Judges’ Biases Can Be Balanced by Random Assignment But Randomness is Uncommon
Whether a judge you get in your Federal case is Democratic or Republican should not matter or, if it does, a judge should be assigned as if it doesn’t. More generally, how a judge feels about a type of case or a party, even absent an appearance of a conflict of interest, should not matter or, if it does, a judge should be assigned so as to even out the odds across the many cases the courts receive. This can be done so as still to maintain a nearly even caseload distribution.
A survey of Federal appellate courts has many statements of random assignment, some indicating that randomness is by computer, but, in the portions I read, nothing was said about how computers or anyone randomize the assignments.
Apparently, Federal trial judges are assigned largely as the promulgators of local rules incline. Rationales I’ve seen in print are whether a judge is local to where a case may be tried or has specialized expertise in a subject. But either may allow a judge to shop for the cases or parties they’d like to have, and especially for one party or lawyer but not the opposition to shop for the judge they’d prefer. If the rationale is harder to justify, the possibility of judge-shopping may be even more damaging to judicial neutrality.
The best system is random assignment from a relatively large list of judges. This has been an option in some courts in the past and may still be in some, but it’s not in all.
Randomness is defined by statisticians and mathematicians. The concept applicable here, the number of judges always being finite, is that of the discrete random variable. When a good method is used, keeping it secret is unnecessary. If a fair coin flip were used, saying so in advance would be harmless. Secrecy is needed only if there is a defect in the randomness and the defect must be hidden, but obscurity is not good security. A common method in the general population is mentally to pick a value (as in picking a number from one to ten), but people tend to seek or impose order and thus the mental process even in that effort tends to be nonrandom even if the people present don’t realize it. Using a method such as the next digit in a long sequence of digits like pi (Π or 3.14159 . . .), even though it’s a nonrepeating sequence and, when long, almost impossible to memorize, fails, because it can be looked up or calculated. Mathematical randomness is required.
Randomness in assigning judges has also been tested, decades ago. One attorney told of learning in law school from a guest-lecturing court clerk how random assignment worked: in that court, it was by picking the next name in the list of available judges. I don’t recall if the list, including its length and its order, was public. If it was known only to the clerk, for a party or attorney filing only one case or no more cases than one more than the number of judges on the list, that would be as good as random. However, an attorney, such as a prosecutor or a representative of a busy law firm, with more cases than one plus the number on the list, filing one case at a time, and not filing the next case until the identity of the previously-assigned judge was known to that attorney, could sort the cases yet to be filed so that particular cases would be assigned to more amenable judges.
Appeals should have a similar principle. If three Federal judges are to hear an appeal, each one should be assigned randomly. Only an en banc appeal would not have randomness, and that is only because all of the court’s judges sit on the case.
Party affiliation may not matter in the same way that it does in the legislative and executive branches, since judges serve for far more years than do legislators and executives between their elections and adherence to political parties’ views is not enforced through the electoral process. Indeed, pressure for political orthodoxy is weakened by the duty, at least implied by the presence of conflicting duties, to refrain from deciding most cases on political grounds. It may even have little to do with fundamental ideological views of a judge. However, it appears that fundamental leanings of many judges generally do correlate with party affiliation at the time of nomination and confirmation and during their reviewable prior history, so they probably continue to correlate after they begin serving and for some years after.
Decades ago, a news report said that a judge had given a speech indicating that that particular judge took an especially dim view of pornography and thereafter a law enforcement agency chose to prepare its case against a pornographer so that the arrest would occur under that judge’s jurisdiction of place. Many of us take a dim view of pornography and of unlawfulness in general. So, the question can be raised, was this a legitimate tactical decision by the government or did the defendant have a right to arbitrariness in judge assignment? Arguably, an arrest in a place does not guarantee which judge will be assigned, but if a nonrandom system of judge assignment allowed the prosecutor to get that favorable judge for that case, then the chain is complete.
Petty juries are supposed to be, in part, randomly assembled. Since challenges are not random and are without compensatory randomness of the post–voir dire jury, the final composition cannot be random. But, at least, the pool is relatively random; when they respond to a summons, probably many cases are pending in the court they report to and prospective jurors are not yet assigned to an actual case.
I examined preliminarily how several courts assign judges. Before applying randomness, I applied some selectors: only Article III Federal courts with subnational boundaries, no bankruptcy court or panel, no two district courts to be from the same circuit, and no circuit court to be over any district court I had chosen. I also considered the court for New Jersey, a Third Circuit court, because it was in the test described above. The sample size is about five percent of district courts and more for appellate courts (five of 94 not counting New Jersey and two of 13, respectively), thus keeping the margin of error low. I randomly chose which courts by using computer spreadsheet software with a function that selected random numbers between a lower numerical value and a higher one, inclusive (I tested the inclusivity), together with a list of the 94 district courts (excluding bankruptcy courts), so that if the spreadsheet randomly gave me the number “1” I’d interpret that as the 1st court in the list of 94. The software was LibreOffice Calc version 18.104.22.168 on a PCLinuxOS FullMonty 2016.03 platform and the Calc function was RANDBETWEEN. The randomly sampled district courts were for Hawaii (which is in the Ninth Circuit), Eastern Pennsylvania (in the Third Circuit), Connecticut (First Circuit), Western Virginia (not West Virginia) (Fourth Circuit), and Northern Illinois (Seventh) and the randomly sampled appellate courts were for the Second and Eighth Circuits. Because of the application of the selectors prior to random selection, the District Court for the Central District of California was excluded due to being in the same appellate circuit, the Ninth, as is the District of Hawaii, which had already been selected.
To gain clarifications, followup communications to the courts were sent, all in the same week, except for courts from which no clarification appeared to be useful. For each court contacted, contact was in my own name and with reference to this website (lawslip.com). It was by email if an email address or form was found on its website; if not, it was by postal mail. Consistency of content among what was sent was approximate, inconsistent especially because the exact questions had to vary due to differences in local court rules. Probably all contacts should have been by the same medium, such as postal mail, and all should have been timed for simultaneous arrival at the respective courts.
In Hawaii, for criminal cases, the rules are silent.
For civil cases, the rules say, “[c]ases will be assigned as determined by the court, in most cases by random draw.” Many people would not know how to make the draw random. For example, mentally picking a number off the top of one’s head, so to speak, is usually not a random method. So, whether the courthouse method said to be random is statistically random should be evaluated.
By letter of to the Office of the Court Clerk, presenting “a judge assignment generic research question on no specific case”, I asked, “[w]hen the method of assigning a case to a judge is by random draw, what is the method for the random draw?”
In reply, a letter by an unnamed Deputy Clerk, who did not sign, on stationery of the Office of the Clerk, Sue Beitia, dated
Assuming that there was only one Deputy Clerk on the letter’s date and serving that court and that the letter is official, it partially clarifies the rules. The Case Management and Electronic Case Files (CM/ECF) system is available for all of the Federal Article III courts, but an initial search through Google, not performed to a major depth, did not reveal how CM/ECF can be used for randomness and by what method it provides randomness. The latter point especially needs further research, in large part because CM/ECF, presumably with this feature, is in use by most or all courts in the Federal judiciary.
Pennsylvania Eastern District
Under Eastern Pennsylvanian rules, both criminal and civil, judges’ names are to be written on cards, which are to be kept nonsequentially. A set of cards is always in a sequence, so being nonsequential is impossible, and this is virtually admitted in the next sentence of the two rules on point. Perhaps what was meant was that the cards were not to be in a certain kind of sequence. But that’s not said, so a clerk could treat the restrictive clause as meaningless and without effect. For civil cases, the judge-name cards are to be kept in “sequences”, while for criminal cases they’re to be kept in one “sequence”; so, for civil cases alone, randomness is possible in that one sequence can randomly follow another, but that, at best, is only a partial solution, since randomness should be at the level of individual judges.
The rules bar everyone from trying to ascertain which judge will be assigned. A court can enforce a rule against saying it aloud; but cannot enforce a rule to prevent an observer from having a subconscious sensitivity to hints. Psychologists already have scientific studies establishing the impossibility. A lawyer filing cases might be just such an observer. Las Vegas has many visitors with just such skills, and, reputedly, they often find it advisable to keep their mouths shut about why they think the next card won’t be an ace while they announce how much they’re betting. A system of randomness would not depend on lawyers with stacks of cases becoming almost comatose during judge assignment, one case at a time.
The cards with judges’ names are to be preserved. That’s either for the record or for reuse. Reuse tends to produce a gradual accumulation of dog-eared corners and other forms of wear and tear that can be noticed even when the names are out of view. Case numbers are to be stamped onto the cards; if that’s not on the same side as the judges’ names (and obviously assuming that a judge’s name is on one side only), case numbers can be memorized without memorizing all of them or all of the digits, and even just the angle, position, incompleteness, and/or color of a stamped number can be enough to recognize a card.
This leaves me with some questions, so I emailed the court clerk’s office on , and I am hopeful that an answer will arrive shortly. My subject was a “judge assignment generic research question (no specific case)”. I asked, “[w]hen the method of assigning a case to a judge is with assignment cards, once a card is stamped with a case number, is that physical card reused for a future assignment or is a new card made with the same judge's name? The court's rules provision specifying preservation of a once-used card does not specify whether preservation is for the record of the past use or for reuse.” I continued, “While for criminal cases assignment cards are to be kept in one sequence, for civil cases the assignment cards are to be kept in multiple sequences within a block. For civil cases, are they actually kept in multiple sequences and, if so, what determines which judge is in which sequence and, since with multiple sequences there is inherently a sequence of sequences, what is the method for sequencing the sequences and, if random, what is the method for the randomness?”
In Connecticut, regarding attorney discipline, randomness is required, but how it is to be achieved is not said. And since the order of assignment is not to be revealed except to judges and the clerk, presumably it’s important to try to hide the order, which means it must not be random.
An email inquiry to the court produced this reply from Joanne Pesta, Operations Supervisor, of the court: “When a case is filed, the computer will randomly select a judge. There is no method to random.” () Of course, there are multiple methods for randomness, but perhaps the court does not understand this.
Regarding assignment to magistrate judges, no randomness is required by the rule.
Regarding criminal matters, the rules specify only that cases will be assigned “in accordance with a general policy . . . .” That is almost as vague as silence. The sentence following that one talks of secrecy of assignment until the judge is assigned, which seems contradictory, but perhaps the drafter knew what they meant to say, and the most likely version of that would imply that the system lets you know in advance which judge will be assigned, and therefore the system is not random. A fair coin toss does not let you know how the coin will land until a fraction of a second before it lands, and that fraction is generally short enough to prevent an exploitable revelation. A random system produces a result that can be immediately used in public.
Regarding special proceedings, which generally were not considered for this study, a similar secrecy provision applies even though only one judge handles them during any given “designated period”. During that period, if one person discovers who the judge is, and one will, the secret can make the rounds. And if that person is not in the court’s personnel, then revealing the judge’s name is not illegal.
For other kinds of civil cases, “a general policy on assignments . . .” is, to the court, enough, but it’s too vague.
Virginia Western District
In the Western District of Virginia, a standing court order speaks of implementing random selection, but the “random” method is explicitly for assignment to one named judge or to the next judge in a rotation of named judges. It depends on other factors, like an address, but, for this court, randomness is only a label, not a description, for the method.
Illinois Northern District
For Northern Illinois, the system is “by lot”. That sounds like something we’d know the meaning of but I wouldn’t be sure, given what other courts are saying. For one circumstance, a rule says, “by lot in accordance with the rules of this Court”, but I have not seen any further rules. In another circumstance, the system is whatever system is applied “solely under the direction of the Executive Committee by the clerk or a deputy clerk who is designated by the clerk as an assignment clerk.” In other words, we don’t know.
To find out, I wrote a letter to the Clerk of Court, by name, on , with “a judge assignment generic research question on no specific case”, which was, “[w]hen a case is to be assigned to a judge ‘by lot’, what is the specific method?”
The court Clerk, Thomas G. Bruton, replied by letter of : “When a case is assigned to a judge ‘by lot,’ it means randomly assigned via computer.” That’s helpful, but specifically what method is used with the computer to be random is not stated.
New Jersey District
For New Jersey, the one court not randomly chosen in this study but included because, according to an attorney, its older system was judicially reviewed on appeal and found to fail as not random, the district court now states information on how judges are to be assigned while the court rules per se now say little on point.
First, considering the rules: A capital habeas corpus case is to be assigned to the next judge in a vicinage list; but next-in-list is not random.
Criminal cases are in some circumstances to be assigned according to a “plan”, but what plan is not said.
Civil cases (other than capital habeas cases, which, being brought on habeas corpus petitions, are civil) are to be assigned under an order from the chief judge, which order I haven’t seen.
However, looking below the rules, the court does provide public information via a website FAQ (usually meaning ‘Frequently Asked Question’) that is probably issued under the authority of the local civil rules: “Judges are randomly chosen by a computer database, similar to a deck of cards. When one judge is assigned, the cards are shuffled again for the next selection. The Clerk's Office has no discretion in the assignment of cases.” Ignoring the technical point in computer science that a database does not choose anything (a database management system (DBMS) or its user might do that) and assuming the shuffling is metaphorical, that is, it’s by the computer and not by hand, which is more work than letting a well-designed computer program do it and thus would tempt someone to skip the step from time to time, well-designed including that the computer program forces the shuffling after each assignment of one judge and that using the program is forced, this may be, for the courts studied here, a system that could be the closest to random, and, except for one omission, the system might indeed be random, the omission being of which judges are in the database at the times of reshuffling and assignment.
Among appeals courts, the one for the Second Circuit uses a random draw for death penalty appeals and, for some other cases, if a three-judge panel finds itself being a two-judge panel which decides it needs a third judge, random selection is to be used by the clerk.
However, how most cases get judges or panels assigned is not stated and, where randomness is required, how randomness is achieved is not stated.
I might find that out. I wrote a letter to the Office of the Court’s Clerk on , stating “a judge assignment generic research question on no specific case”, specifically, “[w]hen a judge is to be assigned to a panel by a random drawing or a judge is to be designated by random selection, what is the method for randomness in either situation?” A reply may be pending.
The Eighth Circuit has the most explicit rule in this sample: “The clerk's office uses software to form the hearing panels and randomly assign the cases. The judges do not participate in the panel-formation or the case-assignment processes.” Which software and how it is set up are not said, but, if this much was thought about and made into a rule by the court, one can find more justification for hoping that the programming was competent. Many programmers are aware of the difference between random and next-in-list selection.
In pursuit of clarity, I wrote a letter to the Clerk of Court, by name, on , offering “a judge assignment generic research question on no specific case”, in particular, “[w]hen a case is to be assigned by software to a judge randomly, what is the specific method for randomness?”
The reply was unsigned and did not quite answer the question. By letter of , the Clerk of Court, Michael E. Gans, wrote: “This is to acknowledge your letter dated , in which you ask about the specific method for randomness in the assignment of judges on the Eighth Circuit Court of Appeals to specific panels. The software uses a series of algorithms taking into account the judges available to sit during certain months and in certain court sites, the numbers of times judges sit with one another to distribute ‘co-sits,’ and the availability of judges who have taken senior status. The software was developed locally.” While this addresses steps prior to the applying of randomness for panelization, how randomness itself is determined for either panelization or case assignment when there is more than one panel is not said.
Courts Administrative Office
The Federal judiciary generally seems to misunderstand randomness. The Administrative Office of the U.S. Courts says, “[t]he majority of courts use some variation of a random drawing. One simple method is to rotate the names of available judges.” But rotation is not randomness. Rotation tells you explicitly who the next judge will be, allowing prediction with complete accuracy, whereas randomness requires that knowing the name of one assigned judge tells you nothing about which judge will be next. In general, the Administrative Office’s explanation of randomness does not meet a definition prevailing in the academic discipline of statistics.
More Information Could be Available
For only one court in this study did I find public information on point from the website for the court other than the rules, that one being the court for New Jersey, which had a FAQ. I do not know if any of these courts has an offline publication, such as an information sheet for pro se litigants, that might answer the question, or if court staff would answer such a question by telephone, in person, or by mail or email; I did not try those possibilities.
Other Venues and Situations
Article I Federal courts and State and local courts and administrative agencies should also be considered. Many operate on an en banc basis, including many having only one adjudicator each, but probably all others should apply randomness. Where assignment rules apply to subsets of cases within the larger categories of civil and criminal, such as those in admiralty, at the grand jury stage, before magistrate judges, or on attorney discipline, they should be considered.
Many nonjudicial situations permit testing past results for likely randomness, but it is questionable whether court records are adequate for testing. Unless precise times of assignments were recorded, with precision being to the second, or finer if required for uniqueness for one assigner, along with who did the assigning if each assigner had their own randomly-ordered list and along with information sufficient to support authorized nonrandomness, so that a long sequence of assignments can be discovered or reconstructed, testing may be impossible. Asking a clerk to remember a sequence is susceptible to self- and judge-serving error against rules and statistics and is generally inadequate.
It is not sufficient for a court to say only that its system is “random”. Either every court needs to define randomness as statisticians would or an authority above the courts must do so. And the definition must be both abstract and operational. The abstraction can be by reference to a pre-existing definition. Operationalism must clarify how a court could implement it so that a resulting assignment of a judge is not nonrandom.
The New Jersey situation described above may suggest that litigation to a level above where a challenged procedure is implemented will have a lasting beneficial effect. It may be too difficult to use a trial level for anything but preservation of an issue, because a trial judge invalidating a procedure used by the same court's other judges without protest from them may hear the invalidation questioned in no uncertain terms, e.g., as merely purported. Thus, a lawyer who did not benefit from nonrandomness might have at least a member of the court’s personnel and a statistician testify, then raise the issue cursorily but sufficiently for preservation, and then bring it to the appeal with full briefing and oral argument. I don’t know if an amicus brief would also be helpful.
Nonrandomness sometimes may be acceptable, when a policy has a good ground for preferring it. Some reasonable objectives are satisfied by nonrandomness. Specialized subject knowledge advances fairness of decisions. Patent is one such field and traditionally lawyers, even when denying that they specialized by subject, did so for patents, likely because much more scientific knowledge was expected than for handling most other kinds of cases. Judges being locally available advances convenience and affordability, and thus justice and speed, for the parties. The solution to those quandaries is not to assign based on subject specialization or locality unless several judges share the specialty or address and are otherwise a cross-section of the judges in the court. Likely, “several” as to judges would be at least three or four.
Relatedness of cases is a debated ground for judges to take on cases formerly assigned to others. It can be abused. But the Church of Scientology decades ago applied for Federal tax deductibility and parties filed over two thousand lawsuits toward getting it. Depending on the law at the times of those cases, that might have been a good situation in which a judge should have grabbed many or all of the cases as “related”, since consolidations of trials and of motion hearings and consistency of judgments on similar facts are likely good ideas for the sake of national justice.
Emergency situations can make randomness impracticable. Perhaps a courthouse has to be open around the clock for matters that can’t wait until the next normal business hour, like requests to stay execution in capital cases in the minutes before midnight. There may not be enough work to justify having more than one judge on hand outside of normal hours, so the identity of the judge can be known before filing, thus can inform a party’s or attorney’s decision on whether to file or which pre-prepared documentation to file. However, for cases that can wait till morning and later, a full line-up of judges may be available and then random selection from that roster can be applied.
Letting a party select a judge may sometimes be appropriate. Sometimes, an application to the United States Supreme Court would be decided on by one justice, who would be the justice permanently allotted to take cases arising from the relevant part of the country. If the party was denied relief, the party is then permitted to approach any other justice of the party’s choice. However, when counting all proceedings in all courts across the nation, there probably are few such cases in which parties may select judges for themselves.
If It Matters
I don’t know if public confidence in judges is waning due to party leanings and the cases they like. Probably, it’s not by much, so far, and won’t be as long as there’s no high-profile scandal. A few questions have been raised in recent years, but not enough to make a major problem.