Donohue sure did lay the smackdown.

Let me simply focus on Prof. Donohue's main criticism. Prof. Donohue's discussion leads the reader to believe that the paper makes no effort to "control" in some way for the attributes of the judges. He provides the following analogy, "Barondes’ error is a bit like concluding that because the death penalty is almost nonexistent in the Northeast, which has the lowest murder rate in the country, and widespread in the South, which has the highest murder rate, this means that the death penalty causes murder."

This is a common theme expressed various ways on numerous blogs. I believe that criticism is inaccurate. The modeling technique used, in estimating the relationships, does seek to account for this issue. Let me quote from the description provided by Stata: "Conditional logistic analysis differs from regular logistic regression in that the data are grouped and the likelihood is calculated relative to each group; i.e., a conditional likelihood is used." (So, to address this at a rudimentary level, the syntax for the estimation command includes the identity of the judge who issued the pertinent opinion and that is incorporated in estimating the model.)

I believe it is fair to say that Prof. Donohue's analogy is false, though I suppose ultimately it is for the reader to decide.

There are various ways one can seek to model a statistical relationship. It is not clear to me there is a better (valid as well as practicable) way to model this relationship using the data set I have available as it is structured, although it is certainly possible there is. Prof. Donohue has suggested he has a better way. Providing that is what fellow academics do when they comment on working papers. A couple of hours after Prof. Donohue's post, I commented, inviting him to send me the code that he would like for me to use to model the relationship in a better way.

I am not aware of any other investigation like this concerning the impact of clerks on reversal rates of district court judges. So, I examined it. The real creativity for a legal scholar here is trying to formulate a good way to test an issue that is of importance--here, whether clerks matter or whether legal education at the best schools emphasizes the correct things, which has been hotly debated in the last decades. Perhaps others citing this work will benefit from it simply by being informed of how one might use data that has been publicly available for a long time to address these or other issues of interest.

As I indicated previously, if Prof. Donohue provides me with the computer code for the model he thinks is perfect (practicable as well as valid) and the results are that Yale clerks (or clerks from the University of Chicago, or any other school for which he has a valid hypothesis, for that matter) are fantastic and hiring them decreases reversal rates, that would be an interesting thing to report in a paper and I would be happy to report it in a paper.

Royce Barondes