Here is one additional thing I found; I don't really know what to make of it, so I invite comment.
It has been shown in multiple papers (here & here & here) that CRT and religiosity are negatively correlated. This finding is treated as evidence that there is a causal link of some sort between religion and the more intuitive, less reflective reasoning style associated with "system 1" in Kahneman's dual process scheme.
In my dataset (a nationally representative panel of 1700 U.S. adults), I find that same negative relationship. But it's moderated by partisan self-identification. The negative impact of religiosity (measured by a scale that combined importance of religion, importance of God, and self-reported church attendance) on CRT gets bigger as respondents' identification with the Democratic Party increases.
Any ideas about what's going on here? I don't really have any.
I'm also not sure what the significance of this relationship is, if any, for the studies that find religion is associated with low-level or system 1 processing. One difficulty for me in that regard is that I'm sort of puzzled by what the psychological theory is behind the religion/low CRT finding generally (I mean, historically, plenty of really highly reflective types have been religious, right Rev. Bayes?); it only seems harder, for me at least!, to articulate a theory if it has to incorporate the religiosity/partisan-identification interaction at the same time.
One other important thing to note is that at least a couple of the studies on religion & cognitive style also included experimental elements, in which manipulations of subjects' reliance on reflection or intuition influenced expressed indicia of religiosity or vice versa. So it's not as if everything about those studies turns on inferences from correlations. But one would still think that interactions between religiosity and other characteristics have to fit with whatever the theory is that connects religiosity to less reflective modes of cognition.
Now I could, of course, go on & try all sorts of additional combinations of demographic variables, including additional cross-product interaction terms. But frankly, I see that sort of approach as pretty mindless. The sorts of demographic variables that predict CRT will tend to co-vary; that goes not double but rather exponential for the various cross-product interactions one can form with them. When all of those get stuck indiscriminately into the regression, it starts to become very unclear what is being modeled (uh, let's see-- how might the simultaneous increase in religion and gender influence CRT holding both race and its interaction with religiosity constant at their "means...")
So if others have suggestions about tests, I'm happy to run them. But before I do, the test requester has to say why the particular combination of variables proposed (including cross-product interaction terms) makes sense. What or who does that combination of variables model given the the sorts of covariances that are being partialed out? Researchers who "over-control" in regressions -- putting everything one can think of onto the right-hand side without any thought of what such a model models -- is something that really gets me steamed!
Well, got quite a number of really interesting off-line responses to this post. A couple of them raise really interesting issues about the right strategy for scaling CRT and the right model for predicting CRT performance based on characteristics like religiosity & partisan self-identification. I'm going to synthesize the responses in some appropriate form & post them -- once the dust settles since the discussants are still going at it.
Also a couple of people offered hypotheses aimed at one possible source of omitted variable bias. I can address there point but will wait to see what emerges from the "modeling" discussion first.
Meanwhile, others are invited to get in on the action. Definitely to feel free to post in comments section--where there's another interesting response--although also to send to me off-line if you are another of these introverted shy, bookish types.