Field of Science

How does religion prevent suicide?

Religious people are less likely to commit suicide. This is pretty much established as fact; the big question is how the effect is mediated. Is it that the spiritual beliefs of religious people help resolve existential anxieties, gives them hope for the future, or perhaps even fear of punishment in the hereafter? Or is it that religion provides a framework for mutual support that helps people through tough times?

There's a new study from a Canadian team that sheds some light on this question. They used data from a massive survey of over 35,000 Canadians (the Canadian Community Health Survey) to take a look at some of the factors associated with thinking about suicide (suicide ideation) or even attempting suicide. It was a nationally representative sample - around two-thirds were spiritual and slightly more were religious - and around half were both religious and spiritual. Religious people were those who went to church (for reasons other than weddings or funerals) at least once a year.

What they found was that people who were either spiritual or attended religious services (or both) were significantly less likely to have attempted suicide. But the problem is that spiritual and religious people are different to atheists. For example, women are more likely to be spiritual, and also more likely to attempt suicide.

So they took this into account in their statistical analysis. Importantly (and uniquely, as far as I know), they also took account of differences in social supports. Religious people might simply have better access to information, more friends, and more emotional support. So their analysis takes this into account too, to try to provide an estimate of the 'pure' effects of spirituality and religion.

What they found was that, taking demographic factors and social support into account, spirituality no longer had any significant effect. Religion, however, did. A top-line interpretation of this is that spirituality doesn't offer any particular help above and beyond what can (in theory at least) be provided by other means. But religion seems to. That's certainly how the Christian Post sees it:
According to the data, the former category [spirituality] did not show a decreased inclination to take their lives, suggesting something more was involved that was related to the actual attendance at a religious event occurring in a church, mosque, temple or other spiritual gathering.
But I think there's something else going on here. Looking at the data closely, what it looks like to me is that spirituality is linked to fewer suicide attempts, even after accounting for demography and social support - it's just that their model become so statistically weak that it can't show this. And the data also seem to show that religious attendance reduces suicide despite being associated with less social support.

The explanation for this is a little bit complicated, but here goes! You'll need to refer to the graph on the right. The blue diamonds show the point estimate - the best guess from the statistical model. The light blue boxes show the 95% confidence intervals. They're an indication of how uncertain the point estimate is.

In short: if the 95% confidence interval crosses the 'zero' line, you can't be confident that what you're testing (spirituality, for example) has any effect - no matter where the point estimate lies.

You can see what's happening with spirituality. Once you expand the model to include demographics and social support, the 95% confidence interval balloons. Although the best estimate for the effect does go down a little (from a 43% reduction to a 37% reduction in attempted suicides), the main reason they don't find any effect is that the more complex model simply doesn't have the power to detect anything but a massive effect.

It's not that the effect goes away if we add demographics and social support - it's more that adding these factors makes the model so weak that we simply can't tell if there's any effect.

Look at the situation with religion, by comparison. Although the confidence interval increases, it's not by nearly so much. And the point estimate also goes down - from a 47% reduction to a massive 62% reduction in the risk of attempted suicide.

In fact the effect of social support is even larger than you would think from the graph. Compared with the model that controls for everything except social support, controlling for social support as well nearly doubles the effect of religion on the risk of attempting suicide.

This is a big surprise! Intuitively, you would expect that religion increases social support, and that this would explain part of the effect of religion on suicide rate. But if this were so, controlling for social support would reduce the estimated effect of religion, not increase it.

In other words, what this model seems to be showing is that religious people get less social support. And how do they define social support?
... informational support (offering of advice or guidance), tangible support (material aid or behavioral support), positive social interaction (available persons to do things with), affection (involving expressions of love and affection) and emotional support (expression of positive affect, understanding and encouragement).
Now, it's impossible to be sure about this without digging into the original data. I've contacted the authors, and will update this if they get back.

There are some other caveats here. First, this was a study on suicide ideation and suicide attempts, not completed suicides. Complete suicides have a very different epidemiology (they are more frequent among men than women, for example. So the conclusions of this study might not hold for actual suicides.

Second, over 40% of Canadians go to religious services less than once a year. That's quite low by global (although not UK) standards. In poorer countries, non-spiritual people are more likely to attend religious services - so the results might not hold elsewhere.

Bearing all that in mind, here's what I conclude from this study. First, spirituality might help reduce suicide attempts, but if it does the effect is small after accounting for other relevant factors.

Second, religion probably does help stop people going down the path towards suicide, but it does it in spite of seemingly reducing other forms of social support.

D Rasic, S Belik, B Elias, L Katz, M Enns, J Sareen (2008). Spirituality, religion and suicidal behavior in a nationally representative sample Journal of Affective Disorders DOI: 10.1016/j.jad.2008.08.007


  1. Power would only be an issue if this were a smaller sample size, but the sample size is simply huge. You have have to have thousands of variables to eat up that many degrees of freedom.

    The fact that the religiosity CI's do not expand an extraordinary amount is a testament to the strength of the effect, not some weird statistical anomalie of spirituality.

    Now, it is definitely possible for religiosity to be related via a different mediator other than social support etc. that wasn't included in the model. The issue with spirituality, however, can't be an issue of statistical power with a sample size that large. If that is the argument then when there is not an effect as predicted for a small sample you can just argue that "well it would totally be right if we had more power, but we didn't, but don't worry we're right."

  2. What's the definition of spirituality used here?

  3. Mark, The CI is an indication of power. A high powered analysis has a small CI. By 'power' I mean the calculation you make when determining how big a sample you need to determine an effect. Typically goes something like: "For an A% chance of seeing an effect of magnitude B, assuming an underlying variance of C, we need to recruit D number of people.

    This study only had the power to detect an enormous effect of spirituality. There would have to have be a 70% reduction in suicide attempts in the general population for them to have a good chance of detecting it with this sample.

    Power can be an issue even in huge studies if the underlying variance is high or if the event rate is low. In my day job I write up clinical trials. We had one recently that followed 22,000 patients for 4 years. It was underpowered! (Not enough people had strokes, because of better background treatment).

    A similar thing has happened here. Although they looked at 35,000 people, only a few hundred had attempted suicide. Then when they added in the 'Social support' variable, the confidence interval balloons - presumably because there is a lot of variance in this measure. It adds a lot of noise to the analysis.

    The situation is similar with religion, but not as bad. Hard to guess why (statistical anomaly, perhaps - or maybe an interaction with religiosity).

    The increase in CI happens when you add Social Support. So it's the Social Support variable that's weakening the power in the second stage of the analysis.

    Bjorn They just asked people "Are you spiritual - yes/no?" I think this is actually a good measure because I think the standard measures of spirituality used in psychology don't reflect common understanding of the term.

    Few normal people would think that atheists are spiritual - but they can be from a psychologists perspective!

    Another factor to consider is that both religion and spirituality are operationalized as dichotomous variables - rather than scalar ones. That also reduces the sensitivity (i.e. power) of the analysis.

  4. I think the yes/no measure of spirituality was a poor measure (i.e. it did not discriminate well between people, or between various 'types' of spirituality). While I agree that psych measures of spirituality can be weird and convoluted, just using a yes/no item is a poor conceptualization. No one knows what the participants mean.

    Similarly, depending on the data set, they should have used a continuous measure of religiosity. They may have found more movement with the addition of social support had they done this. With the yes/no coding of the variable into people who don't go to services and those who go more than once a year the measure isn't separating people who might actually get social support from service (those who go once a week) from those who likely don't (those who go once a year).

    Back to my original comment, I'm still not sure it is an issue of power. It's an issue of social support mediating the relationship between spirituality and suicide variables. If it was purely an issue of the small # of people who thought about suicide (which would have to be approximately less than 200 for this to be a big issue) and a dichotomous spirituality variable than similar patterns would be found for the religiosity variable (same #'s on the DVs and also dichotomous).

    Also, I apologize for appearing rude or angry. Blog comments are such a difficult medium to discuss anything.

  5. Hi there,

    interesting analysis if this research paper. From a sociologists' perspective (that's me), the relation between religion and suicide rates is a classic one. If you're interested in this problem, please do refer to Durkheim's classical work called 'Suicide'.

    In that work, which was later corroborated using more advanced techniques and data, another possible explanation of this relationship is suggested. Durkheim suggested that people (to some extent) internalize the norms of groups with which one is associated. Religion is practiced in churches, which form such intermediary groups. Also, amongst religious groups the norm is present that one should not commit suicide.

    I realize that in this paper these norms were not taken into account, but it might add to the explanation of why spirituality and religious conviction have a different effect (at least in strength): spirituality does not impose norms, church membership does.

  6. Hi Mark, they were constrained by the original survey - which was a general purpose health survey and so didn't capture any detail on spirituality. The religion analysis is less understandable. They survey asked people to give one of 7 answers (never up to every week) for attendance. They scored everyone who attended once a year as religious. I guess this was done to facilitate the analysis, although they're not clear on that.

    Re: power. It depends essentially on two things - sample size and noise. The noise comes from the variance of the samples, but also from the interactions. What the CI inflation suggests is that adding social support to spirituality is increasing the noise without adding to the explanatory power - perhaps because social support is highly correlated with one or more of the existing variables (i.e. spirituality). But it's hard to tell what's going on from the data presented. But one thing you can conclude is that you can't conclude much from it...

    And no, you didn't come across as rude! Don't worry about that. If my reply came across as terse, it was for the same reason - not the best medium for communication!

  7. Rense, I've read a little about Durkheim's work, but had come away with the idea that he was saying the Protestant religious structures provide less defence against suicide compared with catholic. But reading further just now I see it's down to the specifics of the social structure, rather than the theology.

    Another factor that just struck me is the Survey that provided the data was face to face. It's well known that people over-report their church attendance in such face to face questionnaires (or even over the phone). I wonder whether this may have weakened the results of this study.

  8. Very intersenting analysis but I would suggest that the questions are more specific do you believe in God? Do you consider yourself a member of a religious group? If so do you consider yourself to practicing your religion? Where you raise in a God-loving or God fearing-household or both? Where you baptised as a baby or child?

    Do you consider yourself a spiritual or religious person or none? Do you believe in universal and/or family and/and/or community values or none? Is the respecte for ones life one of them? Do you believe in and afterlife, maybe or not? Etc

    There are people that I know with strong values that were inhereted by their strong religious God loving/fearing parents or caretakers that are atheists


Markup Key:
- <b>bold</b> = bold
- <i>italic</i> = italic
- <a href="">FoS</a> = FoS