In November 2015, I reported a study of 1170 children from six countries (Canada, China, Jordan, South Africa, Turkey, and the US) by Jean Ducety and his colleagues. The study appeared to show that children from Christian and Muslim households were less altruistic when playing a laboratory game than children from religiously unaffiliated households. It now appears that their conclusion was incorrect.

When correlating religion with altruism, it is necessary to statistically control unwanted variables that might explain both religiosity and altruism. The Ducety team claimed to have controlled for the age, socioeconomic status, and country of origin of their participants. However, a team of researchers headed by Azim Shariff pointed out that, although Ducety and his colleagues intended to statistically control for country of origin, they used a statistically incorrect procedure. When the data were reanalyzed correctly, the association between religion and altruism was no longer statistically significant. This is primarily due to low levels of generosity among children from South Africa and Turkey, two countries with a high level of religious affiliation.

The correct conclusion, then, is that religion has no effect on altruistic behavior. I’m not sure that religious people will be happy with this conclusion, but at least it’s less embarrassing than Ducety’s conclusion. Shariff and his colleagues also point out the following:

• When nationality was controlled correctly, there was no longer an association between religion and the punitiveness of the children.
• The association between religion and parents’ claims that their children are higher in empathy also disappeared when the data were reanalyzed.
• However, there was still a significant association between family devoutness and the altruism of the children, with children from highly religious families being less generous than children from moderately religious homes.

This is an embarrassment for the Ducety group. Had the data been analyzed correctly, the study would probably not have been published.

In 2015, Sharif reported the results of a meta-analysis of 31 studies showing that, while religious people claim to engage in more prosocial behavior on self-report measures, there is no consistent effect of religion on behavioral tasks measuring altruism, such as the one used by Ducety group. He explains this in two ways. First, religious people are more likely to engage in socially desirable responding in which they exaggerate their good behavior. Secondly, laboratory tasks measuring altruism do not contain the contextual cues that sometimes elicit prosocial behavior in the real world, such as being asked by a clergyman to donate money.

In support of this second explanation, Sharif points to a second, separate meta-analysis of 25 studies of religious priming on prosocial behavior. In these studies, participants perform a task intended to remind them of their religious beliefs, such as reading Biblical passages, and are then given an opportunity to behave more or less generously. These studies find that religious primes increase the altruism of religious people, but have no effect on non-religious people.

Sharif explains the effects of religious primes in two ways. First, some religious rituals such as hymn-singing and prayer may create the emotional conditions which encourage people to behave prosocially. Secondly, these primes may remind religious people that they believe they are being observed by supernatural agents who will punish them if they behave badly.

My takeaway from Sharif’s research is that most opportunities for altruistic behavior in the real world probably do not contain religious primes. If I’m right, we should usually not expect religious people to practice the values that are preached to them.

An optional wonkish addendum:

Any time you do a correlational study, you must consider the possibility that your results are explained by some other variable that accidentally coincides with both of the variables of interest. For example, if you find that people who live near nuclear power plants are more likely to die of cancer, you must consider the possibility that poor people are more likely to live near nuclear power plants, and their poverty is the cause of their death rather than their exposure to radiation.

The usual approach to such alternative explanations is to remove their impact on the data through statistical analysis. However, it is not always clear whether an alternative explanation is a source of error which should be removed, or an integral part of the variable of interest.

The Shariff group seems to be saying that if children’s ungenerous behavior can be explained by their country of origin, it need no longer be attributed to their religion. But in a country like Turkey, where 99.8% of its citizens are Muslims, how can you separate its religion from the rest of its culture? In fact, statistically controlling for Turkish nationality precludes the possibility that the Muslim religion of its children will have any affect on the outcome of the study. Was this the right decision? (The situation in South Africa is less extreme, since only 80% of South Africans are Christians.)

An analogy may help. Suppose I do a survey of the gender gap in the salaries of U. S. adults. I statistically control for variables like age, socioeconomic status, education, work experience, etc., and I find that men are paid more than women for the same job. But suppose a critic maintains that tall people are respected more than short people, and therefore paid more. He argues that I am obligated to statistically control for the height of my respondents. Since men are on average taller than women, when I statistically eliminate the effect of height, the association between gender and salary disappears. Does this mean that women are not discriminated against in the workplace, but only short people are?

You might argue that this is a bad analogy because gender is a more plausible explanation for wage discrimination than height. But is nationality a more plausible explanation for lack of altruism than religion? Or did it only seem that way because the negative effect of religion on altruism was unexpected?

You may also be interested in reading:

More Bad News For Religion