« Less Public Sources of Information About Judges for Empirical Research | Main | Neglected Variables in Empirical Research into Judicial Decisionmaking »

13 December 2006


test person



My is a help rather,comments.Please,Sir,do help me do an 'EMPERICAL REVIEW'on Determinants of security prices on the Nigerian Stock Exchange.

Sir,ur help will reat assistant.


Josh Fischman

I think that the discussion within the ELS community on measurement of ideology is focusing on the wrong question. Party-of-appointing-president and Giles-Hettinger-Peppers (GHP) scores are both useful proxies in a variety of situations, but there are other ways of studying ideology that are often preferable using either of these variables.

Proxy variables are not causal variables; a judge's GHP score does not "cause" her to vote in a particular way. Proxy variables capture ideology (the true causal variable) with measurement error, which means that the regressions will yield biased estimators. In a linear regression, the ideology estimate will be biased toward zero, but in nonlinear regressions (as is usually the case for GHP scores), the bias can be in any direction. My concern is that the measurement error can in many instances be very large. This means that the estimates from a regression based on these proxies can be biased in any direction, and so can the standard errors. I have no idea how severe this problem is in typical applications, and I'm not aware of any papers that have given this question serious thought.

There is another way to measure ideology that is used quite regularly, but doesn't get as much attention as it should: just use dummy variables for each particular judge. This is done all the time in studies of the Supreme Court. We don't just look at Republicans vs. Democrats; we typically examine each justice's votes individually. This is also done occasionally in studies of district judges and administrative judges.

There are many advantages to this approach. We completely avoid the problem of measurement error. We can still examine differences by party, gender, prior experience, or other demographic variables. And sometimes there are interesting results about outlier judges that we might have missed with proxy variables.

When the ideology estimates aren't central to the study, we can treat the individual judges as fixed or random effects. There are many studies of sentencing, for instance, that treat judges as fixed effects. The random effects approach is rarer, but one example is Anderson, Kling, and Stith (JLE 1999), who model district judges as random effects in their study of the impact of the Sentencing Guidelines. In that paper, they are interested in the distribution of judge sentencing severity, not in the individual judges. There is no way they could have gotten the same results using proxy variables.

Empirical studies of circuit courts are a bit trickier because of the collegial nature of the decisions. If you are interested, take a look at my paper that addresses this question. I estimate individual judge's voting propensities in sex discrimination cases, accounting for a "cost" of dissent for both sides. I'm sure that there are other approaches that could be used as well for multi-member courts. You can find the current version of my paper here:


The estimates for individual judges at the end of the paper are derived from the judge's actual voting behavior, but I have found that they have strong predictive power for judges' votes in other areas of law. (Much better than party or GHP scores). So another possibility, that hasn't been exploited much so far, is to use measures derived from judges' voting records in other areas of law to proxy for ideology. This can't be used in all applications, and still leads to problems of measurement error and biased estimators, but at least the problem will be less severe.

Still, there are times that party or GHP scores are the best measures of ideology that wll be available. I agree that GHP scores seem to be on the ascendancy, but I don't think that they are always the superior measure. I have found situations where GHP scores outperform party, but I have also found situations where the opposite is true. There are also several advantages to using party-of-appointing-president. First, it is easily interpretable. The
coefficient on the party variable is the difference between the average Republican and the average Democrat. Do you know off the top of your head what a coefficient of 0.4 means on a GHP score? Second, if all of the right-hand-side variables are discrete, you can use a linear probability model, which is easier to interpret (and you know the direction of the measurement error bias). GHP scores are seldom used in linear regressions, because they generally don't have a linear effect on votes.

Tracy Lightcap

Quantifying what judges are doing during the course of cases is a real problem. The difficulty showed up regularly when I worked in judicial administration. Our office was charged with recommending which circuits would be eligible for new judgeships each year, a recommendation that went straight from us to the state judicial council and on to the legislature. Problem: how do you determine the actual workload burdens of judges in circuits that had differing jurisdictions? We had some general jurisdiction courts that heard traffic cases and some that didn't, ditto with some misdemeanors and juvenile cases. We had some general jurisdiction courts that were being dog-robbed of many of their civil cases by limited jurisdiction courts with some concurrent jurisdiction too. What's a hard working court statistician to do?

We solved this problem by using a Delphi technique to get the judges to estimate how much time they spent on different kinds of case and disposition combinations. This required getting the judges to make individual estimates, then to compare their previous figures with mean figures for other judges in their circuit or administrative district. The second round corrections were then averaged and applied to different case combinations (how many hours for a bench trial misdemeanor, how many for an uncontested divorce, ect.). We could then estimate the number of judge years needed to dispose of the caseload in a circuit based on a mutually agreed judge year standard.

Lot of work, right? True, but think of the alternative. And when we actually went out and surveyed the court records to see if the estimates were on track, lo and behold they were! (We were flabbergasted, btw.)

So how does this figure in on Greg's post? We might, just might, be able to get some sort of national sample of judges using a 3 or 4 stage cluster design and convince the judges to make such estimates. Or, more likely, get a national sample of recently retired judges to do it. Result: estimates of judicial time spent on different phases of cases that are at least as good as the measures of judicial ideology mentioned above and a big fat database with many research uses.

Well, just an idea.

The comments to this entry are closed.


July 2024

Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30 31      

Site Meter

Creative Commons License

  • Creative Commons License
Blog powered by Typepad