Rethinking the Principal–Agent Theory of Judging


Many commentators argue that a principal–agent model is helpful to understanding judicial hierarchies. Under the traditional principal–agent paradigm, the principal, having a set of goals in mind, selects an agent to fulfill them. The agent has a tendency to shirk rather than fulfill the principal’s goals; the principal monitors—and can punish and even discharge—the agent to the extent that the agent observably fails to follow through on the principal’s wishes.

Scholars have tried to fit the principal–agent paradigm to the setting of judicial hierarchies, describing lower courts as agents of their higher court principals.1 A superior court reviews a lower court in order to minimize shirking.2 It also crafts judicial holdings that will constrain the lower court’s freedom to decide cases contrary to the superior court’s preferences.3

In this Essay, we challenge the underpinnings of the principal–agent understanding of judicial hierarchies. Specifically, we argue that the fit of the principal–agent model to this setting is worse than common wisdom would suggest. We begin by questioning the theoretical justification for applying the paradigm to judicial hierarchies. We then empirically examine whether judicial actors behave as the model suggests that they would, in a setting where application of the paradigm should be at its apex. We do not find evidence of such behavior.

Consider initially the weakness of the theoretical argument in favor of importing principal–agent understandings to most judicial hierarchies. There are two reasons that this simple story is insufficient to describe the real relationship between higher and lower court judges. First, while principals ordinarily select their agents, higher court judges usually do not select lower court judges. Rather, in the case of the federal judiciary, higher court judges are stuck with lower court judges that the President has appointed with the advice and consent of the Senate.

Second, lower court judges may also cast votes with an eye to the possibility of elevation to a higher court. But once again, the higher court judges who review the lower court’s decisions usually do not decide whether to elevate a judge from that court to a higher position. In the case of the federal judiciary, that responsibility falls to the President and the Senate. At best, lower court judges may be seen to be potentially responsive to two masters: higher courts and the political branches.

In order to get better purchase on these theoretical problems, we empirically examine the application of the principal–agent paradigm to judicial hierarchies. We identify the federal bankruptcy litigation system as an area that allows for a natural experiment from which to tease out answers to these questions for the following reasons. First, the system provides a more prototypical principal–agent relationship between higher and lower courts than do other systems in the federal law regime. The bankruptcy judges who sit as trial judges at the bankruptcy court level are appointed for renewable time-limited terms by the court of appeals. Moreover, the court of appeals provides a second intermediate level of appellate review of bankruptcy court decisions. Bankruptcy court decisions are appealed first to the district court or, alternatively, to a bankruptcy appellate panel (“BAP”) if the circuit has created one. The circuit’s judicial council, over which the court of appeals has dominant sway, selects BAP judges from among the circuit’s bankruptcy court judges. BAP decisions, in turn, are appealed directly to the court of appeals.

Second, BAP judges simultaneously sit on two courts—the BAP and the bankruptcy court. This provides an opportunity to observe how the same bankruptcy judge may change his voting behavior depending on his voting capacity—that is, as a trial judge or as an appellate judge. One would expect a bankruptcy judge voting in his appellate capacity to be more sensitive to monitoring by the court of appeals. For one thing, BAP decisions are appealed directly to the court of appeals, whereas the appeal of a bankruptcy court decision must wend its way through the first tier of appellate review (i.e., either the district court or the BAP) before reaching the court of appeals, thereby making the bankruptcy judge’s decision as a trial judge further removed from the watchful eye of the court of appeals. Moreover, while bankruptcy court decisions lack precedential effect, BAP decisions have precedential value and thus warrant greater scrutiny by the court of appeals.

To evaluate empirically the principal–agent theory of judging, we have collected data on the voting behavior of circuit court judges and bankruptcy judges (both as trial judges when sitting on the bankruptcy court and as appellate judges when sitting on the BAP) in student-loan-discharge proceedings in consumer bankruptcy cases. While analyses of the data provide support for the proposition that the ideological preferences of the circuit court judges predict their voting behavior, we do not find evidence of voting behavior by bankruptcy judges that would suggest sensitivity to the potential for principal monitoring. Thus, our findings cast some doubt on the principal–agent theory of judging.

This Essay proceeds as follows. In Part I, we elucidate different theories that commentators have advanced regarding the voting behavior of judges. In particular, we explicate the attitudinal model and arguments maintaining that judges consider other actors—both judicial and extrajudicial—in deciding how to cast votes. We also discuss principal–agent models for judicial hierarchical interactions, highlighting shortcomings in the application of these models. In Part II, we provide a summary of the federal bankruptcy litigation system, focusing on aspects of the judicial structure within which our natural experiment arises. In Part III, we present our empirical study, first explaining our research design and then turning to our statistical analyses and results. In Part IV, we interpret these results and consider the implications of our findings.

I.     Models of Judicial Behavior

In this Part, we consider models and theories that purport to describe the voting behavior of judges. We begin with the unconstrained attitudinal model. We then consider arguments that incorporate the possibility of strategic voting by judges that anticipates responses by other judicial and nonjudicial actors. Finally, we discuss principal–agent models of the judicial hierarchy.

A.     The Attitudinal Model

The attitudinal model predicts that judges will vote their sincere preferences.4 The model does not ignore the fact that others—the legislature, the executive, the public, or other judges—may prefer to have judges cast votes different from those that they in fact cast. Instead, the model simply assumes that judicial independence is strong enough that these outside forces wield too little power to influence judges’ votes.

The attitudinal model is at its theoretical apex when dealing with the votes of Supreme Court Justices, who enjoy life tenure and are in all likelihood unconcerned with the ability to ascend to another office. It is therefore not surprising that commentators have found strong evidence of the attitudinal model in Supreme Court voting, especially in certain areas such as criminal justice.5 Commentators have also found evidence of ideological voting by judges on courts of appeals in subject matters ranging from criminal justice6 and employment discrimination7 to administrative law,8 environmental law,9 and patent law.10

B.     Strategic Considerations

Some scholars suggest limits on the attitudinal model, arguing that even Supreme Court Justices take into account how Congress and the President will react to a decision in deciding how to vote. Thus, for example, if a decision in line with a Justice’s true preference will predictably result in Congress and the President enacting a law that replaces the Court’s decision with a legal standard that is even less desirable than the status quo, the Justice may vote for the status quo notwithstanding what his or her pure personal preferences would dictate.11 Such concerns may also affect voting by judges on lower courts.

There are additional reasons to question the applicability of the pure attitudinal model to voting by lower court judges. First, there may be other constituencies that lower court judges wish to please. For one thing, lower court judges may seek elevation to a higher judicial post. On that account, they may seek to curry favor with the current President and Senate in order to facilitate selection for a higher court. For another thing, lower courts are subject to reversal by higher courts. One can say that the costs of reversal to lower court judges are low (or that the likelihood of reversal, especially by the Supreme Court given the minute percentage of cases it hears, is low). Still, a lower court judge may take a reversal as a public rebuke. Reversals may also adversely impact a judge’s reputation. In the end, the specter of reversal may constrain lower court voting. Second, from a legal perspective, lower court judges are seen to be more constrained by precedent—and thus less free to vote ideologically—than are higher court judges.  

C.     Principal–Agent Judicial Hierarchical Models

Commentators have long understood a principal–agent relationship to exist between higher and lower courts in a judicial hierarchy.12 Higher courts review lower court decisions to constrain shirking; lower courts consider reversal a sufficient sanction to deter at least some shirking.13

However, the application of the principal–agent model in the judicial setting is often far from perfect.14 Professor Pauline Kim has elucidated the ways in which “the principal–agent model does not map so neatly onto the structure of the [federal] judicial hierarchy.”15 First, the Supreme Court does not appoint—or otherwise contract with—lower court judges.16 Second, the Court “has none of the usual levers of control to shape the incentives of lower court judges.”17 Finally, whereas a typical “agent has the power to impact the interests of the principal,”18 lower court decisions can only affect the Court if one understands the Court to care about actual case outcomes (in addition to policy pronouncements).19

While building upon Kim’s work, our analysis extends beyond it by exploring uncharted territory. Specifically, commentators have largely overlooked a setting of growing importance where Article III judges on one court select judges to sit on another court, with the former court reviewing decisions of the latter. Although Congress authorizes these judgeships, the lower court judges are answerable to the Article III judges who appoint them.20 Here, the principal–agent judicial relationship is stronger.21 Thus, by shifting focus away from the Supreme Court, our theoretical and empirical analyses seek to provide new insights into the applicability of the principal–agent theory within the federal judicial hierarchy.22

II.     The Federal Bankruptcy Litigation System

The Judicial Code vests original but nonexclusive jurisdiction over bankruptcy matters with the federal district courts.23 At the same time, the Code establishes in each federal judicial district a bankruptcy court that functions as a “unit” of the district court.24 The Code authorizes the district courts to refer bankruptcy cases to the bankruptcy courts in the first instance.25 The district courts have accepted this invitation, with each district court having a “standing” order of referral to the bankruptcy court in its district.26

The bankruptcy courts are staffed by bankruptcy court judges. These non-Article III judges are appointed for fourteen-year terms by the judges of the court of appeals for the circuit in which the district in question is located.27 Moreover, the court of appeals has dominant sway over the multi-stage process that ultimately culminates in the presentation of a single candidate to the court for a vote of appointment.28

The Judicial Code authorizes bankruptcy court judges to “hear and determine” (i.e., enter final judgments) in all “core proceedings” that arise under the Bankruptcy Code or that arise in a case under the Bankruptcy Code.29 Such proceedings lie at the heart of a bankruptcy case and include matters relating to estate administration, the allowance of creditor claims, the dischargeability of debts, and the confirmation of reorganization plans.30 Appeals from a final judgment in a core proceeding lie ordinarily with the federal district court.31 However, the Judicial Code authorizes—indeed mandates absent certain exceptions—circuits to create BAPs: tribunals consisting of bankruptcy court judges that hear appeals from the bankruptcy courts in the circuit.32 BAP judges are selected from sitting bankruptcy court judges by, and for terms set by, each circuit’s judicial council.33 BAP judges continue to hear cases as trial judges in addition to their BAP responsibilities.34

In circuits that have established BAPs (currently the First, Sixth, Eighth, Ninth, and Tenth Circuits),35 litigants appealing a bankruptcy court judgment have the option of having either the district court or a three-judge BAP panel hear their appeal.36 The BAP will automatically hear the appeal unless a party timely files an election to have the district court hear the appeal.37 Whether the district court or the BAP hears this initial appeal, a second appeal as of right lies to the court of appeals,38 with discretionary review by the Supreme Court possible thereafter.39 This appellate structure is summarized in Figure 1.40

Figure 1

Federal Bankruptcy Litigation System for Core Proceedings

Figure 1 use

III.     Empirically Investigating the Principal–Agent Theory of Judging

A.     Theoretical Considerations

We empirically investigate the principal–agent theory of judging in the setting of judges who serve on BAPs even while they continue to serve as bankruptcy court judges. This is a felicitous setting in which to conduct our investigation for two overarching reasons. First, the setting provides a stronger principal–agent relationship than one typically finds in a judicial hierarchy. Thus, if the principal–agent understanding of judging has traction, we ought to find evidence of it here. Further, the absence of such evidence in this setting would cast doubt on the principal–agent paradigm.

The second reason this setting is felicitous is that it allows for a natural experiment. As we explain below, bankruptcy judges reasonably should anticipate greater, and more effective, monitoring by the courts of appeals of their voting behavior as BAP judges as compared to their voting behavior as bankruptcy court judges. Insofar as the judges we study cast votes as bankruptcy court judges even while they cast votes as BAP judges, the setting allows us to tease out the extent to which a stronger principal–agent relationship translates into greater conformity of voting by agents to the preferences of the principals.

1.     The Strength of the Principal–Agent Relationship in the Federal Bankruptcy Litigation System

The setting we study here—bankruptcy court judges and BAP judges as agents of the courts of appeals—is a stronger judicial principal–agent relationship than one ordinarily encounters in other judicial hierarchies. First, as discussed above, the court of appeals either directly appoints these agents (in the case of bankruptcy court judges)41 or has great influence in appointing them (in the case of BAP judges).42 In doing so, they have incentives to select individuals who are ideologically aligned with them.43 Second, not only does the court of appeals have review power over BAP judges (and indirect review power through the hierarchy over bankruptcy courts),44 it also decides whether to reappoint these agents to the bankruptcy court and plays a robust role (through its influence on the circuit judicial council) in deciding whether to reappoint these agents to the BAP.45

2.     The Heightened Incentive and Ability for Courts of Appeals to Monitor BAPs Relative to Bankruptcy Courts

The mere fact that the relationship between circuit courts and bankruptcy judges (whether sitting as trial or appellate judges) meets the theoretical requirements for a paradigmatic principal–agent relationship would not alone be enough for us to conduct a meaningful empirical analysis. Crucial to our investigation is the fact that courts of appeals have heightened incentives and ability to monitor BAPs relative to bankruptcy courts. The potential for increased scrutiny invites a comparison of (1) the votes cast by bankruptcy judges serving on a BAP to (2) the votes cast by the same judges serving on the bankruptcy court to determine whether the expected differential in monitoring results in a difference in voting patterns.

There are four reasons, summarized in Table 1, to expect that BAP judges will be more sensitive to the potential for circuit court review of their decisions as compared to decisions rendered in their capacity as bankruptcy court judges. First, bankruptcy court decisions are not likely to be appealed in the first instance.46 In contrast, in an appeal that has reached the BAP, a strong incentive likely exists in a nontrivial number of those appeals for the losing litigant to subsequently appeal the decision to the court of appeals.47

Second, even leaving aside the question of appeal, to the extent that courts of appeals are likely to pay attention to lower court decisions that do not reach the court on appeal, they are more likely to focus on BAP decisions than bankruptcy court decisions. BAPs are higher in the judicial hierarchy than are bankruptcy courts. Moreover, bankruptcy court opinions do not constitute binding precedent.48 In contrast, BAP decisions are seen, at a minimum, to create binding precedent on future BAP panels, and possibly on all bankruptcy courts within the circuit.49

Third, additional effective monitoring at the BAP level results from the fact that the BAP hears cases in panels of three judges. Scholars have elucidated different ways in which decision-making by a panel may affect the votes of the individual judges comprising the panel. The fact that judges may jointly deliberate could affect individual judges’ votes.50 Further, commentators have documented how a judge’s vote may shift depending upon the ideologies of the other judges on the panel.51 One way or another, then, other judges on a panel might tend to moderate a judge’s tendency otherwise to rule in conflict with the preferences of the court of appeals. In contrast, bankruptcy court judges almost always make decisions individually in their trial capacity.52

Fourth, when bankruptcy judges sit as trial judges, they hear witnesses and evidence directly. They can decide which witnesses and evidence are most credible, and they accordingly have more freedom to couch their decisions in findings of fact that are generally subject to less stringent appellate review.53 In contrast, on the reasonable assumption that BAP judges predominantly resolve legal issues or mixed issues of law and fact,54 they have much less opportunity to couch their decisions in ways that insulate them from appellate review.55

Table 1: Institutional Constraints on the Voting Behavior of Bankruptcy Judges

Table 1 

 The foregoing discussion leads us to promulgate the following hypotheses.


Hypothesis 1: Judges will cast votes according to different ideological valences depending upon whether they are casting a vote as a bankruptcy court judge or as a BAP judge.


Hypothesis 2: The voting behavior of BAP judges will be more ideologically aligned with the ideological leaning of the circuit than will the voting behavior of bankruptcy court judges.


In the following Subparts, we describe our empirical study to test these hypotheses.

B.     Research Design

To empirically investigate the principal–agent theory of judging, we focus on votes cast in student-loan-discharge proceedings in consumer bankruptcy cases. We have chosen this specifically focused approach for the following reasons.

First, confining our investigation to a narrow subset of disputes minimizes concerns over controlling for (1) differences in the procedural nature, litigant identity, and subject matter of bankruptcy disputes; (2) potential selection effects in the decision to bring an appeal across various types of disputes; and (3) differences in burdens of proof and appellate review standards.56 By eschewing such heterogeneity, our fine-grained approach seeks to bolster the reliability of our findings.

Second, student-loan-discharge proceedings involve a nontechnical area of bankruptcy law with a minimal role (if any) for specialized expertise. Prior to October 7, 1998, a debtor could discharge a student loan in bankruptcy on one of two grounds: by showing either (1) that the first payment on the student loan had become due more than seven years before the debtor filed for bankruptcy or (2) that repaying the student loan would impose an undue hardship on the debtor.57 After that date, Congress amended the Bankruptcy Code so that a debtor could discharge a student loan only upon a showing of undue hardship.58 Since then, there has not been any intervening statutory change to the standard.59 Importantly, the overwhelming majority of the votes we observe in our study (i.e., more than ninety percent) were cast in undue-hardship-discharge proceedings. Because the Bankruptcy Code does not define undue hardship,60 “a great deal of residual policymaking inheres in determining the scope of discharge whenever educational debt is at issue.”61 Moreover, such policymaking invites a court to make a general, nontechnical inquiry into the level of sacrifice that is expected of debtors and the threshold at which the sacrifice becomes impermissible.62 Put another way, it is an inquiry that calls upon a judge to draw upon personal notions of fairness.63 Finally, such proceedings are resolved pursuant to procedures that should be familiar to all judges within the federal judiciary—namely, a full-blown federal lawsuit.64

Third, given the opportunity for remarkably broad policymaking in student-loan-discharge proceedings, we expect it to be an area where one would witness ideological voting by judges. And, in fact, there is some evidence to support this proposition. We have previously theorized that circuit court judges would be likely to engage in ideological voting in debt-dischargeability determinations.65 Although our prior efforts did not unearth general evidence to this effect,66 one of our unreported findings was that a statistically significant correlation at the ten-percent level (p = 0.065) existed between the ideology of the voting judge and the direction of the judge’s vote in student-loan-discharge determinations—specifically, that as the judge’s ideology became more conservative, the odds of voting conservatively (i.e., in favor of the creditor) increased.67 Because our prior study involved the votes of circuit court judges in only four regional circuits over a ten-year period,68 we expect to unearth more robust evidence of such ideological voting if we conduct a similar study of student-loan-discharge determinations from all regional circuits over a longer period of time.

Furthermore, prior empirical research has demonstrated that bankruptcy court doctrine interpreting the Code’s undue hardship standard has been inconsistent, as evidenced by differential treatment of similarly situated debtors—an outcome best explained by differing judicial perceptions.69 Additionally, in a study of outcomes of settled and tried undue-hardship-discharge proceedings, it has been documented that legally irrelevant factors that should not bear on the merits of a debtor’s claim for relief, including the identity of the bankruptcy judge assigned to the debtor’s case, were statistically significantly associated with the amount of relief that the debtor obtained, and that such factors were more strongly associated with the legal outcome than the handful of legally relevant factual characteristics that were also statistically significantly associated with the amount of debtor relief.70 Finally, in an experimental study in 2004 involving 113 bankruptcy judges, the authors of the study documented a statistically significant association between a judge’s self-reported political affiliation and the direction of his or her vote in a hypothetical undue-hardship-discharge proceeding, with Democratic judges discharging a greater amount of educational debt (i.e., more relief) than Republican judges (i.e., less relief).71 All of this evidence suggests that bankruptcy judges likely have an opportunity (if not a propensity) to vote ideologically in student-loan-discharge determinations. As such, we deem such determinations to be fertile ground for exploring the principal–agent theory of judging.

To implement our research design, we created two original datasets—one focusing on votes cast by circuit court judges (the “Principal Dataset”) and one focusing on votes cast by BAP judges and bankruptcy court judges (the “Agent Dataset”). More specifically, both datasets focus on votes cast in student-loan-discharge proceedings during the fifteen-year period beginning on January 1, 1997 and ending on December 31, 2011. We chose this time period in order to capture the BAP experience at its apex in terms of participating circuits: BAPs did not become a fixture in the bankruptcy appellate system until 1996. Prior to that time, only the Ninth Circuit had a continuously operating BAP. After the 1994 amendments to the Judicial Code, the First, Second, Sixth, Eighth, and Tenth Circuits established BAPs.72 This is the legal landscape today, with the exception of the Second Circuit, which terminated operation of its BAP on July 1, 2000.73

To constitute the Principal Dataset, we formulated a search query in Westlaw’s FBKR-CS database, which contains reported and unreported decisions and orders relating to bankruptcy issued by various courts, including the federal circuit courts of appeals. The search query consisted of a single term: “523(a)(8),” the provision of the Bankruptcy Code pursuant to which a court determines whether a debtor’s student loan is dischargeable.74 This search term was coupled with (1) a date restriction that limited query retrieval to opinions issued during the fifteen-year period beginning on January 1, 1997 and ending on December 31, 2011, and (2) a field restriction that limited query retrieval to documents whose preliminary field contained the term “court of appeals.”75 The search query produced a total of 115 documents. Because most of these were opinions rather than orders, for ease of reference we will collectively refer to them as “opinions” for the remainder of the Essay.

To constitute the Agent Dataset, we formulated two separate search queries in Westlaw’s FBKR-CS database. The first search query consisted of a single term: “523(a)(8).” This search term was coupled with (1) a date restriction that limited query retrieval to opinions issued during the fifteen-year period beginning on January 1, 1997 and ending on December 31, 2011, and (2) a field restriction that limited query retrieval to opinions whose preliminary field contained the term “bankruptcy appellate panel.”76 The search query produced a total of 87 opinions.

Pursuant to the selection procedures set forth below, after identifying the set of judicial votes by BAP judges in student-loan-discharge proceedings for inclusion in the Agent Dataset, we used the identity of those BAP judges to formulate a second search query that would retrieve the votes cast, if any, by the same set of judges while sitting at the trial level as bankruptcy court judges in student-loan-discharge proceedings. The search query consisted of a single term: “523(a)(8).” This search term was coupled with (1) a date restriction that limited query retrieval to opinions issued during the fifteen-year period beginning on January 1, 1997 and ending on December 31, 2011, (2) a field restriction that limited query retrieval to opinions whose preliminary field contained the term “bankruptcy court,” and (3) a field restriction that limited query retrieval to opinions whose judge field contained the last name of any of the bankruptcy judges who had already been included in the Agent Dataset by virtue of having voted as a BAP judge in a student-loan-discharge proceeding.77 The search query produced a total of 237 opinions.

Both datasets include published and unpublished opinions that involve the resolution of: (1) trial-level dispositions of adversary proceedings involving a determination regarding the dischargeability of educational debt pursuant to Bankruptcy Code § 523(a)(8); and (2) appellate-level dispositions (either by BAPs or courts of appeals) of such proceedings.78 We included opinions that disposed of the appeal on the merits as well as opinions that involved solely procedural dispositions, such as dismissal for lack of jurisdiction, dismissal for lack of standing, and dismissal on mootness grounds. If an opinion involved multiple debt-dischargeability issues, but the court did not address all of them, only those matters addressed by the court were coded. For all consolidated appeals decided by a single opinion, we disaggregated the consolidated appeals for coding purposes.

Pursuant to these selection procedures, our Principal Dataset consists of 138 judicial votes cast by 84 circuit court judges and derived from 46 opinions. Our Agent Dataset consists of 305 judicial votes—more specifically: (1) 168 judicial votes cast by 70 BAP judges and derived from 56 BAP opinions; and (2) 137 judicial votes cast by 39 bankruptcy court judges and derived from 137 bankruptcy court opinions.

C.     Ideology of the Principal: The Voting Behavior
of Circuit Court Judges in Student-Loan-Discharge Determinations

Here, we seek to provide an analysis of ideological voting by circuit court judges in student-loan-discharge determinations by fitting a simple logistic regression model. The dependent variable is whether the judge voted conservatively. Consistent with our prior empirical research regarding the voting behavior of circuit court judges in debt-dischargeability determinations,79 we code a judge’s vote as conservative if it fully favored the creditor and as liberal if the vote partially or fully favored the debtor.80 Approximately 48.9% of the votes in the Principal Dataset were conservative. Put another way, the odds of observing a conservative vote in the dataset were approximately 0.96 to 1. Thus, liberal voting by circuit court judges appears to be the norm in student-loan-discharge determinations, but only slightly so.

To predict whether the judge voted conservatively, the independent variable is the voting judge’s ideology, which we measure using the judge’s Judicial Common Space score (“JCS” score).81 JCS scores range from -1.0, the most liberal score, to 1.0, the most conservative score.82 We were able to assign a JCS score to the voting judge for 136 of the votes in the Principal Dataset.83 The median and mean JCS scores in the Principal Dataset are, respectively, 0.077 and 0.039. The most liberal JCS score is -0.681 and the most conservative JCS score is 0.702.

Table A1 in the Appendix sets forth the results from our regression model. Overall, the simple model (Model 1) is statistically significant as compared to a model without independent variables.84 The voting judge’s JCS score is a statistically significant predictor of the direction of a judge’s vote in the expected direction.85 To examine the effect of a voting judge’s JCS score on the likelihood of a conservative vote, we estimate using the simple model equation the predicted probabilities of a conservative vote as the JCS score changes. As illustrated in Figure 2, as the judge’s JCS score increases (i.e., becomes more conservative), the likelihood of conservative voting increases.86 

Figure 2
Probability of Conservative Vote by JCS Score

Figure 2

D.     Searching for Evidence of Agent-Based Voting:
The Voting Behavior of BAP Judges and Bankruptcy Court
Judges in Student-Loan-Discharge Determinations

Having found evidence suggesting that the ideological preferences of circuit court judges influence their voting behavior in student-loan-discharge determinations, we now seek to ascertain whether evidence exists that the voting behavior of a bankruptcy judge changes based on his or her voting capacity (i.e., as a bankruptcy court judge or as a BAP judge) and whether evidence exists that the voting behavior of BAP judges will be more ideologically aligned with the ideological orientation of the circuit than the voting behavior of bankruptcy court judges. We do not find evidence of such voting behavior.

First, to explore the effect of voting capacity on the voting behavior of bankruptcy judges, we control for whether the judge cast his vote while sitting as a bankruptcy court judge (i.e., at the trial level) or as a BAP judge (i.e., at the appellate level). Approximately 51.8% of the votes in the Agent Dataset were conservative. Thus, in the absence of a relationship between the voting capacity of the judge and the direction of the judge’s vote, we would expect to see the judges in our sample vote conservatively approximately 51.8% of the time. For votes cast by bankruptcy court judges, the judges voted conservatively approximately 54.2% of the time; and for votes cast by BAP judges, the judges voted conservatively approximately 48.9% of the time. The difference between the observed and expected values is not statistically significant (p = 0.360) according to a chi-square test with two degrees of freedom.

Second, to explore the correlation between the ideological composition of the circuit and the voting behavior of the bankruptcy judges, we fit a simple logistic regression model with the direction of the judge’s vote as the dependent variable (coded 1 for a conservative vote and coded 0 for a liberal vote). The independent variable in the model is the ideological composition of the circuit, which we operationalize by using the median JCS score among all active courts-of-appeals judges in the corresponding regional circuit at the time that the bankruptcy judge cast his or her vote.87 As set forth in Table A2 of the Appendix, there is no statistically significant association between the ideological composition of the circuit and the direction of the voting judge’s vote. This finding holds true even when we control for the voting capacity of the judge.

IV.   Interpretation and Implications

Our empirical study focused on an area in which one might have expected the principal–agent theory of judging to be robust. By selecting an area where the theory should have applied—if it applies anywhere—we can be more confident that, to the extent we find that the theory’s predictions do not hold, there is something fundamentally wrong with the theory itself.88 Having found evidence suggesting that ideology influences the voting behavior of circuit court judges in student-loan-discharge determinations, but simultaneously failing to find evidence of conformist voting behavior by BAP judges, we contend that our results cast empirical doubt on the principal–agent theory as a construct for understanding the decision-making process of judges.

But before we offer our thoughts on the implications that flow from discrediting the application of the theory in this particular context, we begin by addressing a challenge that may be raised to our argument that the failure to find evidence supporting Hypothesis 2—that is, that the agents would be responsive to their principals—warrants concluding that, here, the principal–agent theory stands on shaky ground (at best). After all, the absence of evidence is not evidence of absence. There could be a plausible explanation for our lack of findings with respect to Hypothesis 2. If valid, that alternate interpretation would not justify the inference we draw from the lack of findings.

The most plausible alternate story to explain our lack of findings with respect to Hypothesis 2 would be one about how the selection process for bankruptcy judges tends to produce ideological homogenization. If, as has been theorized elsewhere, the specialization and expertise demanded as qualifications for those who would serve as bankruptcy judges make it easier for the circuit courts to identify appointees with a particular ideological bent, the selection pattern that emerges will be one that populates the bankruptcy bench with members who have ideological preferences consistent with those of the circuit court.89 On this account, it should be perfectly understandable why bankruptcy judges do not alter their voting behavior based on the capacity in which they vote—that is, as either bankruptcy court judges (i.e., trial capacity) or BAP judges (i.e., appellate capacity). Given the politicization of the bankruptcy bench in the first instance, when the circuit court subsequently selects from this applicant pool individuals who will serve as BAP judges, those individuals will continue to exhibit the desired type of politicization that prompted the circuit to select them to serve as bankruptcy court judges. Accordingly, one would not witness any difference in the voting pattern of bankruptcy judges on the basis of their voting capacity.

There are a couple of reasons why this alternative interpretation is unconvincing. First, it assumes that the ideological composition of the bankruptcy bench will not fluctuate once the bankruptcy judges have been appointed. It very well could be the case that, once appointed, some bankruptcy judges may not feel constrained to follow the lead of the circuit court in judicial policymaking. After all, with a fourteen-year term that bestows a significant degree of independence, some bankruptcy judges may not feel beholden to hew to the ideology of the circuit court.90 Or, even if all bankruptcy judges were inclined to march in step with the policymaking signals issued by the circuit court, the alternative interpretation fails to account for the possibility that some bankruptcy judges may have a tendency to drift closer to the ideology of the circuit after ascending to the bench. One would expect this group of judges to be among the prime candidates for selection to serve on a BAP. Given the variance in the ideological preferences of bankruptcy judges, it should be that BAP judges will be nonrandomly selected on the basis of the proximity of their preferences to those of the circuit court judges. Thus, once one takes into account the very real possibility of ideological drift (in either direction) by bankruptcy judges, one should expect to see differences in voting behavior of the bankruptcy judges on the basis of voting capacity. But we did not witness such a difference.

Second, even assuming that the ideological homogenization of the pool of bankruptcy judges remained static, the alternative interpretation is incomplete insofar as it does not account for our finding that the ideological composition of the circuit was not statistically significantly associated with the voting behavior of the bankruptcy judges (regardless of their voting capacity). If the circuit court judges had predicted fairly accurately at the time of selection the manner in which the appointed bankruptcy judges would vote, and if the bankruptcy judges had not drifted ideologically, then the ideological preferences of the circuit court should have predicted the voting direction of the bankruptcy judges—whether as an entire group or whether disaggregated by voting capacity (i.e., bankruptcy court judge or BAP judge). But they did not. Our lack of findings is most consistent with a story about the nonresponsiveness of bankruptcy judges (i.e., the agents) to the ideological preferences of the circuit court judges (i.e., the principals). With this in mind, we now turn to the implications of our results.

Should our study be seen as the death knell for principal–agent understandings of judging? Perhaps not entirely. There are two broad implications one can take from our study. One is more problematic for judicial application of principal–agent models than is the other.

One possibility is that we did not observe a difference in ideology based on voting capacity because the sanctions that bankruptcy court judges and BAP judges face from circuit court judges—a decision not to reappoint—is insufficient motivation for judges to hew closer to the circuit’s ideological preferences when they face closer monitoring.91 They do not modulate their votes when voting on the BAP because they do not fear the possible sanction.92

If this is the case, then the application of principal–agent understandings in other judicial settings is especially problematic. After all, the sanction of non-reappointment is one that is available in few judicial hierarchies.93 Most higher court judges do not have such a power. Indeed, circuit court judges have, with respect to BAP judges, the power to vote to deny reappointment in addition to the ordinary “sanctions” that higher courts generally laud over lower courts. Thus, if the potential sanctions that circuit court judges wield are insufficient to generate voting in line with the higher court’s wishes in this setting, then one is left to wonder whether the principal–agent understanding would ever apply in a judicial setting—at least with respect to altering votes to please higher courts once judicial appointment has been secured.

A second interpretation of our findings is less problematic for judicial application of principal–agent models. Perhaps the reason we did not observe modulation in ideological leaning in voting based on voting capacity was not because the courts of appeals wield unconvincing sanctions, but rather because the courts of appeals do not select BAP judges primarily for ideological leaning. Since ideology is not the primary basis for selection to the BAP, it is also not the basis on which circuit court judges predominantly monitor their agents.

Consider that higher courts may not seek a single, identifiable function from their lower court agents. Indeed, higher court judges may select lower court judges with an eye to fulfilling different, and often disparate, goals. Higher court judges may prefer to have lower court judges vote in accordance with the higher court judges’ ideological preferences.94 On the other hand, the higher court may also (or alternatively95) be concerned with minimizing lower court errors.96 Lower courts hear numerous cases, the vast majority of which may not raise many ideologically charged issues, and which can be resolved with fair ease under existing law. To the extent that outright legal errors become too frequent, public confidence in the judiciary may be undermined, perhaps ultimately threatening the power of judges.

Either of these goals is possible in bankruptcy: Circuits might appoint and monitor BAPs to ensconce ideological voting or to minimize errors. While we have focused here upon a subset of bankruptcy cases that seem to appeal to judges’ ideologies, bankruptcy is an area that is perceived to be—if it is not in fact—highly technical.97 To the extent that the possibility for technical error is sizeable, higher court judges may have an enhanced incentive to select lower court judges who are likely to minimize those errors.98

In this light, our findings may suggest not that the prototypical principal–agent story has no traction in bankruptcy. Instead, it may indicate that courts of appeals select bankruptcy court judges and BAP judges not to promote ideological decision-making, but rather with an eye to minimize the need for error correction in a technical area. Circuit court judges would find it costly to resolve cases “correctly” or even to audit lower court decisions on that basis.

If bankruptcy court judges and BAP judges understand that minimizing the need for error-correction is the dominant factor behind appointment and reappointment decisions by the circuit court and the circuit judicial council, then they may feel that closer scrutiny by the court of appeals will not mean that circuit court judges will more closely examine decision-making for ideological fidelity. If that is the case, one might not expect to find that closer scrutiny by the higher court enhances the lower court’s ideological alignment with the higher court. The principal–agent relationship will constrain lower court judges to vote with greater technical accuracy, but not greater ideological fidelity, when subject to greater scrutiny.

To put this story in context here, the main argument would be that circuit court judges (who sit on the circuit judicial council) vote to select BAP judges predominantly for technical expertise, not ideological alignment. Knowing this, bankruptcy court judges do not feel compelled to vote with greater ideological fidelity when casting votes as BAP judges, even in cases that are in fact more open to ideological decision-making.

There are two theoretical explanations that bolster this understanding. One is that BAP judges know that the court of appeals values their service largely because of their technical expertise. Given that, they feel that they may cast votes with substantial ideological freedom for several reasons: First, the court of appeals is not evaluating BAP judges on the basis of their votes in student-loan-discharge proceedings; second, ideologically charged cases make up a comparatively small part of the docket; and last, but not least, the value of BAP judges as technical experts is so great that the court of appeals would be loath to dismiss a BAP judge on the ground of ideological disagreement.99

A problem with this explanation is that there is no reason to think that technical expertise is correlated with ideological position. If that is true, then circuit court judges can presumably identify technically apt bankruptcy court judges to serve as BAP judges who also hew to the circuit’s preferred ideologies. Put another way, technical expertise is not so unique that it provides insulation against any kind of ideological review.

Another explanation for BAP judges’ apparent capacity to vote with ideological freedom in student-loan-discharge proceedings is that BAP judges have an information advantage that gives them great freedom to vote without a strong eye to the circuit’s ideology even in ideologically charged cases. Perhaps the circuit court judges believe that ideological voting is appropriate in the absence of technical expertise, but also assume—even if erroneously—that BAP judges vote based on technical expertise when they decide these cases. If that is true, then circuit court judges might be inclined to defer—and not audit closely—BAP judges’ rulings in student-loan-discharge proceedings (and other ideologically charged bankruptcy proceedings) even if the circuit court judges, alone and left to their own devices, would have voted with a clear ideological influence. Importantly, moreover, even if BAP judges in fact vote ideologically in student-loan-discharge proceedings (and, indeed, even in cases that circuit court judges perceive to be even more ideologically neutral), the BAP judges believe that the circuit court judges, thinking (erroneously) that the BAP judges cast their votes based on expertise rather than ideology, will defer to the BAP judges’ votes.

If this story—or some version of it—is correct, then what we have unearthed is not the utter inapplicability of the principal–agent approach to judicial decision-making. Instead, the results highlight the need to consider, and specify, the dominant goals that motivate a higher court to appoint and reappoint—and in general to monitor—lower court judges.


The extant literature on the relationship between higher and lower courts is a rich and varied one consisting of many strands, including the conceptualization of judicial hierarchies through a principal–agent lens. We elucidated some theoretical problems with applying principal–agent understandings to the typical judicial setting. Prototypical principals have some voice in selecting their agents, and are in a position either to impose some sanction upon agents who behave in an undesirable fashion or to offer a reward to agents who behave well. In contrast, higher courts generally do not select the judges who staff the courts beneath them and may sanction lower court judges only by reversing their decisions.

Taking these theoretical shortcomings into account, our empirical study focused on an area in which one would expect the principal–agent theory of judging nonetheless to be robust: the federal bankruptcy litigation system. We chose this setting because the system is one in which higher court judges actually appoint lower court judges. The fact that the higher court judges further have the discretion to reappoint the lower court judges, who lack life tenure, also affords higher court judges a greater ability to sanction lower court judges than in almost all other judicial settings. We then refined our focus to student-loan-discharge proceedings in consumer bankruptcy cases because of the opportunity for ideological voting by judges in such proceedings at both the trial and appellate levels.

Our study demonstrates that the ideological leanings of circuit court judges predict how they will cast votes in student-loan-discharge proceedings, thus confirming the inherently ideological nature of our area of focus. Given that, and given the fact that circuit court judges have an incentive to select bankruptcy judges who are ideologically aligned with them, one would expect bankruptcy judges to decide cases more in line with the circuit court’s ideology when casting votes in a capacity that subjects them to greater scrutiny by the circuit court. Importantly, our results do not support this hypothesis. Given that our study examined an area where the prototypical principal–agent story was most likely to be applicable, the fact that we have not found evidence that the story applied presents a significant challenge to scholars who have looked to the principal–agent theory of judging as an explanatory tool.


Table A1

Logistic Regression Models for

Conservative Vote by Circuit Court Judges

Table a1


Table A2

Logistic Regression Model for

Conservative Vote by Bankruptcy Judges

Table a2


  1. [1]. See, e.g., Clifford J. Carrubba & Tom S. Clark, Rule Creation in a Political Hierarchy, 106 Am. Pol. Sci. Rev. 622 (2012) (theoretical and empirical analyses); Tom S. Clark, A Principal-Agent Theory of En Banc Review, 25 J.L. Econ. & Org. 55 (2008) (treating en banc court of appeals as principal to individual panels as agents); Micheal W. Giles, Thomas G. Walker & Christopher Zorn, Setting a Judicial Agenda: The Decision to Grant En Banc Review in the U.S. Courts of Appeals, 68 J. Pol. 852 (2006) (same); Donald R. Songer et al., The Hierarchy of Justice: Testing a Principal-Agent Model of Supreme Court–Circuit Court Interactions, 38 Am. J. Pol. Sci. 673 (1994) (describing the Supreme Court as a principal and the lower federal courts as agents); see also Pauline T. Kim, Beyond Principal–Agent Theories: Law and the Judicial Hierarchy, 105 Nw. U. L. Rev. 535, 535 n.1 (2011) (citing examples of judicial politics scholarship that describes the federal judicial hierarchy in terms of a principal–agent relationship).

  2. [2]. E.g., Songer et al., supra note 1.

  3. [3]. E.g., Jeffrey R. Lax, Political Constraints on Legal Doctrine: How Hierarchy Shapes the Law, 74 J. Pol. 765 (2012).

  4. [4]. See generally Jeffrey A. Segal & Harold J. Spaeth, The Supreme Court and the Attitudinal Model Revisited (2002).

  5. [5]. See, e.g., id.

  6. [6]. Frank B. Cross, Decision Making in the U.S. Courts of Appeals 27–29 (2007). But see Cass R. Sunstein et al., Ideological Voting on Federal Courts of Appeals: A Preliminary Investigation, 90 Va. L. Rev. 301, 325–26 (2004) (finding no evidence of ideological voting in criminal cases by circuit court judges).

  7. [7]. Sunstein et al., supra note 6, at 319–21, 324–25.

  8. [8]. Thomas J. Miles & Cass R. Sunstein, The New Legal Realism, 75 U. Chi. L. Rev. 831, 842 (2008).

  9. [9]. Richard L. Revesz, Environmental Regulation, Ideology, and the D.C. Circuit, 83 Va. L. Rev. 1717 (1997); Sunstein et al., supra note 6, at 322–23.

  10. [10]. Banks Miller & Brett Curry, Expertise, Experience, and Ideology on Specialized Courts: The Case of the Court of Appeals for the Federal Circuit, 43 Law & Soc’y Rev. 839, 855–57 (2009) (finding evidence of ideological voting in patent cases at the Federal Circuit).

  11. [11]. Lee Epstein & Jack Knight, The Choices Justices Make 12–17 (1998).

  12. [12]. See supra note 1. In summarizing how “the federal judicial hierarchy resembles the types of relationships fruitfully analyzed under principal–agent theories,” Professor Pauline Kim has noted that the Supreme Court cannot monitor all actions by the courts below it, that there is generally at least some tension between the goals and policy preferences of lower court and higher court judges, and that lower court judges have an informational advantage over the Justices. Kim, supra note 1, at 553–54.

  13. [13]. See, e.g., Tracey E. George & Albert H. Yoon, The Federal Court System: A Principal-Agent Perspective, 47 St. Louis U. L.J. 819, 822 (2003).

  14. [14]. See Lee Epstein, William M. Landes & Richard A. Posner, The Behavior of Federal Judges: A Theoretical and Empirical Study of Rational Choice 33–36 (2013) (noting weaknesses in the traditional principal–agent account of the federal judicial hierarchy, and emphasizing instead restraints that judges may feel by virtue of oversight by the President and the Senate).

  15. [15]. Kim, supra note 1, at 537.

  16. [16]. Id. at 554.

  17. [17]. Id. at 556. Kim argues that “reversal alone is insufficient to ensure compliance with the Supreme Court’s goals where the lower federal courts have differing goals,” and that favorable policy outcomes are an insufficient reward to incentivize conformity by lower court judges. Id. at 556, 558.

  18. [18]. Id. at 558.

  19. [19]. Id. at 559–61.

  20. [20]. See Judith Resnik, Uncle Sam Modernizes His Justice”: Inventing the Federal District Courts of the Twentieth Century for the District of Columbia and the Nation, 90 Geo. L.J. 607, 669 (2002) [hereinafter Resnik, Uncle Sam] (“Can Article III judges enact the values of judicial independence as they carry out the duties of employing other judges? My focus here is on . . . the effects of appointment and reappointment of judges by judges on the conception of a judge as independent.”); Judith Resnik, Interdependent Federal Judiciaries: Puzzling About Why and How to Value the Independence of Which Judges, Daedalus, Fall 2008, at 28, 42.

  21. [21]. Cf. Resnik, Uncle Sam, supra note 20, at 677 (“In short, the process and the criteria used for reappointment [of statutory judges] pose challenges to the ideology of independent judges that Article III promotes. As constitutional judges evaluate the track records of statutory judges by soliciting information from litigants and by reviewing decisions and reversal rates, they may prompt lower level judges to search for supporters, publish little, and keep low profiles. While presumptive reappointment avoids those problems, it results in giving life-tenured judges power to create, de facto, another set of tenured judges.”).

  22. [22]. Kim does acknowledge the Supreme Court-centric nature of her analysis and the potential for limiting its import when considering other interactions within the federal judicial hierarchy. See Kim, supra note 1, at 540–41 (“As a caveat, my focus here is on the vertical relationships within the federal judiciary, specifically between the United States Supreme Court and the lower federal courts. Principal–agent theories have been applied to other interactions in the judicial system as well: for example, the relationships . . . between the federal courts of appeals and district courts. These theories may or may not be apt when applied to those settings, and I do not attempt to address them here.” (footnotes omitted)).

  23. [23]. See 28 U.S.C. § 1334(b) (2006).

  24. [24]. Id. §151.

  25. [25]. Id. §157(a).

  26. [26]. See, e.g., Jonathan Remy Nash & Rafael I. Pardo, An Empirical Investigation into Appellate Structure and the Perceived Quality of Appellate Review, 61 Vand. L. Rev. 1745, 1756 (2008).

  27. [27]. 28 U.S.C. §152(a)(1); see also id. § 152(a)(3) (“Whenever a majority of the judges of any court of appeals cannot agree upon the appointment of a bankruptcy judge, the chief judge of such court shall make such appointment.”).

  28. [28]. See Malia Reddick & Natalie Knowlton, Inst. for the Advancement of the Am. Legal Sys., A Credit to the Courts: The Selection, Appointment, and Reappointment Process for Bankruptcy Judges 1–19 (2013), available at
    wygwam/documents/publications/A_Credit_to_the_Courts.pdf (discussing the selection, appointment, and reappointment process for bankruptcy judges in all federal regional circuits other than the D.C. Circuit); Rafael I. Pardo, The Utility of Opacity in Judicial Selection, 64 N.Y.U. Ann. Surv. Am. L. 633, 645–47 (2009) (discussing process for the appointment of bankruptcy judges in the Ninth Circuit); see also Mary L. Clark, Judges Judging Judicial Candidates: Should Currently Serving Judges Participate in Commissions to Screen and Recommend Article III Candidates Below the Supreme Court Level?, 114 Penn. St. L. Rev. 49, 55–56 (2009) (discussing process for appointment of bankruptcy judges); Resnik, Uncle Sam, supra note 20, at 672 n.284 (“The regulations addressing bankruptcy permit but do not require the appellate courts to use merit selection panels.”).

  29. [29]. 28 U.S.C. § 157(b)(1). A bankruptcy court judge may also “hear and determine” a non-core proceeding if the parties consent to the entry of a final order by the judge. See id.
    § 157(c)(2). Without such consent, the judge may hear the matter but cannot enter a final judgment; instead, the judge must submit “findings of fact and conclusions of law to the district court,” which will provide a de novo review of the findings and conclusions to which a party has specifically objected. See id. § 157(c)(1).

  30. [30]. Id. § 157(b)(2)(A), (B), (I), (L). The constitutional authority of bankruptcy courts to definitively resolve all such proceedings, however, has recently been called into question. See Stern v. Marshall, 131 S. Ct. 2594, 2610–11 (2011) (holding that a bankruptcy court is constitutionally prohibited from entering a final judgment on a state law tortious interference counterclaim); Executive Benefits Ins. Agency v. Arkison (In re Bellingham Ins. Agency, Inc.), 702 F.3d 553, 565 (9th Cir. 2012) (“Taken together,GranfinancieraandSternsettle the question of whether bankruptcy courts have the general authority to enter final judgments on fraudulent conveyance claims asserted against noncreditors to the bankruptcy estate. They do not.”), cert. granted, 133 S. Ct. 2880 (2013).

  31. [31]. See 28 U.S.C. §158(c)(1).

  32. [32]. See id. §158(b)(1).

  33. [33]. See id. §158(b)(1), (3).

  34. [34]. See Thomas E. Carlson, The Case for Bankruptcy Appellate Panels, 1990 BYU L. Rev. 545, 558; Judith A. McKenna & Elizabeth C. Wiggins, Alternative Structures for Bankruptcy Appeals, 76 Am. Bankr. L.J. 625, 661 (2002).

  35. [35]. Nash & Pardo, supra note 26, at 1757.

  36. [36]. 28 U.S.C. §158(b)(1), (c)(1).

  37. [37]. See id. § 158(c)(1). There are further limitations on the ability of a BAP or a BAP judge to hear an appeal. A BAP is not empowered to hear appeals from bankruptcy courts in a given federal judicial district unless a majority of district court judges in that district vote to authorize the BAP to hear such appeals. Id. §158(b)(6). A BAP judge may not participate in a panel hearing of an appeal that originated in the district for which the BAP judge has been appointed as a bankruptcy judge. Id. §158(b)(5).

  38. [38]. Id. § 158(d)(1).

  39. [39]. Id. § 1254(1).

  40. [40]. It should be noted that appellate review of a bankruptcy court’s decision can involve direct appeal from the bankruptcy court to the court of appeals, with the effect of bypassing the first tier of intermediate appellate review—the district court or the BAP. Appeal may proceed directly to the court of appeals pursuant to a certification procedure if one of the following circumstances exists: (1) the appeal involves a question of law unresolved by the court of appeals for the circuit or by the Supreme Court; (2) the appeal involves a matter of public importance; (3) the appeal involves a question of law requiring resolution of conflicting decisions; or (4) the appeal may materially advance the progress of the case or proceeding in which the appeal is taken. Id. § 158(d)(2)(A). For a detailed discussion of the use of this appellate path, see Laura B. Bartell, The Appeal of Direct Appeal—Use of the New 28 U.S.C. §158(d)(2), 84 Am. Bankr. L.J. 145 (2010) and Lindsey Freeman, Comment, BAPCPA and Bankruptcy Direct Appeals: The Impact of Procedural Uncertainty on Predictable Precedent, 159 U. Pa. L. Rev. 543 (2011).

  41. [41]. See supra note 27 and accompanying text.

  42. [42]. Although it is the circuit judicial council that selects BAP judges, it seems reasonable to conclude that the court of appeals has substantial influence over the council given that the chief judge of the circuit presides over the council’s membership, which consists of an even number of circuit court judges and district court judges. See supra note 28 and accompanying text; see also 28 U.S.C. § 332(a)(1).

  43. [43]. See Pardo, supra note 28, at 649–51; cf. Reddick & Knowlton, supra note 28, at 23 (“[T]wo interviewees suggested that extraneous considerations may come into play after the merit selection panel completes its work. One panelist expressed the view that ‘politics come in at the judge level,’ and another cited an instance in which ‘political muscle’ may have led the court of appeals to select an applicant who was further down on the panel’s list.”).

  44. [44]. See supra note 40.

  45. [45]. See 28 U.S.C. § 158(b)(3) (authorizing the reappointment of BAP judges by the circuit judicial council); Federal Courts Improvement Act of 1996, Pub. L. No. 104-317, §303, 110 Stat. 3847, 3852 (“When filling vacancies, the court of appeals may consider reappointing incumbent bankruptcy judges under procedures prescribed by regulations issued by the Judicial Conference of the United States.”).

  46. [46]. For example, during the 2011 fiscal year there were 75,568 adversary proceedings commenced in the bankruptcy courts. Thomas F. Hogan, Admin. Office of the U.S. Courts, Judicial Business of the United States Courts: 2011 Annual Report of the Director app. at 287 tbl.F-8 (2012), available at
    2011/JudicialBusiness2011.pdf. An adversary proceeding is the analogue to a federal civil lawsuit. See Jonathan Remy Nash & Rafael I. Pardo, Does Ideology Matter in Bankruptcy? Voting Behavior on the Courts of Appeals, 53 Wm. & Mary L. Rev. 919, 933 (2012). Also during the 2011 fiscal year, litigants filed 3312 bankruptcy appeals in the first-tier appellate courts (i.e., the BAPs and district courts). Hogan, supra, app. at 114 tbl.B-10, 162 tbl.C-7. Accordingly, during the 2011 fiscal year, first-level appeals constituted 4.4% of total adversary proceedings.

    This figure is meant to give the reader a very rough approximation of the miniscule amount of appeals from the bankruptcy court to a first-tier appellate court. These numbers, however, should not be construed as an accurate measure of the appeal rate to such courts. On the one hand, the overwhelming majority of adversary proceedings settle.
  47. [47]. Cf. McKenna & Wiggins, supra note 34, at 662 (noting the argument by some that bankruptcy appeals from the district court to the court of appeals “are sometimes presumed to be the more serious cases in which a precedential opinion is needed or where reversible error has obviously been committed”). During the 2011 fiscal year, there were 683 bankruptcy appeals filed in the courts of appeals. Hogan, supra note 46, app. at 99 tbl.B-6. Given that during the same period litigants filed 3312 bankruptcy appeals in the first-tier appellate courts, see supra note 46, second-level appeals constituted 20.6% of total first-level appeals. It has been estimated that, during the 1997 fiscal year, litigants appealed merit-based dispositions by the BAPs to the courts of appeals at a rate of 36%. See McKenna & Wiggins, supra note 34, at 668.

  48. [48]. See, e.g., In re 400 Madison Ave. Ltd. P’ship, 213 B.R. 888, 890 n.2 (Bankr. S.D.N.Y. 1997) (stating that the decision of one bankruptcy judge in a multijudge bankruptcy court was not binding on the other bankruptcy judges); In re Jones, 112 B.R. 975, 977 (Bankr. W.D. Mo. 1989) (stating that the decision of a bankruptcy court from one district is not binding on a bankruptcy court from another district).

  49. [49]. See Nash & Pardo, supra note 31, at 1761 & nn.68–70; see also McKenna & Wiggins, supra note 34, at 652–53 (discussing precedential effect of BAP decisions).

  50. [50]. See Adeno Addis & Jonathan Remy Nash, Identitarian Anxieties and the Nature of Inter-Tribunal Deliberations, 9 Chi. J. Int’l L. 613, 615–17 (2009) (discussing judicial deliberation); Harry T. Edwards, The Effects of Collegiality on Judicial Decision Making, 151 U. Pa. L. Rev. 1639, 1644–70 (2003) (discussing judicial collegiality).

  51. [51]. See Revesz, supra note 9, at 1751–56.

  52. [52]. For examples of a bankruptcy court sitting en banc, see In re Outen, 220 B.R. 26 (Bankr. D.S.C. 1998); In re Calzadilla, 151 B.R. 622 (Bankr. S.D. Fla. 1993); Baker v. GFC Corp. of Mo. (In re Baker), 11 B.R. 125 (Bankr. W.D. Mo. 1981); and In re Ford, 3 B.R. 559 (Bankr. D. Md. 1980).

  53. [53]. See, e.g., Fed. R. Bankr. P. 8013 (stating that, on appeal to the district court or to the BAP, “[f]indings of fact, whether based on oral or documentary evidence, shall not be set aside unless clearly erroneous, and due regard shall be given to the opportunity of the bankruptcy court to judge the credibility of the witnesses”); Peklar v. Ikerd (In re Peklar), 260 F.3d 1035, 1037 (9th Cir. 2001) (stating that, for purposes of the appeal of a bankruptcy court’s decision, “[w]e review pure issues of fact for clear error”); Rifino v. United States (In re Rifino), 245 F.3d 1083, 1086 (9th Cir. 2001) (“We review the bankruptcy court’s findings of fact under a clearly erroneous standard.”); see also Kim, supra note 1, at 569 (“[T]he general rule against permitting new evidence to be considered at the appellate level and the standard of review requiring appellate courts to defer to a trial court’s factual findings exacerbate the lower courts’ informational advantage.” (footnotes omitted)); cf. Suzanna Sherry, Logic Without Experience: The Problem of Federal Appellate Courts, 82 Notre Dame L. Rev. 97 (2006) (criticizing existing doctrine for giving federal trial judges too much discretion on procedural issues and too little discretion on the merits).

  54. [54]. See, e.g., Cal. Self-Insurers’ Sec. Fund v. Lorber Indus. of Cal. (In re Lorber Indus. of Cal.), 564 F.3d 1098, 1100 (9th Cir. 2009) (“No questions of fact are at issue in this appeal; the parties disagree only about whether the bankruptcy court and the BAP correctly interpreted
    § 507(a)(8)(E)(ii). Thus, our review is de novo.”).

  55. [55]. See, e.g., In re Peklar, 260 F.3d at 1037 (stating that mixed issues of law and fact will be reviewed de novo); Cool Fuel, Inc. v. Bd. of Equalization (In re Cool Fuel, Inc.), 210 F.3d 999, 1001–02 (9th Cir. 2000) (“We review de novo the BAP’s conclusions of law. Moreover, we independently review a bankruptcy court’s ruling on appeal from the BAP.” (citation omitted)).

  56. [56]. See Nash & Pardo, supra note 46, at 935. Importantly, in a manner that is consistent with our general theory, see supra Part III.A.2, student-loan-discharge proceedings seemingly provide bankruptcy court judges a greater opportunity than BAP judges to couch their decisions in ways that insulate them from appellate review. See Hedlund v. Educ. Res. Inst., Inc., 718 F.3d 848, 854 (9th Cir. 2013) (“[W]e now confirm that a good faith finding under Brunner should be reviewed for clear error.”); Krieger v. Educ. Credit Mgmt. Corp., 713 F.3d 882, 884 (7th Cir. 2013) (“[W]e must remember that the statutory inquiry is ‘undue hardship,’ a case-specific, fact-dominated standard, which implies deferential appellate review.”).

  57. [57]. 11 U.S.C. § 523(a)(8) (1994) (amended 1998 & 2005).

  58. [58]. Higher Education Amendments of 1998, Pub. L. No. 105-244, § 971(a), 112 Stat. 1581, 1837.

  59. [59]. See 11 U.S.C. § 523(a)(8) (2012).

  60. [60]. See id. § 101 (failing to define undue hardship among the terms defined for purposes of the Bankruptcy Code).

  61. [61]. Rafael I. Pardo & Kathryn A. Watts, The Structural Exceptionalism of Bankruptcy Administration, 60 UCLA L. Rev. 384, 403 (2012); see also Roth v. Educ. Credit Mgmt. Corp. (In re Roth), 490 B.R. 908, 920 (B.A.P. 9th Cir. 2013) (Pappas, J., concurring) (“Congress has never defined the circumstances constituting the sort of undue hardship justifying the discharge of an educational debt under § 523(a)(8), apparently preferring that bankruptcy courts craft a working definition.”); id. at 923 (“Under § 523(a)(8), Congress did not draw bright lines, but instead presumably intended that bankruptcy courts have the flexibility to make fact-based decisions in individual cases about the need for student loan debt relief.”).

  62. [62]. Rafael I. Pardo, Illness and Inability to Repay: The Role of Debtor Health in the Discharge of Educational Debt, 35 Fla. St. U. L. Rev. 505, 516–17 (2008) (“At its core, an undue hardship discharge determination seeks to answer whether the debtor requesting relief will have the ability to repay his or her educational debt without suffering impermissible sacrifice—namely, undue hardship. This, of course, requires a court to predict on the basis of a variety of factors the likelihood that the financial distress suffered by the debtor will persist into the future.” (footnote omitted)).

  63. [63]. See, e.g., N.Y. State Higher Educ. Servs. Corp. v. White (In re White), 6 B.R. 26, 29 (Bankr. S.D.N.Y. 1980) (“It is regrettable that Congress shed so inadequate a spotlight on the exculpating phrase ‘undue hardship’. . . . It is also regrettable that so much is therefore left to the individual view of each judge who, after all, brings the sum of who and what he was, what he has become, and what he sees through his own eyes to this basically disagreeable task.” (citation omitted)).

  64. [64]. Rafael I. Pardo & Michelle R. Lacey, The Real Student-Loan Scandal: Undue Hardship Discharge Litigation, 83 Am. Bankr. L.J. 179, 188 (2009).

  65. [65]. See Nash & Pardo, supra note 46, at 931, 946.

  66. [66]. See id. at 954–58.

  67. [67]. For a discussion theorizing why a judge with a conservative ideology would be expected to vote conservatively (i.e., in favor of the creditor) in debt-dischargeability determinations, see id. at 951-54.

  68. [68]. See id. at 948–49.

  69. [69]. See Rafael I. Pardo & Michelle R. Lacey, Undue Hardship in the Bankruptcy Courts: An Empirical Assessment of the Discharge of Educational Debt, 74 U. Cin. L. Rev. 405, 495–509 (2005).

  70. [70]. Pardo & Lacey, supra note 64, at 223–29.

  71. [71]. Jeffrey J. Rachlinski et al., Inside the Bankruptcy Judge’s Mind, 86 B.U. L. Rev. 1227, 1247–48 (2006).

  72. [72]. Nash & Pardo, supra note 26, at 1779 n.122. It should be noted that the First Circuit had established a BAP in 1979 but terminated its operations in 1984 in the wake of the Supreme Court’s decision in Northern Pipeline Construction Co. v. Marathon Pipe Line Co., 458 U.S. 50 (1982). See Nash & Pardo, supra note 26, at 1779 n.122.

  73. [73]. Nash & Pardo, supra note 26, at 1779 n.122.

  74. [74]. 11 U.S.C. § 523(a)(8) (2012); see also United Student Aid Funds, Inc. v. Espinosa, 559 U.S. 260, 274 (2010) (“[Section] 523(a)(8) requires a court to make a certain finding before confirming the discharge of a student loan debt.”).

  75. [75]. The preliminary field in Westlaw is found at the top of caselaw documents and generally contains the name of the court that issued the document. In its entirety, the search query was the following: 523(a)(8) & da(aft 12/31/1996 & bef 01/01/2012) & pr(“court of appeals”).

  76. [76]. In its entirety, the search query was the following: 523(a)(8) & da(aft 12/31/1996 & bef 01/01/2012) & pr(“bankruptcy appellate panel”).

  77. [77]. The judge field in Westlaw is found at the start of the opinion and generally contains the name of the judge who authored the opinion. In its entirety, the search query was the following: 523(a)(8) & da(aft 12/31/1996 & bef 01/01/2012) & pr(“bankruptcy court”) & ju(aug baum baxter bohanon brandt brown bucki carlo clark clarkson cooper cordova deasy dreher dunn federman gallet goodman gregg haines hagan hargrove hill hillman jesus jones jury klein koger kornreich kressel kurtz lamoutte latta lundin mahoney markell marlar mcdonald mcmanus mcniff meyers michael montali morgenstern-clarren nail ninfo nugent o’brien ollason pappas parsons perris pusateri rhodes russell ryan saladino schermer scott shea-stonum smith stosberg tester vaughn venters votolato waldron).

  78. [78]. We excluded trial-level and appellate-level dispositions involving “discharge by declaration” of educational debt, pursuant to which a debtor seeks to discharge his or her student loans through a repayment plan without commencing an adversary proceeding. See, e.g., Ruehle v. Educ. Credit Mgmt. Corp. (In re Ruehle), 412 F.3d 679, 681 (6th Cir. 2005), abrogated by United Student Aid Funds, Inc. v. Espinosa, 559 U.S. 260 (2010) (Chapter 13 repayment plan); Kaufman v. Case W. Reserve Univ. (In re Kaufman), 122 F. App’x 815, 816–17 (6th Cir. 2004). We also excluded trial-level and appellate-level dispositions in which a debtor sought a determination that the discharge injunction, see 11 U.S.C. § 524(a)(2), had been violated by a creditor seeking to recover an educational debt from the debtor subsequent to his or her discharge. See, e.g., McKay v. Ingleson, 558 F.3d 888 (9th Cir. 2009).

  79. [79]. Nash & Pardo, supra note 46, at 954.

  80. [80]. Id. For a discussion explaining this coding protocol, see id. at 951–54.

  81. [81]. See Lee Epstein et al., The Judicial Common Space, 23 J.L. Econ. & Org. 303, 306 (2007). We gratefully acknowledge Professor Epstein and her co-authors for generously making their JCS data available to other researchers. See Lee Epstein, Research, Univ. S. Cal., (last visited Sept. 19, 2013).

  82. [82]. See Epstein et al., supra note 81, at 309.

  83. [83]. The JCS Database does not contain scores for circuit court judges who were appointed after 2008, and one of the votes in the Principal Dataset was cast by such a judge: the Honorable Susan L. Carney of the U.S. Court of Appeals for the Second Circuit, who was appointed in 2011.

    Approximately 5.8% of the votes in the Principal Dataset (8 of 137) were cast by federal district court judges sitting by designation. We assigned these judges the JCS scores provided in Christina L. Boyd,
  84. [84]. To assess model fit, we compare the observed and predicted values for a conservative vote. Using the model equation, we calculate the predicted probability of a conservative vote for each observation in the model given the actual values of the independent variable for that observation. For any predicted probability less than or equal to 0.5, we classify the predicted vote as a liberal vote; and for any predicted probability greater than 0.5, we classify the predicted vote as a conservative vote. The simple model, which only controls for the voting judge’s JCS score, correctly predicts the direction of the judge’s vote in approximately 55% of the observations. See infra Table A1. The proportion of correct predictions is referred to as the count R. J. Scott Long, Regression Models for Categorical and Limited Dependent Variables 107 (1997).

    Of course, without referring to the independent variable in the simple model, one could correctly classify the outcome in some of the observations by assigning the most frequent category of outcome—that is, the marginal distribution of the dependent variable—to all of the observations. In this case, one could correctly classify the outcome in 51.1% of the observations in the simple model by guessing a liberal vote for all observations. Thus, when predicting with a model that includes the voting judge’s JCS score, the error rate drops by 9.1%.
  85. [85]. See supra notes 67–71, 81–83 and accompanying text.

  86. [86]. We find that the statistically significant relationship between the voting judge’s ideology and the direction of his or her vote persists even when controlling for other variables to address competing theories for the determinants of the vote cast by the judge. See infra Table A1. We fit a multiple logistic regression model (Model 2) that, in addition to judicial ideology, controls for the following factors.

    First we control for whether the first-tier appellate court was a BAP. We do so given our prior finding that circuit courts perceive BAPs to provide a higher quality of appellate review than district courts, as evidenced by statistically significantly higher affirmance rates of BAP dispositions relative to district court dispositions.
    Second, we control for whether the bankruptcy court voted conservatively (Conservative Trial Outcome) and for whether the first-tier appellate court voted conservatively (Conservative First-Appeal Outcome). We do so for a variety of reasons. First, appellate review standards create an affirmance bias,
    We control for whether the appeal was decided during a period of recession (Recession) as identified by the Business Cycle Dating Committee of the National Bureau of Economic Research. During the fifteen-year period of our study (1997–2011), there were two recessions: March 2001 to November 2001 and December 2007 to June 2009.
    We also control for whether the appeal was decided subsequent to October 17, 2005—the effective date of the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005 (BAPCPA). Bankruptcy Abuse Prevention and Consumer Protection Act of 2005, Pub. L. No. 109-8, § 1501(a), 119 Stat. 23, 216. And we control for some individual characteristics of the judges, specifically gender (Male Judge) and age (Age). Our prior research has demonstrated that the macroeconomy, the philosophical orientation of bankruptcy law, and certain judge characteristics are determinants of circuit court judicial votes in debt-dischargeability determinations.
  87. [87]. Because the JCS data do not include median JCS scores among all active courts-of-appeals judges after 2008, we are unable to include in our simple logistic regression model the observations in the Agent Dataset corresponding to votes cast during 2009, 2010, and 2011 (i.e., 37 of the 305 observations). See Epstein, supra note 81.

  88. [88]. See Harry Eckstein, Case Study and Theory in Political Science, in 7 Handbook of Political Science 79, 96–123 (Fred I. Greenstein & Nelson W. Polsby eds., 1975).

  89. [89]. See Rachlinski et al., supra note 71, at 1258–59; cf. Pardo, supra note 28, at 637–38 (“In such a case [where the selection process becomes politicized], judicial candidates are likely to be drawn from a group of individuals whose values and thinking reflect those of the dominant political group. Such candidates, if selected, may feel compelled to exercise their judicial function in a manner that comports with the ideology of their political patrons. This will produce an ideological judiciary . . . .” (footnotes omitted)). For a discussion of the incentives of circuit court judges to appoint bankruptcy judges whose ideology aligns with that of the circuit court, see id. at 648–52.

  90. [90]. See Nash & Pardo, supra note 26, at 1766.

  91. [91]. Cf. id. at 1765–66 (“[A]lthough bankruptcy judges are not granted life tenure, their terms last fourteen years. Moreover, their appointments may be renewed, and indeed in most cases are renewed. While judicial independence may be fostered by life tenure, the renewable, fourteen-year term of bankruptcy judges effectively allows them to serve as long as many of their Article III counterparts. Even if the absence of life tenure gives Congress leeway to reduce the term of bankruptcy judges—an option that it has never exercised since it created the bankruptcy courts—the fourteen-year, renewable term still grants a fair amount of judicial independence to bankruptcy judges.” (footnotes omitted)).

  92. [92]. To be sure, it may be that bankruptcy court judges cast votes ideologically in order to secure appointment to BAP service, but that is a story about circuit court judges “prejudging” bankruptcy court judges before appointing them to higher service. See Jonathan Remy Nash, Prejudging Judges, 106 Colum. L. Rev. 2168, 2171–74 (2006) (discussing how prejudging a judicial candidate can be a kind of substitute for enforcing accountability after a judge has ascended to the bench). It is not a story about judges, having already been appointed to a position, behaving in a certain way so as to assure themselves of keeping their posts.

  93. [93]. The sanction of non-reappointment is also available, for example, in the judicial hierarchy involving federal district court judges and federal magistrate judges. See 28 U.S.C.
    § 631(a)–(b) (2006) (providing for the appointment and reappointment of federal magistrate judges by federal district court judges).

  94. [94]. E.g., Songer et al., supra note 1 (analogizing lower courts’ relationship to higher courts to agents acting at the behest of a principal).

  95. [95]. See Matt Spitzer & Eric Talley, Judicial Auditing, 29 J. Legal Stud. 649, 674 (2000) (observing that it can be hard to determine exactly which goal dominates higher courts’ monitoring of lower courts).

  96. [96]. E.g., Steven Shavell, The Appeals Process as a Means of Error Correction, 24 J. Legal Stud. 379 (1995); cf. Larry W. Yackle, Reclaiming the Federal Courts 98 (1994) (“[T]he Supreme Court no longer has the capacity to sit as a court of error in routine cases.”); Charles M. Cameron & Lewis A. Kornhauser, Appeals Mechanisms, Litigant Selection, and the Structure of Judicial Hierarchies, in Institutional Games and the U.S. Supreme Court 173, 190–91 (James R. Rogers et al. eds., 2006) (discussing the “team model” of hierarchical adjudication). But see Carolyn Shapiro, The Limits of the Olympian Court: Common Law Judging Versus Error Correction in the Supreme Court, 63 Wash. & Lee L. Rev. 271, 275–86 (2006) (“[The Supreme Court] attempts to position itself as a source of structure, guidance, and uniformity, not as a traditional court of appeals that reviews the correctness of lower court opinions.”).

  97. [97]. See Nash & Pardo, supra note 46, at 922, 925–28.

  98. [98]. See Resnik, Uncle Sam, supra note 20, at 671 (arguing that, “given the dependence of life-tenured judges on magistrate and bankruptcy judges and the blur from the public perspective of exactly which judge holds what position, Article III judges have incentives to pick stellar candidates,” and observing that “[t]he process appears to have resulted in bankruptcy and magistrate benches of high quality”). But see Pardo, supra note 28, at 648–52 (offering arguments as to why circuit court judges may look to appoint ideologically similar bankruptcy court judges). See generally Chad M. Oldfather, Error Correction, 85 Ind. L.J. 49 (2010)(noting the widespread understanding that courts of appeals focus on error correction, while also noting the ambiguity inherent in the term).

  99. [99]. Cf. SNTL Corp. v. Centre Ins. Co. (In re SNTL Corp.), 571 F.3d 826, 829 (9th Cir. 2009) (per curiam) (“We adopt the BAP opinion as our own and attach it as an appendix to this opinion.” (citation omitted)); Khaligh v. Hadaegh (In re Khaligh), 506 F.3d 956, 957 (9th Cir. 2007) (per curiam) (“We . . . affirm for the reasons cogently set forth in the majority opinion of the Bankruptcy Appellate Panel.”); Ohio Univ. v. Hawkins (In re Hawkins), 469 F.3d 1316, 1317 (9th Cir. 2006) (per curiam) (“We adopt the opinion of the BAP, which is reported at 317 B.R. 104, and affirm its judgment.”).


Professor of Law, Emory University School of Law.


Robert T. Thompson Professor of Law, Emory University School of Law. 

For valuable comments, we are grateful to Tom Clark, Micheal Giles, Margaret Lemos, Cherie Metcalf, Joseph Ura, John Walson, Kathryn Watts, and participants in a graduate seminar on research design in the Emory Political Science Department. This Essay also benefited greatly from the commentary of participants at the 2012 Annual Conference of the Southern Economic Association, the 2012 Annual Meeting of the Canadian Law and Economics Association, the 2012 Annual Meeting of the Midwestern Law and Economics Association, the 2012 Annual Meeting of the Midwest Political Science Association, the 2013 Annual Meeting of the American Law and Economics Association, and faculty colloquia at Emory University School of Law and the University of Florida Levin College of Law.