DISQUS

Neel Krishnaswami Hi Robin,

I think I phrased things in a way that misled you. It's of course true that someone with a consistent preference relation can prefer A to B at time t1 and B to A at time t2.

However, one of the possibilities I wished to draw attention to was the preference relation itself changing over time. This creates a significant difference once we start talking about counterfactual conditionals.

As a concrete example, consider something like a Becker-Murphy style rational addiction model. Here, we assume that the addict is engaging in intertemporal optimization of some utility function. One consequence of this model is that it's regret-free: if you ask rational addicts at some t2 > t1 if they would have preferred not to have started taking drugs at t1, they would say no, because they've chosen a dynamically optimal couse of action. On the other hand, we could also model addiction as something that changes a preference relation. This means that time-inconsistent behavior is now possible.

This then leads to differences in what sorts of treatments the model suggests. In a rational-addiction model, there are no treatments needed: an addict is already doing the optimal thing, and any "treatment" will merely reduce their satisfaction. On the other hand, with a time-varying preference relation, the model can advise you to keep drugs from an addict, because it is possible that the addict prefer counterfactually to be clean, while presently preferring to get high -- that is, the addict can regret their decision.

Coming to your question, I'm not saying "we shouldn't assume that people want their preferences to be complete and consistent." I'm saying, "It's impossible, as a matter of fact, for peoples' preferences to be complete and consistent. Therefore, any use of complete preferences must be justified either positively or normatively."

Note that calling something "standard practice" is not itself a justification! Most disciplines do have standard practices, and also have standard justifications for those practices. But we have to look at those justifications to see whether they actually apply to the situation at hand.

Here, I mean "positive" and "normative" in the sense of Friedman's "The Methodology of Positive Economics".

A normative justification is an argument that some other principle argues for using a particular analytical assumption. For example, a libertarian can argue that respect for individual autonomy means that we should analyze situations under the assumption that people are the ideal judges of their own welfare, and model this with complete, consistent preferences. As another example, J.S. Mill argued that envy is a destructive impulse, and therefore a preference that other people suffer should not count in a utilitarian calculus.

If we look at the rational addiction model, we see how it formalizes a libertarian ethos: we can see, mathematically, how the assumption that people are the best judges of their own good leads to no-interference conclusions, and how weakening that assumption can justify paternalistic interventions. But as an argument, it's not going to be terribly convincing to people who aren't libertarians. So if we want to actually rationally persuade people, we need stronger justifications.

A positive justification is some kind of evidence that the assumption of complete, consistent preferences is a safe analytical tool: that is, it is a calculational device whose use does not change the answer. Friedman advances such an argument for assuming profit-maximizing behavior in competitive markets, based on the loosely evolutionary argument that non-maximizing firms will go broke. (As an aside, algorithmic game theory is very useful as a way of making precise when Friedman's analysis holds. Very lovely!) But it's worth noting that his arguments don't scale down to a theory of individual decision -- we need to advance different arguments for that.

Neel Krishnaswami This is actually a normative, not descriptive, stance, and it's one with very strong and surprising consequences.

So, I was specifically talking preferences themselves, and not inferences about them from actions. For reasons of computational complexity, we know that preferences cannot possibly be complete in general. Likewise, peoples' preferences are endogenously-formed and change over time. These are both simply descriptive facts about preferences, and they both mean that we have no reason to expect that people have a stable, complete, preference relation. Which, of course, means we have no reason to expect that people have preferences that can be described with utility functions.

Therefore, if we want to do cost-benefit analysis, we have to have advance a normative theory of how people should want things and how they should make decisions based upon those wants. We need this in order to rule certain observed actions as irrelevant, and to rule others in, and then we need some deductions to fill in the gaps on what's left over, in order to get a preference relation that can be described with a utility function.

But this is by no means a neutral move! For example, if we assume that everyone has a stable preference relation, then this commits us to the position that inalienable rights (ie, rights that we possess but cannot surrender) are a bad idea. This is because by hypothesis such surrenders can only occur as part of mutually-beneficial exchange, and so inalienability means that we will achieve less efficient outcomes. But, on the other hand, if we don't expect preferences to be stable over time, then strategies based on precommitment can be a good idea (and can be a bad idea in other circumstances).

Neel Krishnaswami You can't possibly mean (only) this, because this entails that you can't actually make policy recommendations.

Actual, physical, people do not have complete or consistent preferences. This means costs and benefits are not well-defined, because there's no reason to expect that a utility function exists which describes their actual preferences. Therefore, you need some notion of an ideal or formal completion of actual preferences in order to get a preference relation that we can derive a utility function from, before you can apply cost-benefit analysis.

That completion is where normative assumptions must be made.

Kevin Assuming that Bryan's recounting of your position is correct, this is why I thought you were corrupt:

"...Robin endorses an endless list of bizarre moral claims. For example, he recently told me that "the main problem" with the Holocaust was that there weren't enough Nazis! After all, if there had been six trillion Nazis willing to pay $1 each to make the Holocaust happen, and a mere six million Jews willing to pay $100,000 each to prevent it, the Holocaust would have generated $5.4 trillion worth of consumers surplus."

http://econlog.econlib.org/archives/2009/04/are...

This seems to be a reductio of your position. If you accept it, I can't figure any good reason that you would except if you were ignoring some crucial set of moral intuitions, ones that we require people to weight appropriately to consider them ... not corrupt. That said, 'corrupt' is a word I wish, in retrospect, to not have used. But that was what I had in mind.

As for the point about moral philosophy - I take it when you're talking about harms, you're talking about something normative, which falls within the domain of moral philosophy.