« "The Company Store: A Deeper Look at Prison Commissaries" | Main | Prez Trump posthumously pardons boxer Jack Johnson »

May 24, 2018

Amazing new empirical research in federal sentencing outcomes detailing disparities based on political background

This week brought this amazing new working paper by Alma Cohen and Crystal Yang titled simply "Judicial Politics and Sentencing Decisions." I did not want to blog about the paper until I had a chance to read it, and doing so make me want to now do dozens of blog posts to capture all the issues the paper covers and raises. The paper's simple abstract provides a hint of why the paper is so interesting and provocative:

This paper investigates whether judge political affiliation contributes to racial and gender disparities in sentencing using data on over 500,000 federal defendants linked to sentencing judge.  Exploiting random case assignment, we find that Republican-appointed judges sentence black defendants to 3.0 more months than similar non-blacks and female defendants to 2.0 fewer months than similar males compared to Democratic-appointed judges, 65 percent of the baseline racial sentence gap and 17 percent of the baseline gender sentence gap, respectively.  These differences cannot be explained by other judge characteristics and grow substantially larger when judges are granted more discretion.

Each of these three sentences could alone justify multiple postings on just research particulars: e.g., I believe a database with over 500,000 sentencings might be the largest ever assembled and analyzed; I wonder if the data looks different for Clinton and Obama judges among the Ds, for Nixon and Reagan and others judges among the Rs; I fear many judge characteristics like prior jobs and connections to certain communities are really hard to control for.  In other words, just the scope and methods of this research is fascinating.

Moreover and more importantly, there is great richness in the findings of the full paper.  For example, the authors find "statistically significant differences in racial gaps in base offense level and final offense level by judge political affiliation."  In other word, the authors have discovered worrisome disparities in how guideline ranges are set/calculated, not just in how judges sentence in reaction to a particular guideline range.   Some additional notable findings are summarized in this recent WonkBlog piece at the Washington Post headlined "Black defendants receive longer prison terms from Republican-appointed judges, study finds."  Here are excerpts:

Federal judges appointed by Republican presidents give black defendants sentences that are, on average, six to seven months longer than the sentences they give to similar white defendants, according to a new working paper from Alma Cohen and Crystal Yang of Harvard Law School.  That racial sentencing disparity is about twice as large as the one observed among judges appointed by Democrats, who give black defendants sentences that are three to four months longer than the sentences they give to white defendants with similar histories who commit similar crimes....

They did find, however, that the gap between sentences for black and white defendants was smaller for more-experienced judges than for less-experienced ones.  They also found that differences between how Republican and Democratic judges treat black and white defendants grew larger after the Supreme Court's 2005 decision in United States v. Booker, which gave federal judges much more leeway to depart from federal sentencing guidelines.

Importantly, however, they found that growing differences between Democratic and Republican judges in the post-Booker era are due to Democratic judges reducing disparities in how they sentence black and white defendants.  Given more discretion, in other words, Democratic judges treated defendants of different races more equally, while Republican judges continued to carry on as they had before.

Cohen and Yang also found one important geographical effect: Black defendants fared particularly poorly in states with high amounts of population-level racial bias, measured here by the percentage of white residents in a given state who believe there should be laws against interracial marriage.  These states tend to be clustered in the South, and previous research has shown a similar racial sentencing bias in these states when it comes to capital punishment.

Finally, they also observed an opposite effect in how Democratic and Republican judges treated female defendants: While all judges tended to hand down shorter sentences to women than to men charged with similar crimes, Republican judges were considerably more lenient to women.  “Overall, these results indicate that judicial ideology may be a source of the persistent and large racial and gender disparities in the criminal justice system,” Cohen and Yang conclude.

Anyone with any experience in the federal sentencing system knows full well how judicial ideology may be a source of the persistent and large disparities in the operation of the system. But reflecting on my own experiences as a defense attorney and expert in a number of federal sentencing settings, I am eager here to highlight how the impact of judicial ideology may be impacted by the work of other actors involved in the federal sentencing process. I often sense that those judges (perhaps disproportionately Republican Appointees) with an earned reputation as a "by the guideline" type may not consistently receive the same type of mitigating information from probation officers and defense attorneys as do those judges known often to depart or now vary.

If readers are as intrigued and engaged by this new paper as I am, please say so in the comments, and I may try to see if I can encourage some folks to write up some guest-postings about this research.

UPDATE: A helpful reader sent me this link to the full paper in case folks are not able to access it via the NEBR site.

May 24, 2018 at 11:55 AM | Permalink


The differences are not meaningful, being tiny. They are smaller than the differences before and after lunch in Israel. This study shows Republican judges care about black victims much more than Demcrat judges. Decent black people should be voting for Republican executives to appoint judges.

Posted by: David Behar | May 24, 2018 12:31:18 PM

I initially filed this into the "More F'ed Up Sh!t That My Brain Cannot Process Right Now" file. I'm glad you did a post on it. I'm going to spend some time re-reading this study.

Posted by: Annie | May 24, 2018 2:12:47 PM

Assume the method is valid. Assume the disparities are important and rise to the level of discrimination. Assume the discrimination is insidious and inherent. Assume higher courts are flooded and cannot remedy the discrimination.

This study is a strong argument for the sentencing robots. They would run algorithm written by the legislature. They would be oblivious to status and would not get hypoglycemic and irritable before lunch.

Posted by: David Behar | May 24, 2018 3:17:29 PM

Thanks for the analysis.

Posted by: Joe | May 24, 2018 4:21:30 PM

Doug, given your experience in the federal criminal justice system, I am curious as to your thoughts on one of the core assumptions of the paper. The authors write:

"The key assumption of our empirical design is that cases are randomly assigned to judges within the same district court, in particular to Republican-appointed versus Democratic-appointed judges (p. 2). ... "In many courts, cases are randomly assigned to federal district court judges after charges are filed in order to “assure equitable distribution of caseloads and avoid judge shopping"" (p. 6).

I know one of the difficult aspects of conducting empirical work on federal judge sentencing is that the Administrative Office of the US Courts has a policy of removing judge identifying information from data prior to making it available to the public. TRAC, based out of Syracuse University (and one of the sources used in the paper) has used FOIA requests to obtain judge-level data and relink it with sentencing data. However, TRAC charges a subscription fee to access the data in raw form. One of the reasons that AOUSC has given for stripping judge-data is to prevent attorneys from judge-shopping based on statistical analysis of sentencing patterns. But, that statement itself presumes that it is in fact possible to "judge shop" in the federal criminal justice system, which would seem to cast some doubt on the author's core assumption that cases are randomly assigned. I don't think it is a stretch to assume that those who work in a district court have a general sense (through experience or word-of-mouth) of the general sentencing proclivities of the judges in that their district.

While I do not have experience with the federal criminal justice system, in my younger days I had my fair share of experience as a defendant in county court. When I paid a premium for a defense attorney one of the advantages I knew I was getting was the attorney's ability, within limits, to strategically navigate scheduling and court contacts in an attempt to get the case before certain judges. Living in a conservative county in which the majority of county judges had been elected running as Republican, I would find myself amazed when my case (using a certain defense attorney) was heard by a judge who was one of only two Democrats and the only female on the bench.

In short then, my question is whether defense attorney and prosecutors in the federal system are able to strategically act in a way that increases the likelihood that they will have their case heard by a specific judge, or at decrease the likelihood that the case is not heard by a specific judge. I could see how this tactic would be beneficial to both the defense and the prosecution.

Posted by: anonuser879 | May 24, 2018 5:55:08 PM

This is indeed a very interesting paper particularly on the sentence outcomes. However, I do have an issue with the assertion that the differences in base and final offense level by race reflect the fact that "offense level manipulation may be endogenous to the sentencing judge." But, at least in terms of the USSC data, it is my understanding that the base offense level is not determined by the judge - this typically comes from the PSR, which is written by the PO; the code book notes PSR/SOR as the origin, so there is some question about the origin) and certainly the adjusted offense level comes only from the PSR. Now the PSR could be revised as a result of objections by either attorney but the quantitative data wouldn't reflect this, or what role the judge played. Related to this, it is often the prosecutor who determines the base offense level through the plea agreement, but again, that wouldn't be reflected in the data (I don't think?) and thus isn't something they are able to look at it in their discussion of prosecutorial discretion. Maybe I'm missing something?

Posted by: Christine Scott-Hayward | May 24, 2018 8:20:48 PM

anonuser879: My sense is that federal district courts do a reasonable job with random assignment, though that can be influenced in various ways by various factors (e.g., senior judges willing only to take certain types of cases; multi-defendant cases being consolidated, etc.)

CSH: Though the PSR does an initial guideline calculation, judges have to make a formal and final determination on any matters that get disputed. But often matters are not disputed (either because this has been settled via plea terms OR there is just no fight over any particular), and so it is sound to note that other actors have a significant impact on the guideline calculation. Again here I wonder about a synergy: when prosecutor and defense know they pulled a tougher judge, they may negotiate to a guideline range (or the PSR author may gravitate to a range) that is a little distinct than when a less-tough judge is in charge.

Both these comments provide great reminders of how dynamic and multi-player the federal sentencing system is. That reality does not mean we ought to look for patterns between judges and certain outcomes, but it does mean we need to keep in mind that all sentencing outcomes involve the input of an array of players.

Posted by: Doug B | May 24, 2018 10:39:46 PM

I am interested in more analysis of their findings and methodology. I am particularly interested in how you think defense attorneys or prosecutors can or should use this data either from an advocacy standpoint or a idealist standpoint.

Posted by: Nick S | May 25, 2018 8:05:37 AM

Christine, respectfully, you are incorrect. The PSR makes a recommendation. The judge determines base offense level based on the PSR, sentencing briefs, any witness testimony at a sentencing hearing, and anything else in the record. The judge often, but not always, adopts many of the findings of fact in the PSR. Some judges adopt them in total, but a more typical outcome in a fiercely litigated case is to adopt only the portions that are undisputed or turn out to be entirely consistent with the judge's conclusion, then make the remaining findings of fact from the bench.

The PSR's calculation is not binding.

Posted by: Nick S | May 25, 2018 8:12:52 AM

Respectfully Nick you are missing my point, which is about where the data come from. I never said the judge had nothing to do with the Offense level. My point was that the data that the study authors use most often come from the PSR, which may or may not reflect anything that has happened at the sentencing hearing itself. It’s potentially a methods issue.

Posted by: Christine Scott-Hayward | May 28, 2018 9:57:20 AM

Post a comment

In the body of your email, please indicate if you are a professor, student, prosecutor, defense attorney, etc. so I can gain a sense of who is reading my blog. Thank you, DAB