pageok
pageok
pageok
The Yale Clerk Effect:

I just read the paper that Stuart just noted; I'm not econometrician enough to tell whether the analysis is sound, but I'd love to hear from those who are.

Note, though, that even if the statistical analysis is correct (which it might be), this still doesn't tell us why this effect is present (as the paper makes clear). One possible reason, which the paper points to, is the conventional wisdom that Yale students learn a lot of theory but not a lot of doctrine.

But there are other possible reasons. For instance, perhaps Yale has such an excellent reputation that Yalies who are relatively far down in the class get more clerkships than comparably ranked students at comparable quality schools. (I have indeed heard that this is the case, though I have no hard data on the subject except for this appellate clerk data.)

Say, for instance, that the Yale district court clerks generally come from the 25th to 75th percentile at Yale, and on average represent the 50th percentile. But say the Harvard district court clerks on average represent the 55th to 85th percentile at Harvard, and on average represent the 70th percentile. And say the Yale and Harvard student bodies are on average comparably good (or perhaps even that Yale is on average better, but only by a little), so that the people at the 50th percentile at Yale are a little less good at legal analysis than the people at the 70th percentile at Harvard.

The higher reversal rate associated with Yale clerks may then just reflect the lower average quality of Yale clerks, not the lower average quality of Yale graduates or of a Yale education. And this lower quality would flow from the school's reputation exceeding its actual merits (even if its merits are very great).

Likewise, the paper itself points to some other possible explanations, for instance "a grading system that is not sufficiently partitioned to allow judges to identify the quality of applicants": "Officially, there's a system of honors, pass, low pass and fail, but three-quarters or more of the class gets a pass, and professors rarely give out low passes ...." That too would be a way in which the average quality of Yale clerks might be lower than the average quality of other clerks (if it is), even though the average quality of the entire Yale class might be as good or better as that of other schools' classes.

frankcross (mail):
There's another possible explanation I would like to see checked. Appellate court reversals, I have found, are disproportionately conservative (even in the case of many more liberal circuit court judges). Perhaps Yale clerks are more likely to push the envelope in a liberal direction, or perhaps judges who hire Yale clerks are more liberal.
4.18.2008 6:25pm
Ashley Higgins (mail):
One hopes that the likelihood of reversal is correlated to judicial, rather than clerk, error. Is it not likely that judges who hire Yale clerks have erroneous judgment with respect to the cases assigned to them and, possibly, the clerks they hire?
4.18.2008 6:26pm
genob:
My colleague down the hall, who is a Yale law grad, offers the follwoing explanation:

"Clerks bitter about the fact that they are not clerking on the Court of Appeals -- and thus refusing to follow it."
4.18.2008 6:41pm
Lonetown (mail):
I would think the layman would conclude Yale clerks take more risk. They are more willing to push the envelope.
4.18.2008 10:25pm
Alex650 (mail):
I wonder about the measures used--a caution signal shows up for other reasons than simple reversal. Could it be that more "academic" or high-profile DJs might actually have a higher caution rate because of other courts feeling the need to distinguish their opinions? In other words, maybe higher-prestige judges are disagreed with, while lower-prestige judges are simply ignored? And no, I didn't go to Yale.
4.19.2008 9:58pm
Ben P (mail):

Say, for instance, that the Yale district court clerks generally come from the 25th to 75th percentile at Yale, and on average represent the 50th percentile. But say the Harvard district court clerks on average represent the 55th to 85th percentile at Harvard, and on average represent the 70th percentile. And say the Yale and Harvard student bodies are on average comparably good (or perhaps even that Yale is on average better, but only by a little), so that the people at the 50th percentile at Yale are a little less good at legal analysis than the people at the 70th percentile at Harvard.


I know you probably made these numbers up, but as someone between the 5th and 10th percentile at a state law school who's considering a clerkship, that made me cry a little inside. Because I know people with similar class ranks to mine who didn't get federal clerkships of any kind.
4.20.2008 12:52pm
TerrencePhilip:
It is quite unlikely that a district judge is getting reversed because of basic errors traceable to poor legal analysis by a law clerk. Believe it or not, the judge might have even learned a little law on his/her own, before the 25-year-old law clerk came aboard.

If it is in the briefs and arguments of the parties, even a dense clerk will at least read the argument. Even minimal diligence by the district judge means he/she would also read the briefs and not blindly sign off on a clerk's draft without knowing what was presented to him. If the law clerk were really doing a poor job, even a judge of average ability would pick up on this very quickly.

In the federal system it is EXTREMELY difficult to get reversed on the basis of a factual finding. You get reversals on questions of law, and that is almost surely what these reversals are about. A Yale degree is so prestigious that the clerks could have made better money elsewhere, and while lower-ranked Yale students may not have the same pick of clerkships the top grads do, they will trend towards clerking with the kind of liberal mavericks who have a bit of a prestige name. I am sure that nearly all of these reversals involve district judges who know what they believe and why they believe it, and take the risk of saying so consciously.

Some analysis of the reversals would have been helpful. Also, if possible a comparison of the judges' reversal rates in years when they did not have Yale clerks would have been nice.
4.20.2008 4:29pm
with all due respect ...:
a comparison of the judges' reversal rates in years when they did not have Yale clerks would have been nice.

I only disagree to the extent that the above language understates the importance of such data. The most obvious objection to the study is that it does not prove anything about the clerks, as opposed to other potential sources of causation. Are these particular judges already prone to a higher reversal rate? Unfortunately, I don't expect anyone (other than a Yale grad annoyed by this study) to look at that data.
4.20.2008 9:17pm
jig:
There's a bit of a black box in their hand wavy "clogit estimater in Stata 10" method. See:

http://www.stata.com/help.cgi?clogit

I'm thinking there may be a correlation between desirability of places to clerk and "hard" cases to decide (cases more likely to be reversed). Many of the topics the author lists have strong or subtle state law dependencies, and in places where the legislature is a constantly moving target, geography becomes an issue. The author attempted to "control" for this through "judge identity clustering" (hey, the author made up the buzz words, not me) and the topic biasing, but that probably wouldn't work on the time scale used in the survey, which is one reason why the author included results with the criminal cases removed. If Yale grads tend to get the jobs everyone wants, then their decisions might be outliers to a large part because of the particular topics that come to them.

Desirable places to clerk would be a pretty easy ranking to survey for, from the hip. For any particular place, it would be some mixture of total number of applicants and the ranking of their respective schools.

It's such low hanging fruit for a criticism of the method (it would have been easy for the author to do the additional work, if not reliably estimate the correction), either I'm missing where the author built it into the model, or the paper has a big hole.

-jig
4.21.2008 5:04am
Royce Barondes:
jig:

"clogit fits maximum likelihood models with a dichotomous dependent variable coded as 0/1.... 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. ...

Economists and other social scientists fitting fixed-effects logit models have data that look exactly like the data biostatisticians and epidemiologists call k1i:k2i matched case-control data." Stata Base Reference Manual 275-76 (release 10).

The grouping (clustering) for the purposes of the paper is the identity of the judge who wrote the opinion. So, I would not describe the phrase you referenced as a "buzz-word." I could have written "grouped by judge identity" or "clustered by judge identity". I'm not sure that one is materially better than the other, or that there is some other, but better, way to say it.

I have in the past seen an econometrician whom I respect criticize papers using empirical techniques for failing to identify precisely how the estimation was done. So, in writing this paper, I sought to communicate unambiguously how I estimated the results.

I have, in assorted other responses on this blog (there are three main posts), sought to describe this estimation method. I gather you would not have any problem with a model that put in a dummy variable for each of the judges--that would "control" for anything, to put it colloquially, that makes the particular judge more or less likely to be reversed. At the moment I am inclined to say that it is better to present in the paper the models that are in the paper. But, as I noted in response to another post, I re-estimated the results reported in model 1 of Table 4 using 90-some dummy variables (one for each judge, other than one). The results for Yale were similar.

In writing a paper, an author has to judge the extent to which the paper should recite background. It seemed to me that this paper was already sufficiently long describing as concisely as I could what was done, and that additional explanation of modeling techniques would not be helpful.

I thank you for your comment, and will consider it in connection with seeking to clarify the paper.
4.21.2008 4:33pm