This topic could be thought a continuation of our last week's topic, the trans fat ban in New York City. We are again dealing with safety regulation. The case for punishing drunk drivers may seem clearer than the case for banning trans fats in restaurant meals because the externality is more pronounced and consumer competence is not in issue, but the appearance is deceptive. Becker's proposal for heavier penalties for drunk driving could be criticized as paternalistic, because it regulates an input rather than an output. If there are 1.4 million annual arrests for drunk driving, and if we assume realistically that this is only a fraction of the actual incidents of drunk driving, yet only 2,000 innocent people are killed by drunk drivers, then it follows that most drunk driving is harmless. Why then punish it with arrests and severe penalties? Why not just punish those drunk drivers who cause deaths or injuries to nonpassengers? In fact we do punish such drivers, under such rubrics as reckless homicide (if the victim dies) or reckless infliction of bodily injury. And the punishments are severe. Why punish the 99+ percent of drunk driving that is harmless? Indeed, if the penalties for reckless homicide are optimal, the implication is that the number of deaths from drunk driving, 17,000 a year, is also optimal.
This is actually a plausible inference. If there are only 2,000 nonpassenger deaths (other than that of the drunk driver himself) caused by drunk driving every year (and how many of the accidents in which a drunk driver is involved are actually caused by the drinking?), then the probability of being killed by a drunk driver is very small, and the value of life estimate that I used in my post on the trans-fat ban should be usable here as well to conduct a cost-benefit analysis of drunk driving. The probability of a drunk driver's killing someone must also be small, given the number of drunk drivers implied by the arrest statistics. Suppose the annual probability that a drunk driver will kill a nonpassenger is .001 (as it would be, given the 2,000 victim figure, if there are 2 million drunk drivers, which is a very modest extrapolation from the arrest figure, since many drunk drivers are not caught). Then the expected injury cost from drunk driving is $7,000 (.001 x $7 million). (This corresponds to Becker's $10,000 figure, which seems to me too high, as it disregards the drunk drivers who are not arrested. Notice that if only a third of drunk drivers are arrested each year, the expected-cost figure drops to $3,333.) This implies that a driver who derives at least $7,000 in utility per year from drinking while driving (more commonly, shortly before driving) is behaving optimally and should not be punished at all.
The larger issue that the drunk-driving question raises is the choice between ex ante and ex post regulation. Health inspections of restaurants and, yes, a ban on trans fats are examples of ex ante regulation. Such regulation prevents dangerous activity rather than waiting for the danger to materialize and using the legal system to punish the injurer. The tort system is an example of ex post regulation. If you drive recklessly but don't injure anybody, you have not committed a tort. Tort law comes into play only when an injury occurs. The theory is that an optimal tort penalty for the injury deters tortious conduct, not perfectly--or there would be no tort cases--but well enough.
Criminal law is a mixed bag. Crimes that result in injury are punished, usually quite severely, but much preparatory conduct--attempts and conspiracies--is punished as well, even when no harm results (the failed attempt, the abandoned conspiracy). Arrests for speeding--and for drunk driving--are examples of ex ante regulation. The economic argument ex ante regulation is that ex post regulation is often inadequate. This is obvious in the trans-fat situation--it would be impossible to figure out which victims of heart disease owed the disease, and to what extent, to which restaurants. In the case of reckless homicide, the answer is less clear. Suppose drunk driving is inefficient--the drunk driver derives less utility than the expected accident cost--and so we want to deter it by punishing the drunk driver who kills or injures a nonpassenger. Suppose the value of life is $7 million and 10 percent of the drivers are not apprehended. Then the optimal penalty would be a fine of $7,780,000 ($7 million √∑ .9). Few drivers could pay that, so the trick would be to impose an equivalent disutility on them by nonpecuniary means, such as imprisonment.
This is not to suggest that punishing drunk drivers who are arrested, the method that Becker endorses, can't achieve the correct deterrence. But punishing just the ones who kill might be more efficient--there wouldn't be as much need for policemen, there would be fewer trials and prison terms, and probably many drunk drivers are quite harmless, for it is unlikely that everyone who drives while drunk has an equal probability of causing an accident. In general, heavy punishment of fewer people is chaper than light punishment of more people.
Thus, only if ex post punishment failed to deter optimally would there be a strong case for punishing drunk drivers who are not involved in accidents with nonpassengers (assuming drunk driving is inefficient, the assumption questioned earlier in this comment)--or at least a strong case based on the simple model of rational choice that underlies my analysis. Maybe drunk drivers systematically underestimate the effect of drinking on the likelihood of an accident, or believe that they can fully compensate for the danger by driving more slowly, or exaggerate the degree to which they can hold their liquor, or exaggerate their driving skill. Maybe their drunk-driving behavior is addictive, and they do not realize, before starting to drink, that they won't be able to avoid drinking before they drive. These might be grounds for ex ante regulation--even for regulations anterior to arrest, such as stiff alcohol taxes.