Its one of the many novel ideas from David K. Levine: the non-journal. You write your papers and you put them on your web site. Congratulations, you just published! Ah, but you want peer review. The editors of NAJ just might read your self-published paper and review it. We supply the peer-review, you supply the publication. Peer-review + publication = peer-reviewed publication. That was easy.
(NAJ is an acronym that stands for NAJ Ain’t a Journal.)
Its been around for a few years with pretty much the same set of editors. Its gone through some very active phases and some slow periods. David is trying to breathe some new life into NAJ by rotating in some new editors. So far so good. Arthur Robson is a new editor and he just reviewed a very cool paper by Emir Kamenica and Matthew Gentzkow called “Bayesian Persuasion.”
The paper tells you how a prosecutor manages to convict the innocent. Suppose that a judge will convict a defendant if he is more than 50% likely to be guilty and suppose that only 30% of all defendants brought to trial are actually guilty. A prosecutor can selectively search for evidence but cannot manufacture evidence and must disclose all the evidence he collects. The judge interprets the evidence as a fully rational Bayesian. What is the maximum conviction rate he can achieve?
The answer is 60%. This is accomplished with an investigation strategy that has two possible outcomes. One outcome is a conclusive signal that the defendant is innocent. Since the judge is Bayesian, the innocent signal occurs with probability zero when the defendant is actually guilty. The other outcome is a partially informative signal. If the prosecutor designs his investigation so that this signal occurs with probability 3/7 when the defendant is innocent (and with probability 1 when guilty) then
- conditional on this signal, the defendant is 50% likely to be guilty (we can make it strictly higher than 50% if you like by changing the numbers slightly)
- 3/7 of the innocent and all of the guilty will get this signal. (3/7 times 70%) + 30% = 60%.
The paper studies the optimal investigation scheme in a general model and uses it in a few applications.