That odd couple, “subjectivity” and “rationality”

Nowadays “Bayesian” is often taken to be a synonym for rationality, and I can see how this can irritate thoughtful philosophers and statisticians alike: To start with, lots of rational thinking—even lots of rational statistical inference—does not occur within the Bayesian formalism. And, to look at it from the other direction, lots of self-proclaimed Bayesian inference hardly seems rational at all. And in what way is “subjective probability” a model for rational scientific inquiry? On the contrary, subjectivity and rationality are in many ways opposites! [emphasis added]

The goal of this paper is to break the link between Bayesian modeling (good, in my opinion) and subjectivity (bad). From this perspective, the irritation of falsificationists regarding exaggerated claims of Bayesian rationality are my ally. . . .

See here for the full article, to appear in the journal Rationality, Markets and Morals.

7 thoughts on “That odd couple, “subjectivity” and “rationality”

  1. Article reads very well.

    Strongly agree that it is key to accelerate the process of identifying features of the model that can be made less wrong, in order of importance, and try hard to make them less wrong. Of course, as you well know, this never really ends.

    But perhaps more focus (in the future) on the elucidation of _power_ of posterior checks to do exactly this and get past the vague (for now) “it has low power for what _may_ not be important to detect and not so bad _power_ for more important detections”?
    K?

  2. Fascinating! Thanks for putting it up! Good read even for a non-statistician in the professional sense!

    I noticed Footnote 3 last line has a type in the web version that I am reading….
    The line says….”second, if the consensus is so clearly wrong,
    how so many could intelligent people hold to it”

    *could* appears misplaced

  3. I see no conflict at all between subjectivity and rationality. The former is about setting the assumptions, while the latter is about deriving the consequences of such assumptions. The empirical question on how to evaluate the consequences of such assumptions is a third factor, and that’s where techniques such as predictive p-values and forms of testing are important.

    • Or as CS Peirce would put it (roughly):
      speculation, deduction, interpretation (i.e. restarting more informed or critical speculation).

      I do think these are often missed up (or at least always worth trying to separate).

  4. Hmmm. Interesting paper, although it goes in the opposite direction of what I’d like to have, namely a frequentism that acknowledges more honestly its subjective aspects… I get from the paper that your approach mixes various approaches in a nonstandard creative (but well thought through!) way based on the specific conditions of the application. To me it seems that this is good data analysis, but it is good *because* it doesn’t bother too much about at least some aspects of “objectivity”. I don’t think, for example, that you could write down general rules of what you do in such a way that it could be reproduced by everyone who just knows the methods. So it seems quite dependent on your personal decisions which may well be based on good reasons. You are even modest enough to state problems with your approach clearly. It’s subjective in the scientifically best sense of the word.

    On a workshop at LSE some years ago Jim Berger gave a presentation about “objective Bayes” and a bright lady, I think a social scientist, commented that “this is all very well, but not particularly objective, given all the personal decisions that you have to make”. To which Berger responded that she may well be right but the researchers for whom he is working are more happy if he calls it “objective”. Are there better reasons than marketing that you want to get away from the “subjective” label?

  5. To my understanding, Bayesianism is about acknowledging subjectivity that isn’t going to go away. It seems to me that just because you use a method that *should* be, or has the *feeling* of being “objective”, that isn’t worth beans if the objectivity is only apparent rather than actual. We can’t have true objectivity, because we don’t have access to the “territory”, only the “map”, but Bayesian-rationality proponents would say that Bayesianism gives you the best possible map.

Comments are closed.