Well, I'm confused, because I thought that the use of the clogit procedure discussed on this board resolved the fixed effect problem identified by Donohue. Maybe it doesn't, but I think he should have addressed this in his criticism.

I have to agree with his second point (I already made it myself), but I didn't think that Barondes rejected that hypothesis.

I want to say that the introduction of "facts" and "statistics" to this debate is really killing my buzz of making fun of Yale law students.

Frankcross is right, and I should have looked at the paper before being snarky.The paper does use conditional logit models, which are basically fixed-effects logits (one cannot directly do a fixed effects logit because of something called the incidental parameters problem). What it comes down to is that the judge fixed-effects Donohue is calling for are already included in the analysis. I believe Donohue's other points remain valid, though.

Even though he may have gone partly wrong here, Donohue's paper with Justin Wolfers in which they demolish the empirical literature on the death penalty - crime rate link is a real treat and highly recommended.

Unless there is a strong reason to only examine one state and looking at it for just one year (2007), it is very hard to take this one observation seriously. Might it hold for other years? Who knows, but was it that hard to get together the data for five years or from 2000 to 2007. There is also no discussion of whether the samples are so small that there is no statistical significance. In any case, there is an old economics problem that Armen Alchian used to give at UCLA. He argued that the farther away a student travels to go to school the better that they would on average be expected to do because of the higher costs that they are bearing to go to school. The same is true for jobs. I am not sure if this relative ranking of Stanford and Yale would hold if they were comparing bar passage rates in say Connecticut.

As to Donohue's claims about the death penalty, I have a discussion here. Commenterlein is wrong about the Donohue and Wolfers paper. It is really strange that they take seriously the notion that the execution rate is defined as death penalty executions divided by the number of people in prison. Can Commenterlein explain this to me? It seems as if we put fewer people in prison for auto theft that will make committing murder riskier. Is that serious? What they do with the graphs at the beginning is also disappointing. They also don't accurately describe the other work that they discuss.