09/04/2010

Hobby Horse Time: The Mistaken "Unintended Consequences" Argument against Interventions Again

[Added: Adam Ozimek replies in the comments.]

Adam Ozimek writes:
We should be extremely cautious about passing [laws designed to reduce negative externalities] if we are uncertain about our ability to identify unintended consequences in advance; by the time we know for sure it could be too late.
One might argue over whether it's still an unintended consequence if it's known in advance, but that's mere semantics; it's clear enough what he means.

I've already explained why I think this argument is wrong:
There is no rule which says that unintended consequences of note need to occur. And although a cursory look at textbook examples or economics blogs might suggest otherwise, unintended consequences, when they do occur, need not be negative. (Here's one example of a positive unintended consequence.)

There currently is no theory predicting how noteworthy the unintended consequences of a law will be or whether their net effect will be positive, negative or neutral.
In other words, it is far from clear that the expected utility of unidentified consequences of a move, such as passing a law, are negative. If that's correct, Ozimek's argument falls flat. Moreover, the intended consequences of a law usually materialize. Given these two arguments, it seems mistaken to argue against passing laws on the basis of Ozimek's argument.

9 comments:

Adam Ozimek said...

If a smoking ban causes drunk driving accidents to go up that is an unintended consequence, whether or not it was anticipated, as long as it is not the intent of the lawmakers to increase drunk driving accidents. So the concept of an unintended known in advance is distinct from one not known in advance. This matters because my point relates to the degree of certainty about unintended consequences.

I don't say that unintended consequences are always bad. Nor do i say that the presence of unintended consequences means you shouldn't pass a law.

But all else equal, we should be more hesitant to pass a law when our ability to identify unintended consequences in advance is poor. To put it another way, the expected utility of passing a law is lower the more uncertainty you have about unexpected consequences.

LemmusLemmus said...

I agree that consequences can be both unintended and anticipated and didn't say otherwise. (When I say "one might", I actually mean "one might".) My remark was acknowledging that the term is most often used for consequences that are both unintended and unanticipated.

"the expected utility of passing a law is lower the more uncertainty you have about unexpected consequences."

I thought I had explained in the post why I think that's wrong, but, rereading it, it seems I may not have been clear enough. My point is that, no, an increase in the uncertainty about the consequences of an action has no bearing on the expected utility (a point estimate) of the action.

If we have reason to believe, that, on average, consequences of actions we didn't anticipate are positive or negative, but I don't see reasons to believe either view.

To take a simple model, if you have a normal distribution and increase its standard deviation (the "uncertainty"), the mean (the expected utility) is still the same.

Adam Ozimek said...

Imagine you have two possible laws that could be passed, laws A and B. Both achieve the intended consequence equally well. Say there also is some measure of unintended consequences that can be negative or positive, and both law A and B have expected values of zero on this measure. However, law A has a huge variance in this measure, meaning that there is a greater chance of really bad or really good outcomes. In contrast, law B has a very small variance, so that you can be almost certain the unintended consequences, whether good or bad, will be close to zero.

The law with more certainty, law B, should have higher expected utility. The only way this isn't true is if people in general are indifferent to the level of uncertainty of unintended consequences in their laws. Would you really disagree with the statement that "we would be better off if we could be certain all our laws achieve only their intended consequences"? In a world where that was true, would be complain that there wasn't enough uncertainty in their legislative outcomes?

LemmusLemmus said...

In the example, both laws yield the same expected utility in intended consequences and the same expected utility in unintended consequences, so the overall expected utility of both laws is by definition the same and we should be indifferent between the two; your statement that law B has higher expected utility is false by definition.

Anonymous said...

I think you are confusing the expected level of unintended consequences with the expected utility. The expected utility depends on both the expected level and the uncertainty of unintended consequences.

Adam said...

Last comment was from me,

LemmusLemmus said...

I may be misunderstanding you, but it seems to me the expected utility, which is a point estimate, does not depend on the uncertainty - the dispersion around the point estimate - at all.

Adam Ozimek said...

I think you are misunderstanding. There are two different expected values: there is the expected level of unintended consequences of a law, and there is the expected utility of a law. The expected utility is a function of the expected level of unintended consequences, but it is also a function of the variance of the level of unintended consequences.

E(c) = expected value of unintended consequences
V(c) = variance of unintended consequences
E(U) = expected utility

E(U) = f(E(c),V(c))

If this weren't true people would be indifferent between otherwise similar laws where on one has unintended consequences that are certain and one with unintended consequences are highly uncertain. This strikes me as obviously untrue.

LemmusLemmus said...

I think we're beginning to clear up the misunderstanding. You model any utility yielded by the uncertainty about the consequences of a law as seperate from the utility yielded by the consequences of the law themselves.

But utility is an inherently subjective concept. I think the problem here may be that the classic paradigm for studying uncertainty aversion has been to pit two choices against each other both of which had the same objective expected payoff (say, $ 2), but a payoff is not the same as utility.

If I understand you correctly, you would, for example, argue the following (Correct me if I'm wrong.): If keeping the status quo in criminal law has an expected homicide rate of X per thousand (which we can be pretty sure about) and a proposed change in criminal law has the same expected homicide rate (but we're pretty usure about that - it might be much higher or much lower), we should prefer to do nothing because the very uncertainty about the outcome of the change yields negative utility.

If this is so, then this would simply be a consequence of the law (which may be unintended or intended, unanticipated or anticipated); hence there'd be no need for a seperate term V(c).

The more interesting question is whether this is, in fact, so. Anecdotally, it seems that people have a strong preference for government to "do something about it" once something is identified as a problem.

I'll also note that an aversion to uncertainty or change was not your argument in your original post at all; rather, it was that uncertainty about laws' outcomes means that laws may have unanticipated negative consequences and we should hence be cautious.