BROADENING DIVERSITY ON THE BENCH: VOTING BEHAVIOR AND PANEL EFFECTS ON THE UNITED STATES COURTS OF APPEALS Diane Hu* This Note seeks to determine the extent to which personal characteristics of judges—namely gender, race, and prior prosecutorial experience—affect individual judicial votes and panel decisions on the United States Courts of Appeals. Although these characteristics do not have a significant effect on the way an individual judge votes, this Note finds that the presence of one of these characteristics on a three-judge panel can influence how the other two judges vote, affecting the overall outcome. The presence of at least one female or black judge on a panel increases the likelihood of a more liberal decision across all cases. However, this effect disappears in cases that are specifically related to gender or race issues, such as employment discrimination cases. The presence of a prior prosecutor on a panel in criminal cases decreases the likelihood of a more liberal ruling (in favor of the defendant). These results shed light on the dynamics of panel decision- making, and allow us to critically examine the federal judicial appointment process and the pursuit of judicial diversity. * J.D. Candidate 2018, Columbia Law School; A.B. 2015, Princeton University. The author would like to thank Professor Olatunde Johnson for her feedback during the writing process, and Professor Douglas Massey for his advice and comments on previous versions of this Note. The author would also like to thank the staff of the Columbia Journal of Race and Law for their editing assistance and contributions. Finally, many thanks to Professor Donald Songer for offering guidance on this study, and creating and providing public access to his U.S. Courts of Appeals database. 342 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 I. Introduction ........................................................ 342 II. How Diversity Affects Voting Outcomes ....... 344 A. Diversity on the United States Courts of Appeals .................................................................. 344 B. Panel Effects on the United States Courts of Appeals .................................................................. 346 C. Empirical Studies of Voting Behavior ....... 349 III. Methodology .................................................... 352 A. Data ............................................................... 352 B. Research Design ........................................... 356 C. Hypotheses ................................................... 363 IV. Findings .......................................................... 365 A. Results .......................................................... 365 1. Individual Judicial Vote .......................... 365 2. Panel Effects ............................................ 371 3. Predicted Probabilities ............................ 378 B. Discussion of Panel Effects Results............ 383 C. Implications .................................................. 388 V. Conclusion ....................................................... 396 I. INTRODUCTION It is common knowledge that members of the executive and legislative branches of the United States government vote and craft policy according to their political beliefs. What is less apparent to the public but has been established through numerous studies is that judges’ voting behavior is also strongly correlated with their political No. 3:341] BROADENING DIVERSITY ON THE BENCH 343 ideology.1 Recent studies have focused on whether judges’ other personal characteristics besides political affiliation, such as race and gender, have an effect on their voting behavior. Results have been mixed and incomprehensive, however. This Note will examine the relationships between judicial voting behavior and the attributes of race, gender, and prior prosecutorial experience on the United States Courts of Appeals. Amidst calls for greater diversity in the judicial branch to better reflect our nation’s people and their backgrounds, this Note supports the notion that a diverse bench can affect judicial outcomes and have tangible benefits on society. This Note tests quantitatively the hypotheses that race, gender, and prior prosecutorial experience affect judges’ individual votes as well as the two other judges’ votes on three-judge panels.2 Part I discusses the numerous benefits of judicial diversity and panel effects—that is, how one judge’s vote on a three-judge panel can affect other judges’ votes—on the United States Courts of Appeals. It also identifies past empirical studies and how their methodologies have fallen short of painting a comprehensive picture of the effects of judges’ attributes. Part II outlines this Note’s methodology of measuring both how personal characteristics affect judges’ individual votes and the votes of the other judges on the panel. Part III summarizes the results and concludes that although race, gender, and prior prosecutorial experience may not affect judges’ individual votes, they do affect the other judges’ votes and thus the overall outcome. More specifically, the presence of a female or black judge increases the likelihood of what is coded as a more “liberal” 1 FRANK B. CROSS, DECISION MAKING IN THE U.S. COURTS OF APPEALS 7 (2007). See, e.g., Sheldon Goldman, Voting Behavior on the United States Courts of Appeals, 1961–1964, 60 THE AM. POL. SCI. REV. 374, 380 (1966); Sheldon Goldman, Voting Behavior on the United States Courts of Appeals Revisited, 69 THE AM. POL. SCI. REV. 491, 503–04 (1975). 2 Although qualitative studies and interviews can tell us a great deal about how judges’ personal experiences and backgrounds have shaped the way they perceive the law, we can make greater general observations from a more comprehensive empirical study. “[A] broad examination of many cases can reveal a systematic pattern of decisions, and statistical analyses add rigor to claims of ideological bias.” Cross, supra note 1, at 14. 344 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 decision across all cases but not in gender- or race-specific cases, respectively. Additionally, the presence of a prior prosecutor decreases the likelihood of a “liberal” decision (that is, one in favor of the defendant) in criminal cases. These significant results help legitimize calls for diversity on the bench and shed more light on the possible dynamics and thought processes on diverse three-judge panels.3 II. HOW DIVERSITY AFFECTS VOTING OUTCOMES A. Diversity on the United States Courts of Appeals The concept of judicial diversity has expanded from political ideology and age to race, gender, and prior professional experiences—as well as numerous other attributes— since the increased diversification of the courts in the 1970s during the Carter Administration.4 In presidential administrations’ efforts to diversify the courts, women have made the greatest inroads compared to other political minorities.5 The movement to increase the proportion of female and minority judges on both state and federal courts has simultaneously brought about efforts to promote merit systems for judicial appointments instead of political elections.6 Calls for diversity in prior career experience have been less prominent, but are just as important for shaping a bench with a variety of perspectives. Even though some older studies have not found that more female or minority judges yield significant differences in judicial outcomes, the “inclusion of these groups is vital in maintaining and even increasing the legitimacy of the 3 Cross, supra note 1, at 70–71. 4 Jason L. Morin, The Voting Behavior of Minority Judges in the U.S. Courts of Appeals: Does the Race of the Claimant Matter?, 42 AM. POL. RES. 34, 35 (2013). 5 Mark S. Hurwitz & Drew Noble Lanier, Women and Minorities on State and Federal Appellate Benches, 1985 and 1999, 85 JUDICATURE 84, 92 (2001). 6 Id. at 84. No. 3:341] BROADENING DIVERSITY ON THE BENCH 345 nation’s judicial tribunals.”7 This symbolic significance of judicial diversity is part of the notion of “descriptive representation,” or the idea that elected officials should represent not only their constituencies but also their descriptive attributes, such as race or gender.8 A diverse courtroom is arguably vital for the symbolic representation as well, for the “intangible psychological benefits”9 induce more faith and confidence in the courts, thereby giving them more legitimacy. These intangible benefits are accrued “in the aggregate, developed through judges’ interactions with one another and the public over time.”10 Some empirical work has shown that “raising citizens’ level of legitimacy towards legal authorities . . . makes people more likely to obey the law,” so a diverse judiciary could potentially lead to a more law-abiding society.11 Even if more diverse benches do not decide cases differently, the public might feel a greater sense of justice and fairness by seeing their own backgrounds reflected in those who hold power in the courtroom. Others suggest that diversifying courts could add new perspectives to important national discussions and perhaps provide remedial justice for past and continued inequalities.12 Thus, by “looking like America” and creating the perception of fairness, diverse courts can influence relevant policy outcomes, effect positive change, and bring about actual fairness.13 In contrast, 7 Id. at 85. 8 Nancy Scherer, Diversifying the Federal Bench: Is Universal Legitimacy for the U.S. Justice System Possible?, 105 NW. U. L. REV. 587, 597 (2011). 9 Morin, supra note 4, at 36–37. 10 Joy Milligan, Pluralism in America: Why Judicial Diversity Improves Legal Decisions About Political Morality, 81 N.Y.U. L. REV. 1206, 1209–10 (2006). 11 Scherer, supra note 8, at 632. 12 Morin, supra note 4, at 36–37. 13 Josh Hsu, Asian American Judges: Identity, Their Narratives, and Diversity on the Bench, 11 ASIAN PAC. AM. L.J. 92, 115 (2006). In this qualitative study, Hsu examined the language in judges’ opinions and rationales, and found that background and personal experiences inform their interpretation of the law, especially in immigration deportation cases. Id. at 107–11. 346 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 “substantive representation” focuses more on the “decisional behavior of governmental officials and is therefore more policy oriented.”14 This Note’s empirical work ultimately examines the effects of substantive representation, as it looks at potential differences in case outcomes. B. Panel Effects on the United States Courts of Appeals This Note focuses exclusively on the U.S. Courts of Appeals, also known as the federal circuit courts.15 The lack of data and information on judicial decision-making on the appeals courts (in contrast to the Supreme Court) used to be a limiting factor.16 Though often overlooked, the circuit courts play an extremely important role in interpreting the law and influencing public policy, especially given the large number of cases they decide each year relative to the Thus, the public must advocate for diversity on the bench despite some studies’ inconclusive statistical results. Id. at 114. The key flaw with past empirical studies is that most only test to see if a single characteristic has an influential effect on judges’ decisions. Id. at 101. If studies can determine that judicial background is an important factor in case outcomes, the case for diversity on the bench would be even more convincing. Cross, supra note 1, at 75. 14 Barbara L. Graham, Toward an Understanding of Judicial Diversity in American Courts, 10 MICH. J. RACE & L. 153, 159 (2004). 15 There are 94 federal judicial districts, with at least one district in each state as well as one in the District of Columbia and one in Puerto Rico. These districts are organized into one of 12 regional circuits, and with the addition of the Federal Circuit—which has nationwide jurisdiction—they make up a total of 13 United States courts of appeals. FEDERAL JUDICIAL CENTER, THE U.S. COURTS OF APPEALS AND THE FEDERAL JUDICIARY, http://www.fjc.gov/history/home.nsf/page/courts_of_appeals.html [https://perma.cc/MH2H-WRBA](last visited May 21, 2018). Each court of appeals hears and decides appeals from the district court situated within its circuit, and has at least three assigned judgeships. Each circuit has between six and twenty-eight judges. Id. 16 Tracey E. George & Reginald S. Sheehan, Circuit Breaker: Deciphering Courts of Appeals Decisions Using the U.S. Courts of Appeals Data Base, 83 JUDICATURE 240, 245 (2000). No. 3:341] BROADENING DIVERSITY ON THE BENCH 347 Supreme Court.17 Furthermore, the Supreme Court rarely reviews the outcomes of lower federal courts, so these decisions are final. Thus, the appeals courts play a large role in settling disputes and determining the “likely direction of the law.”18 Since these circuit courts are often the “last resort” for a substantial number of major legal issues,19 it is vital that we continue to study the possible determinants of the courts’ decisions. When a party appeals a lower district court decision, a panel of three judges either affirms or reverses the decision.20 Some cases are decided en banc, meaning the case is heard before all of the judges on the bench. However, in order to examine panel effects and any existing influence that one judge may have over others, this Note will focus solely on cases decided by three-judge panels. Because appeals court judges are not only swayed by their own legal views but also by those of their colleagues on the panel, the composition of a panel has great potential to 17 Sue Davis, Susan Haire & Donald R. Songer, Voting Behavior and Gender on the U.S. Courts of Appeals, 77 JUDICATURE 129, 130 (1993). The U.S. Courts of Appeals were established by the Judiciary Act of 1891, which is also known as the Evarts Act. They were the first federal courts designed for the sole purpose of hearing cases on appeal from trial courts. This significantly lightened the Supreme Court’s caseload and eliminated the need for U.S. circuit courts, the original intermediate level courts that had both trial court jurisdiction and appellate jurisdiction over the district courts. From then on, justices no longer ruled on appeals of cases that they had seen earlier in trial courts. FEDERAL JUDICIAL CENTER, LANDMARK LEGISLATION: U.S. CIRCUIT COURT OF APPEALS, https://www.fjc.gov/history/legislation/landmark-judicial-legislation-text- document-10 (last visited Mar. 27, 2018). However, it was not until the Judiciary Act of 1925 that the federal circuit courts’ power expanded, for this act gave the Supreme Court more discretionary control over its docket, which in turn increased the appellate docket. George & Sheehan, supra note 16, at 240–241. The larger docket gave the federal circuit courts more responsibility and lawmaking ability in deciding every single case properly appealed to them. Id. 18 CASS R. SUNSTEIN, DAVID SCHKADE, LISA M. ELLMAN & ANDRES SAWICKI, ARE JUDGES POLITICAL? AN EMPIRICAL ANALYSIS OF THE FEDERAL JUDICIARY 3 (2006). 19 Frank B. Cross, Comparative Judicial Databases, 83 JUDICATURE 248, 248 (2000). 20 Cross, supra note 1, at 1. 348 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 change the case outcome. Each circuit of the U.S. Courts of Appeals has an official system of randomly assigning judges to case panels. This method of neutral assignment should prevent judges from being able to select which cases they feel most strongly about, which supposedly “increases the legitimacy of judicial decision making.”21 However, this may also produce unbalanced panels that are more likely to vote a particular way, especially if the issue is ideologically contested. As a result, some final panel decisions might reflect partisan interests,22 which is problematic on a branch whose decisions should not be determined by politics. One study found that Republican judges were more likely to vote more liberally when sitting with two Democratic judges than when sitting with at least one other Republican.23 This Note looks for similar effects with other characteristics like the presence of females and racial minorities. According to some scholars, the traditional model of judicial decision-making abides by the “median voter theorem,”24 in which each judge votes according to his own preference, and the judge with the median preference prevails. However, with panel effects, it is likely that judges can influence others, be influenced by others, and change their minds, as “differing ideological and background characteristics of a single panel member can affect the group’s decision.”25 Research has shown that circuit court judges are influenced by others, and that this “collegiality effect” is at least as strong as a single judge’s own choices.26 As a result, some use the notion of panel effects to argue that circuit courts should have a minimum number of women and 21 Jonathan P. Kastellec, Panel Composition and Voting on the U.S. Courts of Appeals Over Time, 64 POL. RES. Q. 377, 379 (2011). 22 Id. 23 Id. This effect is called ideological dampening. Sunstein, supra note 18, at 9. Conversely, when a judge votes in extreme ways when sitting on a panel with other judges of the same party, it is called ideological amplification or group polarization. Id. 24 Cross, supra note 1, at 148. 25 Id. at 177. 26 Id. at 9. No. 3:341] BROADENING DIVERSITY ON THE BENCH 349 racial minority judges for the effect they may have on the decision-making process.27 C. Empirical Studies of Voting Behavior Some of the first studies that examined U.S. Courts of Appeals judges’ characteristics and voting behavior were done in the 1960s. In his first study, Goldman only found evidence of a relationship between party affiliation and voting patterns on the appeals courts from 1961 through 1964.28 Other demographic variables tested included religious affiliation, place of birth, undergraduate and law school institutions attended, and past federal experience,29 none of which were found to have an effect on the way judges voted. Goldman’s first study did not test for race or gender. Goldman’s second study again found that party affiliation had the strongest direct link to voting behavior.30 Since then, studies have consistently found that party affiliation or political ideology has a statistically significant effect on judicial decisions.31 Several studies on the role of race and gender on the U.S. Courts of Appeals have since been undertaken, and results have been inconsistent. For the most part, scholars find that background factors like race, gender, and past career experience have little impact on judicial outcomes.32 Some have found differences in race while others have not.33 27 Id. at 168. 28 Sheldon Goldman, Voting Behavior on the United States Courts of Appeals, 1961–1964, 60 THE AM. POL. SCI. REV. 374, 380 (1966). 29 Id. at 382. 30 Sheldon Goldman, Voting Behavior on the United States Courts of Appeals Revisited, 69 THE AM. POL. SCI. REV. 491, 503–04 (1975). Goldman coded judicial decisions on a scale of 0 to 2 of liberalness; the higher the number, the more liberal the decision. Id. at 492–93. “This is by now a convention of judicial research on lower courts.” Id. This Note uses a similar scale from 1 to 3. 31 Cross, supra note 1, at 7. These results support the notion of legal realism, a movement among legal academics about the political nature of judicial decision-making. Id. at 11. 32 Id. at 8. 33 Id. at 73. 350 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 There have yet to be any definitive conclusions, but it is undeniable that these factors play a role in forming people’s perspectives on critical issues. The growing number of women and racial and ethnic minorities on the federal courts has allowed other scholars to conduct similar studies that focus on either race or gender, or both. A recent 2013 paper examined employment discrimination claims between 2001 and 2009 on the U.S. Courts of Appeals and found that minority judge voting behavior varied by ethnicity; namely, while African American judges were more likely to vote in favor of black claimants, Latino judges were less likely to vote in favor of claimants in general.34 However, because these studies tested across all cases and focused primarily on the litigants’ characteristics rather than on the judges’ characteristics, our understanding of race and voting behavior is still not fully established. Conclusions on gender on the U.S. Courts of Appeals are, again, quite varied and conflicting; some say female judges are more liberal than their male counterparts, while other empirical studies claim that there are no significant gender differences.35 One study examined gender and race but incorporated and advocated for a different approach that focused on the interaction of different individual characteristics in criminal cases, a method that this Note 34 Morin, supra note 4, at 34–35. 35 See, e.g., Davis et al., supra note 17, at 131–32 (analyzing voting behavior in three areas of employment discrimination and finding that the voting of female judges differed significantly from that of male judges in employment discrimination and criminal search and seizure cases, but not in obscenity cases). The same three scholars expanded on this in a second study, which reconfirmed that employment discrimination cases showed significant discrepancies. Donald R. Songer, Sue Davis & Susan Haire, A Reappraisal of Diversification in the Federal Courts: Gender Effects in the Courts of Appeals, 56 J. POL. 425, 425 (1994). The study also acknowledged the problem with examining only “three narrowly defined issue areas,” and that further research into other area topics is needed, especially because women and minority judges are still relatively new to the legal system. Id. at 437. This Note will endeavor to expand the range of topics in the author’s study by incorporating over 50 types of cases in hopes that this will yield different results. No. 3:341] BROADENING DIVERSITY ON THE BENCH 351 incorporates.36 Mixed findings in the past have soundly indicated that judges are not influenced by just one attribute but rather “by a host of personal traits that collectively impact their behaviors.”37 The joint effects of race and gender were shown through the finding that minority female judges were more likely to vote in ways that supported criminal defendants even after controlling for other factors like region, ideology, and age.38 Overall, past studies show that we must alter our methods in understanding judicial voting behavior. This includes examining overlapping characteristics instead of individual ones39 and examining panel effects rather than individual judicial votes, which is what this Note does.40 Additionally, most studies of panel composition that have been undertaken have focused on a narrow time period, focusing heavily on the past couple decades,41 and tend to focus on only one or two legal issues, “mak[ing] generalizing 36 Todd Collins & Laura Moyer, Gender, Race, and Intersectionality on the Federal Appellate Bench, 61 POL. RES. Q. 219, 221 (2007). 37 Id. at 219. 38 Id. at 222, 224. 39 Id. at 225. 40 Kastellec suggests that because of the low percentage of female and minority federal judges, it would be best to stop studying whether minority judges vote differently from nonminority judges, but rather whether the presence of females or racial minorities on appeals courts influences outcomes and other judges. This new approach is more suitable to the structure of the appeals courts. Kastellec, supra note 21. He discusses the phenomenon in which panel composition affects individual judicial decisions and as a result, the final decisions of three-judge panels. Id. at 377. Most studies described above treat each judge as an individual acting of his own volition. Cross, supra note 1, at 148. This would be accurate if we were looking at lone trial judges in the federal district courts, but it disregards the institutional structure of the U.S. Court of Appeals, where decisions are made by majority vote on three-judge panels. Focusing on individual judicial decisions thus yields a problematic research design that may explain the inconclusive results from past studies. Sean Farhang & Gregory Wawro, Institutional Dynamics on the U.S. Courts of Appeals: Minority Representation Under Panel Decision Making, 20 J.L. ECON. & ORG. 299, 327 (2004). 41 Kastellec, supra note 21, at 380. 352 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 beyond specific issues difficult.”42 This Note uses data that includes information dating back to the 1920s. In his book, Cross stated that one might expect to see a greater gender panel effect if the data were further broken down into particular case types, a method that this Note adopts.43 III. METHODOLOGY A. Data This two-part study first examines how characteristics affect an individual judge’s voting behavior and then how that individual judge’s characteristics affect the voting behavior of the other judges on the panel. Within each test, this Note will see if the results change for specific related types of cases, and how the characteristics affect outcomes across all cases.44 Because this study looks at 42 Id. See, e.g., Jonathan P. Kastellec, Racial Diversity and Judicial Influence on Appellate Courts, 57 AM. J. POL. SCI. 167, 167 (2013) (finding that the presence of one black judge on a panel of three greatly increased the likelihood that the final ruling favored affirmative action programs); Jennifer L. Peresie, Female Judges Matter: Gender and Collegial Decisionmaking in the Federal Appellate Courts, 114 YALE L.J. 1759, 1761 (2005) (examining the presence of female appellate judges on a panel across just three years and finding that plaintiffs were twice as likely to be successful in sexual harassment or sex discrimination cases when a female judge served on the panel). Thus, both blacks and females are able to influence their white or male colleagues in particular types of cases. 43 Cross, supra note 1, at 170. 44 This Note uses three datasets in this study. The first is the original comprehensive multi-user appeals court database created by Dr. Donald Songer, with a sample of cases from each circuit for each year from 1925 to 1996. Drs. Ashlyn Kuersten and Susan Haire (Kuersten and Haire, 2007) then created an updated dataset, coding the cases from 1997 to 2002 in a similar fashion. This Note merged both of these datasets with a third dataset compiled by Drs. Gary Zuk, Deborah Barrow, and Gerard Gryski (Zuk, Barrow, and Gryski, 2009) with information on the political, career, and economic attributes of U.S. Courts of Appeals judges from 1801 to 2000 (and limited information on judges through 2004), making sure that the identifying numerical codes for the judges matched across all datasets. The variables in the Songer Database on appeals court cases are divided into four categories: basic case characteristics, participant No. 3:341] BROADENING DIVERSITY ON THE BENCH 353 panel effects, only cases whose method of decision was by a three-judge panel (as opposed to en banc decisions made by all of the judges on the circuit) were included. There are 220 specific categories of cases. For each case, the directionality of the court’s decision has been determined on a liberalness scale of 1 to 3. The higher the number, the more liberal the decision. For example, decisions in favor of criminal defendants, racial minorities in civil rights claims, labor unions, injured workers, and parties seeking protection from the First Amendment are coded as liberal (“3”). This method of coding “liberalness” is used in public law literature and is similar to how decisions were recorded in the Dr. Harold Spaeth’s U.S. Supreme Court database.45 It is also “now a convention of judicial research on lower courts.”46 Because this coding only captures the final vote, however, it is difficult to capture all the details and nuances from the discussions, opinion, and particular facts of each case.47 characteristics, issues, and judges’ characteristics and votes. Basic case characteristics consist of general case information, such as docket number, decision date, length of opinion, procedural history, circuit, state, district, and method of decision. The participant variables give detailed information on the litigants of the case. First, the number of appellants and appellees are given, and then the first two appellants and appellees are each given a five-digit code based on their nature (e.g. business, nonprofit organization, federal government, state government, or natural person). Each general category is broken down into more specific categories. The database also matches the appellant and appellee to the plaintiff and defendant. Issue variables are first divided into eight general categories: criminal, civil rights, First Amendment, due process, privacy, labor relations, economic activity and regulation, and miscellaneous. Finally, the judges and votes variables denote the identifying codes for each judge and his or her vote on each case for up to 15 judges. Of course, for this study, we will only look at cases with three judges on the panel. 45 Donald R. Songer, United States Courts of Appeals Database Phase 1, INTER-UNIVERSITY CONSORTIUM FOR POLITICAL AND SOCIAL RESEARCH, http://artsandsciences.sc.edu/poli/juri/appct.htm. 46 Goldman, supra note 30, at 492–93. 47 There are, of course, issues with “translating something so amorphous as ideology into a numerical measure for quantitative analysis.” Cross, supra note 1, at 20. However, this most common method 354 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Prior prosecutorial experience was defined as any past experience as an Assistant U.S. Attorney, U.S. Attorney, special prosecutor, state prosecutor, District Attorney, County Attorney, City Attorney, deputy or assistant district attorney, deputy or assistant county attorney, or deputy or assistant city attorney.48 of coding is still more detailed, manageable, and ideal than other possible options. A binary variable that only accounts for liberal or not liberal might simplify tests, but it also would have ignored even more nuance in case decisions. Switching to a scale of, say, 1 to 5 in order to better account for this nuance might then be deemed too arbitrary and would be up for debate amongst legal scholars when analyzing case outcomes. Therefore, the tried-and-true scale of 1 to 3 might be the best way to carry out a quantitative study of legal cases. 48 The attributes dataset contains information on the judges’ race, gender, religion, political party affiliation, appointing president, education, net worth, age, date of appointment and departure, American Bar Association rating, and prior federal experience. To incorporate this information with the Songer Database and match the different judge IDs, the author first created a new “judge vote” variable and three copies of the Songer data. After making sure that each copy of the data included the judge vote for the first judge, second judge, and third judge, respectively, the author combined all three copies of the data together. In order to make the process of testing panel characteristics manageable, this Note recoded the three main attributes of interest as binary variables. Gender was already a binary variable (though the author did change “1” for male and “2” for female into “0” for male and “1” for female). There were five options for race in the attributes dataset, but it made more sense to include only two binary variables due to the extremely low number of cases with Asian American or Native American judges. As said, one variable is for black and one is for white, leaving Hispanics as the reference category. After defining these binary variables for individual judge characteristics, the author then created variables for panel characteristics. Rather than coding these as binary as well—such as whether or not a female, minority, or past prosecutor was on the panel—I decided to designate panel characteristics as the total number of females, minorities, or past prosecutors on the panel per case, ranging from 0 to 2. This way, this Note will be able to see if there are any significant differences between having, say, one female on the panel versus two or three. The “2” category encompasses panels with either two or three judges that fit the characteristic, as the number of panels with three judges of each characteristic is very low, except for the more common characteristics like White. No. 3:341] BROADENING DIVERSITY ON THE BENCH 355 This Note grouped together particular case types as defined in the original database into three larger case categories for the study: gender, race, and prosecution cases. This created a larger pool of cases that past studies lacked. After cases that were decided en banc and not on a three- judge panel have been removed, there were a total of 40,523 observations. Any case type or issue that this Note deemed directly relevant to each independent variable was included. They span across several of the dataset’s general categories, from First Amendment to privacy issues. The gender cases are employment sex discrimination, pregnancy discrimination, other sex discrimination, abortion rights, contraception, and other privacy claims related to marital relations or sexual behavior. There are a total of 414 cases that fall into these categories and are thus designated as “gender cases.” Race cases encompassed voting rights (including race discrimination in voting), desegregation, employment race discrimination, other race discrimination, and the legality of expression through overt acts (such as speeches and picketing) specifically protesting race discrimination. There are 807 race cases total. Categorizing prosecution cases was the most straightforward of the three types. This Note includes all criminal cases (both federal and state offenses): murder, rape, arson, aggravated assault, robbery, burglary, auto theft, larceny, other violent crimes, narcotics, alcohol related crimes, tax fraud, firearm violations, morals charges, criminal violations of government regulations of business, and other white collar crimes. There are 14,884 criminal cases in the dataset. The focus on appeals courts not only makes up for a missing component of the literature, but also gives the study consistency, which can make the statistical analysis and results more reliable. For example, because all cases are federal, the procedures in court are largely the same across all circuits. Additionally, the method of voting on panels of three judges and the uniform appointment system are similar. (Although there are of course substantive differences and circuit splits on certain areas of the law, 356 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 these regional differences are controlled for in the model.) The large size of the database and the fact that there are three judges per case in the appellate courts give us a great deal of information and observations, which increases reliability. Finally, because the U.S. Courts of Appeals do not have the discretion to decide which cases to take on (unlike the Supreme Court), there is less of a concern for selection bias. B. Research Design This Note focuses on race and gender because they are significant traits that can shape one’s thoughts, preferences, and beliefs. This Note also explores whether a judge has had prior prosecutorial experience in criminal cases because this experience can significantly alter the way one views the criminal justice system.49 The first test focuses on judges’ individual votes, with the dependent variable as the directionality of each judge’s vote on a 1 to 3 scale of “liberalness,” with three ordered categories: “liberal,” “mixed,” or “conservative.” For the second test for panel effects, the dependent variable is also from 1 to 3, but only accounts for the directionality of the final panel decision (rather than the vote of each individual judge).50 The equations are as follows:51 1. Individual Judicial Vote = B0 + B1Female + B2White + B3Black + B4Pros + B5GenderCase + B6RaceCase + B7ProsCase + B8GenderCase*Female + 49 Goldman did not find that prior federal experience as a judge, public office candidate, or public prosecutor mattered. Goldman, supra note 30, at 499–500. This Note tests this trait for criminal cases only. 50 Because the dependent variable in this study is ordinal (on a scale of 1 to 3), the author ran ordinal logistic regressions in Stata and included a number of other independent and control variables as shown in the two models above. 51 Please refer to Table 1 for definitions of the abbreviated variables in the two models. No. 3:341] BROADENING DIVERSITY ON THE BENCH 357 B9RaceCase*White + B10RaceCase*Black + B11ProsCase*Pros + B12FemalePanel + B13WhitePanel + B14BlackPanel + B15HispanicPanel + B16ProsPanel + B17AppelFemale + B18AppelMinority + B19RespondFemale + B20RespondMinority + B21Party + B22PresParty + B23Age + B24GradDeg1 + B25GradDeg2 + B26Decade40 + B27Decade50 + B28Decade60 + B29Decade70 + B30Decade80 + B31Decade90 + B32Circuit0 + B33Circuit1 + B34Circuit2 + B35Circuit3 + B36Circuit4 + B37Circuit5 + B38Circuit6 + B39Circuit7 + B40Circuit8 + B41Circuit9 + B42Circuit10 2. Panel Decision = B0 + B1FemalePanel + B2WhitePanel + B3BlackPanel +B4HispanicPanel + B5ProsPanel + B6GenderCase + B7RaceCase + B8ProsCase + B9GenderCase*FemalePanel + B10RaceCase*WhitePanel + B11RaceCase*BlackPanel + B12RaceCase*HispanicPanel + B13ProsCas*ProsPanel + B14AppelFemale + B15AppelMinority + B16RespondFemale + B17RespondMinority + B18Party + B19PresParty + B20Age + B21GradDeg1 + B22GradDeg2 + B23Decade40 + B24Decade50 + B25Decade60 + B26Decade70 + B27Decade80 + B28Decade90 + B29Circuit0 + B30Circuit1 + B31Circuit2 + B32Circuit3 + B33Circuit4 + B34Circuit5 + B35Circuit6 + B36Circuit7 + B37Circuit8 + B38Circuit9 + B39Circuit10 358 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Table 1. Description of Dependent and Independent Variables Variable Name Definition and Coding JudgeVote Directionality of individual judge’s vote (scale of 1 to 3) PanelDecision Directionality of case outcome (scale of 1 to 3) Female 1 if female, 0 otherwise White 1 if white, 0 otherwise Black 1 if black, 0 otherwise Pros 1 if has prior prosecutorial experience, 0 otherwise GenderCase 1 if case is gender-related, 0 otherwise RaceCase 1 if case is race-related, 0 otherwise ProsCase 1 if case is a criminal case, 0 otherwise FemalePanel Number of females on panel (0, 1, or 2+) WhitePanel Number of whites on panel (0, 1, or 2+) BlackPanel Number of blacks on panel (0, 1, or 2+) HispanicPanel Number of Hispanics on panel (0, 1, or 2+) ProsPanel Number of past prosecutors on panel (0, 1, or 2+) AppelFemale Appellant’s gender (1 if female, 0 otherwise) AppelMinority Appellant’s race (1 if minority, 0 otherwise) RespondFemale Respondent’s gender (1 if female, 0 otherwise) RespondMinority Respondent’s race (1 if minority, 0 otherwise) Party Party affiliation of judge (1 if Democrat, 0 otherwise) PresParty Party of appointing president (1 if Democrat, 0 otherwise) Age Age of judge at the time of decision GradDeg1 First graduate degree (other than J.D.) attained No. 3:341] BROADENING DIVERSITY ON THE BENCH 359 GradDeg2 Second graduate degree (other than J.D.) attained Decade40 1 if case occurred in the 1940s, 0 otherwise Decade50 1 if case occurred in the 1950s, 0 otherwise Decade60 1 if case occurred in the 1960s, 0 otherwise Decade70 1 if case occurred in the 1970s, 0 otherwise Decade80 1 if case occurred in the 1980s, 0 otherwise Decade90 1 if case occurred in the 1990s, 0 otherwise Circuit0 1 if case occurred in the D.C. Circuit, 0 otherwise Circuit1 1 if case occurred in the First Circuit, 0 otherwise Circuit2 1 if case occurred in the Second Circuit, 0 otherwise Circuit3 1 if case occurred in the Third Circuit, 0 otherwise Circuit4 1 if case occurred in the Fourth Circuit, 0 otherwise Circuit5 1 if case occurred in the Fifth Circuit, 0 otherwise Circuit6 1 if case occurred in the Sixth Circuit, 0 otherwise Circuit7 1 if case occurred in the Seventh Circuit, 0 otherwise Circuit8 1 if case occurred in the Eighth Circuit, 0 otherwise Circuit9 1 if case occurred in the Ninth Circuit, 0 otherwise Circuit10 1 if case occurred in the Tenth Circuit, 0 otherwise 360 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Table 2. Summary Statistics of Variables Proportion of 0 (%) Proportion of 1 (%) Proportion of 2+ (%) Female 95.13 4.87 White 94.96 5.04 Black 96.58 3.42 Pros 64.71 35.29 GenderCase 99.08 0.92 RaceCase 98.21 1.79 ProsCase 66.99 33.01 FemalePanel 88.13 10.73 1.13 WhitePanel 0.16 4.0 95.85 BlackPanel 90.91 8.75 0.34 HispanicPanel 95.95 3.90 0.16 ProsPanel 28.55 47.75 23.70 AppelFemale 92.96 7.04 AppelMinority 95.10 4.90 RespondFemale 97.33 2.67 RespondMinority 99.01 0.99 Party 50.04 49.96 PresParty 50.83 49.17 GradDeg1 84.71 15.29 GradDeg2 98.85 1.15 Decade40 91.06 8.94 Decade50 91.36 8.64 Decade60 83.17 16.83 Decade70 82.52 17.48 Decade80 82.79 17.21 Decade90 78.18 21.82 Circuit0 90.50 9.50 Circuit1 90.69 9.31 Circuit2 90.47 9.53 Circuit3 91.38 8.62 Circuit4 91.37 8.63 Circuit5 90.67 9.33 Circuit6 90.97 9.03 Circuit7 90.62 9.38 Circuit8 91.35 8.65 No. 3:341] BROADENING DIVERSITY ON THE BENCH 361 Circuit9 91.41 8.59 Circuit10 91.70 8.30 As can be seen in Table 1 and the equations above, the key independent variables for gender, race, and prior prosecutorial experience are bolded. In this model, the author created two binary variables for race (black and white) in order, leaving Hispanics as the reference category. While the original dataset used for this Note also coded for Asian Americans and Native Americans, there was only one Asian American and there were no Native Americans, so those two options were not included in the model. For individual judicial vote, the key independent variables are Female, White, Black, and Pros. For panel decision, the bolded variables are the panel characteristics FemalePanel, WhitePanel, BlackPanel, HispanicPanel, and ProsPanel, each of which is the number of females, whites, blacks, Hispanics, or past prosecutors on a given panel (up to two), respectively. The next few variables in each model are the characteristics of the cases in question.52 If the appellant or appellee is a natural person (as opposed to a business or the government), the model will include that individual’s race and gender. Of course, gender, race, and prior career experience are not the only significant characteristics that influence judges’ preferences. This Note thus includes other attributes that are relevant to judges’ personal backgrounds and may influence the judges’ votes as control variables, namely party affiliation, party of appointing president, age, and any graduate degree education other than a Juris Doctor.53 The variables that follow in the models are dummy 52 The variables Hispanic, GradDeg3 (third graduate degree), Decade30 (the 1930s), and Circuit11 (Eleventh Circuit) are not included in the model or Table 1 to avoid collinearity. 53 Because the original datasets the author worked with accounted for many details, many variables of interest like the race and gender of the appellants and appellees are categorical over five groups. In order for these variables to work with this Notes ordinal logit, the author adjusted their coding to binary. This Note repeated this process for some of 362 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 variables for circuit and decade in order to evaluate any changes over time or across region and circuit.54 Finally, this Note uses interaction terms created by the author to capture interaction effects between the attributes and their corresponding case types. These interaction effects allow the author to examine individual or gender effects with cases deemed relevant to gender, race effects with cases deemed relevant to race, and so on.55 For past prosecutorial experience, this Note looks at all criminal the judges’ other characteristics as well, namely party affiliation, party of appointing president, age, and education. Age is the only one of this Note’s control variables that is not binary. The minimum age in the dataset is 25, the mean is 52, and the maximum is 71. 54 This Note determined the sampling weights according to instructions in the Songer Database codebook. Because the total number of cases varies per circuit per year—in some years 15 cases were selected per year while in others 30 cases were selected per year—the total sample of cases is not a random sample of all appeals court decisions. Thus, using a provided table of weights, this Note calculated a sampling weight for each circuit per year, which was then attached to each case. Each sampling weight was determined by dividing the number of cases per circuit in a given year by the total number of cases across all circuits in that year. 55 The gender interaction term is the product of two binary variables: whether it is a gender case, and whether the judge is female. This Note will then be able to evaluate how both males and females vote in gender cases versus non-gender cases. In the second test, the interaction term is the product of one binary and one continuous variable: whether it is a gender case, and the number of females (up to two). Thus, the first factor is binary, and the options for the second factor are 0, 1, and 2, with two encompassing panels of two or more females. This will provide information on how the presence of females (or lack thereof) affects gender cases as well as non-gender cases. The other interaction terms are calculated in the same way, but instead of the number of females, it is the number of blacks, whites, and past prosecutors (but again, up to two). The reason this Note models interaction effects this way is because if it were a simple count of the number of females up to three, there would not be enough variability due to the small number of cases with panels of three females. A simple run of the ordinal logistic regression and marginal effects illustrated that this greatly skewed the results. For example, there are only 12 out of tens of thousands of observations that have three blacks on one panel; this accounts for less than one percent of all cases. While this was not an issue for whites, this Note coded all three attribute panel variables as either 0, 1, or 2, with two representing cases with either two or three females, blacks, whites, or past prosecutors for the sake of consistency. No. 3:341] BROADENING DIVERSITY ON THE BENCH 363 cases and no civil cases. Different types of cases have different levels of importance when examining background characteristics; for example, a boundary dispute between two states would not be as relevant as, say, First Amendment or labor relations issues.56 The type of case may be even more important than the judge’s ideology to the direction of case outcomes.57 Most studies in the literature either examine too broad a range of cases (all of them) or too specific a range; this Note attempts to modify and correct this. C. Hypotheses This Note hypothesizes that each of the three attributes will influence the directionality of individual judges’ votes on cases that are relevant to each particular characteristic. This Note also hypothesizes that when there is at least one female or minority in the panel in a gender- or race-related case, respectively, there will be a stronger relationship between the panel characteristics and the directionality of the panel’s final decision. Both females and racial minorities bring an important perspective to the bench and think about relevant issues differently from other appeals court judges. Their presence on the bench may make other judges on the panel more aware of gender or race issues. They may also be able to persuade their colleagues to support a particular litigant. This Note thus hypothesize that females are more likely to vote liberally in gender-related cases, as are blacks and Hispanics in race-related cases. Even if the individual votes of females, blacks, and Hispanics may not change substantially, this Note hypothesizes that their presence on panels may influence their white male colleagues and increase the probability of an ultimately more liberal outcome. Finally, this Note predicts that judges with past prosecutorial experience will be more stringent on defendants and will thus vote more conservatively, since a liberal decision is coded as one that supports the position of 56 Cross, supra note 1, at 25. 57 Id. 364 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 the defendant in a criminal procedure case. This Note also hypothesizes that there will be panel effects for prosecutorial experience, for even if past professions are not as observable or prominent as features like race and gender, judges can still influence others during deliberations. Table 3. Hypotheses Individual Judicial Vote Gender Female judges will vote in a more liberal manner in gender cases. Race Black and Hispanic judges will vote in a more liberal manner in race cases. Prior Prosecutorial Experience Judges with prosecutorial experience will vote in a less liberal manner in criminal cases. Panel Decision Gender The presence of at least one female judge on the panel will result in a more liberal outcome in gender cases. Race The presence of at least one black or Hispanic judge on the panel will result in a more liberal outcome in race cases. Prior Prosecutorial Experience The presence of at least one judge with prosecutorial experience on the panel will result in a less liberal outcome in criminal cases. No. 3:341] BROADENING DIVERSITY ON THE BENCH 365 IV. FINDINGS A. Results 1. Individual Judicial Vote Table 4. Ordinal Logistic Regression Results for Individual Judicial Vote (1) (2) VARIABLES Logit coefficient Odds ratio GenderCase 0.105 1.110 (0.100) (0.111) Female -0.0396 0.961 (0.0816) (0.0784) GenderCase*Female -0.0713 0.931 (0.359) (0.335) RaceCase -0.0696 0.933 (0.688) (0.641) White 0.0510 1.052 (0.129) (0.136) RaceCase*White 0.101 1.107 (0.692) (0.766) Black 0.0651 1.067 (0.152) (0.162) RaceCase*Black -0.152 0.859 (0.833) (0.715) ProsCase -1.166*** 0.312** * (0.0328) (0.0102) Pros 0.0126 1.013 (0.0325) (0.0329) ProsCase*Pros -0.0185 0.982 (0.0547) (0.0537) FemalePanel 0.118** 1.125** (0.0496) (0.0558) BlackPanel 0.188*** 1.207** 366 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 * (0.0529) (0.0639) HispanicPanel -0.0245 0.976 (0.0763) (0.0744) WhitePanel 0.000805 1.001 (0.0672) (0.0673) ProsPanel 0.0261 1.026 (0.0198) (0.0203) AppelFemale -0.289*** 0.749** * (0.0472) (0.0354) AppelMinority -0.0320 0.969 (0.0664) (0.0644) RespondFemale 0.141** 1.151** (0.0689) (0.0793) RespondMinority 0.376*** 1.456** * (0.124) (0.180) Party 0.0359 1.037 (0.0504) (0.0522) PresParty 0.135*** 1.145** * (0.0509) (0.0583) Age 0.000193 1.000 (0.00178) (0.00178) GradDeg1 -0.0233 0.977 (0.0334) (0.0326) GradDeg2 -0.242 0.785 (0.226) (0.178) Decade40 0.0548 1.056 (0.0529) (0.0559) Decade50 -0.0284 0.972 (0.0520) (0.0505) Decade60 -0.0453 0.956 (0.0466) (0.0445) Decade70 0.117** 1.125** (0.0468) (0.0526) Decade80 -0.0128 0.987 No. 3:341] BROADENING DIVERSITY ON THE BENCH 367 (0.0480) (0.0474) Decade90 -0.249*** 0.780** * (0.0486) (0.0379) Circuit0 -0.0898 0.914 (0.110) (0.100) Circuit1 -0.128 0.880 (0.106) (0.0935) Circuit2 -2.97e-05 1.000 (0.107) (0.107) Circuit3 0.00750 1.008 (0.107) (0.108) Circuit4 -0.0328 0.968 (0.106) (0.103) Circuit5 -0.102 0.903 (0.107) (0.0965) Circuit6 -0.0176 0.983 (0.106) (0.104) Circuit7 -0.219** 0.804** (0.106) (0.0852) Circuit8 -0.163 0.850 (0.106) (0.0901) Circuit9 -0.0213 0.979 (0.107) (0.105) Circuit10 -0.00440 0.996 (0.107) (0.107) Cut Point 1 Constant -0.165 0.848 (0.214) (0.181) Cut Point 2 Constant 0.145 1.156 (0.214) (0.247) Observations 40,523 40,523 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 368 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 The first notable observation we can make from the results in Table 4 is that the three attributes of interest are not significantly correlated with individual judge voting behavior (the outcome variable in the first model).58 While this is not what we would expect, it is in line with past studies finding that gender, race, and other personal characteristics have little determinative effect on individual decisions.59 This also establishes a baseline from which we can make different conclusions about panel effects. The variables that do have some significance from the first model are female panel characteristics, black panel characteristics, the appellant’s gender, the appellee’s gender, the appellee’s race, and party of the judge’s appointing president. Because both the coefficients for FemalePanel and BlackPanel are positive and significant over all types of cases, this seems to suggest that the presence of females and blacks increases the likelihood that an individual judge’s vote will be on the liberal side regardless of case type. This supports the panel effects theory and hypothesis, and it is reasonable that ProsPanel is not significant here. Gender and race are two more visible and apparent characteristics that may subconsciously influence other judges’ thinking and decision-making, whereas past experience is more difficult to discern. Neither HispanicPanel nor WhitePanel is significant. If we look at the odds ratio for these two variables, we can see that with the addition of one female to a panel of judges, the odds of having a liberal decision versus a conservative or mixed decision are 1.125 times greater. With 58 Although it is difficult to discern much meaning from the raw coefficients given in the ordinal logistic regression results, we can still take an initial, cursory look at the sign of the coefficients and the statistical significance in column 1 of Table 4. Because the column 1 coefficients show the change in the log of the odds, we can then examine the odds ratio in column 2 that displays the proportional odds ratios for the ordered logit model. The odds ratios shown are obtained by exponentiating the coefficients in column 1 (ecoefficient), and give us a much clearer sense of how these variables affect voting behavior. 59 See Goldman, supra note 30. No. 3:341] BROADENING DIVERSITY ON THE BENCH 369 a one unit increase in the number of blacks on the panel, the predicted odds of observing a more liberal decision are 1.207 times greater. Thus, each additional female is associated with a 12.5% increase in the odds of jumping a level in the outcome variable (either from conservative “1” to mixed “2” or liberal “3”, or from conservative “1” or mixed “2” to liberal “3”) and each additional black is associated with about a 20.7% increase in increasing a level.60 Thus, although variables representing individual characteristics like Female and Black are not significant, FemalePanel and BlackPanel are. This bolsters the prediction that their presence may still influence other judges on the same panel even though a judge’s individual attributes may not have an effect on his or her own vote. These results from the first test support past scholars’ statement that “background variables . . . mask too wide a variety of conditioning experiences for us to expect them to be directly and clearly associated with voting behavior and thus to be able to account for a significant portion of the variation of the behavior.”61 However, these scholars also neglected to take panel influences into account, which we found were significant even in the first model. This first test serves as an interesting starting point for interpreting panel effects. It is also worth discussing other significant covariates in this model. The negative sign of the ProsCase coefficient shows that criminal cases decrease the odds of a more liberal decision. The odds ratio here is 0.312, so for an increase in the unit of this independent variable from 0 to 1, there is a 68.8% (1 – 0.312) decreased likelihood of a more liberal decision. This is an interesting observation, as it shows that 60 Notice that throughout this Note, the author asserts that the odds that the decision “is more liberal” rather than “is liberal,” because the odds ratio predicts the probability of jumping a level, either from conservative to mixed or liberal, or from conservative or mixed to liberal. It does not tell us the chances of the decision being a conservative, mixed, or liberal one. 61 Goldman, supra note 30, at 496. 370 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 judges are, as a whole, tougher on the defendant and more likely to vote conservatively in criminal cases.62 It is not too surprising that the party of the judge’s appointing president is significant in determining an individual judge’s vote. This suggests that federal appeals judges vote in ways that align with the political preferences of the president who appointed them. The way the two-party variables were coded make it so that a one unit increase from 0 to 1 is a change from Republican to Democrat. Thus, if the judge’s appointing president is a Democrat, the odds that the individual judge’s vote will be on the more liberal side is 14.5% greater.63 Finally, there are some significant results in the dummy variables this Note included to capture changes over time and region. The 1970s showed positive significance, which could be explained by the substantial increase in diversity of the courts during those years. During that decade, there was a 12.5% increase in the odds that an individual judge’s vote was more liberal. Individual voting behavior in the 1990s, however, had a 22% decreased likelihood of being more liberal. In addition, judges’ votes in the Seventh Circuit (with jurisdiction over the federal district courts in Illinois, Indiana, and Wisconsin) had a 19.6% decreased likelihood of being more liberal. 62 The litigants’ characteristics are important, though the interpretation of these results is more difficult and could be explored in further studies. If the appellee is a female or minority, both of these characteristics increase the likelihood of a more liberal decision by 15.1% and 45.6%, respectively. In contrast, only the appellant’s gender seems to matter, but here, a female appellant actually decreases the odds by around 25% that the outcome will be more liberal. This may tell us something about the way judges perceive those who choose to appeal cases versus those who are being brought back to court. 63 The positive sign of the coefficient (or the fact that the odds ratio is greater than one) makes sense for the most part, for even if liberal and conservative ideologies are not synonymous with the Democratic and Republican parties, they are aligned in many cases given the way the case outcomes are coded. No. 3:341] BROADENING DIVERSITY ON THE BENCH 371 2. Panel Effects Table 5. Ordinal Logistic Regression Results for Panel Decision (1) (2) VARIABLES Logit coefficient Odds ratio GenderCase 0.230** 1.259** (0.107) (0.134) FemalePanel(1) 0.132*** 1.141*** (0.0462) (0.0527) FemalePanel(2+) 0.480*** 1.616*** (0.153) (0.248) GenderCase*FemalePanel(1) -0.648*** 0.523*** (0.244) (0.128) GenderCase*FemalePanel(2+) 13.98*** 1.182e+06* ** (0.470) (555,670) RaceCase 29.61*** 7.219e+12* ** (0.698) (5.036e+12) BlackPanel(1) 0.242*** 1.273*** (0.0477) (0.0607) BlackPanel(2+) 0.442* 1.556* (0.241) (0.376) RaceCase*BlackPanel(1) -0.368 0.692 (0.324) (0.225) RaceCase*BlackPanel(2+) -15.30*** 2.26e- 07*** (0.653) (1.47e-07) HispanicPanel(1) -0.0698 0.933 (0.0677) (0.0631) HispanicPanel(2+) 0.138 1.148 (0.290) (0.333) RaceCase*HispanicPanel(1) 0.162 1.176 (0.408) (0.480) WhitePanel(1) 0.394 1.483 (0.363) (0.538) 372 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 WhitePanel(2+) 0.395 1.484 (0.367) (0.544) RaceCase*WhitePanel(1) -29.22*** 0*** (0.565) (0) RaceCase*WhitePanel(2+) -29.51*** 0*** (0.475) (0) ProsCase -1.113*** 0.328*** (0.0475) (0.0156) ProsPanel(1) 0.111*** 1.117*** (0.0324) (0.0362) ProsPanel(2+) 0.0923** 1.097** (0.0382) (0.0419) ProsCase*ProsPanel(1) -0.156** 0.856** (0.0606) (0.0519) ProsCase*ProsPanel(2+) -0.0926 0.912 (0.0723) (0.0659) AppelFemale -0.310*** 0.733*** (0.0475) (0.0348) AppelMinority -0.0515 0.950 (0.0671) (0.0637) RespondFemale 0.109 1.115 (0.0695) (0.0775) RespondMinority 0.394*** 1.483*** (0.124) (0.183) Party 0.00328 1.003 (0.0506) (0.0508) PresPart 0.0906* 1.095* (0.0510) (0.0558) Age 0.00204 1.002 (0.00176) (0.00176) GradDeg1 -0.0183 0.982 (0.0333) (0.0327) GradDeg2 -0.355 0.701 (0.230) (0.161) Decade40 0.0784 1.082 (0.0531) (0.0575) Decade50 0.00465 1.005 (0.0521) (0.0524) No. 3:341] BROADENING DIVERSITY ON THE BENCH 373 Decade60 -0.0266 0.974 (0.0467) (0.0455) Decade70 0.166*** 1.181*** (0.0469) (0.0554) Decade80 0.00558 1.006 (0.0482) (0.0485) Decade90 -0.266*** 0.767*** (0.0488) (0.0374) Circuit0 -0.0715 0.931 (0.111) (0.103) Circuit1 -0.119 0.887 (0.108) (0.0955) Circuit2 0.00594 1.006 (0.108) (0.108) Circuit3 0.0222 1.022 (0.108) (0.110) Circuit4 -0.0376 0.963 (0.107) (0.103) Circuit5 -0.0934 0.911 (0.108) (0.0986) Circuit6 -0.0205 0.980 (0.107) (0.105) Circuit7 -0.227** 0.797** (0.107) (0.0854) Circuit8 -0.180* 0.835* (0.107) (0.0895) Circuit9 -0.0265 0.974 (0.108) (0.105) Circuit10 -0.00463 0.995 (0.108) (0.108) Cut Point 1 Constant 0.232 1.261 (0.402) (0.507) Cut Point 2 Constant 0.546 1.726 (0.402) (0.694) 374 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Observations 40,523 40,523 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The results from the second model shown in Table 5 have much greater significance for the three primary panel variables of interest. This contrasts greatly with the individual characteristic variables in the first model. Analyzing the results of panel characteristics in the second model is a bit more complex because of the multiple interaction effects for each possible scenario. There is one interaction term for when there is one female on the panel of a gender case, another term for when there are two or more females on the panel of a gender case, and so on for the three races and two other attributes. We are also given the coefficients and odds ratios for when FemalePanel (as well as BlackPanel, WhitePanel, etc.) is equal to 1 and when it is equal to 2. Thus, not only can we evaluate how the presence of a single female, black, white, Hispanic, or past prosecutor affects other judges and case outcomes, but we can also evaluate how the presence of multiple females, blacks, whites, Hispanics, or past prosecutors affects other judges and case outcomes. As anticipated in the hypothesis, the results from the second test show that gender generally affected panel decisions. The presence of one female on a panel of judges increased the likelihood that the final panel ruling would be on the more liberal side by 14.1% across all case types. For two or more females, the odds increased to 61.6% across all case types. This is a significant finding, as it tells us what a large difference gender diversity can make on the courts. A surprising result was that in gender cases alone, the presence of one female decreased the odds that the outcome would be more liberal (by 47.7%), but with two or more females, it would return to an extraordinarily large increase in the likelihood of a more liberal decision. These unusual and unexpected results might be a result of the fact that out of the 414 cases deemed “gender cases,” only 11 of No. 3:341] BROADENING DIVERSITY ON THE BENCH 375 them had two or more females on the panel. Thus, when interpreting the effect of female judicial presence, it might be be wise to rely only on the FemalePanel(1) and FemalePanel(2+) variables or on the gender case interaction term with only one female, because they span across a much larger sample of cases. However, since these two types of variables have opposing results, we must conclude from these results alone that while females influence others to vote more liberally as a whole, they do not necessarily do so in gender-specific cases. Just like with gender, the odds of a more liberal decision are greater (1.273 times to be exact) across all case types with the presence of one black judge. With two or more, this number increases to 1.556. However, when limited to only race cases, this 27.3% and 55.6% increase in the likelihood of a more liberal decision disappears. The only interaction term that is significant here is with a race case and two or more blacks on the panel, and the odds ratio is so small that it is almost completely certain that the case outcome will not be liberal as opposed to mixed or conservative, or liberal or mixed as opposed to conservative. Because the racial diversity of the U.S. Courts of Appeals was not particularly high from 1925 to 2002, it is not surprising to find parallels to the gender issues described earlier. Out of the 807 cases categorized as “race cases,” there are only five cases in which there are two or more blacks on the panel. Although this Note tried to evade this problem by coding panels with two or three females, blacks, and others together (for example, there are 496 cases with two females but just twelve cases with three), the small sample of gender or race cases interfered nonetheless. It is possible that in those five race-related cases, the black judges and their colleagues on the panels voted conservatively; a larger sample might tell us something different. Nonetheless, it is not possible to speculate this given the constraints of the dataset, so we would be better off focusing on the original BlackPanel variables unencumbered by case type specification. Even in spite of the small number of relevant cases, it is interesting to note that the presence of 376 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 females and blacks raises the probability of a more liberal decision overall, but tends to do the opposite on cases that relate to their attributes. Possible explanations of this voting behavior will be explored in Part III. The results for Hispanic judges were not significant, so we cannot make any conclusions about them at this time in this study, except as part of a larger “nonwhite” group when evaluating the WhitePanel variables and interaction terms. The two WhitePanel variables that looked at all cases did not yield any significance, but when interacted with the presence of a race case, a panel with one white (and two nonwhites), two whites (and one nonwhite), or three whites would again greatly decrease the odds of a more liberal decision. Though the tables say the odds ratios are 0, these are rounded figures from the infinitesimal number e-29. This very tiny chance of a more liberal decision makes sense given the results from the RaceCase*BlackPanel interaction terms discussed above (although the nonwhites include Hispanics as well here). The variables on past prosecutorial experience seem to correspond with the hypotheses that past prosecutors vote more conservatively in criminal cases. Although the presence of prosecutors does not have much of a conservative effect on decisions (quite the contrary in fact), it does when it is a criminal case. The ProsPanel(1) and ProsPanel(2+) variables that indicate panels with 1 and 2 or more past prosecutors, respectively, both have positive coefficients and odds ratios greater than one (though not by much). Thus, the presence of prosecutors slightly increases the odds that the decision will be more liberal (by 11.7% if there is one, and by 9.7% if there are two or more). However, in criminal cases, the presence of a past prosecutor decreases the likelihood that the decision is more liberal, which was predicted. With one prosecutor, the probability decreases by 14.4%.64 We can also see that all three types of cases, not just prosecution, are significant. While gender and race cases are 64 The coefficient for two or more prosecutors on a criminal case was also negative, but not significant. No. 3:341] BROADENING DIVERSITY ON THE BENCH 377 associated with more liberal outcomes, criminal cases are associated with more conservative outcomes. Most of the variables that were significant for individual judicial vote are still significant for panel decision, specifically the appellant’s gender, the appellee’s race, the party of the appointing president (though much less significant), the Seventh Circuit, and the decades 1970 and 1990. Once again, a female appellant decreases the likelihood of a more liberal decision, this time by 26.7%, whereas a minority appellee increases it by 48.3%. The only other differences between the results in the two models (besides levels of significance) are that the appellee’s gender is no longer significant in the second test, and that the Eighth Circuit is moderately significant as well.65 In sum, in the first model, individual judges’ personal characteristics do not affect the way they vote. However, having females or blacks on a panel can influence others’ individual votes into being more liberal, regardless of case type. In the second model, we found that panel characteristics are significant all around, though in some unexpected and unclear ways. In non-gender cases, the presence of females greatly increased the odds of a more liberal decision, as did the presence of blacks in non-race 65 It is reasonable that the two significant region variables, Circuit7 and Circuit8, decrease the odds that the decision will be a more liberal one (by 20.3% and 16.5%, respectively), as the Seventh and Eighth Circuits are two of the most conservative circuits “in accordance with standard lore.” Sunstein, supra note 18, at 108. As for results for changes over time, cases in the 1970s increased the odds of a more liberal decision by 18.1%, and cases in the 1990s decreased the probability of a more liberal decision by 23.3%. While these differences can be partly attributed to the changing docket over the years, these significant results may also be explained by the appointing president’s administration and party. In the 1970s, 59% of the federal judiciary was appointed by a Democratic president, and in the 1990s, that percentage decreased to just 33%. Id. at 123. Additionally, the Carter Administration in the late 1970s was the first to implement a policy of affirmative action in his judicial appointments and appointed more women and racial minorities, which might be another reason for the significant results for decade. Jon Gottschall, Carter’s Judicial Appointments: The Influence of Affirmative Action and Merit Selection on Voting on the U.S. Courts of Appeals, 67 JUDICATURE 165, 167 (1983). 378 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 cases. In gender and race cases, results were more mixed. Hispanics had no significant effect, while whites decreased the odds of liberalness in race cases. Finally, judges with past experience as a prosecutor increased the odds of a liberal decision across all cases, but decreased those odds in criminal cases. 3. Predicted Probabilities Figure 1. Probability of Liberal Versus Conservative Rulings in Gender and Non-Gender Related Panel Cases (FemalePanel) 0.566 0.533 0.445 0.509 0.636 0.000 0.359 0.390 0.476 0.414 0.295 1.000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 0 Females 1 Female 2+ Females 0 Females 1 Female 2+ Females Conservative Liberal Non-Gender Gender Cases No. 3:341] BROADENING DIVERSITY ON THE BENCH 379 Figure 2. Probability of Liberal Versus Conservative Rulings in Race and Non-Race Related Panel Cases (BlackPanel) 0.567 0.507 0.454 0.532 0.566 1.000 0.358 0.416 0.467 0.391 0.359 0.000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 0 Blacks 1 Black 2+ Blacks 0 Blacks 1 Black 2+ Blacks Conservative Liberal Non-Race Race 380 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Figure 3. Probability of Liberal Versus Conservative Rulings in Race and Non-Race Related Panel Cases (HispanicPanel)* * There are no race cases with two or more Hispanics on the panel. 0.562 0.579 0.528 0.543 0.520 0.363 0.347 0.395 0.381 0.403 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 0 Hispanics 1 Hispanic 2+ Hispanics 0 Hispanics 1 Hispanic Conservative Liberal Non-Race Race No. 3:341] BROADENING DIVERSITY ON THE BENCH 381 Figure 4. Probability of Liberal Versus Conservative Rulings in Criminal and Non-Criminal Panel Cases (ProsPanel)* When interpreting ordinal logistic regressions, odds ratios can be misleading in some cases. It can be helpful to supplement the information from above with an alternate method of predicted probabilities, which many find makes the results more tangible and meaningful.66 As shown in 66 Using the margins command in Stata this Note evaluated the predicted probabilities for conservative, mixed, and liberal decisions for each of the three key attributes, while every other variable is held at the mean. For each of the variables FemalePanel, BlackPanel, HispanicPanel, WhitePanel, and ProsPanel, there are six possible scenarios. The author has structured it in such a way that there are predicted probabilities for each of these six possible outcomes, which allows us to compare the same characteristic across all scenarios at once. For example, the FemalePanel figure below orders the six scenarios in the 0.478 0.451 0.455 0.736 0.745 0.736 0.444 0.471 0.466 0.207 0.200 0.207 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 Conservative Liberal CriminalNon-Criminal 382 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Figures 1–4, we can see the predicted probabilities of a conservative and liberal decision when we set and hold all other predictor variables at their mean values. Here, the numbers do not tell us the probability of jumping a level to a more liberal decision, but rather the probability of each outcome. Regardless of variable or case type, the predicted probability of a mixed decision is always very low compared to a conservative or liberal decision (never above a 10% probability), which is why they are not included in the figures. The predicted probability of a liberal decision increases from 35.9% to 39% when one female is added to a panel on non-gender cases, and then again from 39% to 47.6% when yet another female or two are added to a panel. Of course, these observations should and do align with our conclusions from the odds ratios above, but predicted probabilities allow us to look at this information from a different angle. For example, we see that the likelihood of a liberal decision decreases from 41.4% to 29.5% when one female is added to a panel on a gender case, but then increases greatly with the presence of two or more females. Similarly, the presence of one black on the panel for a non-race case increases the probability of a liberal decision from 35.8% to 41.6%, and then from 41.6% to 46.7% with two or more blacks, as can be seen in Figure 2. However, this percentage decreases when the case is race-related, and even more so with the addition of blacks for those cases. Although Hispanics do not seem to vote according to any specific pattern in non-race cases, the probability that the case outcome will be liberal on a race case increases slightly from 38.1% to 40.3% when a Hispanic is on the panel (see Figure 3). Overall, however, Hispanics seem to vote on the conservative side, which is something we would not be able to glean from the coefficients and odds ratios alone. following way: (1) a non-gender case with zero females on a panel, (2) a non-gender case with one female on a panel, (3) a non-gender case with two or three females on a panel, (4) a gender case with zero females on a panel, (5) a gender case with one female on a panel, and (6) a gender case with two or three females on a panel. The five other figures are ordered in the same way for each respective characteristic and case type. No. 3:341] BROADENING DIVERSITY ON THE BENCH 383 For judges with prior prosecutorial experience, the difference is even more apparent in the predicted probabilities. For non-criminal (civil) cases, panels with a past prosecutor have just a 45.1% probability of reaching a conservative decision. In contrast, in a criminal case with a prosecutor on the panel, this probability is significantly higher at 74.5%. By looking solely at the predicted probabilities, however, one might think that the presence of prosecutors does not have any effect on the panel decision, as all outcomes in criminal cases are more likely to be conservative. Yet we know from the odds ratios that in criminal cases, one prosecutor decreases the likelihood that the decision is more liberal by 14.4%, and with two or more, the probability decreases by 8.8%. Thus, while predicted probabilities allow us to analyze the specific differences for each scenario, as well as observe the predicted probability not just of a “more liberal” decision but also of a liberal, mixed, and conservative decision, we cannot rely on them alone. Both predicted probabilities and odds ratios are needed to gain a comprehensive picture of the data results and their implications.67 B. Discussion of Panel Effects Results Although not every result matched up with the hypotheses, the results have opened up new avenues for discussion and possible interpretations. The main takeaway, however, is that panel effects and the presence of certain characteristics are strong and highly significant in determining the overall panel decision. In contrast, individual characteristics do not affect one’s own vote, which is consistent with past studies’ conclusions. In a majority of cases, the presence of females and blacks on a panel of appeals court judges increases the odds of a more liberal 67 The figures for predicted probabilities of individual judge characteristics have not been included, as there are very few differences among the probabilities of a conservative, mixed, or liberal vote regardless of case type or the characteristics of the judge. 384 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 ruling. The finding that the presence of a prior prosecutor on a panel decreases the likelihood of a liberal outcome only in criminal cases confirms the hypothesis that one’s past career experiences can shape one’s view of important issues, such as the criminal justice system. Although this Note could have made the panel variables binary to see the effect of the mere presence of certain characteristics on voting behavior, it is useful to look for any discrepancies between panels with more than one female and panels in which the female is outnumbered. As it turns out, we were able to observe these differences for several variables. Perhaps what was most surprising about the panel characteristics results were the variables FemalePanel and BlackPanel when interacted with GenderCase and RaceCase, respectively. Here, the presence of one or more females or blacks significantly decreased the likelihood of a more liberal decision in gender and race cases, which was the opposite of the original hypothesis. This Note’s original hypothesis was that the presence of females and blacks would raise the probability of a more liberal outcome in cases involving, say, employment discrimination, because not only do females and racial minorities bring in a new perspective to the law, but they might also seek different results that benefit litigants similar to them in regards to gender and race. Furthermore, the concept of panel effects implies that the visual presence of female or minority judges makes other judges on the panel more aware of and sensitive to issues of gender or racial discrimination, whether the female or minority judges say something to that effect or not. There are several possible explanations for why these results did not adhere to this theory, other than the coding of gender and race cases. As mentioned, the coding of the dependent variables on a scale of 1 to 3 captures only the final vote. It is thus difficult to account for the details of the actual opinion and case. That there are so few gender or race cases with two or more females or blacks on the panel also explains the odd results in those cases. Additionally, although federal judges have discretion, the law and the facts often dictate a particular outcome in cases and thus No. 3:341] BROADENING DIVERSITY ON THE BENCH 385 limit what judges can do.68 The relatively low reversal rate on the United States Courts of Appeals reveals another possible reason for the results, especially since the dataset and the model did not include this factor.69 Another possible reason for the results is that the effects of personal traits that we might assume to be very strong in forming one’s perception of the law are “washed out by the preappointment screening process.”70 The screening process is a careful one, with each of the President’s nominees undergoing an extensive vetting process by the Senate. It would be difficult to confirm a judge with notoriously strong political stances—especially in regards to gender or race issues—because of some people’s fear that she would allow those beliefs to cloud her judgment. A tight ideological control over the judicial selection process is often the main reason why certain judges are nominated and ultimately confirmed over others. The two characteristics of race and gender are especially sensitive today—they are often deemed diversity “plus factors,” but only when those judges already fit the bill in every other way. Sometimes this means relatively more conservative judges, especially since most have been through the same elite educational and professional processes. 68 Sunstein, supra note 18, at 83. 69 UNITED STATES COURTS, STATISTICAL TABLES FOR THE FEDERAL JUDICIARY, Appeals Terminated on the Merits, http://www.uscourts.gov/sites/default/files/statistics_import_dir/b05dec01.p df; http://www.uscourts.gov/sites/default/files/statistics_import_dir/b05dec02.p df (last visited Mar. 27, 2018). In 2001, the percent of cases reversed was 9.2%. In 2002, the rate was 9.6%. Id. “Using quantitative empirical methods to analyze judicial decisions has some inherent limitations because it is simply impossible to control for all the relevant factors underlying a decision.” Cross, supra note 1, at 6. In fact, it is virtually impossible to even name all the relevant factors underlying each judicial decision. In a perfect world, we would be able to account for everything, but given missing information from datasets and for the sake of keeping this Note’s long model clean and manageable, some factors had to be left out. However, in spite of this inevitable obstacle, this Note still included the most significant judge and case characteristics that might determine policy preferences or otherwise affect the outcome of an appeal. 70 Cross, supra note 1, at 72. 386 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Federal judicial positions are not jobs open to just anyone, as can sometimes be the case with elected office. For example, a conservative president might appoint minorities while still ensuring that they are conservative. Thus, voting similarities between black and white judges either indicate that the courts represent “a traditional, conservative black elite, selected for its behavioral conformity” or that “a combination of institutional, role, and self-imposed demands” pressures black judges to conform to the voting patterns established by a dominantly white legal system.71 This pressure to conform is also known as “collegial concurrence,” which occurs when one judge joins his colleagues and refuses to dissent publicly.72 As mentioned, a major issue with studies like these is not having a sufficient number of female or minority judges, especially in the earlier decades. Because females and blacks may be relatively new to the system, some might be reluctant to dissent from the majority too often or be overly vocal in presenting new perspectives that might advance a particular group’s rights. In an attempt to be perceived favorably by the other judges and preserve the spirit of consensus and unanimity on appellate panels, they might yield when confronted with two other judges already in agreement.73 In other cases, female or minority judges might overcompensate to appear objective and fair in certain cases, thinking that others will expect them to vote liberally. One study found that black males did not vote more liberally than white males in racial discrimination cases, and suggested that “blacks avoid partisan identification with issues so salient to their own career and life opportunities.”74 An analogous situation is when a female or minority politician tries not to allow gender or race to be her identifying feature as a policymaker, and thus does not advocate for her own “interest group.” For example, in the federal district courts, because black judges tend to identify as liberal, some might 71 Thomas M. Uhlman, Black Elite Decision Making: The Case of Trial Judges, 22 AM. J. POL. SCI. 884, 892 (1978). 72 Sunstein, supra note 18, at 148. 73 Id. at 15. 74 Gottschall, supra note 65, at 172. No. 3:341] BROADENING DIVERSITY ON THE BENCH 387 expect that they will vote in ways more sympathetic to criminal defendants, especially black defendants.75 However, studies have shown that there are no significant differences between black and white judges in criminal cases.76 Another explanation for these conformity effects is that one might genuinely become convinced that the other judges have a point if they are all in agreement. Experiments have indeed shown that people are easily swayed by and highly susceptible to fellow colleagues’ opinions.77 Another related explanation is not that the female or black judge wants to appear neutral, but rather that the other two male or white judges on the panel do. They may be wary of the potential influence the female or black judge may have on the decision-making process and thus overcompensate to try not to let that affect their judgment. Finally, there is one more explanation for these results that is especially applicable to the methodology. One study found that in controversial cases involving abortion or the death penalty, judges’ individual votes were “impervious to panel effects,”78 and any possible influence their colleagues might have had were outweighed by their own firm convictions.79 Debating and discussing issues that the judges already felt strongly about may have only pushed them further toward what they initially believed. This entrenchment may help explain why the presence of females and blacks had an overall liberal effect on the decisions, but was not strong enough for gender- or race-specific cases. Thus, although Kastellec stated that panel effects may be apparent only in the more ideological cases as opposed to the “more routine cases,”80 these results show that it might actually be the opposite. 75 Susan Welch, Michael Combs & John Gruhl, Do Black Judges Make a Difference?, 32 AM. J. POL. SCI. 126, 127 (1988). 76 Id. at 128. 77 Sunstein, supra note 18, at 74–75. 78 Id. at 62. 79 Id. 80 Kastellec, supra note 21, at 380. 388 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Whether because of the screening process, selection effects, normative social influences, or case type, there are numerous explanations for why females and blacks vote more liberally across all cases but less so on gender- or race- specific cases. Whatever the reason, it makes sense that the same does not hold true for past prosecutors in criminal cases. The trait of prior prosecutorial experience is not relatively new to the federal bench, and there is no pressure for judges with such experience to act in a particular way or to defy stereotypes. The possibility of pressure to conform may be a new issue with the judicial system that is even worse than simply voting according to one’s ideology or beliefs. C. Implications Depending on how one sees it, the liberal influence female and black judges have on a decision in the court can be both a good and bad thing. Some might find that this only confirms the need for greater diversity on federal courts in order to counter the long-held judicial legacy of white male judges. Other conservatives may fear what changes this may bring. Still, they might take the decreased likelihood of a liberal decision on gender or race cases when females or blacks are on the panel as a sign that the system is not flawed, and judges are not actually being unnecessarily swayed into voting differently. At the very least, it might indicate that we should focus on other flaws within the system. Based on the results for prior prosecutors, advocates for criminal justice reform may want to pay more attention to judicial appointees and support nominees who have more unique career backgrounds. Perhaps it is still too soon to make any of these conclusions about the U.S. Courts of Appeals or federal courts at large. It is difficult to make any definitive statements until the databases are updated to include cases that happened over the past fifteen years, a time period during which much tumultuous social change has occurred, along with many controversial cases covering topics that have everything to do with gender, race, and criminal justice. No. 3:341] BROADENING DIVERSITY ON THE BENCH 389 Even when the bench has diversified more and there are more relevant cases included in the sample, we still must be cautious about making generalizations. After all, we can see from the summary statistics in Table 2 that females, blacks, and Hispanics made up only 3.42%, 4.87%, and 1.56% of the total appeals court judge population in the dataset, respectively. It is thus surprising that past studies have made broad conclusions about female voting behavior when examining only one or two specific types of cases over just a few years. Even if the panel influences were present and meaningful, they may not have been strong enough to outweigh the dominant influence white male judges have, something not captured in the simple coding of a 1, 2, or 3 decision. It is more difficult to apply the same possible explanations given for gender and race to prior prosecutorial experience. The presence of a prior prosecutor on a criminal case increases the chances of a more conservative decision that does not favor the defendant. Past studies have discovered bipartisan consensus and similar voting behavior across party lines for criminal cases.81 This Note found that prior experience as a prosecutor does create differences in voting on criminal cases, which has interesting implications. These significant effects on criminal outcomes suggest that the pursuit of diversity in career background is just as important as the pursuit of diversity of other characteristics in reflecting the entire nation’s views, especially given the homogeneity of most federal judges’ career paths.82 Thus, if a future presidential administration wants to follow 81 Sunstein, supra note 18, at 61. 82 Out of President Clinton’s 61 appointees to the U.S. courts of appeals, 37.7% of had prosecutorial experience, and only 29.5% lacked either prosecutorial or judicial experience). Sheldon Goldman & Elliot E. Slotnick, Introduction: Clinton’s Judicial Legacy, 84 JUDICATURE 227, 244, 249 (2001). See also New York Times Editorial Board, The Homogeneous Federal Bench, NEW YORK TIMES (Feb. 6, 2014), https://www.nytimes.com/2014/02/07/opinion/the-homogeneous-federal- bench.html [https://perma.cc/SC6J-L68F ] (85% of President Obama’s nominees to the federal bench have been corporate attorneys or prosecutors, and fewer than 4% have worked in public interest organizations). 390 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 President Nixon’s lead and appoint judges who are tough on crime, it might look for prior prosecutorial experience. If a presidential administration prioritizes reforming the criminal justice system, it might do well to nominate judges with career backgrounds that differ from the prosecutorial norm. Overall, the results are not as straightforward as the original hypotheses, which were that the presence of females and racial minorities would increase the likelihood of a more liberal decision, particularly in gender- and race-related cases, and that the presence of past prosecutors would decrease the likelihood of a more liberal decision in criminal cases. However, if the hypotheses were sustained, this Note would not have been able to delve deeper into the more detailed and realistic aspects of judicial decision-making. It would have been easy to attribute and reduce deliberations to judges’ personal characteristics, but the results instead showed that there are many other equally important social and psychological factors that cannot be fully accounted for in empirical studies. Phenomena like the pressure to conform, especially as somewhat of a newcomer to the bench, often slip the minds of the public when we discuss federal judges and minority representation. This could mean that more attention should be paid not only to the relationship between personal attributes and voting behavior but also to the other pressures and difficulties that certain judges must face. A possible area of extension in the study of judicial outcomes relates to the process by which state court judges are selected. Comparing voting patterns between federal and state courts would be able to give some insight into differences between decisions made by judges who were appointed and decisions made by judges who were elected by the public. Although this Note alone cannot tell us which method is preferable, one study83 stated that the election of state judges has a significant impact on the religious, 83 Bradley C. Canon, The Impact of Formal Selection Processes on the Characteristics of Judges – Reconsidered, 6 L. & SOC’Y. REV. 579, 588 (1972). No. 3:341] BROADENING DIVERSITY ON THE BENCH 391 educational, and career diversity of elected judges.84 One might then conclude that bringing the selection process to the masses means selecting judges with ideological views and characteristics that align with those of the constituency. This is not to say that the appointment system is devoid of politics, however. According to one study, appointing presidents follow one of two models that determine their judicial nominations.85 The first is the policy model, in which presidents seek to appoint judges who have similar policy preferences—this theory is supported by the results. An example of the policy model is when the Nixon Administration sought to appoint “law and order” judges who were tough on crime.86 The second is a partisan model, in which presidents try to reward party loyalty by nominating judges whose ideological views reflect those of the judge’s population.87 Yet despite the general understanding that many nominated judges can be political activists who reflect “the values and outlook of the appointing administration,”88 judicial appointments are not entirely driven by politics either. Presidents do not have unlimited control, as the judge must be palatable to the Senate. There are structures in place to prevent extreme judges from being appointed to the bench. Judicial appointments are the product of many factors: competence, ideology and party affiliation, and, of course, being in the right place at the right time.89 Nominated judges are drawn from a list of possible candidates that usually consist of a “vast network of friends, acquaintances, and friends of friends.”90 All things 84 Id. 85 Michael W. Giles, Virginia A. Hettinger & Todd Peppers, Picking Federal Judges: A Note on Policy and Partisan Selection Agendas, 54 POL. RES. Q. 623, 627–28 (2001). 86 Gottschall, supra note 65, at 166. 87 Giles et al., supra note 85, at 627. 88 Sheldon Goldman, Judicial Appointments to the United States Courts of Appeals, 1967 WIS. L. REV. 186, 214 (1967). 89 Id. at 186. 90 Id. at 189. 392 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 considered, however, it is still an elite, largely closed-off, and private process that is only open to a select few. Because the study has found that there are indeed panel effects that either increase or decrease the likelihood of a more liberal decision, it is important that we now turn to another one of this study’s original questions, which is whether or not these findings should increase calls for greater diversity on the courts. More simply, now that we have confirmed the influence of the presence of females, racial minorities, and past prosecutors on judicial outcomes, is the pursuit of diversity still desirable? Our initial reaction might be an unqualified yes because of all of diversity’s benefits, both for its descriptive and substantive representation. The news that there are now a greater number of “non-traditional” judges on the U.S. Courts of Appeals has been heralded as a positive development.91 There is the symbolic diversity element that “helps promote trust in the [legal] system.”92 The image of judges who are not white or male can serve as visible role models for young students. There is also, of course, the substantive diversity that comes from the inclusion of new perspectives and competing arguments. This increases the potential for dissent, which increases “the probability that the law will be followed.”93 Diversifying courts can also check extreme, lawless decisions, 91 As of July 2014, the U.S. Courts of Appeals is composed of a majority of “non-traditional” judges with the appointment of Pamela Harris to the Fourth Circuit. Barry J. McMillion, Demographic Diversity on the U.S. Courts of Appeals: An Update, CONGRESSIONAL RESEARCH SERVICE REPORT, GOVERNMENT AND FINANCE DIVISION (2014). “Non- traditional” here refers to those who belong to “demographic groups from which, historically, individuals were seldom, if ever, selected for federal judgeships,” namely females, African Americans, Asian Americans, and Hispanics. Id. Before President Obama took office, non-traditional judges only comprised 39.4% of the active federal appeals court judges, and there were no Asian American appellate court judges. Today, white men are still the plurality of all active appeals court judges, but the percentage of non- traditional judges is now 50.3%, with 86 non-traditional and 85 traditional judges. Id. 92 Hsu, supra note 13, at 115. 93 Sunstein, supra note 18, at 135. No. 3:341] BROADENING DIVERSITY ON THE BENCH 393 something especially critical in cases when the law is ambiguous. One qualitative study uses the example of Japanese American internment to support the claim that appointing more Asian American and other racial minorities to the bench can “ensure that the legislature never passes similar legislation.”94 A better understanding of racial issues can lead to a more inclusive rule of law that takes all perspectives into account. All of this assumes, however, that conformity effects will not be as prevalent, and that minorities on the courts will voice their true opinions. Scholars claim that the ultimate goal here is “diversity of reasonable views,”95 and that the President and Senate should actively pursue such a range of opinion. Without this diversity, we risk facing unequal treatment and application of the law, and panels that give extreme or unfair rulings. Judicial appointments are thus an opportunity for presidents to leave a legacy that outlasts their tenure and shape the court system into one that better resembles the changing society it serves.96 However, this simple view has made “diversity” a buzzword used to justify numerous policies and settle all sorts of arguments. After all, it is difficult to criticize a principle that is a fundamental American value. Is it an objectively “good” thing that the presence of females and minorities increases the probability of a more liberal decision, and is it an objectively “good” thing that the presence of past prosecutors increases the probability of a conservative decision? If the law is truly to be unbiased and binding, judges should vote the same way regardless of who else is on the panel. Symbolic and substantive diversity aside, increased diversification and the subsequent shift toward the left over time may not necessarily be a positive 94 Hsu, supra note 13, at 119. 95 Sunstein, supra note 18, at 138. 96 Philip Rucker, Obama Pushing to Diversify Federal Judiciary Amid GOP Delays, WASHINGTON POST, Mar. 3, 2013, https://www.washingtonpost.com/politics/obama-pushing-to-diversify- federal-judiciary-amid-gop-delays/2013/03/03/16f7d206-7aab-11e2-9a75- dab0201670da_story.html [https://perma.cc/LJ3N-ZW6P]. 394 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 development. Regardless, arguments that diversity on the bench leads to judicial impartiality because it “ensure[s] that a single set of values or views do not dominate judicial decision-making” are compelling and worth considering.97 Another argument for the continued support of diversification within the judicial branch is that this is an overdue process to counteract years of judges who may not have had certain groups’ best interests at heart in the past. There have also been calls for other kinds of diversity in religion, sexual orientation, and professional experience. While this Note looked at prior prosecutorial experience, many liberal groups have also urged the government to nominate fewer prior corporate lawyers and more attorneys with experience in academia or the public interest, such as public defenders. Most of the time, federal judges do not come from civil rights organizations or small practices.98 Prior public defenders might counteract a bias against criminal defendants that some prior prosecutors may hold, and judges who have advocated for clients at lower levels of socioeconomic status might place more emphasis on achieving justice for all Americans, and not just for a select few. The next question, then, is how to feasibly achieve greater diversity, especially if citizens, let alone future presidents, do not see the issue as a major priority. It is difficult to advocate the importance of judicial diversity as there is no specific group directly hurt by the lack of action, and because there are no feasible short-term solutions. Furthermore, the ultimate decision is beyond the public’s access and control, as there is usually a multitude of concealed reasons for confirming or rejecting a candidate. The process could be modified to one that gives the public a greater role in the selection process, or people could demand that deliberations be made public, but this might too closely resemble an election method of selection. 97 Sherrilyn A. Ifill, Racial Diversity on the Bench: Beyond Role Models and Public Confidence, 57 WASH. & LEE L. REV 405, 411 (2000). 98 Id. at 407. No. 3:341] BROADENING DIVERSITY ON THE BENCH 395 One suggestion made in the past is to require at least one Democrat and Republican on each panel.99 While this might prevent any ideologically extreme panels from forming, it becomes more complicated as we learn that other background characteristics have a significant effect as well. Requiring that there be, say, at least one female or one black judge on every panel would be complex and controversial, but worth exploring. Furthermore, the practice goes against the very idea that judges are not policymakers, and whose duty consists solely of following the law. There are thus many practical questions to consider, one of which begins with the fact that the public is relatively unfamiliar and unconcerned with the lower federal courts. It is hard to imagine a scenario in the near future in which people become so invested in the issue that a legislative act is passed to transform the way the U.S. Courts of Appeals currently operates. There has also been much opposition to efforts to broaden diversity on the courts. According to critics, “ideologues have their place, just not on the bench.”100 Some find the push for more women and racial minorities on the bench distasteful because they believe this affirmative action practice is tantamount to lowering one’s standards for qualified judges, falsely assuming that there are not any available competent female or minority judges.101 These political obstacles show that appointees can be easily rejected and that reaching a critical mass of minority judges in federal courts will be a long, arduous process.102 However, as law school populations become increasingly diverse, there will also be a more diverse pool of legal talent and ability from which to choose future judges. These students can also raise awareness of the process of judicial selection at both state and federal levels. 99 Kastellec, supra note 21, at 379. 100 Rucker, supra note 96. 101 Id. 102 Sylvia R. Lazos Vargas, Does a Diverse Judiciary Attain a Rule of Law That is Inclusive?: What Grutter v. Bollinger Has to Say About Diversity on the Bench, 10 MICH. J. RACE & L. 101, 109 (2004). 396 COLUMBIA JOURNAL OF RACE AND LAW [Vol. 8:2 Despite the disagreements over diversity on the judicial branch, every American should view the appointment of federal judges as an important process that affects us all, rather than a clandestine and trifling matter that concerns only presidents and Congress. We can conclude that greater diversity can influence decisions in positive ways. We can also conclude that while judges do not always follow the law in a completely objective manner, their discretion is still often restricted by the rule of law. Thus, although judges should not be appointed to represent the interests of certain groups, they should still interpret the law in a way that reflects the diverse composition of our nation. Personal characteristics seem to make a difference on panels, and what we decide to do with that information is critical to future laws and policies. As much as we strive for objectivity on the courts, we must accept that the legal system can never be fully removed from ideological policymaking. We should also understand that this reality is not necessarily an unfortunate one and can be used to help our nation reach the ideal of equal justice. V. CONCLUSION This Note has examined the voting behavior of judges on the U.S. Courts of Appeals, and has found that panel effects are present and influential in the decision-making process. While personal characteristics like race, gender, and prior prosecutorial experience do not appear to affect individual judicial voting, the presence of a judge with one of these attributes can influence the way the other two judges on the panel vote, which then affects the final outcome. Although the results do not completely match up with the initial hypotheses, the findings reveal something even more interesting about the dynamic of federal circuit courts and three-judge panels. Empirical evidence cannot tell us how panels should be organized or what federal judges should do, but it has provided much insight into the way judges think. This Note has contributed to the current literature on voting behavior on federal appellate courts, particularly regarding panel effects, by filling in some of the holes that No. 3:341] BROADENING DIVERSITY ON THE BENCH 397 exist from past studies’ methodologies. Specifically, this Note increased the number of cases as well as the types of cases included in the study, and examined a greater time span of 77 years. The judge attributes this Note focused on also went beyond party and ideology, which has been the focus of most key studies on judicial voting. Prior prosecutorial experience has not been a characteristic of major interest in the past, and although both race and gender have been explored to some extent, this Note’s adjusted methodology allowed for new observations and conclusions about the effects of having females, racial minorities, and prior prosecutors on the bench. Possible areas to explore in future studies include looking at judges on the federal district courts; elected state judges; other attributes such as wealth, educational background, or other professional experiences; and recent data that incorporates cases dating from 2002 until the present, especially given the increased diversification of the federal courts over the past decade.103 In sum, this Note finds that the presence of females and blacks on panels generally has a strong liberal effect on the panel’s final decision across all cases. However, for gender or race cases alone, this effect is less clear and may operate in the reverse direction. There are several possible explanations for these mixed results, namely conformity effects, clear and binding law that must be followed, or strong convictions on certain issues that will not budge regardless of who else is on the same panel. Females’ and blacks’ personal policy preferences are either constrained for the many possible reasons mentioned above, or minority judges may not have the inclination to vote in favor of the female or racial minority litigant in gender and race cases in the first place. Perhaps these judges were deemed “safe,” conservative judicial choices and thus nominated and confirmed for this very reason. In any event, this analysis has helped bring into light other factors that may be more important than personal traits when it comes to making decisions on the federal courts. 103 See McMillion, supra note 91. I. Introduction II. How Diversity Affects Voting Outcomes A. Diversity on the United States Courts of Appeals B. Panel Effects on the United States Courts of Appeals C. Empirical Studies of Voting Behavior III. Methodology A. Data B. Research Design C. Hypotheses IV. Findings A. Results 1. Individual Judicial Vote 2. Panel Effects 3. Predicted Probabilities B. Discussion of Panel Effects Results C. Implications V. Conclusion