Friday, August 31, 2007

The effect of education on leaving religion

Here are more data from the Uecker study that I posted on yesterday. Again the outcome is whether adolescents leave religion during a 7-year period, and here the analyses look at different educational situations.
It's long been assumed that college decreases religious behavior and beliefs, but these data suggest otherwise. Any substantive interpretations as to why religious disaffiliation is highest among those who didn't go to college?


Markus Watson said...

I wonder if college grads are more likely to end up with a relatively traditional family (i.e., husband, wife, 2.7 kids--or whatever that statistic is). Once the kids start coming the parents start feeling they need some sort of spiritual foundation for the kids.

Corey said...

Brad, at the risk of inflicting a technical discussion onto the blog.... Table 2 (pg 12 of "Losing My Religion") reports the odds ratios from multivariate logistic regression models that try to predict decreases in religious service attendance. While these models don't convincingly tells us much, they do show that the education effect seems to decompose when you hold the other factors constant.

Yes, as the level of educational attainment increases, the odds of attendance decline decreases. But for the groups beyond the 2 year degree, the odds ratios are not statistically significant [meaning that the confidence interval around the odds ratios include zero; we can't be confident that there really is an effect there].

All of that is to say that there are probably other unmeasured factors at play. My current pet theory is that religion carries with it cultural capital. But it's late and I can't really articulate that right now.

S.S.Stone said...

I agree with the fact that once parents have children they have that desire to give them the same type of upbringing they had.

Dave P. said...

I would just like to add a plug for my grad school mate Ryan Cragun (now on the faculty at U of Tampa), who did an excellent dissertation examining the factors behind adults leaving religion (Rhys Williams was chair). I'm sure we'll be hearing more about Ryan's results once he settles in and has time to publish.

Good stuff, Brad, thanks.

Brad Wright said...

Markus, I was thinking a similar thing, that graduating from college indicates becoming more conventional and it extends to other areas as well.

Corey, I'm focusing pretty much on disaffiliation, which they report in Table 4. It's interesting, though, to think of why the three different outcomes might have different predictors.

Sarah, that makes a lot of sense about kids; in fact, several studies of deconversion have shown that having kids means less of it.

Dave, thanks for the heads up on Ryan Cragun's work... It sounds interesting, and I'll see if I can find some it.

kent said...

Corey - What??? Huh????

Corey said...

Kent... (and anyone who I confused). Brad reported percentages showing how educational achievement correlates with religious persistence. These percentages are bivariate (meaning they take only two measures into account: education level and religious resistance). When we find a bivariate relationship we have to make sure that this relationship is really not being driven by some third (or fourth, or fifth) factor. Statistically, you want to control for other factors and see if the bivariate relationship holds.

The authors of the study that Brad cites did exactly that. They used a technique called logistic regression that lets us estimate the effects of the bivariate relationships, holding other factors constant.

I pointed to one model where the education effect (higher education achievement leads to a lower likelihood of desistence) seems to disappear when several other factors are considered. [More accurately, the estimate seems to support the original bivariate relationship, but we can't be confident in the estimate because it has a high standard error]. What looks like a clear relationship when we consider only two variables becomes unclear when we consider many variables.

My guess is that the bivariate relationship is capturing differences in cultural capital. But I still can't clearly articulate that and I'm afraid I'm just going to confuse this conversation further.

Note I could be entirely wrong; and I'm not really a stat-head. I'm confident that other readers of the blog with a better grasp on statistics than me could interpret these data differently.