April 13, 2006

Norms of Guessing

via Marginal Revolution, Bryan Caplan writes:

One of the most frustrating things about non-economists is their reluctance to guess. Latest example: Today at the repair shop.

Mechanic: The freon's going to leak out unless we replace the compressor.

Me: How fast?

Mechanic: Don't know.

Me: Could you take a guess?

Mechanic: Nope. Could be tomorrow, could be a year.

You could say that the mechanic doesn't want to convey a false sense of certainty. But surely I let him off the hook when I asked him to guess, didn't I? You'd have to have an awfully short fuse to get mad at someone for being wrong after explicitly asking them to guess.

This strikes me as a straightforward case of the mechanic conforming to norms of assertion, where in this case guessing is a species of assertion. Or if you like, guessing is continuous with assertion and the norms of guessing are continuous with the norms of assertion.

For whatever the mechanic may guess, his guess had better be some sort of guide to the truth. If not, it's a bad guess. Caplan is presumably relying on the mechanic's superior expertise to guide him in some sort of decision; if the mechanic's guess leaves Caplan no better informed than he was before, it won't have been helpful. (And it'll have destroyed what we might call super-Socratic ignorance: Caplan's knowledge that he has no idea.)

We can go even farther. Caplan presumably wants to use the mechanic's guess to make a decision about what to do with the appliance. The mechanic may know that, even if he has some degree of support for a belief about how fast the freon will leak out, that degree of support won't be high enough for any action Caplan might take. So there's good reason for the mechanic to refuse to say anything. Whatever slight degree of support Caplan might get from his guess, he's still better off acting as if his decision were being made under conditions of total ignorance.

Note that the dread knowledge account of assertion plays no role here. The mechanic is explicitly asked to make a guess and permitted to disclaim knowledge. Still, the norm of this guess is that it should be true, or at least likely enough to be true enough for the current purposes. Since it is a guess, those purposes don't require it to be very likely to be true; but it's still possible that any guess won't meet those very low standards.

My own guess is that Caplan wants the mechanic to guess because he thinks preferences should conform to the Savage axioms, so probabilities can be assigned to everything. That's why he complains that it's non-economists who won't guess. On this view there's no difference between risk and ignorance, and you can always adjust your probabilities a micron to reflect whatever new information you have. The mechanic also ought to be able to come up with an expected value for when the freon will leak out. So if you assign probabilities to everything, you should be able to communicate to other your guesses as to what the probabilites should be. But you shouldn't assign probabilities to everything, so sometimes you shouldn't even be prepared to venture a guess.

Posted by Matt Weiner at April 13, 2006 10:13 PM

It is obvious that that the freeon will leak out in 183 days, by a corrolary of the Hotel Axiom.

Posted by: Allan at April 14, 2006 11:07 AM

I've been meaning to post on the Hotel Axiom....

Posted by: Matt Weiner at April 14, 2006 12:42 PM

When it comes to predictive assertions, there are two types of people in the world: mechanics and economists. The mechanic won't guess because he knows that if he says six months and it only lasts one week, the customer will be mad, even though the customer knew the mechanic was guessing. That's customer service. If somebody asks you a question and your best guess is wild-ass, you can keep your mouth shut. This is a good strategy for non-economists, but not for economists. For economists, refusal to blindly opine, unlike guessing, does not lead to professional advancement.

Caplan complains non-economists won't guess (hopefully jokingly); the unspoken corollary is that when economists guess wrong, there is not as great a penalty for failure.


P.S. Do you think I can parlay "there are two types of people in the world: mechanics and economists" into a shallow but best-selling book in the style of dichotomists David Brooks and Tom Friedman?

Posted by: Ben at April 14, 2006 08:33 PM

I think Caplan is confusing, maybe for the sake of making a point, blind guessing with informed guessing. Economists love the idea of even a minimally informed guess (which mechanics may still dislike, for the reasons Ben gives). This is what Caplan indicates when he asks,

Doesn't a person who refuses to guess show a lack of confidence in his own skill?

But that depends. Some kinds of guesses reveal skill and experience. This would be something like, I know your freon started leaking last Thursday, and that a system like this holds x much freon, so with a puncture that size, you could expect the remaining freon to last you n weeks, because freon leaks about that fast in my experience.

But without knowing when the leak started, it's just a complete stab in the dark. That's nothing to do with skill at all.

Posted by: slolernr at April 14, 2006 09:20 PM

Or rather, not a complete stab in the dark, i.e., not any random number will do, but you can say, well, the volume of freon in the tank is x and it could be nearly empty or it could be nearly full. Or, you know, "Could be tomorrow, could be a year."

Posted by: slolernr at April 15, 2006 07:58 AM

Another norm of assertion in this conversation:

The speaker isn't asking just anyone how long it will take for the freon to leak, but a trained mechanic. Asking the mechanic in his capacity as a mechanic indicates a belief that his opinion is likely to hold more weight than asking a non-professional.

'Taking a guess' when uncertain carries more weight when it's coming from a mechanic than it would from a professional academic. I think this would shove the standards for that assertion higher.

Posted by: Cala at April 15, 2006 08:50 AM

My inside-joke detector may be broken, but is there such a thing as the Hotel axiom? There is such a thing as Hotelling's law, but that's unrelated.

Looking at that linked page, I'm oddly pleased that Wikipedia has a category "Eponymous laws."

Posted by: washerdreyer at April 16, 2006 05:43 PM

Hotel Axiom refers to this discussion. I doubt that it is in wider use.

Posted by: Matt Weiner at April 16, 2006 07:25 PM