The existence of moral religions is a multifaceted phenomenon requiring explanations across various levels: historical, ecological, economic, social, and psychological. One influential psychological theory suggests that religious beliefs in supernatural punishment enhance prosocial attitudes and large-scale cooperation, leading to the spread of moralizing religions. Alternatively, religiosity may serve as an individual strategy stemming from high social mistrust and a need for individuals to control others' behaviors through moralizing, particularly concerning sexual behaviors as suggested by the Reproductive Religiosity Model. Existing research lacks sufficient data to definitively resolve this debate. This study addresses this limitation by using two large datasets: the European Value Study (EVS) and the World Value Survey (WVS), encompassing over 295,000 individuals from 108 countries. These datasets allow for individual-level analysis, avoiding the ecological fallacy inherent in solely comparing aggregate national data. The study also addresses the issues of arbitrary selection of aggregate factors and the lack of generalizability attempts in previous studies by applying stratified k-fold cross-validation to multivariate structural equation models. This approach assesses the predictive accuracy of models on independent subsets of the data, ensuring the robustness and generalizability of findings across various populations and contexts.

The paper reviews two competing psychological explanations for the existence of moral religions. The first posits that belief in supernatural punishment promotes large-scale cooperation, leading to the evolutionary success of religious groups. This view emphasizes the prosocial benefits of religion. The second explanation suggests that religiosity is an individual-level strategy used to manage social mistrust and control the behavior of others, specifically focusing on the regulation of sexual behavior. The “Reproductive Religiosity Model” highlights the potential role of religion in enforcing monogamous strategies. Existing empirical evidence is mixed, hampered by sample size limitations and issues of generalizability. The researchers highlight shortcomings in previous studies, including reliance on limited samples, the ecological fallacy (drawing inferences about individuals from aggregate group data), and a lack of attempts to assess generalizability. This study aims to overcome these limitations using large-scale datasets and robust statistical methods.

The study utilizes data from the European Value Study (EVS) and the World Value Survey (WVS), containing information on values, beliefs, and behaviors of more than 295,000 individuals across numerous countries. To mitigate the impact of missing data, multiple imputation techniques were employed, generating multiple complete datasets. Structural Equation Modeling (SEM) was used to test two hypotheses: (1) Religiosity leads to increased social trust and large-scale cooperation, and (2) High social mistrust leads to increased religiosity through increased moralizing of others' behaviors. Three SEM models were developed for each dataset: Model 1 tested Hypothesis 1; Model 2.1 tested Hypothesis 2 specifically regarding sexual promiscuity; and Model 2.2 tested Hypothesis 2 concerning free-riding. The measurement models defined latent variables (Religiosity, Social Mistrust, Large-Scale Cooperation, Moralizing Sexual Promiscuity, Moralizing Free-Riding) using multiple indicators. The structural models then tested the hypothesized relationships between these latent variables. Stratified k-fold cross-validation (1000 iterations) was used to assess the models' predictive accuracy and robustness across different subsamples. This method ensures the generalizability of the results and helps prevent overfitting. The study employed the WLSMV estimator in SEM, which is robust to deviations from normality. Goodness-of-fit indices (χ², CFI, RMSEA, SRMR) were used to evaluate model fit. Effect sizes were assessed through standardized coefficients, coefficients of determination (R²), and Wald tests. The predictive accuracy was evaluated by comparing model performance on real test data with performance on randomly permuted data. This approach ensured that observed effects were genuine and not artifacts of data structure.

The study's findings contradict the first hypothesis. Both the EVS and WVS Model 1 results indicate a weak but robust negative association between religiosity and large-scale cooperation. Higher religiosity is associated with lower involvement in collective actions and political engagement, and higher levels of social mistrust. In contrast, the findings largely support the second hypothesis, particularly regarding moralizing sexual behaviors (EVS and WVS Model 2.1). The models reveal that higher social mistrust is strongly linked to increased moralizing of sexual promiscuity, which in turn predicts higher religiosity. This suggests a mediating effect of moralizing sexual behavior. However, this pattern is less evident when examining moralizing attitudes toward free-riding (EVS and WVS Model 2.2). Cross-validation results confirm the robustness of these findings; the models successfully predict out-of-sample data and fail to predict randomly permuted data, demonstrating the validity of the observed relationships. The explained variance in religiosity is noteworthy: approximately 28% in the EVS Model 2.1 and 17% in the WVS Model 2.1, compared to much lower percentages in Model 2.2. The models show an excellent fit (CFI > 0.95, RMSEA < 0.05, SRMR < 0.08).

The study's findings challenge the conventional view that religiosity promotes large-scale cooperation and social trust. Instead, it suggests that religiosity, particularly its moralizing aspects concerning sexual behavior, might be a response to social mistrust rather than a driver of prosociality. Individuals in environments with lower social trust may be more inclined to engage in moralizing behaviors, including religious practices, as a means of controlling others' actions. The limited explained variance in religiosity highlights the influence of non-psychological factors that warrant further investigation. The results align with existing literature showing a correlation between low social trust, weaker institutions, and higher religiosity. The study's focus on moralizing within the context of world religions, rather than small-scale societal religions, is important to note. The lack of a clear association between religiosity and large-scale cooperation does not imply an absence of cooperation among religious individuals; cooperation might operate within smaller social circles or through different mechanisms not captured in the study's measures. The researchers acknowledge the limitations of their operationalization of 'large-scale cooperation,' recognizing the potential for their indicators to reflect other traits or biases.

This study, using a large-scale dataset and robust statistical methods, challenges prevailing assumptions about the relationship between religiosity, prosociality, and social trust. It demonstrates a significant association between social mistrust and religiosity, particularly mediated by moralizing sexual behavior. The findings highlight the complexity of religiosity and its role in social dynamics. Future research should explore the interplay of psychological and non-psychological factors in shaping religiosity, focusing on more nuanced measures of cooperation and examining the specific contexts in which religious beliefs might facilitate or hinder prosocial behaviors. Further exploration of different types of religious beliefs and practices, particularly the differences between world religions and small-scale religious systems, is needed.

The study acknowledges limitations in its operationalization of large-scale cooperation, potentially confounding the measure with other personality traits. The relatively low explained variance in religiosity highlights the role of other, non-psychological factors in determining religiosity. The focus on world religions might limit the generalizability of findings to small-scale societies or other religious systems. Further research with refined measurements of prosociality, incorporating a broader range of factors, is essential for a more comprehensive understanding of the complex relationships explored in this study.