Party Capability and Split Outcomes

In his now classic article, Galanter “put forward some conjectures about the way in which the basic architecture of the legal system creates and limits the possibilities of using the system as a means of redistributive (that is, systematically equalizing) change” (1974:94). In doing so, Galanter developed a typology of parties based on a distinction between those who litigate only on rare occasions and those who do so regularly. The former (the so-called one-shotters) are generally litigants such as tenants, bankrupt consumers, welfare clients, or injury victims. The latter (those dubbed repeat players), on the other hand, are typically parties such as landlords, creditors, government agencies, or insurance companies. Repeat players have a presumed advantage relative to one-shotters for a host of reasons. For example, repeat players tend to have greater financial resources that enable them to hire more-competent legal counsel and expert witnesses. Such resources also provide the “staying power” necessary to achieve successful outcomes through the appellate process and the wherewithal to settle unfavorable cases while pursuing more promising ones. In addition, repeat players benefit from their experience in litigation, enabling them to shop for more advantageous forums as well as develop a more comprehensive litigation strategy, and thus to navigate the litigation process more successfully. In short, repeat players are high-capability litigants.

Galanter's theory has given rise to an extensive body of research about winners and losers in a variety of courts running the gamut from domestic to foreign, a line of inquiry often referred to as party capability research. For example, studies by Reference WannerWanner (1975) and Reference YarnoldYarnold (1995) found evidence to support the proposition that parties with greater resources are more likely to win in the trial courts. In the context of state courts of last resort, Reference WheelerWheeler et alia (1987) evaluated classes of litigants in terms of their relative resources and experience, finding that, where repeat players were matched against one-shotters, the repeat players were more likely to win than when matched against litigants with equally high resources (see also Reference FaroleFarole 1999; Reference Brace and HallBrace & Hall 2001).Footnote ¹ In the U. S. Courts of Appeals, Reference Songer and SheehanSonger and Sheehan (1992) and Reference Songer and SheehanSonger, Sheehan, Reference Songer and Sheehanet alia (1999, 2000) found that governments and businesses were far more likely to succeed on appeal than were individuals (see also Reference SheehanSheehan 1992). Studies of appellate courts in other countries have similarly found support for the premise that high-capability parties are advantaged in the courts (Reference AtkinsAtkins 1991; Reference McCormickMcCormick 1993; Reference HaynieHaynie et al. 2005).Footnote ²

We can easily link this line of research regarding party capability to the phenomenon of mixed outcomes in the U. S. Courts of Appeals. Where high-capability parties meet low-capability parties on appeal, it is likely that the former will prevail in toto. In contrast, where two high-capability parties are involved in an appeal, both have the capacity to make credible arguments that may ultimately lead the appellate panel to hand both a partial victory. More formally:

Not all high-capability parties are created equally, however. A governmental litigant, in particular, can be considered a special kind of high-capability party. Reference Kritzer, Kritzer and SilbeyKritzer (2003) argues that the advantage government has is twofold:

Since the government sets the rules of the legal game, it can stop suits against itself by invoking sovereign immunity, restricting its liability for damages, and imposing handicaps on erstwhile litigants by complicating or limiting legal fees, among other things (Reference Kritzer, Kritzer and SilbeyKritzer 2003:351–6). Moreover, the very fact that judges are themselves a part of the ruling governmental regime can induce deference when judges are in the position of resolving disputes involving challenges to the government (Reference Kritzer, Kritzer and SilbeyKritzer 2003:358–62; see also Reference RosenbergRosenberg 1991).

The empirical evidence regarding the advantage the government has in terms of securing legal outcomes it desires is extensive. For example, both Reference WheelerWheeler and his colleagues (1987) and Reference FaroleFarole (1999) found governmental litigants advantaged in state courts of last resort. Government parties (in particular, state and big city governments) had the greatest overall success rate and strongest record against every other type of litigant (e.g., individuals, businesses). Reference McCormickMcCormick's (1993) analysis of the Supreme Court of Canada similarly found that governmental parties (federal, provincial, municipal) were the most successful, as did Reference AtkinsAtkins (1991) in the case of the English Court of Appeals. Despite finding little support for the party capability thesis in general in the case of the Australian High Court, Reference SmythSmyth (2000) did find that the federal government has an advantage. Reference SheehanSheehan et alia (1992) found similarly in the case of the U. S. Supreme Court. More recently, Reference HaynieHaynie et alia (2005) found a consistent pattern of government advantage in their analysis of winners and losers in the highest appellate courts across nine countries, including Australia, Canada, Great Britain, India, South Africa, Tanzania, and Zambia. More directly on point for our purposes is the evidence regarding government litigants in the U. S. Courts of Appeals. Reference Songer and SheehanSonger and Sheehan (1992) examined decisions decided in three circuits (the Fourth, Seventh, and Eleventh) in 1986 and found that the success rate for governments was fourfold that of individuals and greater than that of businesses by half. In subsequent bivariate (2000) and multivariate (1999) analyses, Songer, Sheehan, et alia provided even more evidence about the advantage that accrues to governmental litigants. In the context of mixed outcomes, this line of research suggests that government parties will be more likely to prevail in toto, regardless of the nature of the opposing litigant (i.e., whether the litigant is one with fewer or greater resources). Accordingly, we hypothesize:

Strategic Behavior and Mixed Outcomes

In addition to litigant resources, intrapanel dynamics may also affect the likelihood that a case will produce a mixed outcome, particularly the extent to which judges act strategically in negotiating over the final disposition of an appeal. The strategic model of judicial decisionmaking posits that “to achieve the outcome most compatible with their policy preferences, judges consider the impact of other judges' likely actions as well as their own. Depending on the preferences of other relevant actors and how those actors are likely to behave, a strategic judge might act differently than if his behavior was driven by attitudinal considerations alone” (Reference HettingerHettinger et al. 2006:74). Extensive (and compelling) evidence exists demonstrating that justices of the U. S. Supreme Court behave strategically (e.g., Reference Brenner and KrolBrenner & Krol 1989; Reference Krol and BrennerKrol & Brenner 1990; Reference WahlbeckWahlbeck et al. 1998, Reference Wahlbeck1999; Reference CaldeiraCaldeira et al. 1999; Reference MaltzmanMaltzman et al. 2000). The same is true for judges serving on state courts of last resort (e.g., Reference HallHall 1992; Reference Brace and HallBrace & Hall 1993, Reference Brace and Hall1995; Reference Hall, Brace, Clayton and GillmanHall & Brace 1999; Reference LangerLanger 2002).

Of special interest for our purposes is the extant work on the ideological composition of the panel and opinion writing behavior. Existing studies have found that decisionmaking on these courts is influenced in important ways by the policy preferences of those judges (e.g., Reference GoldmanGoldman 1966, Reference Goldman1975; Reference Songer and SegalSonger, Segal, et al. 1994; Reference ReddickReddick 1997; Reference BeneshBenesh 2002). We have no reason to suspect that ideology is related in a direct fashion to mixed outcomes, but judges motivated by attitudinal considerations may act strategically to obtain their preferred result in light of the preferences of other judges on a panel. Indeed, evidence suggests that the ideological composition of circuit panels has a substantial influence on decisionmaking dynamics (Reference SongerSonger 1982; Reference ReveszRevesz 1997; Reference Van WinkleVan Winkle 1997; Reference Cross and TillerCross & Tiller 1998; Reference SunsteinSunstein 2003). These studies suggest that some form of accommodation may take place among members of a panel, perhaps to maintain collegiality or to avoid a dissent that has the potential to dilute the impact of the majority opinion. One such accommodation mechanism available to the majority opinion writer is to moderate the ideological tone of the opinion (Reference MaltzmanMaltzman et al. 2000). In addition, however, judges may accommodate their colleagues by choosing to render a mixed decision—one that provides some benefit to both parties to the litigation. By making some accommodation to both parties via a mixed outcome, a majority on the court may entice a judge who is otherwise predisposed to write a separate opinion to join the majority opinion. Obviously, the greater the ideological differences among panel members, the more such accommodation may be necessary. Hence, we hypothesize:

Of course, the choice to accommodate will be conditioned by the actions of the other judges on the panel. When a judge (the potential dissenter discussed above) ultimately decides to dissent (or even concur) rather than join the majority opinion, he or she has signaled disagreement with the majority's conclusion (Reference HettingerHettinger et al. 2004, Reference Hettinger2006). In those circumstances, the motivation to accommodate through a mixed outcome is obviated given that the judge whom the mixed outcome was intended to accommodate has, by filing a separate opinion, rebuffed the accommodation proffered by his or her colleague. In other words, the presence of a separate opinion will condition the likelihood of a mixed outcome. More formally:

Court Legitimacy and Mixed Outcomes

Courts exist to resolve disputes between litigants. This fundamental axiom will raise few eyebrows, especially for those in established democracies where dispute resolution via court processes is a well-established and accepted means for addressing conflict in society. Yet the continued viability of courts depends on their legitimacy in the eyes of the disputing parties, which in turn affects the willingness of those parties to comply with judicial rulings. Without legitimacy and compliance, courts lose their raison dê'tre (Reference Gibson and CaldeiraGibson & Caldeira 1995).Footnote ³ Convincing the losing party that the court decision was fairly rendered is particularly critical to legitimacy and compliance.

In his seminal work on conflict resolution, Reference ShapiroShapiro (1981) explained that the social logic of courts stems from the need for two disputing parties to call upon a third party to assist in resolving their conflict. According to Shapiro, this “common sense” solution to dispute resolution is so “universal across both time and space” as to render it practically fundamental to social ordering processes (1981:1). In the most primitive circumstances, the parties consent to the designation of the third party/judge. Ultimately, however, courts have largely become institutionalized by political regimes that find it useful to establish judiciaries. Once “the substitution of office and law for consent” takes place (Reference ShapiroShapiro 1981:5–8), the dispute-resolver becomes even more vulnerable because the losing party did not play a role in selecting the dispute-resolver by consent. Thus, as institutionalized third parties with the power to decide in favor of one party over the other, courts risk their legitimacy in the eyes of the loser every time the court actually resolves a dispute. At that point, the loser feels no longer involved in a triad, to use Shapiro's term, of two adversaries and a neutral arbiter. Instead of being viewed as an honest broker, the judge may be seen by the losing party as joining the judge's opponent, in effect “ganging up” on the losing party. Since “two against one” is fundamentally at odds with our commonly accepted notions of fair play and justice, the legitimacy of the ruling can then easily come under attack. How can courts diffuse this problem and enhance their legitimacy and compliance with court judgments?

Courts have several means at their disposal to enhance legitimacy, even in the eyes of the loser. First, judges can enhance their standing before the citizenry by adopting ceremonial trappings, such as the wearing of wigs and robes and the use of a raised platform from which to announce their decisions.Footnote ⁴ Second, appellate courts provide normative justifications for their decisions (opinions) to convince the loser that “the law” mandated the outcome, not the whim of the individual decision maker (Reference ShapiroShapiro 1994).Footnote ⁵ Third, judges can anticipate reactions to their decisions and tailor them in such a way as to avoid the declaration of a clear winner or loser. Reference Stone SweetStone Sweet (2000) suggested that court outcomes exist along a continuum, with positions P and R representing the preferred outcome for the petitioner (appellant) and respondent (appellee), respectively:

If a judge renders a judgment that falls at P or R, he or she thereby clearly (and completely) favors one party over the other. If, however, he or she is able to fashion a judgment between P and R (e.g., P¹ or R¹), the judge can avoid the declaration of a definitive winner or loser (Reference Stone SweetStone Sweet 2000:16–7; Reference Shapiro and Stone SweetShapiro & Stone Sweet 2002:63). To be sure, P¹ is more preferable for the petitioner than for the respondent, and R¹ is more preferable for the respondent than for the petitioner. But neither P¹ nor R¹ represents a total victory for either party. Fashioning a judgment like this is not possible in all cases, but where it is, it can help ensure compliance by the losing party.Footnote ⁶ Such a judgment has the potential to do so because it can be seen as “giving a little to both sides” and, further, ameliorating either litigant's perception that the judge has unfairly sided with his or her adversary.

This supposition is related to research focusing on “procedural justice.” Some scholars have claimed that citizens' perceptions regarding the legitimacy of judicial institutions arise from their sense that the courts follow fair procedures in rendering outcomes; when citizens feel that they have been treated fairly, they are more likely to accept an unfavorable substantive outcome (Reference TylerTyler 1990; but see Reference GibsonGibson 1989; Reference MondakMondak 1993). Though to date the empirical results are mixed on this point, the theory does suggest that the manner in which judges render decisions contributes to public support for judicial institutions.

With this in mind, we contend that mixed outcomes are more likely when judges have reason to be concerned about the effect of a decision on the legitimacy of the court and compliance with its decisions. We further speculate that salient cases are ones in which courts will be particularly sensitive to issues of legitimacy and compliance. Salient (or controversial) cases are those that will garner the greatest attention from other political actors and be placed more squarely in the public eye. We can think of capturing the notion of a case's saliency or visibility in a variety of ways. Three means, in particular, are promising in the present context: subject matter, amicus participation, and opinion publication. With regard to subject matter, the basic argument is that cases dealing with some issues will attract more interest than others. As a general rule, for example, cases dealing with First Amendment rights regarding freedom of speech or freedom of religion garner more attention than those dealing with arcane aspects of the tax code, no matter how important the latter may be in terms of its consequences for the public (Reference HettingerHettinger et al. 2006:58). Similarly, where a case attracts amicus participation, it suggests that the issue is more salient for the public at large (Reference Caldeira and WrightCaldeira & Wright 1988, Reference Caldeira and Wright1990; Reference HettingerHettinger et al. 2003b). Finally, opinion publication by the district court is a relatively rare phenomenon reserved for cases of particular importance (Reference WasbyWasby 2001; Reference GerkenGerken 2004). Accordingly, we hypothesize as follows:

We also speculate that some judicial actors will be more likely to be concerned with court legitimacy than others. In particular, given their institutional position, chief judges are in a position to view the judiciary from a more holistic perspective and may therefore be more concerned with the public's perceptions regarding the court's performance (Reference HowardHoward 1981:225–32). This idea is supported by existing research regarding separate opinion writing by chief judges. As Cohen wrote, “Court culture teaches that a court that presents a unified face has fewer fragmented opinions, has a higher degree of civility among its judges, speaks with a higher degree of moral authority, and enjoys a higher degree of legitimacy” (2002:173; emphasis added). This idea led Hettinger, Lindquist, and Martinek (Reference HettingerHettinger et al. 2003a) to empirically evaluate whether, due to their special sensitivity to institutional concerns such as safeguarding the legitimacy of the institution, chief judges would be less likely to author separate opinions themselves as well as induce others to suppress such inclinations. And indeed, they found evidence supportive of this premise. Likewise, “splitting the difference” can be a mechanism for safeguarding the legitimacy of the institution and, hence, we might well expect that when chief judges, those actors with the greatest responsibility for and sensitivity to issues of institutional legitimacy, author majority opinions, mixed outcomes are more likely. Stated formally:

Of course, these are not the only factors likely to structure the probability of a mixed outcome. Other variables related to both the nature of the decision makers (in this case, three-judge U. S. Courts of Appeals panels) and the nature of the case no doubt matter as well. In the following section, we articulate a set of more routine explanations for mixed outcomes.

Controlling for Alternative Explanations of Mixed Outcomes

A more mundane set of explanations for mixed outcomes has to do with the opportunity presented by each case. First, cases characterized by greater complexity offer more opportunities for mixed outcomes. Complex cases offer more grounds for disagreement both between litigants and among judges. Second, the presence of cross-appeals is likely to be associated with mixed outcomes because it reflects the fact that both parties are unhappy with some element of the judgment below. In that situation, circuit court judges have more opportunity to render alternative dispositions for separate issues on appeal; e.g., a court may find for a petitioner in terms of liability but for the respondent in terms of the amount of damages. Third, cases come to the U. S. Courts of Appeals after having received different treatments below. Some cases involve appeals after a jury or bench trial. In other cases, however, the decision being appealed is more limited in the sense of being an appeal from a summary judgment or some sort of pretrial judgment, dismissal, or injunction (or denial thereof). In these latter cases, the issue under appeal is, on average, really quite limited and therefore offers less of an opportunity for a mixed outcome compared to those cases appealed after a trial (in which a full set of issues has been litigated). For example, a summary judgment generally involves the question of whether sufficient factual issues exist to take the case to trial, or whether one or the other of the parties should win as a matter of law. By contrast, trials may raise any number of legal issues related to evidentiary rulings, jury instructions, and pretrial motions, among many others. Hence, they are more likely to generate appeals where split outcomes may result.

Collectively, the considerations discussed above with regard to opportunity can be more formally stated as follows:

Finally, we consider the issue of workload and its potential to decrease the likelihood of mixed outcomes. The U. S. Courts of Appeals handles a huge (and seemingly ever-increasing) volume of cases each year (Reference HowardHoward 1981; Reference PosnerPosner 1985). For example, from October 2003 through the end of September 2004, the courts of appeals collectively disposed of 27,438 appeals on the merits, with another 28,943 appeals handled via consolidation or procedural termination, for a total of more than 56,000 appeals terminated (Administrative Office of the Courts 2004: Table B-1). During this same period, there were almost 63,000 cases commenced. Compare this with the figures from 1984, when 31,490 appeals were filed and 31,185 appeals were terminated (Administrative Office of the Courts 1984: Table B-1). The heavy workloads with which circuit court judges contend are likely to reduce the probability of mixed outcomes because structuring such an outcome is likely to be more time-consuming. Hence, we hypothesize that:

Having now specified a comprehensive model of mixed outcomes, we turn in the next section to a discussion of the data and methods we use to subject this model to empirical verification.

Data, Methods, and Estimation

Independent Variables

We generated most of our independent variables from information contained in the U. S. Courts of Appeals Database as well. First, guided by the literature on party capability and litigant resources (Reference GalanterGalanter 1974; Reference WheelerWheeler et al. 1987; Reference SheehanSheehan et al. 1992; Reference Songer and KuerstenSonger, Kuersten, et al. 2000; Reference CollinsCollins 2004), we considered private businesses, the federal government, and state governments to be high-capability parties. We identified cases in which both appellant and respondent were one of these high-capability parties using the GENAPEL and GENRESP variables in the U. S. Court of Appeals Database.Footnote ⁹ Where both appellant and respondent were a private business, the federal government, or a state government, we coded this variable coded as 1 (and 0 otherwise). The variable reflecting the presence of a governmental litigant was likewise taken from the GENAPEL and GENRESP variables. For our purposes, we used that information to create a variable coded as 1 if the case involved a government party (and 0 otherwise).

Turning to ideological divergence, we measured the ideological divergence on a panel as the absolute difference between the most liberal and the most conservative judge on the decisionmaking panel based on scores developed by Reference GilesGiles et alia (2002). These scores are derived from Poole's common-space scores (see Reference PoolePoole 1998). A judge's score is based on the appointing president's score when senatorial courtesy is absent and on the senator's (or senators', as the case may be) score(s) when senatorial courtesy is present, thus more accurately representing the process by which lower federal court judges are selected compared to more traditional measures (such as the party of the appointing president). Given how we used the Giles-Hettinger-Peppers scores to measure ideological diversity on a panel, greater values represent greater ideological diversity. To evaluate the effect of separate opinions on ideological divergence, we created a dichotomous variable coded as 1 when a separate opinion was present and 0 otherwise, based on information reported in the U. S. Courts of Appeals Database.Footnote ¹⁰ We then created a multiplicative term including these two variables to evaluate the conditioning effect of a separate opinion on the impact of ideological disagreement on the likelihood of a mixed outcome.

Recall that one of our theoretical arguments is that concern for legitimacy may induce split decisions. Certain kinds of cases, especially those that are highly salient, are the sort of case most likely to induce sensitivity to issues of legitimacy for judges. As discussed above, some cases simply deal with issues that are typically of greater salience (to judges, politicians, and the public). Cases involving a civil rights or civil liberties claim are just these sorts of cases. Thus we created a simple dichotomous variable reflecting whether the case raised such an issue (coded as 1) or not (0).Footnote ¹¹ We also created a dichotomous variable indicating the presence of amicus curiae briefs (1 if there were amicus briefs filed, and 0 if not), given that such briefs are also indicators of salience.Footnote ¹² In terms of visibility (another aspect of salience), we relied on a variable indicating a prior court's decision to publish an opinion in the case. Given that district court judges publish only a small percentage of their opinions in the West Reporter system, we expect that trial court judges are more likely to publish opinions in the most important cases. We thus created a variable to reflect whether the district court or any other lower court (such as the tax or bankruptcy court) published an opinion, coded as 1 if such a published opinion existed and 0 otherwise.Footnote ¹³ We also used information from the Federal Judicial CenterFootnote ¹⁴ to identify chief judges and then used that material in conjunction with information in the U. S. Courts of Appeals Database regarding the majority opinion writer in each case to create a variable indicating whether the chief judge authored the majority opinion (coded as 1 if the chief judge did author the majority opinion, and 0 if not).

Finally, in coding the various factors that presumably simply provide greater opportunities for disagreement, we used a variety of information (transformed as necessary) coded in the U. S. Courts of Appeals Database. First, we created an index of case complexity based on the number of issues raised and the length of the opinion using factor analysis. (See Appendix A for the details regarding how we created this index and its constituent components.) Second, we likewise easily derived the presence of cross-appeals from a variable in the database specifically designed to identify cross-appeals (CROSSAPP). The resulting variable was coded as 1 if cross-appeals were present and 0 otherwise. Third, to assess our hypothesis regarding complexity as signified by the nature of the proceedings below, we used a variable in the U. S. Courts of Appeals Database that reports exactly the nature of that proceeding (APPLFROM). Recall our argument that, when the case goes to trial, complex evidentiary and other legal issues are likely to increase the complexity of any appeal from that proceeding. To capture this notion, we used the APPLFROM variable to create a dichotomous variable coded 1 if the case below went to trial (bench or jury) and 0 otherwise. Finally, we took our measure of workload—calculated as the number of merits terminations per judge by circuit and year—from information available from the Administrative Office of the United States Courts. We report complete descriptive statistics for all of the variables discussed above in Table 1.

Table 1. Descriptive Statistics

Statistical Method

Since our dependent variable is dichotomous—either there was or was not a mixed outcome—we selected logit as our estimation technique (Reference Aldrich and NelsonAldrich & Nelson 1984). Past research has amply demonstrated that decisionmaking by a panel is not entirely independent in either a spatial or temporal sense. That is, decisionmaking by one panel in a given circuit is related to decisionmaking by other panels in that circuit, and each circuit manifests trends in decisionmaking over time (see, for example, Reference Songer and SheehanSonger, Sheehan, et al. 2000; Reference CohenCohen 2002). To account for this, we used robust standard errors clustering on circuit-year. We also employed the weights required for use with the U. S. Courts of Appeals Database given its sampling structure.Footnote ¹⁵ The results of our logit estimation are reported in Table 2, while changes in predicted probabilities based on changes in the values of each of our statistically significant variables are reported in Table 3. While some of the changes in predicted probability appear modest, we note that given the large population of appeals court decisions (as well as the large sample size in our database), a change of even 2 percent can translate into a shift in outcome in hundreds of cases.

Table 2. Logit Estimation of Mixed Outcomes at the U. S. Courts of Appeals

^a Robust standard error clustering on circuit-year.

^b One-tailed test.

Table 3. Changes in Predicted Probabilities of a Mixed Outcome^a

^a Changes in predicted probabilities are calculated based on parameter estimates reported in Table 2, with all variables set at their mean or modal values unless otherwise specified.

^b Δs=change in predicted probability as variable value changes from one standard deviation below the mean to one standard deviation above the mean (for continuous variables).

^c 0→1=change in predicted probability as variable value changes from 0 to 1 (for dichotomous variables).

Results

The results of our logit model provide empirical support for the majority of the relationships we hypothesized. First, consider the results pertaining to our party capability hypotheses. When there are two high resource litigants facing off against one another, the likelihood of a mixed outcome is enhanced, as indicated by the positive coefficient (0.181) and the associated statistical significance (albeit at a more generous 0.064 level than the conventional 0.05 level). In substantive terms, if all the other variables in the model are set at their mean or modal category, the predicted probability of a mixed outcome increases by 0.11 when two high-resource litigants face one another compared to when that is not the case, as reported in Table 3. However, when one of the litigants is a government party, the likelihood of a mixed outcome is diminished, with the probability decreasing by 0.02.

Second, with regard to our hypotheses vis-à-vis strategy and mixed opinions, the empirical evidence is consistent with the idea that ideological diversity increases the likelihood of a mixed outcome but that the impact of ideological diversity is conditioned on the presence of a separate opinion. As reported in Table 3 (calculated on the basis of the relevant parameter coefficients appearing in Table 2), in cases where no separate opinion is filed, increasing ideological diversity is associated with an increase in the likelihood of a separate opinion. On the other hand, when a separate opinion is present, the impact of ideological diversity on the likelihood of a split decision decreases, indicating that in those situations, the majority has not sought to accommodate the minority member on the panel by splitting the difference between the two parties. As reported in Table 3, as ideological diversity is allowed to vary from one standard deviation below its mean to one standard deviation above its mean, it decreases the likelihood of a mixed outcome by 0.02 when a separate opinion is filed but increases the likelihood of such an outcome by 0.01 when no separate opinion is filed.Footnote ¹⁶

Turning to the variables intended to evaluate situations in which legitimacy is likely to be of greater concern, three of the four variables related to salience (the presence of a civil rights or liberties claim, prior publication, and chief judge as the opinion writer) perform as expected. In each case, the effect is to enhance the likelihood of a mixed opinion by approximately 0.02, 0.01, and 0.2, respectively. The influence of having the chief judge as the majority opinion writer is comparable to that of the presence of a civil rights or liberties claim, increasing the probability by 0.02. This comports with our a priori speculation that the chief judge is the actor who should be most likely to be sensitive to issues of legitimacy.

In addition to these results, all but one of the variables intended to capture the opportunity for mixed outcomes do, in fact, affect the likelihood of a mixed outcome in the hypothesized direction. More complex cases, those with cross-appeals, and those that are before the appellate court after having been disposed of by trial in the lower court are all more likely to result in mixed outcomes. For example, when the complexity variable is allowed to vary from one standard deviation below its mean to one standard deviation above its mean, the likelihood of a mixed outcome increases by 0.05. The change in predicted probability associated with the presence of cross-appeals and appeal after a trial are 0.10 and 0.04, respectively. As for workload, it produces a significant coefficient, but in the opposite direction than originally hypothesized. That is, increased workload actually increases the likelihood of a mixed outcome: as workload is allowed to vary from one standard deviation below its mean to one standard deviation above its mean, the likelihood of a mixed outcome increases by 0.03. This result suggests that perhaps mixed outcomes provide a more expeditious way to dispose of appeals as judges, pressed for time, seek to quickly accommodate all members of the panel as well as the litigants rather than thrashing out a single result that favors one party only.

Though our measure of case complexity takes into account both threshold and substantive issues, to be certain that the results reported in Table 2 are not purely a function of threshold issues, we re-estimated our model excluding those cases involving threshold issues. The results of that analysis (reported in Appendix B) demonstrate that the results are not an artifact of the threshold issue, as all of the parameter coefficients remain signed in the correct direction, though the parameter coefficients corresponding to high-resource opponents, civil rights/civil liberties claims, and the chief justice as the majority opinion writer are no longer statistically significant at conventional levels.Footnote ¹⁷

Though most of the variables that achieve statistical significance change the likelihood of a mixed outcome somewhat modestly (although given the number of cases decided in the courts of appeals, their impact may nonetheless affect hundreds of cases), when we consider the likelihood of a mixed outcome under various substantively interesting scenarios, the differences can be quite remarkable. For example, when conditions are not especially conducive for a mixed outcome (i.e., dichotomous variables are set at the value that makes them less likely to result in a mixed outcome and continuous variables are set one standard deviation below their mean values), the likelihood of a mixed outcome is a mere 0.02. When, however, conditions are more favorable for a mixed outcome (i.e., dichotomous variables are set at the value that makes them more likely to result in a mixed outcome and continuous variables are set one standard deviation above their mean values) the likelihood of a mixed outcome skyrockets to 0.80. We can also make comparisons of the predicted probabilities that arise from less extreme scenarios. For example, when two high-resource parties square off and the majority opinion writer is a chief judge (with all other variables at their mean or modal values), the likelihood of a mixed outcome is 0.22, a probability more than 0.15 greater than that when those two conditions do not obtain. On the other hand, when one of the litigants is a governmental party and the majority opinion writer is not a chief judge, the probability of a mixed outcome is a mere 0.04.

Discussion and Conclusion

The results of our model reveal, first, that the decision to affirm in part and reverse in part is, to some degree, a predictable event based on the characteristics of individual cases. Mixed outcomes, which have generally been ignored by scholars of judicial politics so as to construct dichotomous dependent variables, are themselves interesting behavioral phenomena on these federal appellate courts. Second, these outcomes are predictable in substantively interesting ways. To explain these unique case dispositions, we relied on several theoretical foundations to generate hypotheses regarding the likelihood that an appellate panel would rule in this more “ambiguous” fashion.

First, we drew on the party capability literature to consider how litigant types might matter. In particular, we considered the fact that, when two litigants that enjoy high levels of capability compete, each has the resources at its disposal to craft strong and persuasive arguments. Hence, each side has the potential wherewithal to wrest a partial victory (i.e., a mixed outcome), however much less desirable than a complete victory, from the decisionmaking panel. We also took into account the extensive evidence in the extant literature on party capability that demonstrates the particular strength of governmental litigants, a strength that gives governmental litigants a decisive edge against all opponents and, accordingly, makes it more likely for them to win in toto (and concurrently less likely for a mixed outcome to occur). Our empirical evidence is strongly consistent with both of these propositions.

Second, we considered the insights from the literature on strategic behavior on the part of judges. As on the U. S. Supreme Court, judges on the federal appeals courts often accommodate the other judges on the panel so as to preserve a three-judge majority (Reference CohenCohen 2002). In the Supreme Court, this accommodation may take on certain strategic dimensions (Reference MaltzmanMaltzman et al. 2000). Thus we speculated that the same may hold true in the appeals courts, by judges accommodating ideologically diverse colleagues through production of a mixed outcome, which may represent a “middle ground” more palatable to judges of alternative ideological viewpoints. Indeed we did find this to be the case, with greater ideological diversity on a panel leading to a greater likelihood of a mixed outcome. But that effect is conditional in the sense that, when a separate opinion has been filed, ideological diversity decreases rather than increases the likelihood of a mixed outcome since that separate opinion represents a failure in the accommodation attempt.

Third, we relied upon a theory of judicial behavior primarily advanced by Shapiro and Stone Sweet suggesting that judges will seek to preserve their institutional legitimacy and ensure compliance with their decisions by rendering judgments that mollify the loser in some fashion (Reference Stone SweetStone Sweet 2000, Reference Stone Sweet and Stone Sweet2002; Reference Shapiro and Stone SweetShapiro & Stone Sweet 2002). Although judges have other means at their disposal to enhance their legitimacy, one method they may choose is to render moderated judgments that do not completely satisfy one party's position at the total expense of the other party. That is, judges may render judgments that favor one party or the other only in part. Such judgments fall into the category of the mixed outcomes that we have modeled on the U. S. Courts of Appeals.

While the Shapiro/Stone Sweet theoretical proposition seems reasonable, empirical tests of the theory using large-scale data sets have not yet been undertaken. Our research design and model estimation therefore present an important first step in evaluating this intriguing theory. Of course, because not all decisions rendered by the federal appeals courts result in such mixed outcomes, we extended the theory so as to generate hypotheses concerning the circumstances in which such outcomes will most likely arise given underlying theoretical assumptions involving legitimacy and compliance. Accordingly, we identified those cases in which we surmised judges would be most sensitive or concerned about the loser's reactions to the court's decision. Such situations would, we believe, arise in salient cases, and the evidence we report supports that theoretical story. Further, we also speculated (and found evidence consistent with the proposition) that chief judges, with their presumed greater sensitivity to issues of institutional legitimacy in light of their institutional responsibilities, would be more likely to author opinions with mixed outcomes when they serve as majority opinion author.

Do our results mean that judges think consciously about producing outcomes that will reduce the risk of noncompliance in highly salient cases? Not necessarily. Nor do we have any direct evidence to suggest that judges choose to render such outcomes with legitimacy or compliance specifically in mind. Yet it does seem reasonable to assume that judges are sensitive to the delicate balance they must accommodate as unelected dispute-resolvers in the federal system. Certainly, in rendering the decision in Brown v. Board of Education (347 U.S. 483 [1954]), Supreme Court justices were cognizant of the importance of a united front when presenting a controversial ruling to the public. Judges on the courts of appeals are no doubt similarly cognizant that case dispositions have the potential to affect litigants' perceptions of the court's legitimacy. Rendering a mixed outcome may be one mechanism such courts use—consciously or unconsciously—to preserve that critical institutional resource.

In conclusion, our results serve to further strengthen integrated theories of judicial decisionmaking as the most feasible approach to understanding judges' behavior. In the U. S. Courts of Appeals, where judges must handle thousands of appeals each year, where they must write opinions in shifting panels of three judges, and where their docket is nondiscretionary, the decisions they produce are likely to reflect the influence of a variety of institutional, behavioral, and contextual variables. Our findings indicate that appeals court judges appear to be sensitive to the nature of the litigants before them and the arguments they make, to the interpersonal dynamics among the three judges on the panel, and to the legitimacy of their own institution. Within this complex decisionmaking environment, judges fashion results that allow them to negotiate a variety of goals—including their personal preferences, their colleagues' preferences, pressure from litigants, and enhanced institutional legitimacy.