We often see institutional design that deliberately encourages partisan advocates when it comes to information aggregation: lawyers defend and prosecutors try to convict, defense ministries angle for defense spending and education ministries for school spending. This is the case even though ex-ante the information collectors are indifferent about the outcome – until he’s given the case, surely the lawyer doesn’t care what the verdict on some random criminal proceeding will be. Why not just have a judge collect information and render a verdict, or have all government ministers collect information on what’s best for government as a whole?
Dewatripont and Tirole offer a nice little model of where we might optimally see advocates. Consider a policy status quo for which there may be good arguments to expand or contract the policy, or perhaps neither or both. Any agent can look for one of those pieces of news at cost K, or both simultaneously at cost 2K; with probability x, when an argument is looked for, it will be found. From a decisionmaking standpoint, getting arguments for both expansion and contraction is equivalent to receiving no new arguments at all. Regardless of who collects the information, it is ex-post verifiable for free. For now, also assume that information cannot be concealed; if the agent’s research program finds both expansionary and contractionary arguments, she must present both. Assume that benefits from good information are such that the principal always wants the agent to look for both arguments. Also assume that payments must be conditional on the decision, and not on the precise information presented: that is, payments if two contradicting arguments are offered must be the same as if no arguments are offered. Tirole and Dewatripont give some arguments, not terribly convincing, for why this restriction is often seen in practice, but for our purposes, just note that it is empirically true that laywers, for instance, are often paid based on what verdict is given, and not on the inherent quality of their rhetoric.
The problem with nonpartisan information gathering is the following: if an agent is incentivized to exert effort on two causes rather than one, he must be paid more than twice what he is paid to look for one piece of information. This is because if the first search finds an argument (say, to expand), then there is a possibility the second search will contradict that information. Since rewards are based on final decisions only, and the final decision with contradicting arguments is just to retain the status quo, the agent will be paid zero if he presents contradicting arguments. In particular, if the probability of discovering a given argument, x, is greater than .5, then there is no incentive compatible way to induce the agent to search for both pieces of information since the probability of contradicting oneself is too high; if x is less than .5, there are incentive compatible payments, but the agent is able to extract a rent.
What of partisan advocates? If I hire two agents, and pay each only if the final decision supports “their side”, then a payment of K/x(1-x) induces each agent to search and exhausts all their rents. In equilibrium the other agent is searching, so my probability of finding an argument and getting paid for it is x(1-x), so incentive compatibility just requires x(1-x)w-K>=0.
In standard Tirole fashion, a number of simple extensions are offered. If an agent can, for free, sometimes conceal “bad news”, then a single nonpartisan advocate can be optimal if there is a high cost of maintaining the status quo when evidence favors a change in policy. If the decision-maker – a judge, for instance – is potentially biased, advocacy can also be optimal. Consider a world where very costly appeals courts can be used if some party does like a verdict. If appeals are very costly, then they should only be used as an off-the-equilibrium path threat. The IC payment scheme mentioned earlier pays the nonpartisan advocate only if a decision moves away from the status quo, and further pays the advocate identically no matter which way the decision goes. So if judges are biased one way or the other, the nonpartisan advocate will never appeal since she wants the decision to shift policy from the status quo. On the other hand, if there are two advocates, and the decisionmaker picks Expand when the evidence favors Contract, the advocate for Contract will appeal. Since appeals are costly, this bias in judgment will never appear in equilibrium.
(An aside: how does Tirole still not have his Nobel? Some old Fed buddies and I have a Econ Nobel Draft every year, and Tirole has gone in the top 3 each of the last four years. It seems to me the only no-question micro prizes yet to be given are Tirole and Holmstrom – perhaps adding Laffont – and a Milgrom/Roth market design prize. You may like something like Alesina-Tabellini for political theory as well. In any case, Tirole is right at the top of this list. Give him the prize!)
http://www.nyu.edu/econ/user/bisina/advocates.pdf (Final JPE version)