe how previous studies have tested this hypothesis and then explain the cultural dependence of the relationship between analytic thinking and religious disbelief.
1.1. Supporting evidence and cultural dependence of analytic atheism
Previous studies have found empirical evidence supporting the analytic atheism hypothesis. For example, some experimental studies have reported that priming analytic thinking or inducing reflection promotes religious disbelief (e.g., Gervais & Norenzayan 2012; Shenhav et al. 2012; Yilmaz et al. 2016). However, there is counterevidence and concern regarding the robustness and replicability of these experimental findings (Camerer et al. 2018; Gervais & Norenzayan 2018; Sanchez et al. 2017; Yonker et al. 2016). Consequently, the experimental results are mixed. Meanwhile, many correlational studies support the analytic atheism hypothesis (e.g., Browne et al. 2014; Gervais et al. 2021; Jack et al. 2016; Pennycook et al. 2012, 2013, 2014; Piazza & Sousa 2014; Razmyar & Reeve 2013). Pennycook et al. (2016) conducted a meta-analysis on the association between religious belief and performance on the cognitive reflection test (CRT), which is a standard measure of analytic thinking, and found a modest but negative overall correlation (r = –.183). These findings suggest that analytic or intuitive thinking plays a role in explaining individual differences in religious (dis)beliefs.
Recent studies have explored the cultural dependence of the association between CRT performance and religious beliefs. For example, Gervais et al. (2018) examined this association across 13 culturally diverse societies and found an overall negative correlation (r = –.187),1 which was nearly identical to Pennycook et al. (2016). However, they also found significant variability across countries as predicted by each country’s religiosity level. The negative association between CRT performance and religious beliefs was stronger in more religious countries (e.g., Australia, Singapore, and the United States), while it was almost non-existent (Czech Republic, Netherlands, and New Zealand) or even positive (the United Kingdom) in less religious countries. This finding suggests that in highly religious countries, social cues (e.g., speech, behaviour, and practices) reinforce the belief that religion is true, leading people to accept religious beliefs naturally. However, highly analytic thinkers may critically evaluate these socially accepted norms, leading them to hold counter-beliefs. Meanwhile, in non-religious societies where religion is not socially dominant, analytic thinking does not play a strong role in countering religious norms, thus eliminating the association.
The findings of Gervais et al. (2018) were tested in subsequent studies. For instance, Stagnaro et al. (2019) found a negative correlation (r = –.176) between CRT performance and religious belief in India, a religious country. However, they also found a negative correlation (r = –.291) in a British sample, contradicting the findings of Gervais et al. (2018). Weiss et al. (2021) examined this relationship in Germany (a low-religiosity country) and found a small negative correlation (average r = –.127, ranging from –.076 to –.205), supporting the findings of Gervais et al. (2018). Further research across diverse regions is necessary to better understand the link between analytic thinking and religious beliefs and their role in the spread of atheism.
1.2. Present study
This pre-registered study (https://doi.org/10.17605/OSF.IO/J7M53) aimed to examine the association between analytic thinking and religious disbelief. Specifically, this study was conducted in Japan, which is known for its low religiosity. For example, Kavanagh and Jong (2020) found that only 10% of Japanese individuals described themselves as ‘religious’. In another study, Japan ranked 44th among 65 countries in terms of religiosity (Gebauer et al. 2017). At the same time, however, Kavanagh and Jong (2020) reported that more than half of Japanese individuals endorsed broad supernatural beliefs. Scores on the Supernatural Belief Scale (SBS; Jong et al. 2013), a standard measure of belief in supernatural agents, were distributed around the midpoint. Similar patterns have been observed in several studies conducted in Japan, where beliefs were measured considering the Japanese religious context, such as respect for deities and buddhas, animistic views of nature, and karmic notions of moral causality (e.g., Ishii & Watanabe 2021, 2023). In addition, Japan has numerous religious institutions, such as Buddhist temples and Shinto shrines, and many religious practices are deeply ingrained in daily life.
Given that such religious beliefs are culturally accepted in Japan, testing the analytic atheism hypothesis in this context contributes to the literature in two main ways. First, although this hypothesis has been tested across countries and regions, most studies have been conducted in Western cultural contexts. While multiple studies have examined this hypothesis in other regions, including the Middle East, Southeast Asia, and East Asia (e.g., Byrd et al. 2025; Gervais et al. 2018; Ghasemi et al. 2025; Stagnaro & Pennycook 2025); research in Japan remains limited. This study therefore fills this empirical gap by utilizing a large Japanese sample. Second, testing the analytic atheism hypothesis in the Japanese religious context allows us to assess the generalizability of the hypothesis. According to Gervais et al. (2018), analytic thinking is believed to play a role in critically evaluating culturally accepted beliefs. If this is the case, we can predict that the same mechanism will operate in the Japanese context; that is, analytic thinkers in Japan may be more critical of religious traditions, leading them to reject religious beliefs. Therefore, we hypothesize that analytic thinking would show a negative association with religious belief in Japan. Examining this theoretical prediction should deepen our understanding of the role of analytic thinking in shaping religious disbelief across cultures.
To test the hypothesis, performance on the CRT (Frederick 2005; Toplak et al. 2014) was used to measure the individual differences in analytic thinking. Two different scales were used to measure religious belief: a 13-item religious belief scale in Study 1 and the Japanese version of the six-item SBS (Jong et al., 2013, 2019) in Study 2. The 13-item scale was developed in a previous study (Ishii & Watanabe 2021) conducted in Japan, and consists of items reflecting Japanese religious traditions, such as animistic thinking rooted in Shinto and beliefs in supernatural agents familiar to Japanese individuals, including gods, Buddha, and ancestral spirits. Thus, it was deemed suitable for measuring religious beliefs in Japan. The SBS is a standard measure of belief in supernatural beings, including an omniscient and omnipotent god, the soul, afterlife, and miracles, and has been used in previous studies to test analytic atheism (e.g., Gervais et al. 2021; Stagnaro et al. 2019; Weiss et al. 2021). Using both scales, which assess different aspects of religious beliefs, we aimed to more accurately assess the association between analytic thinking and religious beliefs. Based on our hypothesis, CRT scores were expected to show small and negative correlations with religious belief scores.
This study also examined variables related to other hypotheses about atheism, namely, the mind-blind and inCREDulous hypotheses. As described earlier, the mind-blind hypothesis suggests that the social cognitive abilities that enable individuals to represent supernatural religious agents are linked to religious beliefs. We focused on empathy, specifically, other-oriented empathy (empathic concern), as a measure of social cognitive ability. This was based on previous studies showing that empathic concern had a stronger association with religious beliefs than other social cognitive abilities, such as accurate mentalising (Ishii & Watanabe 2021; Łowicki et al. 2020; Vonk & Pitzen 2017) and perspective-taking (Jack et al. 2016). The inCREDulous hypothesis suggests that exposure to others’ religious CREDs facilitates the learning of religious beliefs. Because the CREDs framework offers a general mechanism for cultural learning (Henrich 2009), it should also be applicable to the transmission of religious beliefs in the Japanese context. We focused on caregivers’ religious CREDs (vertical transmission) because caregivers are likely to provide the first religious CREDs to which people are exposed, which can have a long-term influence. To measure other-oriented empathy and caregivers’ religious CREDs, we used the Empathic Concern subscale of the Interpersonal Reactivity Index (IRI-EC; Davis 1983) and the CREDs Exposure Scale (Lanman & Buhrmester 2017). Based on previous findings, we expected that both the IRI-EC and CREDs exposure scores would be positively correlated with religious belief scores.
Additionally, we examined which of our target variables (CRT performance, IRI-EC scores, and CREDs exposure scores) better explained individual differences in religious belief scores by simultaneously entering them into a multiple regression model. Thus, this study not only provides data on the analytic atheism hypothesis but also assesses the relative explanatory power of different hypotheses about atheism. Previous research (e.g., Gervais et al. 2021; Jack et al. 2016) suggests that CREDs exposure and empathy are more strongly linked to religious beliefs than analytic thinking. Therefore, we predicted that religious belief scores would be more strongly associated with CREDs exposure and IRI-EC scores than CRT performance. The materials, data, and analysis scripts used for this article can be accessed from the Open Science Framework repository (https://doi.org/10.17605/OSF.IO/P8HGM).
2. Study 1
2.1. Methods
2.1.1. Participants
The sample size was determined prior to data collection. Pennycook et al. (2016) found an overall negative correlation of –.183 between CRT performance and religious beliefs. Although this correlation coefficient could have been used as the expected effect size, our hypothesis suggested a weaker correlation. Therefore, we referred to the findings of Weiss et al. (2021), who examined this association in Germany, a country with low religiosity. Their study reported a correlation coefficient as low as –.076 between CRT performance and religious belief. Based on this, we determined the sample size necessary to detect a correlation of this magnitude with 80% power using the pwr package (Champely 2020) in R (R Core Team 2024). The results indicated that approximately 1,356 participants were required. To ensure a sufficiently large sample size, we recruited at least 1,500 Japanese adults aged 18–60 years via Yahoo! Crowdsourcing in April 2021.
According to our pre-registered criteria, participants who reported their gender as ‘unknown’ or provided unreliable responses (e.g., selecting the same rating for all items in each measure) were excluded. The final sample comprised 1,470 participants (1,002 men and 468 women, Mage = 30.95, SD = 6.68).
2.1.2. Materials and procedure
The participants completed all the questionnaires online via Qualtrics. After completing an informed consent form, the participants answered questions about demographic variables (age, gender, education level, and monthly income level). Next, they rated their level of agreement on a seven-point scale, with 13 items measuring individual differences in religious beliefs (Ishii & Watanabe 2021). Sample items included, ‘I think gods dwell in big trees and rocks that exist in nature’ and ‘I believe in gods or Buddha’. The average rating of the 13 items was calculated as each participant’s religious belief score (α = .91, ω = .92, M = 3.94, SD = 1.23). Subsequently, they completed assessments of analytic thinking, empathic concern, and CREDs exposure. Analytic thinking was measured using the seven-item CRT translated and adapted into Japanese by Harada et al. (2018). This test included three original CRT items (Frederick 2005) and four CRT items developed by Toplak et al. (2014). The total number of correct responses to the seven items was used as each participant’s analytic thinking score (α = .72, ω = .72, M = 4.11, SD = 2.06). We also calculated scores for the original three items (α = .62, ω = .63, M = 1.57, SD = 1.10) and used them in our analysis to ensure comparability with previous studies that used the original CRT.
Empathic concern was measured using the Japanese version of the IRI-EC (Himichi et al. 2017) on a seven-point Likert scale, which included the following sample item: ‘I often have tender, concerned feelings for people less fortunate than me’. The average rating of the seven items was used for each participant’s IRI-EC score (α = .82, ω = .82, M = 4.55, SD = 0.92). Finally, CREDs exposure was assessed using the CREDs Exposure Scale (Lanman & Buhrmester 2017), which included items such as, ‘To what extent did your caregiver(s) attend religious services or meetings?’ A seven-point Likert scale was used, and the mean of the seven items was calculated as each participant’s CREDs score (α = .87, ω = .87, M = 3.27, SD = 1.21).
2.1.3. Data analysis
A hierarchical multiple regression analysis was conducted. All variables were standardised except for gender (coded as female = –0.5 and male = 0.5). In the first model, demographic variables (age, gender, education, and income) were included as control variables, and religious belief scores were included as the response variable. In the second model, CRT scores were added as an explanatory variable to examine whether they were negatively associated with religious belief scores (i.e., the analytic atheism hypothesis). In the third model, IRI-EC and CREDs scores were added to examine whether they were positively associated with religious belief scores (i.e., the mind-blind and CREDs hypotheses). The third model was expected to reveal the explanatory variables that better explained the religious belief scores.
2.2. Results and discussion
The descriptive statistics are presented in Table 1. First, bivariate correlation analyses showed small negative correlations between the 13-item religious belief scores and both the seven- (r = –.156, 95% CI [–.206, –.106], t(1468) = –6.06, p < .001) and the three-item (r = –.147, 95% CI [–.197, –.097], t(1468) = –5.71, p < .001) CRT scores. Religious belief scores were also positively correlated with the IRI-EC score (r = .237, 95% CI [.188, .284], t(1468) = 9.33, p < .001) and the CREDs score (r = .261, 95% CI [.212, .308], t(1468) = 10.34, p < .001 scores. These results were consistent with prior results.
Table 1
Reliability Coefficients, Descriptive Statistics, and Correlation Matrix for Religious Belief and Target Variables in Study 1.
| α/ω | M | SD | CRT (7-ITEM) | CRT (3-ITEM) | IRI-EC | CREDs | |
|---|---|---|---|---|---|---|---|
| Religious belief | .91/.92 | 3.94 | 1.23 | –.156 | –.147 | .237 | .261 |
| CRT (7-item) | .72/.72 | 4.11 | 2.06 | – | .858 | –.011 | –.117 |
| CRT (3-item) | .62/.63 | 1.57 | 1.10 | – | .001 | –.117 | |
| IRI-EC | .82/.82 | 4.55 | 0.92 | – | .132 | ||
| CREDs | .87/.87 | 3.27 | 1.21 | – |
Next, a three-step hierarchical multiple regression analysis was conducted using the seven-item CRT score as a measure of analytic thinking (Table 2). In the first step, demographic variables explained 4.4% of the variance in religious belief scores (R2 = .044, F(4, 1465) = 16.90, p < .001), with a small effect size (Cohen’s f2 = 0.046). Entering the seven-item CRT score in the second step improved the model (ΔR2 = .015, ΔF(1, 1464) = 26.62, p < .001, f2 = 0.063), suggesting analytic thinking is negatively associated with religious belief. Figure 1 illustrates this association using a boxplot with jittered individual data points, allowing visualization of both the overall pattern and individual variability. Finally, including the IRI-EC score and the CREDs score in the third step also improved the model (ΔR2 = .096, ΔF(2, 1462) = 82.99, p < .001), yielding a medium effect size (f2 = 0.184). All target variables were associated with the religious belief score in the third step model. In terms of coefficient values, the religious belief score was better explained by the CREDs score (β = .224, 95% CI [.175, .272], t = 9.09, p < .001), followed by the IRI-EC score (β = .194, 95% CI [.146, .242], t = 7.93, p < .001), whereas the seven-item CRT score had the smallest coefficient (β = –.097, 95% CI [–.147, –.048], t = –3.86, p < .001).
Table 2
Results of Multiple Regression Analysis for the 7-item CRT score in Study 1 (Classical and Bayesian inference).
| STEP 1 | STEP 2 | STEP 3 | BAYESIAN INFERENCE | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| β | 95% CI | t | p | β | 95% CI | t | p | β | 95% CI | t | p | β | 89% HPDI | |
| Age | .067 | [.016, .117] | 2.58 | .010 | .072 | [.021, .122] | 2.80 | .005 | .029 | [–.019, .077] | 1.17 | .241 | .029 | [–.010, .068] |
| Gender | –.454 | [–.572, –.336] | –7.52 | < .001 | –.415 | [–.533, –.296] | –6.86 | < .001 | –.400 | [–.512, –.287] | –6.96 | < .001 | –.397 | [–.488, –.305] |
| Education | –.062 | [–.114, –.010] | –2.32 | .021 | –.033 | [–.086, .020] | –1.21 | .228 | –.037 | [–.088, .013] | –1.44 | .149 | –.037 | [–.078, .003] |
| Monthly Income | .090 | [.033, .147] | 3.09 | .002 | .085 | [.029, .142] | 2.95 | .003 | .069 | [.015, .122] | 2.49 | .013 | .068 | [.024, .112] |
| CRT (7-item) | –.129 | [–.181, –.077] | –4.89 | < .001 | –.097 | [–.147, –.048] | –3.86 | < .001 | –.097 | [–.138, –.058] | ||||
| IRI-EC | .194 | [.146, .242] | 7.93 | < .001 | .194 | [.155, .232] | ||||||||
| CREDs | .224 | [.175, .272] | 9.09 | < .001 | .224 | [.184, .263] | ||||||||
| R2 | .044 | .059 | .155 | |||||||||||
| ΔR2 | .015 | .096 | ||||||||||||
Figure 1
Boxplots with Jittered Individual Data Points Showing Religious Belief Score as a Function of Correct Responses on the 7-item CRT in Study 1.
In addition, we examined the interaction effect between analytic thinking and CREDs exposure because Gervais et al. (2021) found such an interaction, suggesting that analytic thinking is negatively associated with religious beliefs only among those highly exposed to caregivers’ religious CREDs. We added this interaction to the third model; however, it did not improve the model (ΔR2 = .001, ΔF(1, 1461) = 2.41, p = .121), and the coefficient value suggested no interaction (β = .036, 95% CI [–.010, .082], t = 1.55, p = .121).
To further confirm the robustness of the third-step model results, we used Bayesian inference with the ulam function in the rethinking package (McElreath 2020). We assumed that the priors for the regression models (i.e., intercept and slopes) followed a standard normal distribution (μ = 0, σ = 1), whereas the prior for the response variable’s SD followed an exponential distribution with a rate of 1. Markov Chain Monte Carlo (MCMC) samples were obtained from a single chain of 10,000 iterations, with the initial 500 samples for warm-up. This analysis converged successfully, with Rhat values below 1.001 and large effective sample sizes (8132–11059). The means of the posterior distributions for each parameter are presented in Table 2, along with the 89% highest posterior density intervals (HPDI).2 The results revealed that religious belief scores were associated with all the target variables, with the strongest association observed for the CREDs score (β = .224, 89% HPDI [.184, .263]), followed by the IRI-EC score (β = .194, 89% HPDI [.155, .232]) and the seven-item CRT scores (β = –.097, 89% HPDI [–.138, –.058]).
Finally, we conducted the same hierarchical multiple regression and Bayesian analyses using the three-item CRT score as a measure of analytic thinking. The results, which are shown in the supplementary information, were almost identical to the results obtained with the seven-item CRT score.
In summary, Study 1 supported the analytic atheism hypothesis. The CRT scores were negatively correlated with religious belief scores, and this negative relationship persisted even after controlling for demographic variables (age, gender, educational level, and income level) in a multiple regression analysis. Additionally, Study 1 provided evidence supporting the mind-blind and inCREDulous hypotheses, as the IRI-EC and CREDs scores were more strongly associated with religious belief scores than with CRT scores, which is consistent with recent discussions (Gervais et al. 2021; Łowicki & Zajenkowski 2019) emphasising the importance of empathy and CREDs in explaining individual differences in religious beliefs. Study 2 sought to confirm the findings of Study 1 using the SBS as a measure of religious beliefs.
3. Study 2
3.1. Methods
Similar to Study 1, 1,500 participants were recruited through a crowdsourcing service (Yahoo! Crowdsourcing) in January 2022. Following the pre-registered exclusion criteria, we removed participants who reported their gender as ‘unknown’ or provided unreliable responses (e.g., selecting the same rating for all items in each measure), resulting in a final sample of 1,306 Japanese adults (896 men and 410 women, Mage = 31.36, SD = 6.75) in Study 2. The procedures and materials were identical to those used in Study 1 except for the measure of religious beliefs. Study 2 employed the Japanese version of the SBS (Jong et al., 2013; Jong et al., 2019) using a seven-point Likert scale. Each participant’s SBS score was calculated as the average rating of the seven items (α = .90, ω = .91, M = 3.83, SD = 1.40).
3.2. Results and Discussion
The SBS score had small negative correlations with the seven- (r = –.093, 95% CI [–.147, –.039], t(1304) = –3.38, p < .001) and the three-item (r = –.084, 95% CI [–.138, –.030], t(1304) = –3.05, p = .002) CRT scores. The SBS score also correlated with the IRI-EC (r = .192, 95% CI [.139, .244], t(1304) = 7.06, p < .001) and CREDs score (r = .167, 95% CI [.114, .220], t(1304) = 6.13, p < .001) scores. The descriptive statistics are presented in Table 3.
Table 3
Reliability Coefficients, Descriptive Statistics, and Correlation Matrix for Supernatural Belief and Target Variables in Study 2.
| α/ω | M | SD | CRT (7-ITEM) | CRT (3-ITEM) | IRI-EC | CREDs | |
|---|---|---|---|---|---|---|---|
| SBS | .90/.91 | 3.83 | 1.4 | –.093 | –.084 | .192 | .167 |
| CRT (7-item) | .72/.73 | 4.26 | 2.06 | – | .865 | –.050 | –.066 |
| CRT (3-item) | .65/.65 | 1.64 | 1.12 | – | –.053 | –.052 | |
| IRI-EC | .82/.82 | 4.56 | 0.9 | – | .146 | ||
| CREDs | .87/.87 | 3.27 | 1.19 | – |
The same three-step hierarchical multiple regression analysis was performed as in Study 1 (Table 4). The first model, which included only the control variables, accounted for 3.6% of the variance of the religious belief score with a small effect size (R2 = .036, F(4, 1301) = 12.17, p < .001, f2 = 0.037). Adding the seven-item CRT score in the second step improved the model (ΔR2 = .005, ΔF(1, 1300) = 7.22, p = .007, f2 = 0.043), suggesting that analytic thinking is associated with decreasing religious beliefs (Figure 2). The third step further improved the model (ΔR2 = .045, ΔF(2, 1298) = 32.18, p < .001, f2 = 0.095). In this model, the seven-item CRT score (β = –.059, 95% CI [–.112, –.005], t = –2.14, p = .033) was negatively associated with the SBS score, as predicted. The IRI-EC and CREDs scores were positively associated with the SBS score. However, in contrast to Study 1, the strength of the association was stronger for the IRI-EC score (β = .150, 95% CI [.097, .203], t = 5.51, p < .001) than for the CREDs score (β = .136, 95% CI [.084, .189], t = 5.06, p < .001). Adding the interaction between the CRT and CREDs scores did not improve the third-step model (ΔR2 = .001, ΔF(1, 1297) = 1.19, p = .275), suggesting again that there is little evidence of an interaction (β = –.029, 95% CI [–.082, .023], t = –1.09, p = .276).
Table 4
Results of Multiple Regression Analysis for the 7-item CRT score in Study 2 (Classical and Bayesian inference).
| STEP 1 | STEP 2 | STEP 3 | BAYESIAN INFERENCE | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| β | 95% CI | t | p | β | 95% CI | t | p | β | 95% CI | t | p | β | 89% HPDI | |
| Age | .091 | [.037, .145] | 3.31 | .001 | .092 | [.038, .146] | 3.35 | .001 | .053 | [.000, .107] | 1.96 | .050 | .053 | [.011, .096] |
| Gender | –.390 | [–.515, –.266] | –6.15 | < .001 | –.368 | [–.493, –.242] | –5.75 | < .001 | –.343 | [–.466, –.220] | –5.48 | < .001 | –.342 | [–.441, –.243] |
| Education | –.011 | [–.066, .045] | –0.37 | .709 | .005 | [–.052, .062] | 0.18 | .855 | .007 | [–.049, .062] | 0.24 | .810 | .007 | [–.039, .052] |
| Monthly Income | .021 | [–.031, .081] | 0.70 | .486 | .015 | [–.045, .075] | 0.49 | .624 | .018 | [–.041, .077] | 0.61 | .544 | .018 | [–.030, .066] |
| CRT (7-item) | –.074 | [–.128, –.019] | –2.62 | .009 | –.059 | [–.112, –.005] | –2.14 | .033 | –.059 | [–.103, –.014] | ||||
| IRI-EC | .150 | [.097, .203] | 5.51 | < .001 | .150 | [.107, .193] | ||||||||
| CREDs | .136 | [.084, .189] | 5.06 | < .001 | .137 | [.093, .180] | ||||||||
| R2 | .036 | .041 | .086 | |||||||||||
| ΔR2 | .005 | .045 | ||||||||||||
Figure 2
Boxplots with Jittered Individual Data Points Showing Supernatural Belief Scale Score as a Function of Correct Responses on the 7-item CRT in Study 2.
We also conducted the same Bayesian analysis as in Study 1, and the results of the third-step model were confirmed: the seven-item CRT scores were negatively associated with SBS scores (β = –.059, 89% HPDI [–.103, –.014]), whereas the IRI-EC (β = .150, 89% HPDI [.107, .193]) and CREDs (β = .137, 89% HPDI [.093, .180]) scores were positively associated with SBS scores. The means of the posterior distributions and 89% HPDI for each parameter are listed in Table 4.
Multiple regression analyses (classical and Bayesian inferences) using the three-item CRT score as a measure of analytic thinking yielded results that were almost identical to those in Study 1 (see Supplemental information).
Overall, Study 2 replicated and extended the findings of Study 1 by using the SBS as a measure of religious beliefs. Consistent with Study 1, analytic thinking, as measured by the CRT performance, and CREDs exposure were associated with religious beliefs, though the effect sizes (correlation coefficients) were smaller than those in Study 1. This may be because the 13-item scale used in Study 1 was designed to reflect Japanese religious traditions better. Study 2 also confirmed positive associations between religious beliefs and empathic concern as well as CREDs exposure. Thus, Study 2 provided further support not only for the analytic atheism hypothesis, but also for the mind-blind and inCREDulous hypotheses.
4. General Discussion
This study tested the analytic atheism hypothesis in Japan. The results from two large-sample surveys using two different measures of religious beliefs (the 13-item scale and the SBS) consistently showed that CRT performance was negatively associated with both measures, supporting the analytic atheism hypothesis. This study also tested two additional hypotheses regarding atheism: the mind-blind and inCREDulous hypotheses. Consistent with predictions derived from the hypotheses, both empathic concern and exposure to caregivers’ religious CREDs were positively associated with religious beliefs. Furthermore, multiple regression analyses with CRT performance, empathic concern, and CREDs exposure as explanatory variables demonstrated that all three explained religious beliefs even after controlling for demographic variables. These findings indicate that each variable contributes independently to explaining individual differences in religious beliefs, highlighting the need to consider the three hypotheses as complementary frameworks to understand the formation, acquisition, and reinforcement of religious beliefs.
The first contribution of this study is that it provides evidence for the analytic atheism hypothesis by using Japanese data. According to Gervais et al. (2018), the association between analytic thinking and religious disbelief arises because analytic thinking promotes scepticism towards socially accepted religious norms. Although Japanese society is deeply rooted in religious traditions, its overall religiosity is relatively low compared to other countries (e.g. ranked 44th out of 65 countries; Gebauer et al., 2017). Consequently, the role of analytic thinking in the rejection of religious beliefs in Japan is expected to be relatively small. The negative correlation between the CRT and religious belief scores found in this study was –.120 on average (ranging from –.084 to –.156), which is slightly weaker than the correlation reported in previous meta-analyses (–.183; Pennycook et al. 2016) and cross-cultural studies (–.187; Gervais et al. 2018). Additionally, this correlation coefficient closely resembles the value observed in a previous study conducted in Germany (–0.127 on average, ranging from –0.076 to –0.205; Weiss et al. 2021), a country with similarly low religiosity (ranked 54th, Gebauer et al. 2017). These findings support the argument of Gervais et al. (2018) regarding the role of analytic thinking in shaping religious disbelief. Moreover, our results are consistent with the latest findings of studies that included participants from Asian countries. For example, Byrd et al. (2025) reported that the three-item CRT score was negatively correlated with religiosity in diverse countries, including the Philippines; similarly, Ghasemi et al. (2025) and Stagnaro and Pennycook (2025) showed the same pattern with respondents from countries such as China and Malaysia.
The second contribution of this study was that it showed that empathic concern and caregiver’s religious CREDs were both positively associated with religious beliefs. Thus, we provided evidence for the mind-blind and inCREDulous hypotheses. The positive correlation between the IRI-EC and religious belief scores has already been shown in previous studies in Japan (Ishii & Watanabe 2021, 2023), and this study further supports these findings. However, with a few exceptions (Gervais & Najle 2015; Ishii & Watanabe 2024), this is the first study to examine the association between religious CREDs and beliefs in Japan. For instance, Ishii and Watanabe (2024) conducted a secondary analysis of data from 43 countries and regions collected by the International Social Survey Programme and found that greater exposure to parents’ religious CREDs (e.g., frequent participation in religious activities) was associated with stronger religious beliefs. They included data from Japan. In this study, we re-examined this association using the CREDs exposure scale (Lanman & Buhrmester 2017) and two different measures of religious beliefs. The results from Studies 1 and 2 also indicate that greater exposure to CREDs is associated with increased religious beliefs, further supporting the argument that religious beliefs are culturally transmitted through CREDs.
This study allowed us to discuss the relative contributions of the three hypotheses regarding atheism. In both studies, the correlation with religious belief scores was stronger for IRI-EC and CREDs scores than for CRT scores. These results suggest that analytic thinking has relatively low explanatory power for individual differences in religious beliefs. However, as previously mentioned, the strength of the relationship between analytic thinking and religious beliefs depends on the overall level of religiosity in a given society. Thus, it is premature to conclude that analytic thinking has inherently low explanatory power based solely on the present findings. Notably, in a previous study conducted in the United States (Gervais et al. 2021), a country with relatively high religiosity (ranked 26th, Gebauer et al. 2017), exposure to CREDs was more strongly associated with religious beliefs than with analytic thinking. This suggests that the stronger explanatory power of CREDs exposure over analytic thinking may be a cross-cultural phenomenon. To draw firm conclusions, future studies should compare data from countries with different religiosity levels.
Finally, the interaction between analytic thinking and CREDs exposure reported by Gervais et al. (2021) was not consistently observed in the present study. Gervais et al. (2021) suggested that a negative association between analytic thinking and religious beliefs emerged primarily among individuals with low exposure to their parents’ religious CREDs. In other words, high exposure to parents’ religious CREDs may lead to deep devotional beliefs that remain resistant to analytic thinking. The absence of this interaction in the present study may be related to the nature of religion in Japan. As previously noted, religious practices are interwoven with daily customs in Japan, but few individuals regularly visit temples or shrines or participate in religious events. The average CREDs score in our sample was below the midpoint, suggesting that parents’ religious CREDs may be less prevalent in Japan. If so, the absence of this interaction in our study may be due to the limited variability in CREDs exposure. This raises the possibility that the interaction between analytic thinking and CREDs exposure may depend on the society’s overall level of religiosity. These findings warrant further investigation in future research.
Because this study employed a correlational design and a self-report measure of exposure to caregivers’ religious CREDs, we could not directly address the causal relationship that our hypotheses assumed. Moreover, researchers have noted that participants’ familiarity with CRT can be problematic. For example, Byrd (2023) found that familiarity with CRT, rather than task performance itself, was associated with certain beliefs. Although CRT does not appear to be widely used in Japan, this study may have nevertheless been affected by this familiarity problem. Future research should address this issue to confirm the robustness of our findings.
Despite these limitations, by adopting a well-established research paradigm, this study provides robust data on the associations among analytic thinking, empathic concern, exposure to CREDs, and religious beliefs in Japan. In this regard, our work contributes to broader discussions on the causes and dissemination of atheism. Future research should aim to replicate these findings in different countries and regions and explore additional hypotheses that were beyond the scope of this study, such as the apatheism hypothesis.
Additional File
The additional file for this article can be found as follows:
Supplementary Information
Figures S1 and S2 present partial residual plots showing the relationship between the 7-item CRT and religious belief in Studies 1 and 2, respectively. Tables S1 and S2 present the results of multiple regression analyses using the 3-item CRT score in Studies 1 and 2, respectively. DOI: https://doi.org/10.5334/snr.229.s1
Notes
[1] The correlation coefficient was calculated using the data available at https://osf.io/v53c4/files/osfstorage.
Ethics and Consent
The study was approved by the institutional review board (IRB) of Waseda University (2017-090). All procedures were carried out in accordance with the ethical principles involving human subjects stated in the Declaration of Helsinki. Written informed consent was obtained from all participants in advance.
Data Accessibility Statement
The data used in this paper is publicly available at https://doi.org/10.17605/OSF.IO/P8HGM.
Author Contributions
Tatsunori Ishii: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, writing – original draft, writing – review & editing, visualization, funding acquisition. Katsumi Watanabe: resources, writing – review & editing, supervision.