Do Empathetic People Have Strong Religious Beliefs? Survey Studies with Large Japanese Samples

ABSTRACT

The exploration of personality factors to explain individual differences in religiosity has demonstrated a link between empathic concern and religious beliefs using the Empathic Concern subscale of the Interpersonal Reactivity Index (IRI-EC). Research in the cognitive science of religion emphasized the role of empathizing ability related to mentalizing in acquisition of religious belief and has demonstrated the relationship between the Empathy Quotient (EQ) and religious belief. The current study was designed to compare the strength of relationships between religious belief and two representative measures of empathy (the IRI-EC and the EQ). Study 1 aimed to statistically evaluate the strength of the relationship between the EQ/IRI-EC and religious belief with four Japanese samples (Ns = 207, 155, 208, 183). The mini meta-analysis results with random effect model indicated that the effect size (semi partial correlation, r_sp) of the IRI-EC (r_sp = .120, 95%CI [.0002, .237]) was larger than that of the EQ (r_sp = .074, 95%CI [−.0001, .147]). Moreover, these results were confirmed by Study 2 (N = 1440). Thus, the present study provided reliable evidence of the link between empathy and religious belief in non-Western samples. We discuss how empathic concern and mentalizing-related empathy contribute to acquiring religious beliefs.

Abbreviations: EQ: Empathy Quotient; IRI-EC: Interpersonal Reactivity Index-Empathic Concern; WEIRD: Western, educated, industrialized, rich, and democratic

Introduction

Religion is a ubiquitous part of human cultures, with most societies having some form of religious practices. This may be, in part, because religion has functions that help humans survive (e.g., Fincher & Thornhill, 2012; Terrizzi et al., 2012) and maintain well-being (e.g., Ebert et al., 2020; Lun & Bond, 2013; Routledge et al., 2017; Steger & Frazier, 2005). However, there are individual differences in the extent of religiosity, with some individuals being more religious than others. These individual differences in religiosity have received much attention from personality and social psychologists (e.g., M. W. Eysenck, 1998), leading to personality being regarded as a significant factor in religiosity.

For example, the classical studies of H. J. Eysenck (e.g., H. J. Eysenck, 1975, 1976) found two personality factors (radicalism/conservatism and tough-/tender-mindedness) that explained individual differences in various social attitudes. They reported that religious attitudes (e.g., “whether or not do you agree with the divine law?”) were more strongly associated with tender-mindedness, but not to radicalism. Subsequent studies have found similar evidence (e.g., Francis & Katz, 1992; White et al., 1995; Wilde & Joseph, 1997; Williams & Francis, 2006). White et al. (1995) asked their adult participants to answer various questionnaires, including the Eysenck Personality Questionnaire and Francis Scale of Attitudes toward Christianity, and found a negative association between psychoticism and religiosity (i.e., loaded on the same principal component). Using the same scales, Williams and Francis (2006) reported partial correlations between personality and religiosity after controlling for gender. Religiosity was neither correlated with Extraversion (r = .01) nor Neuroticism (r = −.08); however, it correlated negatively with Psychoticism (r = −.22). This negative correlation between religiosity and psychoticism was also found with non-Christian samples (Israeli: Francis & Katz, 1992, r = −.15; Muslim: Wilde & Joseph, 1997, r = −.32).

Recently studies that focused on the “dark” personality (e.g., narcissism, Machiavellianism, psychopathy) have also reported similar results (Aghababaei et al., 2014; Haddad et al., 2016; Lilienfeld et al., 2014; Łowicki & Zajenkowski, 2017). For instance, people with religious affiliations scored lower in psychopathy measured by the Dark Triad Dirty Dozen (Haddad et al., 2016) or Psychopathic Personality Inventory-Revised (Lilienfeld et al., 2014) than those with no religious affiliations. Aghababaei et al. (2014) found that intrinsic religious orientation was negatively correlated with the Dark Triad personality, especially Psychopathy (r = −.16) and Machiavellianism (r = −.14). Łowicki and Zajenkowski (2017) also reported similar correlations (r = −.18, −.27, respectively).

Empathic concern and religious belief

These findings support the idea that particular personality traits contribute to individual differences in religiosity. However, why is psychoticism or dark personality linked to religiosity? One possibility is that empathy mediates this link (e.g., M. W. Eysenck, 1998; Łowicki & Zajenkowski, 2017). For example, M. W. Eysenck (1998) summarized the existing data at the time and concluded that the inverse link between Psychoticism and religiosity reflects “a general lack of concern about the welfare of others” (p. 14). Regarding the dark personality, Jonason and Kroll (2015) examined the relationship between the Dark Triad Dirty Dozen and the Interpersonal Reactivity Index (IRI, Davis, 1983). They found that both Psychopathy and Machiavellianism can be characterized as a lack of empathic concern, which means “other-oriented feelings of sympathy and concern for unfortunate others” (Davis, 1983, p. 114). Recently, Łowicki and Zajenkowski (2017) showed that the inverse links between Psychopathy or Machiavellianism and belief in God were mediated by the score of empathy items (e.g., “I feel others’ emotions,” “I anticipate the needs of others”).

The link between empathy and religious belief was also suggested in areas outside of personality psychology. For example, several religious studies showed a positive correlation between the empathic concern subscale of the IRI (IRI-EC) and religiosity. For example, Khan et al. (2005) found a positive correlation between the IRI-EC and intrinsic religious motivation (r = .32) among a sample of Pakistani students (N = 168). Hardy et al. (2012) examined the relationship among American adolescents (N = 502). They found that the IRI-EC was positively correlated with religious commitment (“I spend time trying to grow in understanding of my faith,” r = .35) and with religious involvement (“How often do you pray or meditate about spiritual or religious topics?,” r = .20). Jack and colleagues (e.g., Jack et al., 2013, 2016; Jack & Robbins, 2012) argued, based on their neuroscientific studies, that the spiritual worldview, which is deeply associated with religiosity, reflects brain activity of the default mode network (DMN). Moreover, they contended that the DMN has a reciprocal inhibitory relationship with the task-positive network (TPN), which is involved in mechanical worldview. They also claimed that individual differences in prosocial behavior, moral reasoning, and notably – empathic moral concern – are related to DMN activity; thus, empathic concern is linked to religiosity. Subsequently, Jack et al. (2016) demonstrated that the IRI-EC positively predicted belief in gods even after other types of empathy (e.g., perspective-taking) and analytic thinking were statistically controlled. In the psychology of self, Fetterman et al. (2020) hypothesized that “heart-locators” (people who perceive the self as located in their heart) are more religious than “head-locators” (who rely on logic). This is because heart-locators are considered to rely more heavily on their feelings and intuition, and are also “warm-hearted” toward others compared to head-locators. They showed that belief in gods were stronger among heart-locators than head-locators, and more importantly, that this relationship was mediated by empathic moral concern, which was higher among heart-locators.

Perspective-taking refers to the “tendency to spontaneously adopt the psychological point of view of others” (Davis, 1983, p. 113). Notably, within the IRI, it could be less related to religious belief. For example, Jack et al. (2016, Studies 4 and 6) reported that the IRI-EC score was positively correlated with belief in gods (r = .212/.209); meanwhile, the perspective-taking subscale (IRI-PT) was not (r = .019/068). Another report concluded that the IRI-PT was correlated with religiosity as strongly as the IRI-EC in a Pakistani sample (r = .28–34, Khan et al., 2005). Conversely, however, a recent large survey conducted in the United States (Gervais et al., 2021) showed almost zero correlation between the IRI-PT and religious belief (r = .04).

Thus, the moderate relationship between empathy, especially empathic concern and religious belief, has been theoretically and empirically supported for years. Meanwhile, a related, but different argument regarding this link has been proposed in an emerging research area: the cognitive science of religion.

Mentalizing and religious belief

The cognitive science of religion has explored the psychological foundation of religion from the perspective of cognitive and evolutionary science (e.g., Kirkpatrick, 1999; Norenzayan et al., 2016). For example, belief in supernatural agents (i.e., gods, spirits), is a fundamental part of religious belief. The cultural evolutionary theory of the origins of prosocial religions (Norenzayan et al., 2016) assumed that this belief is rooted in social cognitive abilities, such as mentalizing, which is also known as “theory of mind.” This ability allows one to impute mental states to others, animals, and even non-living things (Heider & Simmel, 1944; Premack & Woodruff, 1978). According to the theory, mentalizing allows people to treat supernatural agents as having human-like mental states and enables people to have imaginary interactions with those agents (e.g., prayer, rituals). Through this process, people come to believe in supernatural agents, which is known as the mentalizing hypothesis.

This hypothesis has typically been tested using the Empathy Quotient (EQ; Baron-Cohen & Wheelwright, 2004). The EQ assesses two main components of empathy: a cognitive component (i.e., “understanding and/or predicting what someone else might think, feel, or do”), and an affective component (i.e., “feeling an appropriate emotion triggered by seeing/learning of another’s emotion”). The latter includes sympathy (i.e., “feeling an emotion triggered by seeing/learning of someone else’s distress which moves you to want to alleviate their suffering”). The EQ was originally designed as a single dimension scale because both components “co-occur and cannot be easily disentangled (Baron-Cohen & Wheelwright, 2004, p. 166).” Empirically, the unidimensionality of EQ has been demonstrated in large data sets including healthy subjects (e.g., Allison et al., 2011). Usually, only the cognitive component of EQ is referred to as mentalizing (e.g., Baron-Cohen & Wheelwright, 2004). However, items of both components of the EQ are associated with sensitivity to what others are thinking and feeling (e.g., Apperly, 2012). Thus, the entire EQ score reflects individual differences in our tendency to understand others using mentalizing ability.

Many previous studies successfully demonstrated the positive relationship between EQ and religious belief. For example, Norenzayan et al. (2012) showed that individual differences in people with autism (which is presumably associated with lack of mentalizing ability) and healthy people, were inversely related to belief in God. Moreover, EQ mediated this relationship. Subsequent studies also demonstrated that the positive correlation between the EQ and religious belief in Western (e.g., Lindeman et al., 2015; Norenzayan et al., 2012; Routledge et al., 2017; Willard & Norenzayan, 2013; Willard et al., 2019) and East Asian samples (Ishii, 2017; Ishii & Watanabe, 2021).

Norenzayan et al. (2012) further revealed that individual differences in the Reading the Minds in the Eyes test (RMET; Baron‐Cohen et al., 2001) was positively related to belief in God, along with EQ. The RMET assesses how accurately people match mental state words (e.g., playful, upset, anticipatory) to facial expressions in the eye region, and thus reflects the accuracy of mental state attribution. However, in more recent studies, the correlation between RMET scores and religious belief was negligible (Jack et al., 2016, r = .04; Lindeman et al., 2015, r = .02; Łowicki et al., 2020, r = −.08), or slightly negative (Vonk & Pitzen, 2017, r = from −.23 to .07). Similarly, Ishii and Watanabe (2021) reported that the RMET did not predict religious belief in a large sample of Japanese participants (r = −.01, .02).

Thus, we can conclude that the mentalizing-related broader empathizing ability is linked to religious belief. It should be noted that because some researchers failed to demonstrate the relationship between the EQ and religious belief (e.g., Maij et al., 2017; Vonk & Pitzen, 2017), this relationship is not so strong.¹

Empathic concern and mentalizing

As discussed thus far, the link between empathic concern (the IRI-EC) and religious belief has been established; meanwhile, the link between mentalizing-related empathy (EQ) and religious belief has been emphasized in the cognitive and evolutionary science literature.

The IRI-EC and the EQ are both measures of empathy, and thus involve common psychological function (e.g., general empathy and sympathy). In fact, they are often positively correlated (e.g., Jack et al., 2016; Łowicki et al., 2020). However, the IRI-EC and EQ should still be considered as distinct measures for several reasons. First, their focuses are different. Specifically, the IRI-EC focuses on the extent to which respondents care about others welfare or well-being, and the latter focuses on the extent to which respondents tend to understand others psychologically. Second, there is little overlap between the items in both measures. For example, a typical item of the IRI-EC is, “I often have tender, concerned feelings for people less fortunate than me.” This differs widely from the EQ items, except for “I get upset if I see people suffering on news programmes.” Especially, no such items appear in the short version of the EQ (Wakabayashi, Baron-Cohen, Wheelwright, Goldenfeld et al., 2006). Moreover, considerable evidence demonstrates the dissociation between the mentalizing and empathic concerns in clinical samples (e.g., autism spectrum disorders and psychopathy).

Thus, the IRI-EC–religious belief and the EQ–religious belief links should be considered as distinct psychological processes. Specifically, empathic concern is associated with intuitive and spiritual thinking (e.g., Fetterman et al., 2020; Jack et al., 2016). Thus, people with high empathic concern readily accept the existence of supernatural agents. On the other hand, mentalizing-related empathy facilitates the anthropomorphism of supernatural agents and imaginary interactions with them (e.g., Norenzayan et al., 2016). Although most previous studies have focused on one of the processes, we addressed both to further understand the relationship between empathy and religious belief. Thus, the present study examined the extent to which the EQ and IRI-EC predicts individual differences in religious belief.

Present study

To our knowledge, only a few studies have compared the predictive powers of the EQ and the IRI-EC in terms of religious belief. For instance, Jack et al. (2016) recruited 526 online participants and showed that both the EQ (r = .16) and the IRI-EC (r = .21) were positively correlated with religious belief (i.e., “Do you believe in the existence of either God or a universal spirit?”). However, when both empathy measures were entered as explanatory variables, the IRI-EC (β = .63), not the EQ (β = .04), was the statistically significant predictor in their multiple regression model. Although their model included several other variables outside our scope (e.g., analytic thinking measures), Łowicki et al. (2020) employed the Centrality of Religiosity Scale (Huber & Huber, 2012) as a measure of religious belief, and showed similar results (N = 236). They demonstrated that both the EQ (r = .24) and the IRI-EC (r = .34) were correlated with religious belief. Moreover, in the regression model with both measures predicting religious belief, only the IRI-EC (β = .30), but not the EQ (β = .10), was statistically significant.

Based on these results, we hypothesized that mentalizing would be less related to religious belief than empathic concern. Thus, we predicted that the IRI-EC would have substantially more predictive power than the EQ. We examined this prediction with large Japanese samples to accumulate further evidence on the relationship between the IRI-EC, the EQ, and religious beliefs. Our prediction is not a novel one. However, considering the diversity of religious cultures worldwide, it is crucial for the psychological study of religion to examine the robustness and generalizability of previous findings. Especially, previous studies were conducted in Western countries, where Christian culture is predominant. In contrast, the current studies were conducted in East Asia, Japan.

We believe it is worth examining the relationship between empathy and religious belief among such samples. Japan’s religious culture differs from Western societies; with a long tradition of Shinto or Buddhist religions and is known as a less religious country (e.g., Staufenberg, 2016). For instance, Kavanagh and Jong (2020) reported that only 10% of Japanese respondents (N = 1000) described themselves as “religious,” while 55% responded “non-religious,” 22% “atheist,” and 13% “don’t know.” Ishii (2017) also reported that only 9 out of 140 undergraduate participants affirmed having faith. However, like Western countries, over half of the Japanese population endorsed broad religious/supernatural beliefs. That is, scores on measures of supernatural belief (e.g., the Supernatural Belief Scale, Jong et al., 2013; the belief in God items, Norenzayan et al., 2012) were normally distributed around the midpoints of the scales (Kavanagh & Jong, 2020). Relatedly, Stark et al. (2005), analyzed data from the World Value Survey (1992, 1995). They found that 57% of Japanese people believed in the supernatural, and 72% recognized such beliefs as important, although most of them (76%) did not consider themselves as religious. Similarly, Hayashi (2006) found that only 30% of Japanese participants had religious faith, while 70% considered religious feeling as important. In sum, while most Japanese individuals believe in supernatural agents at the same level as in Western societies, they differ in religious cultures and (low) levels of religiosity. Hence, using Japanese samples, we can examine how having different religious cultures/levels of religiosity is related to the link between empathy and religious belief. This is important because some findings in Western culture could be context-sensitive (e.g., Ebert et al., 2020; Henrich et al., 2010). There is an added advantage to employing Japanese samples given the WIERD (Western, Educated, Industrialized, Rich, and Democratic) problem (Henrich et al., 2010). Japan is a non-Western, but still well-educated, industrialized, rich, and democratic society, the same as Western societies. Thus, the educational, economic, and political state need not be considered as potentially excessive confounding factors when comparing our results with that of previous studies.

We tested our predictions with two studies. Previous studies conducted in Japan (Ishii, 2017; Ishii & Watanabe, 2021) have consistently shown a positive correlation between EQ and religious belief, employing the Belief in God Scale (Norenzayan et al., 2012). Thus, it is expected that empathy is also related to religious belief in Japanese samples. However, the extent to which the EQ and the IRI-EC predicts religious belief has remained overlooked in Japan. Therefore, in Study 1, we gathered four independent sample groups to examine the relationship between the two measures of empathy and religious belief. We further assessed the strength of the relationship using the meta-analytic method (i.e., computed the effect sizes of the two empathy measures). Moreover, we confirmed these results in Study 2; wherein, the sample size was determined by power analysis using the effect sizes computed in Study 1.

Study 1

Study 1 aimed to evaluate how closely the two measures of empathy were tied to religious belief. Four independent samples were collected, and a mini meta-analysis was conducted. Notably, our initial intention was to analyze the four samples as independent studies. However, the results of the samples were not consistent among them (see: 2.2. Results). Thus, we conducted a mini meta-analysis of all samples together to avoid redundancy.

Methods

Samples

A power analysis was conducted to determine the sample size. A previous study conducted in Japan (Ishii, 2017) reported the bivariable correlations between the EQ and religious belief were .19–.22. Because the IRI-EC was expected to show a larger correlation with religious belief (e.g., Jack et al., 2016), we calculated the sample size based on the report, using R with pwr package (α = .05, 1−β = .80). The results indicated that an optimum of 158–215 participants were needed. We then recruited the recommended number of participants.

The first sample was 207 undergraduate students (M_age = 19.04 years, SD_age = 1.42, 85 men, 121 women, one unknown) from a university in Tokyo, Japan. They participated in Study 1 in exchange for course credits in an introductory psychology class in May 2018. The second sample comprised 156 undergraduate students (M_age = 18.96, SD_age = 0.61, 122 men, 33 women, one unknown) attending a cognitive science class at another university in Tokyo and participated in this study as volunteers in January 2019. The third and fourth samples were collected through a crowdsourcing service (Yahoo! Crowdsourcing) in March and May 2019, respectively. We collected data from 208 adults (M_age = 43.21, SD_age = 9.47, 156 men, 52 women) in the third sample and 185 younger adults (M_age = 25.65, SD_age = 3.50, 88 men, 95 women, two unknown) in the fourth sample.

Materials and procedure

Similar to Ishii (2017), another Japanese study, we used the 13-item measure of religious belief. This measure consisted of three items measuring “the apotheosis of natural products,” a subscale of the Animism Scale for Adults (Ikeuchi, 2010), six items testing “belief in spirituality,” a subscale of the short edition of the Attitudes Toward Paranormal Phenomena Scale (Sakata et al., 2012), and four items from the Belief in God measure used in Norenzayan et al. (2012). The original belief in god measure has five items. However, one of them (“I believe in God”) was excluded, as a similar item (“I believe in gods and buddhas”) was used from belief in spirituality subscale. Some examples of items used are, “I think gods dwell in big trees and rocks that exist in nature” (the apotheosis of natural products), “I believe in gods and buddhas” (belief in spirituality), “When I am in trouble, I find myself wanting to ask god for help” (belief in god). All items were rated on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The average of the ratings (1–5) for each item was calculated for each participant’s religious belief score (Sample 1: α = .85, M = 3.19, SD = 0.68; Sample 2: α = .87, M = 2.87, SD = 0.86; Sample 3: α = .92, M = 2.96, SD = 0.85; Sample 4: α = .89, M = 2.98, SD = 0.85).

The 13-item religious belief measure included items that reflect not only a belief in gods but also paranormal belief (e.g., the afterlife, a guardian spirit, and previous life). Although both beliefs can be unified into a broader supernatural belief, which is a core aspect of religious belief (e.g., the supernatural belief scale; Jong et al., 2013), some differences between these two beliefs in terms of their relationship with empathy have been reported (e.g., Willard & Norenzayan, 2013). Moreover, our focus was mainly the belief in supernatural agents, rather than paranormal belief itself. Thus, in addition to the full 13-item measure, we also conducted analyses using the 5-item belief in god measure (Norenzayan et al., 2012) as a dependent variable although the internal consistency of this measure was sometimes low (see Cronbach’s alphas below). An item from the belief in spirituality scale (“I believe in gods and buddhas”) was employed as a substitute for the excluded one (“I believe in God”). The average of the ratings for the five items was calculated for each participant’s 5-item religious belief score (Sample 1: α = .57, M = 3.32, SD = 0.71; Sample 2: α = .39, M = 3.02, SD = 0.74; Sample 3: α = .69, M = 3.05, SD = 0.76; Sample 4: α = .66, M = 3.02, SD = 0.82).

As mentioned, two scales were used to assess individual differences in empathy. The first one was the Japanese version of the short form of the EQ (EQ-short; Wakabayashi, Baron-Cohen, Wheelwright, 2006; Wakabayashi, Baron-Cohen, Wheelwright, Goldenfeld et al., 2006). The EQ-short consists of 22 items and employs a 4-point Likert scale ranging from 1 (definitely disagree) to 4 (definitely agree). Sample items were “I can easily tell if someone else wants to enter a conversation” and “I find it hard to know what to do in a social situation” (reverse-scored item). Each of the items were given two points if the participants strongly displayed an empathizing response (i.e., definitely agree, and definitely disagree in reverse-items), and one point if they slightly displayed an empathizing response (i.e., agree, and disagree in reverse-item). Then, the sum of these points was calculated for a total EQ score (Sample 1: α = .85, M = 18.32, SD = 7.33; Sample 2: α = .84, M = 16.35, SD = 7.16; Sample 3: α = .91, M = 13.38, SD = 8.02; Sample 4: α = .87, M = 14.38, SD = 7.69).

The second measure of empathy was the 7-item Empathic Concern subscale of the Japanese version of the Interpersonal Reactivity Index (IRI-EC; Himichi et al., 2017), which had items like “I often have tender, concerned feelings for people less fortunate than me.” A 5-point Likert scale was used for the IRC-EC in Samples 1, 3, and 4; and a 7-point Likert scale for Sample 2. There was no specific reason for this difference, additionally, as the IRC-EC was standardized in our analysis, it is unlikely to have affected the results. The average rating of the seven items was calculated for each participant’s IRI-EC score (Sample 1: α = .77, M = 3.47, SD = 0.62; Sample 2: α = .72, M = 3.86, SD = 0.90; Sample 3: α = .83, M = 3.37, SD = 0.69; Sample 4: α = .85, M = 3.35, SD = 0.81).

Participants completed these measures on paper (Sample 1) or on the web (Samples 2, 3, 4; Google Forms) in the following order: religious belief measure, the EQ, the IRI-EC. They also provided the demographic information (age and gender) at the beginning of the study (Samples 1, 3, 4) or the end (Sample 2).

Results

Multiple linear regression analyses were used to evaluate the strength of the relationships between the two empathy measures (the EQ and IRI-EC) and religious belief. The EQ score, the IRI-EC score, and the religious belief score were standardized and entered into the model. Simultaneously, gender (0 = male, 1 = female) and age (standardized) were also entered as independent variables. Previous studies (e.g., Jack et al., 2016; Norenzayan et al., 2012) demonstrated that being female or being older was associated with having firmer religious beliefs, additionally, gender was associated with empathic response toward others. Hence, the effects of these demographic variables on religious belief should be controlled to evaluate the strength of the unique relationship between the two empathy measures and religious belief. Therefore, data from participants who reported their gender as “unknown” were excluded from the analysis, including the descriptive statistics (cf. 2.1.2. materials and procedure). We also excluded data from participants whose answers were unreliable (i.e., all of the ratings for each item of each measure were the same). As a result, 18, 4, 15, and 10 participants (8.7%, 2.6%, 7.2%, and 5.4% of data) were excluded from Samples 1–4, respectively. Descriptive statistics are summarized in Table 1 and the analysis results in Table 2.

Table 1. Descriptive statistics for each sample.

	EQ			IRI-EC			Religious belief (13-item)			Religious belief (5-item)
	α	M	SD	α	M	SD	α	M	SD	α	M	SD
Study 1
Sample 1 (N = 190)	.85	18.32	7.33	.77	3.47	0.62	.85	3.19	0.68	.57	3.32	0.71
Sample 2 (N = 152)	.84	16.35	7.16	.72	3.86	0.90	.87	2.87	0.86	.39	3.02	0.74
Sample 3 (N = 193)	.91	13.38	8.02	.83	3.37	0.69	.92	2.96	0.85	.69	3.05	0.76
Sample 4 (N = 175)	.89	14.38	7.69	.85	3.35	0.81	.89	2.98	0.85	.66	3.02	0.82
Study 2 (N = 1319)	.89	14.90	7.72	.82	3.41	0.65	.91	3.02	0.86	.69	3.10	0.80

Table 2. Multiple linear regression analysis predicting religious belief with gender, age, the EQ score, and the IRI-EC score in Study 1.

	Gender				Age				EQ				IRI-EC
	β	SE	t	p	β	SE	t	p	β	SE	t	p	β	SE	t	p
Study 1: Sample 1
13-item measure	.39	0.15	2.62	.010	.07	0.07	0.92	.358	.04	0.07	0.47	.642	.06	0.07	0.74	.462
5-item measure	.53	0.15	3.62	<.001	−.03	0.07	−0.48	.635	.01	0.07	0.16	.876	−.01	0.07	−0.09	.932
Study 1: Sample 2
13-item measure	.14	0.20	0.68	.498	−.05	0.08	−0.55	.580	.19	0.09	2.21	.028	−.01	0.08	−0.05	.957
5-item measure	.26	0.20	1.28	.204	−.06	0.08	−0.75	.457	.18	0.09	2.10	.038	.004	0.08	0.05	.960
Study 1: Sample 3
13-item measure	.28	0.16	1.75	.082	.11	0.07	1.52	.131	.04	0.08	0.55	.586	.30	0.08	3.98	<.001
5-item measure	.21	0.16	1.31	.194	.13	0.07	1.84	.067	.06	0.08	0.75	.454	.28	0.08	3.63	<.001
Study 1: Sample 4
13-item measure	.26	0.15	1.70	.091	−.09	0.07	−1.22	.226	.08	0.08	0.93	.355	.17	0.08	1.99	.049
5-item measure	.15	0.15	0.99	.325	−.07	0.07	−1.01	.313	.09	0.08	1.15	.252	.22	0.08	2.64	.009

Sample 1

Bivariable correlations showed that the 13-item religious belief score (i.e., supernatural and paranormal belief) was neither correlated with the EQ score (r = .07, p = .331) nor the IRI-EC score (r = .08, p = .252). The same results were observed when the 5-item religious belief score (i.e., personal belief in gods) was used; no significant correlation with the EQ (r = .04, p = .588) and the IRI-EC score (r = .03, p = .669) was found. The EQ score was positively correlated with the IRI-EC score (r = .24, p = .001).

In a multiple regression model with gender, age, the EQ score, and the IRI-EC score predicting the 13-item religious belief score, gender was significant (β = .39, SE = 0.15, t = 2.62, p = .010), meaning female religious belief scores were higher than male scores. However, the EQ score (β = .04, SE = 0.07, t = 0.47, p = .642) and the IRI-EC score (β = .06, SE = 0.07, t = 0.74, p = .462), as well as age (β = .07, SE = 0.07, t = 0.92, p = .358) were not significant. Similar results were obtained with the 5-item religious belief score as a dependent variable (gender: β = .53, SE = 0.15, t = 3.62, p < .001, age: β = −.03, SE = 0.07, t = −0.48, p = .635, the EQ: β = .01, SE = 0.07, t = 0.16, p = .876, the IRI-EC: β = −.01, SE = 0.07, t = −0.09, p = .932). Thus, both empathy measures were not related to religious belief in Sample 1.

Sample 2

In Sample 2, the 13-item religious belief score was positively correlated with the EQ score (r = .20, p = .014), but not with the IRI-EC score (r = .04, p = .605). Though the internal consistency was low (.39), a similar pattern was shown with the 5-item religious belief score (EQ score r = .20, p = .013; IRI-EC score r = .05, p = .548). The two measures of empathy were positively correlated with each other (r = .25, p = .002).

In the multiple regression model, the EQ score significantly predicted religious belief (β = .19, SE = 0.09, t = 2.21, p = .028), but gender (β = 0.14, SE = 0.20, t = 0.68, p = .498), age (β = −.05, SE = 0.08, t = −0.55, p = .580), and the IRI-EC score (β = −0.01, SE = 0.08, t = −0.05, p = .957) were not significant. In the same model using the 5-item religious belief score as dependent variable, the result was similar; the EQ score was the only significant predictor of religious belief (β = .18, SE = 0.09, t = 2.10, p = .038), and not others (gender: β = 0.26, SE = 0.20, t = 1.28, p = .204, age: β = −.06, SE = 0.08, t = −0.75, p = .457, the IRI-EC: β = .004, SE = 0.08, t = 0.05, p = .960). Thus, the EQ was more strongly related to religious belief than the IRI-EC in Sample 2.

Sample 3

The 13-item religious belief score was positively correlated with both the EQ (r = .18, p = .011) and the IRI-EC score (r = .33, p < .001). The 5-item belief score showed a similar correlation with the EQ (r = .19, p = .009) and the IRI-EC (r = .32, p < .001). The two measures of empathy were positively correlated with each other (r = .42, p < .001).

Multiple regression analysis with the 13-item religious belief measure revealed that the IRI-EC score was the only significant and positive predictor (β = .30, SE = 0.08, t = 3.98, p < .001), while gender (β = .28, SE = 0.16, t = 1.75, p = .082), age (β = .11, SE = 0.07, t = 1.52, p = .131), and the EQ score (β = .04, SE = 0.08, t = 0.55, p = .586) were not significant. The results for the 5-item measure were almost identical; the IRI-EC score significantly predicted religious belief (β = .28, SE = 0.08, t = 3.63, p < .001), but not the other variables (gender: β = .21, SE = 0.16, t = 1.31, p = .194, age: β =  = .13, SE = 0.07, t = 1.84, p = .067, the EQ score: β = .06, SE = 0.08, t = 0.75, p = .454). Unlike in Samples 1 and 2, the IRI-EC, rather than the EQ, was related to religious belief in Sample 3.

Sample 4

Both religious belief scores were positively correlated with the EQ score (13-item: r = .17, p = .031, 5-item: r = .20, p = .008). They were also correlated with the IRI-EC score (13-item: r = .22, p = .003, 5-item: r = .27, p < .001). A positive correlation was observed between the EQ and IRI-EC scores (r = .45, p < .001).

The regression analysis showed that the IRI-EC significantly predicted the 13-item religious belief score (β = .17, SE = 0.08, t = 1.99, p = .049), but gender (β = .26, SE = 0.15, t = 1.70, p = .091), age (β = −.09, SE = 0.07, t = −1.22, p = .226), and the EQ score (β = .08, SE = 0.08, t = 0.93, p = .355) did not. The IRI-EC was also the only significant predictor for the 5-item measure (β = 0.22, SE = 0.08, t = 2.64, p = .009), while the other variables were not (gender: β = .15, SE = 0.15, t = 0.99, p = .325, age: b = −.07, SE = 0.07, t = −1.01, p = .313, the EQ score: β = .09, SE = 0.08, t = 1.15, p = .252). Consistent with the result in Sample 3, religious belief was closely related with the IRI-EC than the EQ.

Mini meta-analysis

Overall, Study 1 demonstrated that empathy predicted religious belief after controlling for the effect of age and gender. However, there was variability in the relationship between the two measures of empathy and religious belief in Samples 1–4. In Sample 1, both the EQ and IRI-EC showed no relationship with religious belief. Only the EQ showed a significant positive relationship with religious belief in Sample 2. Conversely, in Samples 3 and 4, both the EQ and IRI-EC were positively correlated with religious belief; however, only the IRI-EC was a significant predictor of religious belief in the multiple regression model.

Given this variability, we decided to use these four samples to assess the strength of the relationship between the two measures of empathy and religious belief. Thus, a mini meta-analysis was conducted to compute the average effect sizes of the EQ and the IRI-EC scores on religious belief. We first assessed the unique effect sizes of both the empathy measures on the 13-item religious belief measure in each sample. This was done using semi-partial correlation coefficients (r_sp) that “represent the predictive power of an independent variable from a multiple regression model” (Aloe & Becker, 2012, p. 278). We further computed the summary effect using the dmetar package (version 0.0.9000; Harrer et al., 2019). The r_sp of the EQ score was .033, .178, .037, and .068 in Samples 1–4, respectively. Similarly, the r_sp of the IRI-EC score was .053, −.004, .270, and .146.

Figures 1 and 2 show the results of the mini meta-analysis. The random effect model suggested that the summary estimate of r_sp of the EQ was .0740 with 95% confidence interval (CI) of −.0001 to .1472 (z = 1.96, p = .0502), while that of the IRI-EC was .1203 with 95% CI of .0002 to .2370 (z = 1.96, p = . 0496).² A mini-meta-analysis using a five-item measure of religious beliefs yielded similar results; the summary estimate of r_sp of the EQ was .0780 with 95% CI of .0040 to .1512 (z = 2.07, p = .039), while that of the IRI-EC was .1128 with 95% CI of −.018 to .2396 (z = 1.70, p = .090). That is, both empathy measures had weak effect sizes on religious belief. More importantly, the IRI-EC appeared to be more strongly related to religious belief. However, the 95% CIs for the summary effect sizes of the two measures overlapped; thus, we could not conclude that the IRI-EC was more closely related to religious belief than the EQ. Therefore, for further confirmation, we collected another sample of data using these effect sizes to conduct a power analysis.

Figure 1. The results of mini meta-analysis for the effect of the EQ on religious belief (13-item).

Figure 2. The results of mini meta-analysis for the effect of the IRI-EC on religious belief (13-item).

Study 2

Study 2 aimed to confirm the results of Study 1, i.e., the IRI-EC is more strongly linked to religious belief than the EQ. We calculated the optimal sample size to detect the relationship between empathy and religious belief, as estimated in the mini meta-analysis, and examined the relationship between the two measures of empathy and religious belief. Study 2 was preregistered at OSF (https://osf.io/g89te).

Methods

Participants

The power analysis estimated the required sample size as 1431 when r = .0740, which was the lowest effect size obtained in the mini meta-analysis. Hence, we decided to recruit 1431 participants through a web survey service (Yahoo! crowdsourcing and google form). As a result, 1440 adults (M_age = 48.13, SD_age = 10.87, 994 men, 428 women, 18 unknown) participated in this study. We should note that there are small differences in the power analysis between Study 2 and its preregistration, specifically the summary estimate of r_sp. This is because the sizes of Samples 1–4 in Study 1 decreased after the exclusion criteria was employed (i.e., data from participants whose answers were unreliable had not been excluded in the pre registration), and the estimations were thus adjusted.

Materials and procedure

Materials were the same as Study 1. Participants were given Google Forms to complete the questionnaires in the following order: the demographic questions (age and gender), religious belief measure (13-item: α = .91, M = 3.02, SD = 0.86, 5-item: α = .69, M = 3.10, SD = 0.80), the EQ (α = .89, M = 14.90, SD = 7.72), and the IRI-EC (α = .82, M = 3.41, SD = 0.65).

Results and discussion

One hundred and twenty-one participants (8.4%) were excluded using identical exclusion criteria to Study 1. The 13-item religious belief score was positively correlated with the EQ score (r = .20, p < .001) and the IRI-EC score (r = .27, p < .001). The 5-item belief score showed a similar correlation with the EQ scores (r = .15, p < .001) and the IRI-EC (r = .28, p < .001). The two measures of empathy were positively correlated with each other (r = .41, p < .001).

The same multiple regression model as Study 1 was employed to confirm the results. We note that although this analytic strategy was not identical to the preregistered one (i.e., a hierarchical regression),³ both the strategies were same in their purpose (examining if the IRI-EC is more strongly linked to religious belief than the EQ).

The analysis showed that the EQ (β = .10, SE = 0.03, t = 3.35, p = .001) and the IRI-EC scores (β = .22, SE = 0.03, t = 7.55, p < .001) were both significant predictors of the 13-item religious belief score. Importantly, there was no overlap between the 95% CIs of the EQ ([.04, .15]) and the IRI-EC ([.16, .28]). Gender was also significant (β = .29, SE = 0.06, t = 4.89, p < .001), meaning women showed higher religious belief scores than men. However, age was not significant (β = .01, SE = 0.03, t = 0.17, p = .863). The same analysis with the 5-item religious belief score as the dependent variable yielded slightly different results: the IRI-EC (β = .25, SE = 0.03, t = 8.55, p < .001) and gender (β = .20, SE = 0.06, t = 3.45, p = .001) significantly predicted religious belief, but age (β = −.09, SE = 0.07, t = −1.22, p = .226) and the EQ score (b = .04, SE = 0.03, t = 1.37, p = .170) did not. These results were consistent with the mini meta-analysis; the IRI-EC is more strongly linked to religious belief compared to the EQ.

As preregistered, we also used a Bayesian linear regression model to confirm that the results of the final step of each model was not dependent on the analysis method. The posterior distributions of parameters were approximated by the Markov Chain Monte Carlo (MCMC) method using the R 3.6.3 and the ulam function in the rethinking package (McElreath, 2020). The priors for all the linear model parameters (i.e., intercept and slopes) followed the standard normal distribution (μ = 0, σ = 1). The prior for the explanatory variable’s SD followed the exponential distribution with a rate of 1. Based on 10,000 iterations with the initial 500 samples for warmup, the post-random value was generated via a single chain. The model successfully converged with all R-hat below 1.001 and the sizable effective sample sizes (6484–12185). Complete information is available at [https://osf.io/m7hty/].

The means of the posterior distributions for each parameter are shown in Table 3 with their SD and 95% highest posterior density interval (HPDI). The results indicated that gender (β = .27, 95%HPDI = [.17, .41]) and the IRI-EC score (β = .22, [.16, .28]) were good predictors of the 13-item religious belief score in our model. The EQ score also positively, but weakly, predicted religious belief scores (β = .10, [.04, .15]). Finally, age did not predict religious belief scores (β = .005, [−.05, .06]). Again, there was no overlap between the 95% HPDIs of the EQ and the IRI-EC. The same Bayesian analysis with the 5-item religious belief score showed that gender (β = .20, [.09, .32]) and the IRI-EC score (β = .25, [.19, .31]) were both predictors of religious belief score, while age (β = .03, [−.02, 08.]) and the EQ score (β = .04, [−.02, 10.]) did not. These results were almost consistent with that of the classical regression analysis.

Table 3. Results of the linear regression analysis (Classical and Bayesian inference).

	Classical inference				Bayesian inference
	β	SE	t	p	β	95% HPDI
13-item measure
Gender	.29	0.06	4.89	<.001	.286	[.171, .408]
Age	.005	0.03	0.17	.863	.005	[–.047, .058]
EQ	.10	0.03	3.35	.001	.097	[.038, .150]
IRI-EC	.22	0.03	7.55	<.001	.219	[.161, .278]
5-item measure
Gender	.20	0.06	3.45	.001	.203	[.089, .318]
Age	.03	0.03	1.07	.286	.029	[–.023, .083]
EQ	.04	0.03	1.37	.170	.040	[–.019, .096]
IRI-EC	.25	0.03	8.55	<.001	.249	[.192, .307]

Altogether, we showed that the IRI-EC was more strongly related to religious belief than the EQ in Study 2, which confirmed the results of Study 1. It is important to note the difference between the results of analyses considering the differences in the religious belief measures. While the 13-item scale measured broader supernatural beliefs (belief in gods and paranormal belief), the 5-item scale measured only belief in gods. Hence, the results of the analyses in Study 2 suggested that the EQ had no significant unique relationship with the belief in gods when the IRI-EC was statistically controlled in a multiple linear regression model. In fact, in a model with gender, age, and the EQ score predicting the 5-item religious belief score, the EQ was a significant predictor (β = .27, SE = 0.03, t = 9.91, p < .001) along with gender (β = .21, SE = 0.06, t = 3.56, p < .001). This result was consistent with previous studies (Jack et al., 2016; Łowicki et al., 2020) demonstrating a robust and positive association between the IRI-EC and belief in gods, but no close association between measures of mentalizing (including the EQ) and belief in gods. Thus, Study 2 successfully replicated these previous findings among a non-Western sample.

General discussion

The present study examined the relationship between empathy and religious belief. Based on literature in personality psychology and the cognitive and evolutionary science of religion, we hypothesized that the IRI-EC (i.e., empathic concern) is more strongly linked to religious belief than the EQ (i.e., mentalizing-related empathizing ability). A mini meta-analysis using the results of four samples (Study 1) and a survey with another large sample (Study 2) confirmed this hypothesis; the IRI-EC was a better predictor of religious belief compared to the EQ.

The first contribution of the present study is that we demonstrated the robustness and generalizability of the link between empathy and religiosity. Although previous studies have shown this relationship, they did so with mostly Western samples (e.g., Jack et al., 2016; Lindeman et al., 2015; Łowicki & Zajenkowski, 2017; Norenzayan et al., 2012; Routledge et al., 2017; Willard & Norenzayan, 2013). A few studies have been conducted with samples from other cultures (Islamic: Khan et al., 2005; Japanese: Ishii, 2017; Ishii & Watanabe, 2021). The present study demonstrated the same relationship in large samples of Japanese undergraduates and non-student adults (i.e., educated, industrialized, rich, democratic, and non-Western samples). Interestingly, people in Japan naively believe in supernatural agents, although Japan’s religious culture differs from that of Western countries and is also known as a less religious country (e.g., Kavanagh & Jong, 2020; Staufenberg, 2016). Therefore, our findings suggested that the relationship between empathy and religious belief is not conditional on religious culture and faith, thus suggesting that findings in personality psychology regarding this relationship are culturally independent.

The current study also contributed to extending the discussion about psychological foundations of religious belief by showing the close relationship between religious belief and other-oriented aspect of empathy, that is, empathic concern. The early works in the cognitive science of religion have theorized that the key for the emergence of religious belief is to perceive supernatural agents as human-like agents (e.g., Barrett, 2000; Barrett & Keil, 1996). Subsequent studies assumed that our social cognitive ability such as agent detection (i.e., hyper-active agent detective device) or mentalizing allows us to do so and demonstrated that the measures of such ability are linked to religious belief (e.g., Gervais, 2013; Norenzayan & Gervais, 2013, for reviews). Especially, the EQ has been considered as a good measure of such a socio-cognitive ability (e.g., Lindeman et al., 2015; Norenzayan et al., 2012; Willard & Norenzayan, 2013) based on works of Baron-Cohen (e.g., Baron-Cohen & Wheelwright, 2004), who argued that people with autism disorder spectrum tend to lack the ability to attribute mental states to others and subsequently developed the EQ to measure this tendency.

The current study showed that EQ has a small association with religious belief, especially regarding belief in supernatural agents. This suggested that the link between mentalizing and religious belief is weaker than previously assumed. However, this finding should not be interpreted to mean that mentalizing has no link with religious beliefs. It follows that mentalizing enables us to imagine the existence of supernatural agents, as stated in the mentalizing hypothesis. However, imagining such agents may be distinguished from believing in them. For example, people have imagined extraterrestrial life in various ways, but they may not necessarily believe in its existence. Thus, mentalizing may play a role in imagining supernatural agents, while empathic concern contributes to seriously believing those imagined agents’ existence. To test this argument, we must employ different dependent variables that reflect, for example, how quickly and clearly participants can imagine supernatural agents. Regarding empathic concern, it may foster beliefs in those agents in several ways. For instance, people with high empathic concern tend to employ intuitive and spiritual thinking and thus believe in supernatural agents (e.g., Fetterman et al., 2020; Jack et al., 2016). People who are high in empathic concern may also respond unintentionally and emotionally to the imagined mental state of the supernatural agents (e.g., becoming pleased when hearing that God is pleased). Such responses would reinforce their belief in the existence of those agents. Moreover, people with greater empathic concern are more sensitive to others’ undeserved suffering (e.g., unfair treatment) and are more likely to seek remedies. Thus, they may be more likely to accept and believe in supernatural agents, who can resolve such situations. These possibilities should be examined in further research.

Notably, the strength of the relationships between empathy and religious belief in our findings were small. For instance, the correlations between the empathy measures (the EQ and the IRI-EC) and religious belief were all under .30 in Study 2. This is not surprising because previous studies have reported similar results (e.g., Hardy et al., 2012; Ishii, 2017; Jack et al., 2016; Lindeman et al., 2015; Routledge et al., 2017; Willard & Norenzayan, 2013). However, these results suggest that other factors contributed to individual differences in religious belief, such as analytic cognitive style (e.g., Gervais et al., 2018; Pennycook et al., 2016) and cultural learning (e.g., Gervais & Najle, 2015; Lanman & Buhrmester, 2016; Maij et al., 2017). Especially, it may be helpful to examine the interaction between empathy and other factors. For example, empathic concern might facilitate learning about others’ beliefs, because people with high empathy tend to focus on what others feel, think, and believe. This approach may be fruitful for exploring how religious beliefs are acquired.

Some limitations of the current work should be acknowledged. First, our data were correlational. Even though we discussed the possibility that empathic concern may be important for believing in supernatural agents, we could not test this possibility directly. Many previous studies have also employed a correlational method similar to our study or employed quasi-experimental designs (e.g., comparing those who have empathy and those not) because empathy is a personality trait and is difficult to manipulate. In future, we need better methodological designs that overcome this limitation. For example, a longitudinal study that tracks developmental changes in empathy and religious belief.

Second, we employed self-report measures to evaluate individual differences in empathy and religious belief. Even though self-report measures are a useful tool in psychological studies, social desirability may affect the responses. This may especially be the case for the current study as empathic concern and religious belief may involve moral connotations for some people (e.g., to sympathize and be concerned for others’ misery are morally good in general, so is to respect gods). Thus, the correlation between empathy and religious could be a spurious correlation caused by social desirability. Studies using physiological and behavioral responses (e.g., reaction time paradigm, Cohen et al., 2008; the Implicit Association Test of religious belief, Jong et al., 2012) would better elucidate this relationship.

These shortcomings notwithstanding, the current work provides reliable evidence of the close link between empathy and religious belief in non-Western samples. Our work also underscores the possibility that empathic concern plays an essential role in developing religious belief.

Authors’ contributions

Tatsunori Ishii: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, writing - original draft, writing - review & editing, visualization, funding acquisition. Katsumi Watanabe: Conceptualization, resources, writing - review & editing, supervision, project administration.

Ethical approval

The study was approved by the institutional review board (IRB) of Waseda University (2019-357). 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.

Acknowledgments

We would like to thank Editage (http://www.editage.com) for editing and reviewing this manuscript for English language.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The materials, data and analysis scripts used for this article can be accessed at the Open Science Framework repository (https://osf.io/m7hty/).

Additional information

Funding

This work was supported by the Japan Society for the Promotion of Science KAKENHI [17K13905, 19K14370]; Scientific Research on Innovative Areas [17H06344]; JST-Mirai program [20349063]; and JST-CREST [16817876].

Notes

¹ Previous studies demonstrated the relationship between empathy and religious belief, despite various measures of the belief. However, the strength of the relationship was small. Norenzayan and his colleague often used belief in God. For example, Norenzayan et al. (2012) used mainly five items (e.g., “I believe in God,” “When I am in trouble, I find myself wanting to ask God for help”) to measure belief in God. They reported that the effect size of the association between EQ and the belief was small (e.g., odds ratio was 1.64 in their Study 3). Similarly, Willard and Norenzayan (2013) and Willard et al. (2019) used three items (e.g., “I believe in God,” “I believe in a divine being who is involved in my life”) to measure the belief in God. They reported that the zero-order correlations between the EQ and belief scores were small. In Willard and Norenzayan (2013), the correlations were .10 among both 492 Canadian students and 920 American adults. In Willard et al. (2019), it was .17 in 1010 Czech and 1012 Slovak samples. Ishii (2017) and Ishii and Watanabe (2021) reported that EQ was positively correlated with the score of the same five items as Norenzayan et al. (2012) at .20–21. Lindeman et al. (2015), with a large Finnish sample (N = 2789), showed EQ was positively correlated with the Supernatural Belief Scale (r = .21). Routledge et al. (2017) employed three items (e.g., “In general, would you say you are a religious person?”) to measure religious belief. They also showed a positive correlation (r = .30) between EQ and religious belief. Some studies reported a negative or no relationship between EQ and religious belief. Maij et al. (2017; Study 2) examined the relationship using a large Dutch sample (N = 588). They reported that EQ did not predict religiosity (e.g., odds ratio = −.01), which was measured by three items (e.g., “To what extent do you consider yourself religious?”). Vonk and Pitzen (2017) used various measures of religiosity (e.g., Religious Fundamentalism, Intrinsic and Extrinsic Orientations, Loving God Scales) and found no consistent relationship between EQ and religiosity. Specifically, they found positive correlations with some measures, and negative correlations with others.

² It should be discussed whether any potential factors explain the variability of results between the samples. Although the mini meta-analysis revealed the general pattern, there were differences in the pattern of results between the samples. For example, EQ was highly associated with religious belief while the IRI-EC was not in Sample 2. This pattern differed from the results of other samples. Nevertheless, it is difficult to specify any potential factors. In the mini meta-analysis of the relationship between EQ and religious belief, the heterogeneity indexes (τ²: the variance of the true effects, I²: the proportion of observed dispersion that is real, rather than spurious, Borenstein et al., 2009) were 0. This is because the between-sample variance was larger than the within-sample variance. Thus, the effect sizes of the four samples were consistent rather than dispersed. On the other hand, in the analysis of the IRI-EC–religious belief relationship, τ² was not 0 and I² was 62%; thus, the observed between-sample variance may not be random. The relationship may have been low when the participants were students (Samples 1 and 2), while it may have been high when the participants were recruited using crowdsourcing. However, the computation of heterogeneity was affected by the number of samples (or df), and the indexes can be biased (Von Hippel, 2015). Our sample size (four) is too small to conclude whether there were potential factors.

³ In the preregistration format, it was planned to conduct a three-step hierarchical linear regression analysis by entering the demographic variables (age and gender: 0 = male, 1 = female) into the first model, the EQ score in the second model, and the IRI-EC score in the final model. This is because we initially considered the IRI-EC to be identical to the affective component of the EQ. However, as discussed in the Introduction section, we realized that EQ and IRI-EC should be viewed as distinct measures. We thus decided to conduct a different analysis than the one we had pre-registered. The results of the pre-registered analysis were as follows: entering age and gender in the first step for the 13-item religious belief score produced a statistically significant model (R2 = .03, F(2, 1316) = 18.27, p < .004), although the effect size was small (Cohen’s f2 = 0.03). Adding the EQ score in the second step and the IRI-EC in the final step both improved the model, and small effect sizes were produced (the second step: ΔR2 = .03, ΔF(1, 1315) = 47.91, p < .001, f2 = 0.06; the final step: ΔR2 = .04, ΔF(1, 1314) = 57.4, p < .001, f2 = 0.11). In the final step, gender (β = .29, SE = 0.06, t = 4.89, p < .001), EQ (β = .10, SE = 0.03, t = 3.35, p < .001), and the IRI-EC (β = .22, SE = 0.03, t = 7.55, p < .001) were significant predictors. The same analysis was conducted using the 5-item religious belief score. Entering age and gender in the first step produced a statistically significant model with a small effect size (R2 = .02, F(2, 1316) = 11.8, p < .001, f2 = 0.02). Both adding the EQ in the second step and the IRI-EC score improved the model with small effect sizes (the second step: ΔR2 = .02, ΔF(1, 1315) = 27.02, p < .001, f2 = 0.04; the final step: ΔR2 = .05, ΔF(1, 1314) = 73.10, p < .001, f2 = 0.10). In the final step, gender (β = .20, SE = 0.06, t = 3.45, p < .001) and the IRI-EC (β = .25, SE = 0.03, t = 8.55, p < .001) were significant predictors, but the EQ was not (β = .04, SE = 0.03, t = 1.37, p = .169).