Applied Statistics

The newest way to slam a belief you disagree with–or maybe it’s not so new–is to call it “religious.” For example, “Market Fundamentalism is a quasi-religious faith that unregulated markets will somehow always produce the best possible results,” and so is global warming (“The only difference between the religions right and the religious left, is that the religious right worships a man, and the religious left worships . . . Mother Nature”). As is evidence-based medicine (“as religious as possible . . . just another excuse, really–to sneer at people”). And then there’s the religion of Darwinism.

I encountered an extreme example of this sort of thing recently, from columnist Rod Dreher, who writes disapprovingly of “(Climate) science as religion”–on a religious website called Beliefnet (which has, under the heading, “My Faith,” the options Christianity, Buddhism, Catholic, Hinduism, Mormon, Judiasm, Islam, Holistic, and Angels. Dreher actually appears to be a supporter of climate science here; he’s criticizing a dissent-suppressing attitude that he sees, not the actual work that’s being done by the scientists in the field.

Maybe it’s time to retire use of the term “religion” to mean “uncritical belief in something I disagree with.” Now that actual religious people are using the term in this way, it would seem to have no meaning left.


Perhaps I’m a little sensitive about this because back when I started doing statistics, people often referred to Bayesianism as a religion. At one point, when I was doing work on Bayesian predictive checking, one of my (ultra-classical) colleagues at Berkeley said that he was not a Bayesian. But if he was, he’d go the full subjective route. So he didn’t understand what I was doing.

One of my Berkeley colleagues who studied probability–really, a brilliant guy–commented once that “of course” he was a Bayesian, but he was puzzled by how Bayesian inference worked in an example he’d seen. My feeling was: Bayes is a method, not a religion! Can’t we evaluate it based on how it works?

And, a few years ago, someone from the computer science department came over and gave a lecture in the stat dept at Columbia. His talk was fascinating, bu the irritated me by saying how his method gave all the benefits of Bayesian inference “without having to believe in it.” I don’t believe in logistic regression either, but it sure is useful!


  1. #1 lylebot
    January 1, 2010

    Would you say you have to “believe in” the priors to accept a Bayesian argument? I’ve certainly seen Bayesian arguments that seem plausible until you look at the choices of priors (the worst offenders I’ve seen are “anthropic principle” arguments that use one data point—life on Earth—along with ridiculous priors to make strong inferences about distributions of ideal conditions for life).

  2. #2 bob koepp
    January 1, 2010

    Ernie: I don’t believe in logistic regression either, but it sure is useful!

    Bert: Oh, I get it, Ernie, your god is Utility.

    Ernie: No, no! I don’t worship utility (see? I didn’t even capitalize it!). It’s just useful to value utility.

    Bert: My head hurts.

  3. #3 bill r
    January 2, 2010
  4. #4 Keith O'Rourke
    January 2, 2010

    I should check, but don’t believe there is that much written on the pragmatics of using Bayesian approaches.

    One awkwardness might be that if “the essence of something is taken as what it repeatedly does” then _some_ sort of frequency evaluation necessarily seems to be involved.

    But (because of the former or latter or both) I am not surprised that others often don’t get your pragmatic Bayesianism(and you may just have to write your way out of that)


New comments have been temporarily disabled. Please check back soon.