A new study published in the Archive for the Psychology of Religion provides evidence that attending religious services actually causes an increase in charitable actions, like donating money and volunteering time. While the study found this causal effect to be more modest than simple observations might suggest, the researchers demonstrated that even a slight rise in regular religious service attendance across a population could lead to a noticeable and significant increase in overall charitable contributions.
For many years, scientists who study religion have been interested in understanding whether religious beliefs and practices make people more inclined to act kindly and generously towards others. This idea, that religion promotes prosocial behavior, has been discussed for centuries. However, it is surprisingly difficult to definitively prove that religion causes people to be more helpful.
Many past studies have simply shown that religious people tend to be more prosocial, but this doesn’t tell us if their religion is the reason for their generosity. It could be that people who are already kind and giving are simply more drawn to religious communities, or that other factors are responsible for both religious participation and prosocial actions.
Therefore, a team of researchers set out to investigate this question using a large dataset and sophisticated statistical techniques designed to get closer to understanding cause and effect. They aimed to move beyond simply observing a connection between religion and prosociality and to explore whether changing religious service attendance would actually lead to changes in helpful behaviors in the real world.
“Humans are the religious species. I’ve long been interested in what religion does for us, for better and worse,” said study author Joseph Bulbulia, a professor of psychology at the Victoria University of Wellington.
“My training was in the philosophy and history of religions, and my early contributions were to the evolution of religion. About fifteen years ago, I was fortunate to become involved in national scale longitudinal data collection for the New Zealand Attitudes and Values Study, which my friend and collaborator Chris G. Sibley started in 2009. National longitudinal data can provide insights into causality, but only if you analyse the data correctly.”
“I spent nearly a decade acquiring advanced skills in longitudinal statistical methods, however, for a long time I lacked a clear understanding about what these methods delivered. It wasn’t until I encountered the literatures on causal inference (around 2019/2020) that I understood how to leverage longitudinal data to address causal questions.”
The New Zealand Attitudes and Values Study surveys a wide range of New Zealand residents every year, asking them about their social attitudes, personalities, beliefs, health, and behaviors. For this particular study, the researchers focused on information collected from over 33,000 New Zealanders between the years 2018 and 2021. To ensure the sample was as accurate a reflection of the country as possible, the researchers used statistical adjustments based on the 2018 New Zealand Census data for age, gender, and ethnicity.
The study followed participants over time, tracking their religious service attendance and prosocial behaviors across multiple years. Religious service attendance was measured by asking participants if they identified with a religion or spiritual group and, if so, how many times they had attended a church or place of worship in the past month.
For prosocial behavior, the researchers looked at several different measures. First, they examined self-reported charitable activities, asking participants about the number of hours they volunteered in a week and how much money they had donated to charity in the past year. Second, in a novel approach to measuring prosociality, they also looked at whether participants had received help from others. They asked participants if they had received help in the form of time or money from family, friends, and the wider community in the past week.
The researchers reasoned that if religious service attendance truly fosters prosociality within a community, then increased attendance should lead to a more generally helpful environment where people are more likely to offer and receive support. Using this measure of help received aimed to provide a less biased perspective on prosociality, as it focuses on actual experiences of receiving assistance rather than just self-reported generous actions.
Bulbulia and his colleagues then used advanced statistical modeling to explore the causal effects of religious service attendance on these prosocial behaviors. Instead of simply looking at correlations, they designed their models to simulate different scenarios. They considered three hypothetical situations: what would happen if everyone in New Zealand attended religious services regularly (at least four times a month), what would happen if no one attended religious services, and what would happen if current attendance levels remained unchanged.
By comparing the predicted prosocial outcomes in these different scenarios, the researchers could estimate the potential causal impact of increasing or decreasing religious service attendance on a national scale. To ensure their findings were reliable, they also conducted sensitivity analyses, which helped to assess how much their results might be affected by any unmeasured factors that could influence both religious attendance and prosocial behavior. Finally, they compared their causal findings to results from simpler statistical methods that are often used in research, to highlight the differences and demonstrate the importance of using methods designed for causal inference.
The study’s findings indicated that increasing religious service attendance does have a positive effect on charitable giving and volunteering. When comparing a scenario where everyone attended religious services regularly to one where no one attended, the researchers found a statistically significant increase in both charitable donations and volunteering hours. Similarly, comparing regular attendance to the current situation also showed a significant positive effect.
However, when they looked at the opposite scenario – comparing no religious service attendance to the current situation – the results for charitable donations were less clear, suggesting the effect of eliminating religious services might be less pronounced. For volunteering, there was a small but statistically detectable decrease if religious services were eliminated. While the effects were statistically significant, the researchers noted that the size of these effects was modest.
Nevertheless, when considering the entire population of New Zealand, even these modest effects could translate into substantial real-world impacts. For example, they estimated that if regular religious service attendance became widespread, the increase in charitable donations across the country could be equivalent to about 4% of the New Zealand government’s annual budget.
“There is plenty of evidence linking religious participation to pro-sociality, but most of it rests on simple associations, whether cross-sectional or longitudinal,” Bulbulia told PsyPost. “In this study, we use repeated-measures data from 33,198 New Zealanders alongside ‘doubly robust’ machine learning estimators to clarify the causal effects of clearly specified interventions on religious service attendance for charity and volunteering.”
“We found that if every adult in New Zealand attended services regularly, charitable donations could jump by about NZD 2.4 billion, roughly 4% of the government’s budget. On the other hand, removing religious services wouldn’t reliably reduce giving in the short term, likely because current attendance levels are already low. By specifying clear interventions (e.g., ‘What if everyone attended services four times a month?’ vs. ‘What if no one attended?’), we could pin down the expected consequences of each scenario, and by contrasting these scenarios, clarify causality.”
“The central takeaway is this: before you run any analysis, precisely define your question as a contrast between at least two distinct interventions in a well-defined population,” Bulbulia continued. “Of course, from there, you must evaluate assumptions, check your data, choose appropriate estimators, compute the counterfactual contrasts, and carry out sensitivity analyses for robustness to failures in assumptions… & etc. However, to answer a causal question, you must first ask a causal question. If you skip that first step, confusion is practically guaranteed.”
In terms of receiving help from others, the study found that regular religious service attendance was associated with a higher likelihood of receiving help, both in time and money, from friends and the broader community. This suggests that increased religious service attendance might contribute to a more supportive and helpful social environment beyond just formal charitable giving. Interestingly, they did not find a reliable effect on receiving help from family, suggesting that the impact of religious attendance may be more focused on community-level support networks.
When the researchers compared their causal inference results to those from more common, simpler statistical analyses, they found that the simpler methods tended to overestimate the relationship between religious service attendance and prosociality. This highlights a key point: simply observing a correlation between religion and helpfulness does not necessarily mean that religion is the cause, and methods designed for causal inference are essential for gaining a more accurate understanding of the relationship.
“In New Zealand, religious institutions operate as public charities, accounting for about 40% of the charitable sector,” Bulbulia said. “So we were not surprised to find causal evidence for religious charity. However, what did catch us off guard was how far simpler cross-sectional analyses overstated the charitable effects of religious service. (Note that in another context, such naive estimates might have just as easily understated causal effects.) Many social scientists — myself included, before I learned causal inference — underestimate how misleading correlations can be when evaluating the real-world implications of interventions.”
But as with all research, there are caveats to consider. The findings are specific to New Zealand, a country with its own unique cultural and social context, and it is not yet clear whether these results would be the same in other parts of the world.
“Part of stating a clearly defined causal question is stating the population for whom answers are meant to generalize,” Bulbulia noted. “Under the assumptions that we describe in this study, our results generalize to contemporary New Zealanders. We offer no guarantees that results transport elsewhere. To evaluate the transportability of our findings would require additional assumptions and data.”
“Our group is currently investigating the causal effects of interventions on religious beliefs and behaviors, and other features of psychology, over longer time frames. These studies require special methods that account for time-varying confounding. We’re also using methods that allow us to address heterogeneity in effects – for whom are effects strongest, and for whom are they weak, or reversed? Such questions are easy to formulate, yet remarkably challenging to answer.”
“Fortunately, the past five years or so have seen strong methodological advances, particularly in the areas of causal machine learning,” Bulbulia said. “We’re combining these methods with national-scale longitudinal data to obtain more confident answers to causal questions that cannot be addressed using experiments.”
According to Bulbulia, there are a growing number of excellent introductions to causal inference. For those just getting started, he highly recommends Miguel Hernán’s free Harvard course available at edX. Bulbulia also suggests an accessible textbook for social scientists: Morgan S. L. and Winship C.’s (2015) Counterfactuals and Causal Inference (Cambridge University Press). For those interested in mediation and moderation analysis from a causal perspective, he recommends Vanderweele’s (2015) Explanation in Causal Inference: Methods for Mediation and Interaction (Oxford University Press).
In addition, Bulbulia has published a series of open access tutorials that gather key aspects of causal inference into one convenient resource. These tutorials, which address topics such as causal diagrams and confounding (DOI: 10.1017/ehs.2024.35), interaction, mediation, and time-varying treatments (DOI: 10.1017/ehs.2024.32), measurement error and external validity threats (DOI: 10.1017/ehs.2024.33), and confounding in experiments (DOI: 10.1017/ehs.2024.34), provide a comprehensive starting point for anyone looking to deepen their understanding of causal inference.
The study, “The causal effects of religious service attendance on prosocial behaviours in New Zealand: A national longitudinal study,” was authored by Joseph A. Bulbulia, Don E. Davis, Kenneth G. Rice, Chris G. Sibley, and Geoffrey Troughton.
