Violence against women, religion, and the dark underbelly of statistics

In a paper which shows that women in Chile raised in a non-religious environment are less likely to be a victim of so called ‘intimate partner’ violence, the authors conclude that there is evidence that a moderate dose of religion actually protects against it! How can that be? And, more importantly, are they right? Let me take you on a short journey into the dark side of statistics…

‘Cross-sectional’ studies are ones that measure a whole bunch of stats about a group of people, and then see which ones correlate. And almost everyone knows the basic trap here – that correlation does not mean causation.

Just because A correlates with B, that doesn’t mean that A causes B. It could be that B causes A, or that C causes both A and B, or any one of a myriad other arrangements!

A basic, and very common, tool used to try to strip away some of these complicating factors is multivariate analysis. The idea behind it is very straightforward.

Imagine you have a situation where you’re looking at the relationship between religion (or the lack of it) and intimate partner violence (IPV). In this case, what the researchers did was take a sample of female students at university in Chile, and ask them if they had ever been a victim of violence from a spouse or boyfriend.

They also asked them how often they went to Church when they were 14, and split them into 3 groups – never attended, low/moderate attendance, and high attendance. Now, as you might expect there are some important differences between these groups.

They were concerned with a few factors, in particular, that they knew from previous research reduced the likelihood that a woman is subjected to IPV. Women who come from a big city, were better educated, or who had mothers who had jobs and earned their own money – all these factors reduce the risk.

So they used multivariate analysis to correct for any differences between the groups on these measures. What this analysis does is to tweak the data so that it is as if the three groups are the same. In other words, what come out of their model was the risk for IPV for women of different levels of family religion if all other factors are held constant.

What they found was that, working with these assumptions, moderate/low attendance significantly reduced the risk of IPV compared with the ‘no religion’ group. High levels of religion did not (although there was a slight trend in that direction).

So, they conclude, a little bit of religion is a good thing – but too much and you cross a boundary, where women’s risk starts to increase (probably because of their diminished social status). All this is good, standard analysis, and perfectly valid.

But, and it’s a big ‘but’ that applies to an awful lot of similar research, it does not mean that encouraging religion is the way to reduce IPV. The bugbear in the analysis has a great name: joint endogeneity.

Here’s the problem: the three big factors they corrected for – urban life, education, and mother’s employment – are actually all directly related to ‘no religion’. They all contribute to (or result from) diverse, multicultural, emancipated worldview, and this actually undercuts religion. What’s more, undercutting religion opens the flood gates to just such a world view.

Is it possible to separate them? I don’t think that it is, in practice. It isn’t a coincidence that these factors are better in the non-religious. It’s a direct, causal effect – they have joint endogeneity. And this has important ramifications for the best way to tackle IPV in places like Chile.

To show you what I mean, you need to know that when they ran their model, using all the different factors that characterise the three difference groups (not just religion), they found that the group with the lowest predicted risk was the no religion group. And the difference was big – the risk for this group was 7 percentage points lower than for the low moderate group. They comment:

The size of this protective effect – 7 percentage points – is the same as that of never having witnessed domestic violence as a child, one of the major predictors of vulnerability to dating violence

If all three groups were the same in every way except religion, this analysis shows that a modest dose of religion is linked to less IPV risk. But in the real world, there is no way these groups are going to be the same.

You cannot, in practice, encourage religion without challenging the broad, open outlook the world that permits female enfranchisement. And this open outlook is the best and most important way to empower women and reduce their risks of being subjected to violence from their intimate partners.

And it just so happens that it also leads to atheism.

_______________________________________________________________________________________

Lehrer, E., Lehrer, V., & Krauss, R. (2009). Religion and intimate partner violence in Chile: Macro- and micro-level influences Social Science Research, 38 (3), 635-643 DOI: 10.1016/j.ssresearch.2009.03.001

This work by Tom Rees is licensed under a Creative Commons Attribution-Share Alike 2.0 UK: England & Wales License.