Inference vs. Regulating Inference

In this post I want to focus on a distinction that has come up in both comment threads. It’s the distinction between inferences, and factors that regulate inferences.

Here’s an example. It’s about wishful thinking.

Start with two people: Wishful S and Less-wishful S. Both have excellent evidence for believing a proposition P. But they differ in how sensitive they are to changes in that evidence.

Wishful S appreciates the evidence she has for P. She can see that it supports P. But any weakening of the evidence, short of overwhelming counterevidence to P, would leave Wishful-S still believing P. When you look at the modal profile of Wishful-S’s belief across evidential situations, you can see that her belief is not sensitive to the evidence across those situations. She simply has the good fortune to be in a situation in which clarity on her part about what the evidence supports does not interfere with her preference to keep believing P.

I hope you can see from the case of Wishful-S the differences between two things that sensitivity to evidence can be. Wishful-S could talk a blue streak in this congenial world about the ways that her evidence supports P. That’s one kind of sensitivity to evidence. But there’s another kind of sensitivity to evidence that shows up in a modal profile. And Wishful-S’s believe is insensitive to evidence in that second way. (And by extension, Wishful-S herself is evidentially insensitive in that way).

From now on, I’ll use “evidential sensitivity” to label just the modal kind of evidential sensitivity. (Plain old ‘sensitivity’ in epistemology, which is sensitivity to truth, is also a modal notion). We can call the other kind “appreciation of evidence”.

Now compare Wishful-S to Less-wishful-S. Less-wishful S appreciates the evidence just as much as Wishful-S does. But Less-wishful S is a bit more evidentially sensitive than Wishful-S is. Some weakenings of the evidence, short of overwhelming counterevidence to P, would leave him (let’s say he’s a boy) still believing P. But not all weakenings would do this. If the initially excellent evidence weakens a little, his will back off in proportion. But if the initially excellent evidence weakens a lot, his desire will kick in, and he will keep believing P in response to the weakened evidence. Faced with overwhelming counterevidence to P, he would give up his belief that P.

Both Wishful-S and Less-wishful-S contrast with Perfect S. Like the wishful subjects, Perfect S appreciates the strong evidence they all have for P. Unlike the wishful subjects, Perfect S’s belief is perfectly sensitive to changes in that evidence.

Now that we have the notion of evidential sensitivity on the table, we can ask: what kinds of factors influence how evidentially insensitive a belief is? You can picture all sorts of factors, including fears, suspicions, agitation, anxiety, hopes. But underneath them any other route to evidential insensitivity in case I’ve described is a preference to hold on the belief that P.

(If I had more space, I’d talk about the difference between preferring P, and preferring to keep believing P. These can come apart. What’s at issue here is the preference to keep believing P. You might not want P to be true at all, yet still be invested in believing it. You’d prefer to keep believing it.)

And this brings us to the distinction between inferring and regulating inference.

Any factors are the influence evidential sensitivity are factors that regulate which inferences you make.

Factors that regulate evidential sensitivity are just one kind of factor regulating inference. Other factors have come up in the comments on previous posts. Nick described the Availability heuristic, which regulates inference by highlighting information that one will be draw inferences from.

Inference can also be regulated by regulating when you transition from inquiry to belief (or to answering a question). Think of inductive contexts. You’re looking all over the house for your passport. At what point do you ‘decide’ that it’s lost? At first you just can’t find it, but you think it’s around somewhere. Then after you still don’t find it, it becomes an open question for you whether it’s there or not. Eventually you convince yourself, or near enough, that it is really not there. (Anecdote: Over the summer I was writing about this example. And then, it came true! I couldn’t find my passport anywhere and I had to order a new one. Moral: Beware of Fate’s perversity, and select your examples with care!)

Why is the distinction between inferences and factors that regulate inference important? It’s important for evaluating inferences epistemically. Of the three beliefs, Wishful’S belief is epistemically the worst, the Perfect-S’s belief is the best, and Less-wishful S’s belief is in between.

If evidential sensitivity makes a difference to the epistemic stauts of a belief, then we already have two dimensions to well-foundedness (or “doxastic justifcation” in the jargon). There’s appreciation, and then there’s evidential sensitivity. Evidentialist theories that focus exclusively on appreciation are forced to say that the epistemic status of Wishful, Less-wishful, and Perfect S’s belief that P are all on a par.

I’m interested in whether any factor that affects evidential sensitivity has this epistemic impact, or only some do. I’m tempted to say that any factor does. If God messed you to make your beliefs evidentially insensitive in the way that Wishful S’s belief is, that would be a way of God giving you a preference to hold on to your belief that P. And that would be an epistemically inappropriate form of wishful thinking.

If God’s messing with you has that effect, then does control of your belief by fear, suspicion, hopes, and so on.

[Image credit: Jessica Tam]


  1. In related news:

    They proposed to identify “commensuration bias,” or the idea that bias can creep in when evaluators combine several ratings in different categories to come up with an overall score. They proposed to see whether black and white investigators receiving comparable scores in the different categories (significance, innovation, investigator, environment) ended up with significantly different overall impact scores.

    “Research in social psychology suggests that, when evaluating job applicants along multiple criteria (like education and experience), evaluators prioritize whichever criterion favors the in-group applicant (white/male) versus the out-group (black/female) applicant, which has the effect of boosting the white/male applicant’s overall score,” Lee explained. “Analogously, we hypothesized that at NIH, white grant applicants receive higher overall impact scores than minority applicants in cases where they have received identical (or sufficiently similar) scores on sub-criteria.”

    Whole story here:

    • Except that the title of that article is a lie! The award is not for detecting bias but for proposing a study to see if it exists. Since one of the participants is apparently a philosopher I might have expected better attention to that small(?) detail.

      • More precisely, the UW proposed study is intended to look for a particular kind of bias which may or may not explain the fact (observed in a different study) that blacks are underfunded compared to whites of the same “educational background, country of origin, training, previous research awards, publication record, and employer characteristics”.

  2. Robert Long

    Hi Susanna et al.,

    Thanks so much for these posts and the great discussion. I had a quick clarificatory question about Less-Wishful. Is there a particular reason that L-W is described as having a bell-curved shaped response pattern: i.e. he backs off appropriately when the evidence is either a little or overwhelmingly weaker, but not when it’s moderately weaker?

    As described:

    A little weaker: Backs off
    A lot weaker: Believes p
    Overwhelmingly weaker: Backs off

    A little weaker: Believes p
    A lot weaker: Believes p
    Overwhelmingly weaker: Backs off

    Would the example work the same way if it was described like this–LW just differs from Wishful in that he just backs off “sooner” than Wishful?

    A little weaker: Believes p
    A lot weaker: Backs off
    Overwhelmingly weaker: Backs off

    A little weaker: Believes p
    A lot weaker: Believes p
    Overwhelmingly weaker: Backs off

    I take it that what matters is simply that Less-Wishful has the appropriate response in more cases than Wishful does. But I wondered if there was some point being made with Less-Wishful’s particular pattern of response (perhaps it’s just more psychologically apt).

  3. Susanna Siegel

    hi rob, thanks. yes, i wanted the example to be realistic. less-wishful’s (LW) curve is what you find in the classic dissonance studies. you’re right that LW* is still evidentially insensitive. along that dimension, LW* is epistemically worse than the perfect subject. my thought was that my LW is more evidentially insensitive than your LW*. so the relative epistemic badness seems to be: Perfect>LW*>LW>Wishful.


    • Thanks for clarifying this. When I first read the OP I was inclined to imagine that the description of LW was really intended to describe LW* and that I was somehow having difficulty reading it correctly (perhaps in the same way that I was with the later sentence “Any factors are the influence evidential sensitivity are factors that regulate which inferences you make.”)

      The patterns of both W and LW* can sometimes arise, I think, purely as a result of mental “friction”, or something like hysteresis in magnetism, without any actual “wish” or preference being involved (in the sense that once the opinion is reversed the new view is also insensitive to evidence back in the original direction). Perhaps the pattern of LW could also be something that just happens, without intent and equally in both directions, but it is harder to imagine an explanatory model that way and so seems more indicative of an actual “wish”.

Comments are closed.

Back to Top