Tuesday, November 30, 2010

Living an Extra Seven Years - Adjustment

Part 1  Part 2  Part 3  Part 4  Part 5
My grandfather came to California from Oklahoma at 18 with a nickel in his pocket. He got a blue collar job working at Ralphs. It was good work that paid well enough, but the hours were long. The managers, wanting to be kind to their employees, gave them a smoke break. My grandfather, a nonsmoker, tried to take a break with his coworkers, only to be told that it was a smoke break, not a do-whatever-you-like break. So he took up smoking.

Imagine a researcher asked, “Does working at Ralphs cause lung cancer?” There are two ways to answer the question. You could say yes: working at Ralphs causes smoking, and smoking causes cancer. Statistically, this is called “un-adjusted risk.” Un-adjusted risks tend to be a less specific and so less convincing answer. The 87% increase in risk of dying from never going to church is unadjusted.

But maybe it wasn’t Ralphs’ smoke-break policies, but some other factor which causes lung cancer (maybe they used dangerous chemicals in the canning process). How could you tell the difference? You’d look at the non-smokers. Were non-smokers getting lung cancer, too? If so, you’d suspect that there was some other factor at the cannery. Using the data from the nonsmokers, you can “subtract” the effect of smoking from the smokers and find out for someone like my grandpa, if there was any “adjusted” risk of working at Ralphs. That is, you can find out if there is any effect outside of the known effect of smoking.

So can we tell if the 87% increase in life expectancy is just from Christians reducing known risk factors? The researchers in the papers subtracted the effects of age, gender, race, socioeconomic status, social support (e.g. marital status, number of friends, relatives), health risks (e.g. smoking, alcohol consumption). If church improved health by working through these things, after we’d subtracted the effects of bad behavior, there would be no difference between the churchgoers and the abstainers. We would conclude that the church is a really good way of changing unhealthy behaviors. But even after these risk factors were taken out, the effect remains. Church does indeed lead to healthier lifestyles (less smoking, drinking, etc.). After subtracting all these from the 87%, someone who never goes to church still has a 50% higher chance of dropping dead compared to a frequent churchgoer (6, 15). Church attendees live longer and we don’t know why.

There is speculation, but no solid data to support them. It might be volunteerism. Maybe it’s a more positive outlook on life. Perhaps there is some unidentified good of being in community (“psychosocial effect”). Or maybe there is actually something supernatural going on in church (7). For all we know, it could be the potlucks. But does it matter how it works? Or more accurately, for whom does it matter? Certainly researchers ought to pursue the Truth wherever it goes. But for the average American, mechanism is much less important. And for the average physician, an understood mechanism rarely comes before a treatment.

We have a very strong case for church attendance. Our case for church attendance today is stronger than our case against smoking was in 1950. Our statistical sophistication and our data collection have improved a lot since the 1950’s. But for the family practice physician, does it matter? To the policy makers, does it matter? We started lecturing patients and slapping warning labels on cigarettes before we knew how they were bad.

[All references are listed under the first post]


  1. Good Stuff David! Thanks for sharing. Looking forward to seeing more of your research.

  2. Thanks for the comment! I'm glad you've enjoyed it!

    Please ask questions, so I can clarify points, or do further research to answer them.

  3. Do you mean that the researchers literally subtracted the odds ratios for those things from the odds ratio for church-going or did they control for them in a linear regression model?

  4. Good question! In an effort to make this accessible to a laypeople, I did not make that clear. They controlled for them in a linear regression model (or models). My account of 'subtraction' here is a great (and probably unjust) oversimplification of what the authors actually did: controlling for the slope contribution of each of many variables (using several models) on the effect in question. Nevertheless, I think that the way to conceptualized this is by some concept of 'subtraction' of the effect.

    Thanks for letting me clarify that!

  5. Do these studies refer only to church-goers in America? If so, do you know of similar studies conducted internationally?

    Also, what about synagogue or mosque attendees?

    Great work, and thanks!

  6. The ones I referred to are all in America. Some asked 'religious service' non-specifically, but in America, that's mostly going to be church anyways. I don't know of any studies that looked at synagogue or mosque specifically. I'd guess this has been reproduced in other countries. It might be a great follow up article!