Crime doesn’t pay, but lots of people do it anyway, and keep doing it even after they’ve been repeatedly caught and punished. That seems like a puzzle. One possibility is to think that people who commit crimes aren’t reacting rationally to the opportunity set they face, which may well be true in many cases.
Another possibility is that the circumstances surrounding offending — immediate, psychologically “available,” high-probability rewards and delayed, obscure, relatively low-probability (for any given offense) punishments — are exactly the sorts of circumstances in which lots of people’s behavior deviates systematically from the canons of expected utility maximization. That is, behavioral economics may have a lot to say about why crime keeps happening and how to control it. I’ve been at work on that idea for a couple of years now.
One difficulty that arises is how to value punishment from the punishee’s viewpoint. How bad is a one-year prison sentence? Is a two-year sentence twice as bad, more than twice as bad, or less than twice as bad? (Operationally: If you were the punishee, and had to choose between the certainty of a one-year sentence and an even chance between probation and a two-year sentence, which would it be rational for you to choose?)
Risk aversion suggests that each additional year is worse than the one before it, because it takes a larger share of your remaining stock of non-incarcerated years. But that clashes with the idea, which goes back as far as Beccaria, that certainty is more important than severity in determining the deterrent effect of a punishment strategy.
Enter Daniel Kahnemann. He has studied how people evaluate experiences, both pleasant and unpleasant, retrospectively, compared to the way they evaluate individual moments of those experiences.
You might expect that the overall rating of an experience would approximate, more or less closely, the average rating of the moments composing that experience times the duration: that is, that the value of an experience would be the integral of the values of its component moments. But if you expected that, you would be spectacularly wrong.
In fact, the retrospectively evaluated pleasantness or unpleasantness of a situation is fairly well approximated by the average of its peak (for good or ill) and a moment near its end. [Headline: Scientists discover that you should leave ’em laughing.] The big surprise: Duration is almost entirely irrelevant.
The famous experiments here involve randomly assigning subjects to one of two groups. Group A gets an unpleasant experience. Group B gets the same unpleasant experience plus an additional period that is somewhat less unpleasant. By very large margins, Group B will rate the total experience as less bad than will Group A.
That’s true not only if people are asked to score an experience in the abstract, but if they are asked to make decisions about having another experience like it: Group B will be less reluctant than Group A to have it again. (Note the very interesting ethical problems raised if the procedure involved is a very unpleasant process a patient needs to go through repeatedly to save his life, as for example some forms of chemotherapy for cancer.)
Now back to the problem of punishment. Kahnemann’s work suggests that a five-year sentence may be no more unpleasant, in retrospect, than a four-year sentence, and that four years in a maximum-security prison followed by a year in a minimum-security work-release program may be recalled as less aversive than four years in maximum security followed by immediate release.
Of course, most people haven’t been in prison, and since there’s no reason to think that anticipated utility tracks remembered utility, longer terms may still have greater deterrent value than shorter terms for them. (That’s “general deterrence.”) But it’s possible that severity as measured by duration makes little or no contribution to “specific deterrence,” or to the vehemence with which those who have gone to prison warn others about how bad it is.
This looks to me like a fairly straightforward research project with potentially large policy implications. Anyone interested?