Friday, October 14, 2005

Random public policy

What is the role of randomness in public policy? Not in terms of the issues that come up, or the solutions, but in terms of using randomness in public policy.

The obvious example is in law enforcement where enforcement is random in order to reduce the costs of enforcement while keeping enforcement effective which includes things like IRS audits. In these cases, you get all kinds of debates about the appropriate underlying distributions that law enforcement should use. For example, driving while black and concerns about the IRS spending too much energy auditing the poor. But these are cases where randomness is used to reduce cost. That is, in an ideal world we would have enough resources to audit all returns, and stop all cars that are exceeding the speed limit. Instead, we economize. We recognize the first best policy as the one where everyone is examined, and the resort to randomness as a second best option. Yet there is one policy where randomness isn't a cost-saving measure, but is instead the point of the policy: a draft with a lottery.

Here we are statistically equal, but not in actuality equal. Even if across all possible worlds people are treated equally by the government, in this world people are not treated equally. Some get drafted and some do not, though all (of appropriate age) have equal odds of being drafted. The pragmatic justification for such a policy is obvious: we need more soldiers than we can get from volunteers but not as many as we get from universal conscription. Since we don't want a surplus of soldiers, a lottery is a way of getting that intermediate amount. So randomness as a first best policy seems fine. The moral justification is trickier. Is a lottery more or less fair than universal conscription?

Because we are all equal before the draft happens in both cases, they seem equally fair. And the procedure by which we are selected is equally fair. But because the outcome is not, strictly speaking, equal, we might think of it as less fair. Yet this has the same form as any situation where the process is fair, even if, like in the market, it leads to discrepant outcomes. We who focus on procedural things want to call this fair.

There are then two conclusions to be had: if randomness is a second best cost-saving measure in enforcement, the debate will be over what is the underlying distribution of breaking the law. And if randomness is a first best measure, then even though the ex post impact of the policy is unfair, because the ex ante probability of impact is fair, we should consider it fair.

Somehow I thought this would provide a way to think about statistical lives, but apparently not.


Blogger henry said...

I sort of think that randomness is the first-best in both situations. If you don't take cost into consideration in the first, why do it in the second? (The cost being enlisting extra soldiers...)

7:56 PM  

Post a Comment

<< Home