Cognitive reflection and “belief in” evolution: critically engaging the evidence

Why do 45% or so of Americans consistently say they don’t “believe” humans evolved from an earlier species?

How come about only one-third of them say they accept a conception of evolution—science’s conception—that features  mechanisms of natural selection, random mutation, and genetic variance  (the modern synthesis) as opposed to an alternative religious one that asserts a “supreme being guided the evolution of living things for the purpose of creating humans and other life in the form it exists today?”

These questions get asked constantly. Makes sense: they’re complicated,  and also extremely consequential for the status of science in a liberal democratic society.

One popular answer attributes “disbelief in” evolution to a defecit in critical reasoning that interferes with people’s ability to recognize or accept scientific evidence.  I’ve referred to this in other contexts as the “public irrationality thesis” (PIT) (Kahan in press).

Actually, I think PIT, while a plausible enough conjecture, is itself contrary to weight of the scientific evidence on who believes what and why about human evolution.

It’s well established that there is no meaningful correlation between what a person says he or she “believes” about evolution and having the rudimentary understanding of natural selection, random mutation, and genetic variance necessary to pass a high school biology exam (Bishop & Anderson 1990; Shtulman 2006).

There is a correlation between “belief” in evolution and possession of the kinds of substantive knowledge and reasoning skills essential to science comprehension generally.

But what the correlation is depends on religiosity: a relatively nonreligious person is more likely to say he or she “believes in” evolution, but a relatively religious person less likely to do so, as their science comprehension capacity goes up (Kahan 2015).

That’s what “belief in” evolution of the sort measured in a survey item signifies: who one is, not what one knows.

Americans don’t disagree about evolution because they have different understandings of or commitments to science.  They disagree because they subscribe to competing cultural worldviews that invest positions on evolution with identity-expressive significance.

As with the climate change debate, the contours and depth of the divide on evolution are a testament not to defects in human rationality but to the adroit use of it by individuals to conform their “beliefs” to the ones that signal their allegiance to groups engaged in a (demeaning, illiberal, and unnecessary) form of cultural status compeitition.

Call this the “expressive rationality thesis” (ERT). It’s what I believe—on the basis of my understanding of the best currently available evidence (Kahan 2015).

2. New evidence for PIT?

But if one gets how science works, then one knows that all one’s positions—all of one’s “beliefs”—about empirical issues are provisional.  If I encounter evidence contrary to the view I just stated, I’ll revise my beliefs on that accordingly (I’ve done it before; it doesn’t hurt!).

So I happily sat down last weekend to read Gervais., W, “Override the controversy: Analytic thinking predicts endorsement of religion,” Cognition, 142, 312-321 (2015).

Gervais is a super smart psychologist at the University of Kentucky. He’s done a number of interesting and important studies that I think are really cool, including one  that shows that people engage in biased information processing to gratify their animus against atheists (Garvais, Shariff & Norenzayan 2011), and another that reports a negative association between critical reasoning and religiosity (Gervaise & Norenzayan 2012). 

In this latest study, Gervais correlated the scores of two samples of Univ. of Kentucky undergrads on the Cognitive Reflection Test (CRT) and their beliefs on evolution.

As discussed in 327 previous posts, the CRT  is regarded as the premiere measure of the capacity and disposition to use conscious, effortful, “System 2” information processing as opposed to unconscious, heuristic “System 1” processing, the sort that tends to be at the root of various cognitive miscues, from confirmation bias to the gambler’s fallacy, from base rate neglect to covariance non-detection (Frederick 2005).

Gervais hypothesized that disbelief in evolution is associated with overreliance on “intuitive” or heuristic “System 1” forms of information processing as opposed to conscious or “analytic” “System 2” forms.

“[M]any scientific concepts are difficult for people to grasp intuitively while supernatural concepts may come more easily,” he explains.

From a young age, children view things in the world as existing for a reason; they view objects as serving functions. This promiscuous teleology persists into adulthood, even among those with advanced scientific training. Further, functionally specialized features of animals (such as a zebra’s stripes or a kangaroo’s tail) are viewed as inherently characteristics of an animal’s ‘‘kind,’’ perhaps implying a deeper and more temporally stable essence of the animal. If objects in the world, including living things, are intuitively imbued with function and purpose, it seems a small step to viewing them as intentionally designed by some external agent. . ..

Given that children and adults alike share the intuition that objects in the world, including living things, serve functions and exist for purposes, they may infer intentional agency behind intuited purpose.

Finding a negative correlation between CRT and belief in evolution, he treats the results of his study as supporting the hypothesis that “analytic thinking consistently predicts endorsement of evolution.”

Because the influence of CRT persists after the inclusion of religiosity covariates, Gervais concludes that the “cultural” influence of religiosity, while not irrelevant, is “less robust” an explanation for “disbelief in” evolution than overreliance on heuristic reasoning.

In sum, Gervais is offering up what he regards as strong evidence for PIT.

3. Weighing Gervais’s evidence

So what do I think now?

I think Gervais’s data are really cool and add to the stock of evidence that it makes sense to assess in connection with competing conjectures on the source of variance in belief in evolution.

But in fact, I don’t think the study results furnish any support for PIT! On the contrary, on close examination I think they more strongly support the alternative expressive rationality thesis (ERT).

a. Just look at the data. To begin, the correlation that Gervais reports between CRT scores and disbelief in evolution  actually belies his conclusion.

Sure, the correlation is “statistically significant.” But that just tells us we wouldn’t expect to find an effect as big as or bigger than that if the true correlation were zero.  The question we are interested in is whether the effect is as big as PIT implies it should be.

The answer is no way!

People familiar with logistic regression would probably have an inkling of this when Gervais reports that the “odds ratio” coefficient for CRT is a mere 1.3. An odds ratio of “1” means that there is no effect—and 1.3 isn’t much different from 1.

But researcher shouldn’t presuppose readers have “inklings,” much less leave them with nothing more to go on.  They should graphically display the data in a way that makes their practical effect amenable to reasoned assessment by any reflective person.

The simplest way to do that is to look at the raw data here.

Admirably, Gervais posted his data to his website.  Here’s a scatterplot that helps convey what the “OR = 1.3” finding means as a practical matter:

These scatter plots relate CRT to endorsement of the modern synthesis position as opposed to either “new earth creationism” or a “divine agency” conception of evolution in which a “supreme being guided the evolution of living things for the purpose of creating humans and other life in the form it exists today.”

I think that’s the right comparison if we are trying to assess Gervais’s conjecture that overreliance on System 1 reasoning accounts for the stubbornness of “the intuition that objects in the world, including living things, serve functions and exist for purposes” reflecting “intentional agency.” But the picture is pretty much the same when we look at how CRT relates to endorsement of the proposition that “God created human beings pretty much in the present form at one no part in guided the present time within the last 10,000 years or so.”

Sure, there’s a modest uptick in belief in evolution as CRT increases.

But even those extremely reflective “3’s”–a decided majority of whom attribute the natural history of human beings to divine agency– don’t exactly look like a sample of Richard Dawkinses to me!

Gervais states that these “results suggest that it does not take a great deal of analytic thinking to overcome creationist intuitions.”

But in fact they show that, at least for the overwhelming majority of University of Kentucky undergrads, it would take an amount that far exceeds the maximum value on the CRT scale!

This just isn’t the picture one would expect to see if resistance to science’s account of evolution was a consequence of overreliance on heuristic or System 1 reasoning.

b. Test the alternative hypothesis!  Even more important, the data do look like what you’d expect if the expressive rationality thesis (ERT) explained “belief”/“disbelief” in evolution.

ERT posits that individuals will use their reason to fit their beliefs to the ones that predominate in their cultural group (Kahan 2013).  As explained, existing evidence is consistent with that: it shows that individuals who have a cultural style that features modest Mmm raw data! Always demand a helping of it when served statistically processed datareligiosity become more likely, but those with one that features strong religiosity less likely, to profess belief in evolution.

The way to test for such an effect is not to put religion into a multivariate model as a “control” as Gervais did,  but to examine whether there is an interaction between religiosity and CRT such that the effect of the latter depends on the level of the former.

Here’s what what that interaction looks like in a regression model of “belief in evolution” for a general population sample, in which religiosity is measured with a composite scale reflecting self-reported church attendance, frequency of prayer, and importance of religion in one’s life (α = 0.80):

If we look, we can find the same interaction in Gervais’s data.

This figure graphically displays output of a regression model that uses Study 1’s 7-point “belief in God” scale.

The modest impact of CRT in the sample as a whole is driven entirely by its effect on relatively less religious subjects.

Study 2 has a “belief in God” measure, too, scaled 1-100.  One-hundred point measures are a very bad idea; they aren’t going to measure variance any better than a 10-point (or probably even 7-point) one, but are going to have tons of noise in them.

The study also had a 7-point church attendance measure, so I combined these two into a scale.

Here’s what the raw data look like when we examine how CRT relates to acceptance of the modern synthesis position on evolution in Study 2:

Once more, it’s plain to see that CRT isn’t having any effect on subjects above average in religiosity.  The interaction is there in the regression model, too, but because of the wobbly religiosity measure and smaller sample the model is underpowered (b = –.36, p = 0.07, for “theistic evoluition vs “creationism”; b = –.39, p = 0.19, for “naturalistic” vs. “creationism”). (Actually, if one just uses the 100-point “belief in God” measure, there it is, “statistically significant”–for those who view p < 0.05 as having talismanic significance!)

Contrary to what Gervais concludes from his analyses, then, the evidence doesn’t in fact show a “consistent pattern whereby individuals who are more prone and/or able to engage in analytic thinking” use that capacity to “override” the “intuition objects in the world, including living things, serve functions and exist for purposes” reflecting “intentional agency” in their creation.

We see that “pattern” consistently only in non-religious individuals.

That’s what ERT predicts: as individuals become more cognitively proficient, they become even more successful at forming and persisting in beliefs that express their identity.

I think Gervais missed this because he didn’t structure his analyeses to assess the relative support of his data for the most important rival hypothesis to his own.

In fairness, Gervais does advert to some analyses in his footnotes that might have led him to believe he could ruled out this view. E.g., he didn’t find an interaction between the predictors, he reports, when he regressed belief in evolution on CRT and a “religious upbringing” variable in study 1.  But that’s hardly surprising: that variable was dichotomous and answered affirmatively by 75% of the subjects; it doesn’t have enough variance, and hence enough statistical power, to detect a meaningful interaction.

In study 2, Gervais administered a nonstandard collection of variables he calls “CREDS,” or “credibility enhancing displays.”  Unfortunately, the item wording wasn’t specified in the paper, but Gervaise describes them as measuring variance in “believing” in and “acting” on “supernatural beliefs.”

Gervais reports that the CREDS had only a modest correlation with disbelief in evolution, and also didn’t interact with religiosity when included as predictors of CRT.  I really don’t know what to say about that, except that the discrepancy in the performance of the CRED items, on the one hand, and the Belief in God and church attendance ones, on the other, make me skeptical about what the former is measuring.

I think Gervais should have displayed a bit more skepticism too before he concluded that his data supported PIT.

5.  Limits of Yucky NHT

One last point, this one on methods.

The problem I have with Gervais’s paper is that it relies on an analytical strategy that doesn’t test the weight of the evidence in his data in relation to hypotheses of consequence.  He tells us that he has found a “significant” correlation—but doesn’t show us that the effect observed supports the inference that his hypothesis depends on or rules out a contrary inference supportive of an alternative hypothesis .

These problems are intrinsic to so-called “null hypothesis testing.” Because the “null” is not usually a plausible hypothesis, and because “rejecting” it is often perfectly consistent with multiple competing hypotheses that are plausible, a testing strategy that aims only to “reject the null” will rarely give us any reason to revise our prior assessments of how the world works.

Good studies pit opposing hypotheses against each other in designs where the result, whatever it is, is highly likely to give us more reason than we had before for crediting one over the other.

Gervais is a very good psychologist, whose previous studies definitely reflect this strategy. This one, in my view, wasn’t as well designed—or at least as well analyzed—as his previous ones.

Or maybe I’m missing something, and he or someone else will helpfully tell me what that is!

But no matter what, given the balance of the evidence, I remain as convinced that Gervais is a superb scholar as I am that PIT doesn’t explain conflicts over evolution, climate change, and other culturally contested science issues in the U.S.


Bishop, B.A. & Anderson, C.W. Student conceptions of natural selection and its role in evolution. Journal of Research in Science Teaching 27, 415-427 (1990).

Frederick, S. (2005). Cognitive Reflection and Decision Making. Journal of Economic Perspectives, 19(4), 25-42.

Gervais, W. M. (2015). Override the controversy: Analytic thinking predicts endorsement of evolution. Cognition, 142, 312-321. doi:

Gervais, W. M., & Norenzayan, A. (2012). Analytic Thinking Promotes Religious Disbelief. Science, 336(6080), 493-496. doi: 10.1126/science.1215647

Gervais, W. M., Shariff, A. F., & Norenzayan, A. (2011). Do you believe in atheists? Distrust is central to anti-atheist prejudice. Journal of Personality and Social Psychology, 101(6), 1189.

Kahan, D.M. Climate-Science Communication and the Measurement Problem. Advances in Political Psychology 36, 1-43 (2015).

Kahan, D.M. Ideology, Motivated Reasoning, and Cognitive Reflection. Judgment and Decision Making 8, 407-424 (2013).

Kahan, D.M. What is the science of science communication?” J. Sci. Comm. (in press).

Shtulman, A. Qualitative differences between naïve and scientific theories of evolution. Cognitive Psychology 52, 170-194 (2006).

Leave a Comment