BRINGING THE LAW BACK IN: FINDING A ROLE FOR LAW

IN MODELS OF SUPREME COURT DECISION-MAKING

Herbert M. Kritzer

J. Mitchell Pickerill

Mark Richards

INTRODUCTION

Joseph Oncale worked on an offshore oil rig. In 1991 he filed suit against his employer, claiming that the words and conduct of three other men who also worked on the oil rig constituted sexual harassment which was prohibited as discrimination under Title VII of the Civil Rights Act of 1964. The trial court and the Fifth Circuit Court of Appeals rejected his suit, ruling that Title VII did not extend to same-sex harassment. The Appeals Court was reversed by the Supreme Court in Oncale v. Sundown Offshore Services, Inc. (1998). The Court unanimously ruled that the prohibition against workplace harassment under Title VII is broad enough to include same-sex harassment. Justice Scalia wrote the majority opinion, which all eight of the other Justices joined (Justice Thomas also wrote a one sentence concurrence). Justice Scalia based his reasoning on the plain meaning of Title VII: It prohibits discrimination "because of . . . sex." Thus, for Scalia and the rest of the Court, the statute is clear and there is no reason for excluding harassment because the harasser happens to be the same sex as the harassee.

Wait a minute. Have not political scientists shown that this cannot be? A conservative justice writing an opinion for a liberal decision in which all members of the Court - conservative, moderate and liberal - join? Say it ain't so, Joe (or Harold and Jeff). The attitudinal model posits that "attitudes" explain the Justices' votes. Conservatives vote for conservative outcomes, and liberals vote for liberal outcomes. Moreover, the attitudinal model's proponents argue that it is strongest when applied to cases involving civil liberties and civil rights. According to the attitudinal model framework, unanimous decisions arise only for a case at the extreme end of the ideological perspective. In this case this would mean that the lower courts' decisions were so extremely conservative that even Scalia and Thomas found the lower courts' position as unpalatable; arguing that the lower courts decisions were far to the right of Scalia and Thomas would stretch the meaning of conservatism to the breaking point.

So how did Oncale happen?⁽¹⁾ Did the conservative Court fear reversal by the Republican Congress? Or, might it be possible that the justices on the Court felt constrained by the "plain meaning" of the statute? That is, might it be the case that law matters and, at least sometimes, trumps attitudes? That is the question we address in this paper.

There is no doubt that in the Court's written opinions, and in the briefs and arguments presented to the Court by lawyers appearing before it, the law is discussed as if it were the central determinant of how decisions are made. Despite this, some political scientists who apply empirical research tools to the study of Supreme Court decisions argue forcefully that it is the attitudes of the justices that matter not the law (Segal and Spaeth 1993; Segal and Spaeth 1996). In this paper we present an argument that law is not simply window dressing. While we do not reject the importance of attitudes, we show that attitudes are limited in their ability to explain a significant portion of the votes cast by justices, which could readily be explained on the basis of law. More importantly, we show two ways in which law does impact the justices' votes in systematic ways that are amenable to empirical tests.

For 50 years, starting with the work of C. Herman Pritchett (1948), political scientists have argued that attitudes are an important determinant of justices' votes, and on this point there is what can fairly be described as a consensus. The disagreement is over the question of whether attitudes comprise "a complete and adequate model of the Supreme Court's decisions on the merits," the position advanced by Segal and Spaeth in The Supreme Court and the Attitudinal Model (1993, 11). Segal and Spaeth argue, both in their book and in later writing (Segal and Spaeth 1996) that in the absence of evidence showing that competing explanations, particularly what they label the "legal model," have any efficacy, one must accept the adequacy of the attitudinal model. Segal and Spaeth do not claim that only attitudes matter in the votes of justices. Rather, they contend that there is no evidence of any other systematic influence on the justices' decisions (Segal and Spaeth 1994, 11).

The unabashed nature of the claims for the attitudinal model leaves many political scientists (to say nothing of many, if not most, legal academics, political leaders, and attentive citizens) very troubled (Brenner and Stier 1996; Brisbin 1996; Knight and Epstein 1996; Lawrence 1994; Songer and Lindquist 1996). Where is law in all of this? The courts, including the Supreme Court, are supposed to be guided by the law. Segal and Spaeth go to substantial lengths to show that standard elements of legal explanations--plain meaning, legislative/framer's intent, precedent, and balancing--fail to provide consistent, systematic explanations of decisions (Segal and Spaeth 1993, 33-53; Segal and Spaeth 1996) . For purposes of the discussion that follows, we presume that their argument about these images of the "legal model" is in fact correct, although there is some evidence that suggests, that at least with regard to precedent (stare decisis) their data may not support their argument as well as is apparent from extant discussions (see Nelson 1998). Segal and Spaeth answer the "where is law" question very directly: it is not to be found in the Supreme Court's decisions on the merits, even though the justices go to great lengths to dress up their decisions so that law appears to matter. For Segal and Spaeth, attitudes are not simply one of several systematic influences on Supreme Court decision-making on the merits; rather, attitudes are the only systematic influence: the justices simply follow their attitudes where ever those attitudes lead them. We label this the unconstrained attitudinal model.

While some may choose to confront Segal and Spaeth's argument straight-on (Brenner and Stier 1996; Epstein and Kobylka 1992; Knight and Epstein 1996; Songer and Lindquist 1996) (Nelson 1998), that is not our purpose here. In fact, we simply assume that their argument about the legal model as they have defined it is in fact correct. In the discussion that follows we seek to identify other ways in which we can bring something reasonably labeled "law" back into the understanding of the Court's decision process. We do not reject even the primacy of attitudes. Rather we ask whether the unconstrained attitudinal model should be modified by adding law-based constraints that improve our understanding of the decision-making process? We believe that it is extremely helpful to think of law as constraint. Furthermore, where it was difficult, if not impossible, to formulate systematic, empirical tests of the legal model defined in terms of precedent, plain meaning, etc., we develop and present some clear empirical tests for the law-constrained attitudinal model. The results of those tests support the argument that factors other than attitudes have a systematic role in decisions, and these factors are consistent with the idea of law as constraint on the justices' decisions.

One way to think about what we are doing is that we seek to do for law what Epstein and Wright (1998) have done for institutional constraints. Epstein and Wright do not reject the importance of attitudes, but try to show that justices are not able to pursue their policy goals in an unconstrained fashion. For Epstein and Wright the core constraints are institutional: e.g., justices must secure majority agreement to realize their policy goals, and this results in a need for strategic considerations. We should note that Epstein and Wright do treat "law," particularly precedent (stare decisis) as constituting a type of institutional constraint; they show that the Supreme Court has overruled only a minuscule fraction of the cases it has decided (1998, 176-7; see also Knight and Epstein 1996, 1031), and that justices in their communications with one another also rely very heavily on precedent (pp. 165-72). While these are strong arguments that stare decisis is an important element in the process of deciding cases, Knight and Epstein's analysis does not show in a systematic way that justices are deciding cases differently than they would have in the absence of specific precedents or in the absence of the doctrine of stare decisis.

THE LIMITS OF THE ATTITUDINAL MODEL

The facts of a case and strategic concerns can serve to constrain attitudes in the decision making process. The strategic concerns identified by Epstein and Knight are a good example of how difficult it can be to demonstrate empirically factors that may limit attitudes. Hence, as they argue, the rational choice approach to studying interactions and voting on the Court may be promising in adding to our knowledge how it reaches decisions in cases. Attitudinalists began by arguing that ideology explains votes. They do accept the effect that fact patterns have on votes, but see the facts as positioning cases in the justice's attitudinal space (Rohde and Spaeth 1976). They are reluctant to accept that various factors may act to constrain attitudes. If attitudes go unconstrained, we should not be able to find voting patterns which are inconsistent with the attitudinal model. Unanimous cases provide one example of a voting pattern which seems to run counter to the logic of the attitudinal model. How is it that Justices with varying "attitudes" reach unanimous outcomes if those attitudes go unconstrained?

We thus begin our inquiry into the constrained attitudinal model by analyzing Supreme Court outcomes which provide difficulties for the unconstrained attitudinal model. If law and other factors do not constrain the Justices, then we should be able to find evidence that attitudes and attitudes alone explain unanimous cases as well as they explain nonunanimous cases alike.

Unanimity and the Attitudinal Model

It has been argued that unanimous cases occur because "presumably the facts and the law are so clear that no opportunity is allowed for the [attitudes] of the justices to lead them to opposing conclusions" (Pritchett 1941, 890). This has also been called the "two situation model," where Justices elect to hear a case on the merits or actually decide the merits of the case depending upon how clear of a legal position is presented or how concerned they are about the direction of the outcome (Brenner and Arrington 1987; Perry 1991). Modern Courts have consisted of a mix of liberal, conservative and moderate justices (see Segal and Cover 1989; Segal et al. 1995). And yet, they consistently reach unanimous decisions in more than one-third of all decisions on the merits; in fact, in the 1995 term, 45% of the decisions on the merits were unanimous.⁽²⁾

The attitudinal logic regarding unanimous cases suggests that the outcomes in unanimous cases are simply extreme; in other words where Justice Scalia, the most conservative member of the Court, joins the other Justices in a unanimous liberal decision, he is doing so because the conservative outcome would have been so extremely conservative so as to fall far to the right of him on a unidimensional scale of conservative and liberal values.The line in Figure 1 represents the position of the Justices according to the attitudinal model. Justices will, according to the attitudinal model vote for the outcome (J1 or J2) which comes closest to their "ideal point," typically referred to as "I" points (and shown in Figure 1 by the justices' names) - so in the above case, Scalia would choose J1, as would the other Justices, even though it is the more "liberal" outcome (see Baum 1992; Spaeth 1995). Consequently, all nine Justices are voting in accordance with their attitudes, according to the above argument. Nonetheless, to date, no one has systematically analyzed unanimous cases to determine whether this is so. Probably, some cases do in fact fall to the right of Scalia or to the left of Brennan, but it seems unlikely that 45% (or even a third) of the cases a Court hears on the merits, most of which the Court has voluntarily chosen to hear, fall to the extremes of the attitudinalists' ideological scale.

For a unanimous outcome to occur, the alternative outcome, or "J2" must be more than twice the distance from J1 than it is from the nearest extreme Justice. Figure 2 represents this scenario using the scales developed in the Segal et al. (1989; 1995) ratings of Justices' attitudes. In essence, if J1 is a 0, it is one spatial unit from the most extreme Justices. Thus, for the most extreme Justices to vote for J1, the outcome represented by J2 must be more than 2 units from J1.

Either we have to believe that four Justices are voting to hear extreme cases (in deciding writs for certiorari) or that the spatial distance between the "J" points is extreme. Either situation begs the question, why would the Justices vote to hear the case? It is possible that there are some lower courts and judges which themselves are ideologically extreme outliers, and that Justices may hear cases to "correct" those extreme lower courts. But this position only makes sense for cases in which the Supreme Court reverses the lower court, and actually does "correct" the decision - it cannot account for cases in which the Court affirms the lower court's decision.

Additionally, the attitudinalists' charge that the legal model is NOT empirically falsifiable now applies to the attitudinal model in unanimous cases. Somehow, the "J" points, or the alternative case outcome (reverse or affirm, for instance) would need to be capable of being coded on a scale that corresponds to the Segal et al. scale, which places Justices between -1 and 1 on a liberal to conservative dimension (See Figure 2). It is difficult to imagine a satisfactory measure of the degree of conservativeness/liberalness of case outcomes. Because the distance among the J and I points is relative - thus we do not know how far apart they truly are and to date do not have a numerical scale separate from actual decisions for the J points (and may never have such a scale) - the theory is not falsifiable for unanimous cases. This is a charge the attitudinalists have leveled against the orthodox legal model, and now, in the case of unanimous cases, applies to the unconstrained attitudinal model.

Another argument regarding unanimous cases notes that many unanimous cases are not consensual; that is, Justices may agree on the outcome, but concurring opinions are filed suggesting differences in reasoning or application of the law in the case (Spaeth 1989). However, it is not clear whether the concurrences are filed because of the application of the law or for other reasons. For instance, a Justice may agree with the majority reasoning but also think that there is an additional issue presented in the case which the court is ignoring, or a Justice may be concerned about how the case may be applied by lower courts, and thus file a concurrence in order to clarify, primarily by dicta, how the case is intended to be applied by lower courts. In any event, the lack of consensus does not logically lead to the conclusion that it is attitudes and attitudes only that explains the outcome. In fact, the attitudinal model seems unconcerned with consensus or lack of it. It simply posits that attitudes explain and predict votes. Irrespective of the reasoning applied, then, if a conservative Justice joins a unanimous court in a liberal outcome, contrary to the attitudinal model, we must try to determine what it is that might be causing him or her to do so.

Attitudinalists have argued merely that unanimous cases cannot be explained by the legal model (Brenner and Arrington 1987; Spaeth 1989). They argue that the "two situation" model, wherein Justices sometime operate in an outcome mode and sometime in a legal mode, does not adequately account for unanimous cases. While unanimous cases may not be explained by the "two situation model," the only alternative is not an unconstrained attitudinal model. To the contrary, unanimous cases do not fit nicely into an unconstrained attitudinal model.

Unanimous cases thus provide a potential problem for attitudinalists. Some unanimous cases may be explained by the "extreme ideological outlier" theory. We do not suggest that unanimous cases can never be explained by attitudes. For our analysis, we give the unconstrained attitudinal model the benefit of the doubt by exploring which unanimous cases may really be explained by the unconstrained attitudinal model.

It is certainly conceivable, and likely, that lower court judges reach "extreme" outcomes, which the Supreme Court must "correct." This would mean, however, that the Supreme Court would reverse the lower court's ruling. If a lower court's ruling was as extreme as the spatial model suggests it would have to be for all nine Supreme Court Justices with significantly different "I" points to agree on the outcome, a unanimous affirmation of the lower court should almost never occur. So for step one, we accept for the sake of argument that the logic of the unconstrained attitudinal model may explain those cases which the Court reverses.

Where a conflict in circuits exists, it may be that the Supreme Court is affirming the more moderate decision, and in effect, reversing the court whose decision was ideologically extreme, but not formally decided by the Supreme Court. For instance, if the Ninth Circuit decides issue X in Smith v. United States with a moderately liberal decision, but the Fifth Circuit conflicts with the Ninth Circuit and decides issue X in United States v. Jones with an extremely conservative decision, the Supreme Court could correct the Fifth Circuit's extreme outcome by either hearing Smith and affirming, OR by hearing Jones and reversing. It's simply a matter of which case gets to the Court first. Therefore, for step two, we accept for the sake of argument that the logic of the unconstrained attitudinal model may explain those cases where the Supreme Court affirms a case for which a conflict in circuits exists.

Finally, given the fact that the Supreme Court docket is so discretionary, it makes sense that the Court may vote to take a case on certiorari because the members of the Court believe that the extreme decision of the lower court needs to be corrected. So for step three, we posit that the subset of cases which pose the greatest challenge to the logic of the unconstrained attitudinal model are cases selected by the Court on cert., with no conflicts in circuits, which the Court unanimously affirms. That is, it is possible that the unconstrained model explains cases, however they reach the Court, which are reversed OR which involve conflicts in circuits and are affirmed. However, if justices are exercising their votes with no constraints on their attitudes, we should not observe unanimous affirm cases which are heard on cert and do not involved conflicts in circuits.

Unanimous Cases: Data and Analyses

The data are taken primarily from the United States Supreme Court Judicial Database, compiled by Harold Spaeth; for one table we did supplement Spaeth's data with pre-1953 data from the Schubert data set.⁽³⁾ Most of our analysis focuses on decisions from the beginning of the Warren Court through the 1995 term of the Rehnquist Court, with one table extending back to through the Vinson court . We look at all orally argued cases with signed opinions, including per curium opinions; we limit some of our analyses to cases which the Court accepted on certiorari. There are 6098 observations for all orally argued cases in that period (not counting the Vinson Court cases), and 4722 observations for those taken on cert. only.

Since the beginning of the Warren Court, 38% of all cases orally argued and decided on the merits with full opinion have been unanimous. There is some variance across natural courts, but the split is fairly consistently one of approximately 2/3 nonunanimous and 1/3 unanimous where a significant number of cases were heard within the natural court (See Table 1). Moreover, the Warren, Burger and Rehnquist courts also share very comparable unanimity percentages (37%, 37% and 40%, respectively; see Table 2).

The occurrence of unanimous cases across issue areas does vary, but substantial percentages of outcomes in each issue area are unanimous (See Table 3). The difference among issue areas raises some important questions. Do some issue areas provoke more intense ideological responses, and conversely, do other issue areas raise constraints to attitudes, such as clear legal language or institutional legitimacy? Most issue areas which have higher percentages of unanimity are non-civil liberties issues (The first five issue areas in Table 3 are usually included in "civil liberties and civil rights" variables used by political scientists for analysis).

As Brenner and Arrington found, most unanimous cases are decided in the liberal direction. It is clear in Table 4 that conservative justices are much more likely to join a unanimous court in a liberal outcome than liberal Justices are to join a unanimous court in a conservative outcome.⁽⁴⁾ Additionally, when we look at the proportion of unanimous liberal votes by individual Justice from the Vinson Court through the Rehnquist Court, in Table 5 we can see that even the most conservative Justices vote for a liberal outcome with a unanimous Court far more frequently than they vote for a liberal outcome in nonunanimous cases. In fact, Figure 3 shows fairly dramatically that there is no relationship between liberalism in nonunanimous civil liberties cases and unanimous civil liberties cases, and that part of the reason for this is that no justice is predominantly conservative in his or her unanimous civil liberties decisions.

Next, we examine the cases in which the Court affirmed or reversed the lower court whose decision was being reviewed. We selected only those cases the Court heard on writs of certiorari, in order to test the hypothesis that unanimous cases result when members of the Court select cases on cert in order to "correct" the extreme lower court decisions.⁽⁵⁾ If the unconstrained attitudinal model is correct, the Court should not be selecting cases and unanimously voting to affirm them. As we can see in Table 6, the Court does unanimously affirm lower courts. 513 (28.5%) of the 1798 unanimous cases which were selected by writs of certiorari were affirmed, compared to 1014 (34.7%) of the 2924 nonunanimous cases.

We also examined the data after controlling for conflicts in circuits. As noted above, it might be argued that in a case where there is a conflict in the circuit courts, the Supreme Court heard the case with the less extreme lower court decision, and affirmed the less extreme outcome, and thus in effect, it reversed the extreme court whose decision the Justices did not directly review. We selected those cases for which a conflict was noted. Once again, the patterns of nonunanimous cases and unanimous cases vary slightly (see Table 7), but still of those cases where no conflict in circuits is reported, over 25% of all unanimous cases are affirmed, and of all the cases the Court hears on cert which have no reported conflicts in circuits, over 30% of the affirmed cases are unanimous. The pattern is consistent with the other unanimity patterns described above.

Finally, we look at how unanimity varies in issue areas when controlled for conflicts in circuit. Even advocates of the unconstrained attitudinal model admit that the model explains some cases better than other. The unconstrained attitudinal model is at its strongest for cases involving civil liberties and civil rights. Therefore, we should observe few unanimous outcomes in cases which involve civil liberties or civil rights, report no conflict in circuits and are affirmed. However, Table 8 shows that these cases do in fact occur with some frequency. Of the 1964 civil rights cases taken on cert with no reported conflict in circuits from the Warren Court through the 1995 term of the Rehnquist Court, 607 (24%) have been unanimously affirmed. The total percentage of unanimous cases in this subset of cases is slightly smaller than the pattern observed for other types of cases, but this is a subset of cases which the unconstrained attitudinal model predicts should not occur.

Discussion

Unanimity results pose problems for the unconstrained attitudinal model. Although the attitudinal model may, in theory, account for some unanimous cases, it is counterintuitive, at a minimum, that it accounts for nearly 40% of all cases the Court hears. Moreover, since the attitudinalists have been so demanding of the orthodox legal model, insisting that it be falsified, we should be equally demanding of the unconstrained attitudinal model with respect to unanimous cases. In our analysis, we look for different patterns of unanimity and look for explanations - attitudinal or other - which may account those patterns. ⁽⁶⁾ We even try to be as generous as we can by eliminating all the cases which the attitudinal model might account for by controlling for cases accepted on cert., conflicts in circuits and affirm/reverse outcomes. Unanimous decisions are consistently reached in over one-third, and nearly 40%, of all cases heard on the merits - from Chief Justice to Chief Justice, from natural court to natural court. It is clear that Courts with memberships that include the most liberal and the most conservative justices produce an abundant number of unanimous cases, contrary to what we would expect if the unconstrained attitudinal model held true.

As we note above, Segal and Spaeth admit that the attitudinal model does not do a good job of explaining cases in certain issue areas, like federalism or judicial power. In their analysis, the attitudinal model is strongest when applied to civil liberties cases. In fact, the Court appears the least likely to reach unanimous decisions in civil liberties cases. This could be for a number of reasons. Civil liberties invoke more intense attitudes than the other issue areas, or ideological preferences regarding civil liberties may be more defined. However, even in civil liberties and civil rights cases, the most liberal justices join with the most conservative justices in unanimous cases in about one-third of the time, depending on the specific issue area. While this pattern may not prove that law is constraining attitudes, it raises questions for the unconstrained attitudinal model. According to its proponents, the unconstrained attitudinal model best explains civil liberties case; and yet the Court consistently reaches unanimous decisions in these cases.

Next, we puzzle over the fact that most unanimous cases are decided in the liberal direction. For all cases decided on the merits from the Warren Court through the 1995 term of the Rehnquist Court, 44% of liberal decisions are unanimous while only 31% of conservative decisions are unanimous. And when we break it down by individual Justice, going all the way back to the Vinson Court, it is clear that conservative justices are more likely to vote for a liberal unanimous decision than a liberal unanimous one, while more liberal justices are more likely to vote liberal regardless of whether the case is unanimous or not. Once again, we do not claim that we've found the smoking gun for law, but instead we argue that the unconstrained attitudinal model cannot account for this pattern in votes.

We believe the overall pattern of unanimity is inconsistent with and unconstrained attitudinal model.⁽⁷⁾ But to give the unconstrained attitudinal model its due, we also selected out the cases which it seems plausible might be explained by the unconstrained attitudinal model. This is not an admission that those cases are so explained, but a precaution.

As we have argued, if unanimous cases really do result from extreme lower court decisions and are thus "attitudinal adjustments" to lower court decisions, the Court should never, or at least very rarely, vote to take a case on cert. then turn around and affirm the case. Yet, the unanimity pattern for those cases is consistent with the general pattern - of cases taken on cert, and subsequently affirmed by the Court, 34% are unanimous. Of cert cases that were reversed, 40% were unanimous. Moreover, since some cases may involve conflicts in circuits, which means that the particular case the Court hears might be the less extreme of multiple lower court decisions, others of which might be extreme, the Supreme Court's affirmance of such a case might really be reversing the extreme outcome of another lower court case. Once again, therefore, in an effort to be as generous as possible to the unconstrained attitudinal model, we control for reported conflicts in circuits. While the percentage of unanimous cases drops slightly when we look at cases with no reported conflicts in circuits, it remains consistent with the overall pattern of unanimity. Of those cases taken on cert with no reported conflicts in circuits, 30% have been unanimous. What we have tried to do here is narrow down the types of Supreme Court outcomes that should not occur if the unconstrained attitudinal model has the power its proponents suggest. It is not clear to us the extent to which the unconstrained version explains the cases we controlled for, but attitudes alone cannot explain Affirmed cert. cases with no reported conflicts in circuits. Even when we stack the deck in favor of the attitudinal model, we find that the Court reaches unanimous decisions when the attitudinal model say it should not. However, the data show that the unanimity pattern for cases selected on cert, with no reported conflicts in circuits, which are unanimously affirmed, is consistent with the general pattern we have observed. Once again, while we believe law and other factors are serving as constraints on Justices' attitudes in these case, we acknowledge that our findings do not prove that law is affecting outcome. However, we have identified a subset of cases which defies the logic of the unconstrained attitudinal model. Put simply, attitudes alone cannot explain the pattern we are seeing here.

As we have maintained throughout this paper, we are not arguing for a legal model, or against the attitudinal model, per se. We do not partake of the debate between an unconstrained attitudinal model and an orthodox legal model. Rather, our analyses suggest that a better explanation of Supreme Court decision making might involve constrained attitudinal model. In this section, we have analyzed unanimous cases, and reach the conclusion that attitudes alone cannot describe all unanimous cases. Our analyses of unanimous cases does not show law at work in Supreme Court decision making, but it does raise questions about the extent to which unconstrained attitudes alone can explain Supreme Court decision making. We now go on to explore how attitudes might be constrained.

A CONSTRAINED MODEL OF DECISION-MAKING

We start from the presumption that attitudes are important in decisions made by judges, and probably the single most important factor in Supreme Court decisions. The most simplistic form of the unconstrained attitudinal model can be represented by Figure 4a. This figure is too simplistic because it leaves out elements that Segal and Spaeth include in their statistical models; in particular, they do include case facts, but drawing from Rohde and Spaeth (1976), they argue that facts are part of the attitudinal model, because they can be seen as defining how a judge or justice perceives a case within his or her attitudinal frame (Segal and Spaeth 1993, 215-221). Under the unconstrained attitudinal model, each justice combines the facts of a case to determine where that case falls in his or her preference space. With the addition of the factual elements, the unconstrained attitudinal model should be drawn as in Figure 4b. The dotted box suggests that facts serve as a screen through which attitudes pass in route to formulating a decision. Fundamental to the role of facts in the attitudinal model is that facts function in an individualized fashion for each of the justices (see Segal 1986).

Figure 4c depicts a constrained attitudinal model. The idea of constraint is suggested by the large arrows that serve to clamp down on the impact of attitudes on decisions; the direct arrows from attitudes and law indicate that they directly influence decisions in their own right. In the case of the Supreme Court, it may be that law serves more a constraining function than having a direct impacts. In lower courts, law directly influences decisions (see Ashenfelter, Eisenberg and Schwab 1995; Songer 1987; Songer, Segal and Cameron 1994; Songer and Sheehan 1990). The curved, double-headed arrow between law and facts indicates that these two elements interrelate; that is, law defines which facts are relevant and facts may trigger different aspects of the law. When legal and/or factual constraints are maximal, they clamp down on the connection between attitudes and decision, thus closing off the usual influence. When the constraints are not maximal, they can be thought of as structuring the way that attitudes influence decisions. Note also that the factual elements are not treated here as a filtering element for attitudes, but as an independent factor. This analysis suggests a way to determine whether facts are simply a part of the attitudinal model (as argued by Segal and Spaeth (1993, 255), or whether they exert influences separable from attitudes and determined by legal factors. If facts are attitude driven, then the role of facts should be justice-specific: each justice will weigh the facts differently reflecting what the justice's attitudinal structure dictates as being relevant. In contrast, if facts are law-driven, the relevance of facts will depend on what legal doctrine says should or should not be considered, and the influence of facts on decisions should shift systematically with major decisions that establish the relevance and weighting of key facts.

We label the situation in which law defines the relevant facts to be considered in the decision process as the "doctrinal model." That is, legal doctrine serves to structure judicial decisions. This is different than saying that doctrine determines decisions. Doctrine is a decisional principle established in case law. Doctrine can be very broad, applying generally to cases before a court. One such doctrine is stare decisis. More often, however, the Supreme Court establishes specific doctrines to guide judicial decisions in a particular area of law. We are concerned with doctrine in this sense, which we define as a specific formulation of the analytic principle which should guide Court decisions in a particular area. Thus, doctrine is a decisional principle established in case law which is expected to influence judicial decisions through a structuring process. As a basic principle established to adjudicate the case in which the doctrine was established as well as future cases, a doctrine is not likely to be declared superfluous obiter dicta. The doctrine is established as a principle which should apply to the adjudication of all cases within a substantive area, so it is less likely to be distinguished than is ordinary precedent. Doctrine is not merely an analytic principle; it also indicates support for a normative position in a substantive area of law. Doctrine constitutes precedent but not in the same sense that we usually think of precedent as in principle binding the decisions of judges. But, as with other types of precedent, doctrine can be overturned or substantially limited in principle. If this happens to a doctrine, then it loses its status as doctrine.

It is critical to emphasize that the doctrine-constrained attitudinal model is a revision, rather than a rejection, of the attitudinal model. Doctrine does not determine the votes of Supreme Court justices. Doctrine influences those votes, but this model allows room for attitudes to play a major role. As suggested in the introduction, this conception of doctrine shares something in common with the work of Knight and Epstein (see also Epstein and Knight 1998, 163-77; 1996) on the norm of stare decisis. In their view, justices' decisions are motivated by their preferences, but they are constrained in efforts to establish their preferences by a norm favoring respect for stare decisis (Knight and Epstein 1996, 1032). The evidence they cite includes the citations of precedent by attorneys and justices, and the low percentage of cases overruled as a percentage of cases available for overruling.

Looking to contemporary work in legal theory, we see connections between our argument that doctrine structures the decisions of Supreme Court decisions but does not determine them due to the important role of the justices political attitudes, and the "moral reading" of the Constitution advocated by legal theorist Ronald Dworkin (1996). Like Segal and Spaeth, Dworkin allows substantial room for the attitudes of justices to play a role in decision making. (It should be noted here, however, that while Dworkin's arguments apply to Supreme Court justices, his analysis does not end there. It extends to other judges as well, unlike the position we advance here and the analysis of Segal and Spaeth, which are limited to the Supreme Court.) Dworkin contends that many constitutional clauses, including the free speech clause of the First Amendment, declare individual rights in abtract language. In interpreting these abstract clauses, Dworkin's moral reading proposes that justices and others apply these abstract clauses on the understanding that they invoke moral principles about political decency and justice. . . . The moral reading therefore brings political morality into the heart of constitutional law (1996, 2). This "political morality" concept is compatible with Segal and Spaeth's notion of the political attitudes of the justices. (However, Dworkin's political morality has a more principled character than Segal and Spaeth's concept of attitudes.) In addition, Dworkin contends that the moral reading, like the attitudinal model, actually explains what both liberal and conservative judges do. "So far as American lawyers and judges follow any coherent strategy of interpreting the Constitution at all, they already use the moral reading. . . . That explains why both scholars and journalists find it reasonably easy to classify judges as 'liberal' or 'conservative' (1996, 2)." We contend that justices' attitudes do play a substantial role in Supreme Court decision making, as do Dworkin and Segal and Spaeth. But where does our concept of doctrine fit into Dworkin's argument?

Like Dworkin, we suggest that the political values of justices matter in constitutional interpretation, but doctrine can operate to "structure" or "constrain" those attitudes. Dworkin is careful to defend his "moral reading" position against the argument that it frees justices to impose whatever values in which they believe on the Constitution when deciding issues of constitutionality. "Our constitution is law, and like all law it is anchored in history, practice, and integrity (1996, 11)." The constraints of history, practice, and integrity fit approximately with our conception of doctrine, because Dworkin stresses that judges must offer moral readings which comport with the structure of the Constitution as a whole, as well as with the dominant lines of past constitutional interpretation by other judges (1996, 10). Doctrines are dominant lines of constitutional interpretation.

What distinguishes doctrine from other types of precedent? First, a precedent should present a formulation of an analytic principle. A precedent which establishes only a narrow holding that is particular to one case fails to establish a decisional principle. By contrast, a doctrinal principle is formulated to identify the relevant case facts or decisional principles (i.e., level of scrutiny) that should structure decisions in an area of law. Secondly, a doctrine should have been used to guide adjudication in subsequent cases. It needs to be cited in future decisions. If it is never mentioned subsequent cases, it has not achieved doctrinal status. Thirdly, a doctrine should be recognized by scholars as a precedent which is a formulation of an analytic principle and has guided the decision making of subsequent cases. If scholars have recognized the doctrinal character of a particular precedent, this provides more evidence of its influential status.

Take for example the issue of what factors courts (including the Supreme Court) should consider in issues of curtailing or punishing speech dealing with advocacy of illegal action. In such cases, there are a variety of factors the Court could consider: the identity of the speaker, the level of government acting against the speaker, the setting of the speech, whether the speaker called for violence. A case dealing with these issues which established what we label doctrine is Brandenburg v. Ohio (399 US 444 [1969]). First, it formulates an analytic principle in the area of speech which may incite violence or lawless action. The Court observed that the principle is the constitutional guarantees of free speech and free press do not permit a state to forbid or proscribe advocacy of the use of force or of law violation except where such advocacy is directed to inciting or producing imminent lawless action and is likely to incite or produce such action.⁽⁸⁾ Secondly, the doctrine has been used to guide adjudication in subsequent cases. Siegel (1987) performed a jurisprudential analysis of the influence of Brandenburg on its progeny and found that it was influential. Shepard's citations indicates that Brandenburg has been cited in 39 Supreme Court opinions since it was published, and although it was used as the determining precedent only twice, it has not been overruled. Of course, there is also a progeny in the lower courts. Here, the influence of Brandenburg is likely to be stronger and more extensive. Third, scholars have recognized both the analytic formulation of Brandenburg as well as its status as a doctrine guiding future decisions. The Lockhart (1996) constitutional law case book considers the principle from Brandenburg quoted above to be a modern restatement of law in this area. Siegel (1987, 136) observes, "The current test used by the Supreme Court in determining whether or not a particular act of advocacy should enjoy first amendment protection was set forth in Brandenburg v. Ohio." In an article entitled "The Emergence of Modern First Amendment Doctrine," Rabban (1983) notes that the analytic principle of Brandenburg combined the most protective elements of earlier incitement and "clear and present danger" tests. The influence of Brandenburg has also been recognized. Rabban indicates that the Burger court reaffirmed the doctrine in later cases. In his concluding remarks, Siegel comments on the influence of Brandenburg. "It is comforting to see that the courts continue to treat the 1969 pronouncement in Brandenburg v. Ohio as a reminder that political speech is at the heart of our democratic system of self-government (Siegel 1987, 153)." According to the criteria established here, Brandenburg qualifies as a doctrine. What remains to be assessed is whether it systematically influences the votes of the justices in similar cases. We examine that issue empirically later in this paper.

In one form, the doctrinal version of the law-constrained model predicts that justices' votes in cases decided after Brandenburg should be more likely to be in the liberal direction reflecting the speech protective character of the doctrine laid out in Brandenburg. In a second form, the model predicts differences in how other factors (e.g., factual elements such as government party, threat of violence or lawless action, setting, and type of speaker) influence votes.

Thus, Brandenburg established a decisional principle but it did not establish a precedent that served to determine the outcome of subsequent cases. In fact, a Lexis search indicated that Brandenburg was followed as the determining principle in only two of the 39 subsequent Supreme Court cases that cite Brandenburg, through March, 1997. Or argument is that decisions like Brandenburg do not determine subsequent decisions in the sense that stare decisis is seen as creating binding precedents, but rather such decisions serve to structure future decisions by telling justices to look at facts and parties in a particular manner, or by telling justices how to weigh conflicting elements.

Testable Hypotheses

What are the testable implications of this constrained attitudinal model, particularly in contrast to the unconstrained attitudinal model? There are at least three, all of which suggest an empirical test.

First, facts work differently in the unconstrained attitudinal model and the law-constrained attitudinal model. There has in fact been substantial confusion over whether the influence of facts on decisions reflects legal considerations (George and Epstein 1992, 327-328; Segal 1984) or attitudinal considerations (Baum (in Lawrence 1994, 4; Segal and Spaeth 1993, 255). In fact, one can readily make arguments either way [Segal, 1984 #7145, 892) (Segal and Spaeth 1993, 220). The distinction between the unconstrained and law-constrained attitudinal models provides a way to obtain evidence on the issue of whether facts are primarily attitude-relevant ("[f)acts explain why a particular [emphasis added] justice votes one way in certain cases, and differently in others"; Segal and Spaeth 1993, 299) or primarily law-relevant (facts are "legal factors impinging on the probability of a conservative decision"; George and Epstein 1992, 327). In the unconstrained attitudinal model, facts are relevant for the attitudes of the individual justices, and we should expect that the impact of facts for a given set of cases will differ from justice to justice. In the law-constrained attitudinal model, facts are primarily a legal element with the implication that facts should work similarly across the justices.

Second, to the extent that law does in fact serve a structuring function, one should find that a major doctrinal shift in a given area will result in a change in the way that factors such as the justices' attitudes and/or the case facts influence individual decisions. In the previous section, we explicated this argument in the context of Brandenburg v. Ohio (1969). The most extreme would be cases which essentially ruled issues in and out of the Supreme Court's purview. The most obvious example would be Baker v Carr which ruled that the federal courts (and the Supreme Court) could consider issues of legislative apportionment. Prior to this Baker v. Carr, such issues had been deemed to be political questions which were nonjusticiable; Baker v. Carr rejected this principle vis-a-vis apportionment cases, and from 1963 through the 1995 term, the Supreme Court has decided over 40 cases dealing with legislative apportionment.⁽⁹⁾ In the empirical analysis presented below, we will focus on the impact of the Brandenburg decision; we leave for future papers extension of this type of analysis to other substantive areas.

FACTS AS LAW OR FACTS AS ATTITUDES?

The first student of the Supreme Court to apply a "fact pattern" approach to a quantitative analysis of court decisions was Fred Kort (1957); unfortunately, his analysis suffered from such severe shortcomings that in the end it provided little more than guidance on what to avoid for those who came after. The first solid analysis of this sort was Segal's (1984; Segal 1986) consideration of search and seizure cases. Since then, there have been similar studies focusing on self-incrimination cases (Aliotta 1995), obscenity cases (Hagle 1991; McGuire 1990), establishment clause cases (Aliotta 1995; Ignagni 1993), free exercise clause cases (Ignagni 1994) gender discrimination cases (Wolpert 1991), equal protection cases (Aliotta 1995), and death penalty cases (George and Epstein 1992). These studies have generally found that legally-relevant case facts help predict votes and outcomes of Supreme Court cases.

Unfortunately, none of the analyses to date really allow us to determine whether the facts should be considered primarily supportive of legal explanations or primarily supportive of attitudinal explanations. In his first presentation based on search and seizure cases, Segal described his analysis of factual elements as constituting a "legal" analysis rather than an attitudinal analysis. However, in The Supreme Court and the Attitudinal Model, Segal and Spaeth describe a very similar analysis as supportive of the attitudinal model. The key difference between these two analyses is that the first used the court decision as the dependent variable while the later analysis used the individual justice.⁽¹⁰⁾ While the latter analysis is not inconsistent with the attitudinal model, it is not inconsistent with the legal model either.

We have suggested a way that we might choose between the two understandings of facts: if the influence of facts is largely dependent upon the individual justice, then facts are more consistent with attitudes, but if the influence of facts is independent of the individual justice, then facts are more consistent with a law-related explanation of decision-making. One test of this argument would be to replicate Segal and Spaeth's analysis and then modify the analysis so that the influence of facts can vary by justice. That is in fact the approach we have used.⁽¹¹⁾

Segal and Spaeth report having included 1,550 votes in their analysis (1993, 229), including the votes of 17 different justices over the period 1962 through 1989. The number of individual votes ranged from under 20 to almost 200. Because the approach we used involves getting estimates for individual justices, we excluded from the analysis those justices for whom fewer than 100 votes were available (Warren, Fortas, Black, Harlan, Douglas, O'Connor, Scalia, and Kennedy). Table 9 shows both Segal and Spaeth's original results (for 1,550 cases and 17 justices) and the results we obtained for the same model using only the 9 justices with 100 or more votes. Dropping the 8 justices with fewer than 100 votes has little or no impact on the results.

The analysis reported by Segal and Spaeth includes a measure of attitudes which is a factor score that combines a measure of general attitudes based on newspaper editorials at the time of the justice's appointment (1989), the judicial ideology of the appointing president (Tate and Handberg 1991), and the party identification of the justice. In both their analysis, and the analysis restricted to nine justices, this variable has a very strong effect. However, including this variable when we introduce controls that allow the impact of the other variables to differ for each individual justice creates problems for the statistical analysis. A solution to this problem is to replace the attitude score with dummy variables for the individual justices. This dummy variable allows for attitudinal differences among the justices, and in fact does a better job of it because it simply allows each justice to have his or her own unique predilection in search and seizure cases without attempting to impose any scaling on that predilection. Table 10 shows three different equations for the nine-judge data set: the analysis shown in Table 9, an analysis with just individual judge dummy variables, and an analysis with both the judge dummies and the variables other than attitude. While the results in Table 10 are only a preliminary step, there are several aspects of those results that are worth noting. First, using dummy variables in place of the attitude measure produces a sharply improved fit. Second, the factual variables add significantly to the fit over and above the dummy variables for the individual justices (chi square = 118.86, df=11). Third, the actual values of the coefficients for the variables change little from specification to specification.

The test of whether the factual variables differ from justice to justice if we estimate separate values for each justice can be computed either by creating a set of interaction terms (multiplying each fact variable by each justice dummy variable) or be simply estimating separate models for each justice and combining the results. We used the latter approach. Table 11 shows both the last column of Table 10 which forces all the fact variables to have the same coefficient across justices and individual coefficients estimated separately for each justice. Looking at the individual coefficients, there appears to be substantial variation. However, the central question is whether that variation is attributable to sampling? The test for this compares the sum of the fitted "-2 log likelihoods" (1189.57) to the corresponding statistic for the model that includes justice dummies but constrains the factual variables to be constant across the justices (1281.98) which is the model shown in the first column of Table 11 (and the last column of Table 10). The result, 92.41, is a chi square statistic with 80 degrees of freedom; this chi square indicates that there is no statistically discernible variation (p-value=.1619) in the fact variable coefficients from justice to justice.

This analysis shows that, at least for the search and seizure decisions used by Segal and Spaeth in their analysis, the influence of factual elements does not vary by justice which is what one would expect of facts were primarily related to how justices' attitudes account for their decisions. The constancy of the factual elements across the justices points strongly toward those elements as being law-based.

As a single legal area, this analysis is only a first step in the argument that law is relevant for understanding the votes of Supreme Court justices. Replicating this analysis with other areas of law would provide further support. However, to do this replication, one needs a sizeable group of cases to insure that results for individual justices are reasonably stable, and there must be a clearly definable set of factual elements that are thought to be relevant for those cases. None of the other areas that have to date been subjected to a fact-element analysis (establishment clause cases, obscenity cases, business cases, and death penalty cases) have sufficient cases for a solid analysis. (It is probably no accident that Segal and Spaeth restricted their fact-based analysis to search and seizure cases.)

TESTING THE DOCTRINAL MODEL OF LAW

Above we described what we label the "doctrinal model" of law. By this we mean that law establishes decisional principles that guide, but do not determine, justices' (and judges') decisions. In our discussion of Brandenburg, we provide an example of what we mean: a key decision can establish a doctrinal principle that changes the elements that go into the decision process. In the case of Brandenburg, this means that factors such as setting, government party, speaker, and the threat of violence or lawless action should have different impacts on justices' votes before and after Brandenburg

Modeling Free Speech Decisions

Which facts have the potential to be relevant in the area of subversive speech? First legal facts will be addressed, and then other factual variables. Looking at the Brandenburg formula, several relevant legal facts emerge. Was there an imminent risk of lawlessness? Was the speech likely to result in lawlessness? Was the speech abstractly directed toward producing lawless action? Abstract advocacy by itself should not be enough to convict the speaker. Another variable relevant to the doctrine is whether the speech is political or not. Brandenburg reaffirms the Court's support for protecting speech which may advocate violence in the abstract or be unpatriotic, racist, unpopular or distasteful, but is nonetheless political. Considering first amendment doctrine more generally, the forum in which the speech occurs is also relevant. Traditional public forums such as streets, sidewalks and parks are areas where speakers can expect the most protection, although speech which violates the Brandenburg formula could still be regulated (International Society for Krishna Consciousness, Inc. v. Lee, 505 U.S.672 [1992]). Speakers expect less protection in semi-public forums such as libraries and schools, which require a quiet atmosphere in order to serve their primary purposes. Speakers will have almost no protection in forums which have been declared "non-public" in the case law. These forums include airports, jails, military bases, the interior of courthouses, and government workplaces.

Another factual variable considers the party to the suit. It is possible that the Court may be more deferential to the federal government than to state and local governments and universities. Finally, the identity of the speaker could also matter, as critical legal studies and critical race studies scholars have argued (Kairys 1990; Matsuda and et al. 1993). Which speakers receive more protection: racists, racial minorities, military protesters, or communists?

Hypotheses to be Tested

Models

We test this argument using data drawn from the Spaeth dataset supplemented by our own coding of variables elements not in that data set. We estimate and compare a series of logistic regression models. The key question that we ultimately want to evaluate is whether the factors influencing decisions changed in the wake of the Brandenburg decision. Technically, the statistical question is whether or not there is an interaction between the Brandenburg decision and some or all of the other predictor variables. We employed standard statistical procedures, a "Chow test" (Hanushek and Jackson 1977, 127-129), to test for interaction effects; for clarity we have presented the results of these tests as a set of "conditional" models (Wright 1976), where separate parameter estimates are shown for before and after Brandenburg.

One possible threat to the validity of our analysis is that the composition of the court changed over the period of our analysis. It is conceivable that what appear to be changes associated with the Brandenburg are really nothing more than changes in Court personnel. To assess this we also estimate models using only votes from justices on the Court at the time of the Brandenburg decision.

Dependent variable

Votes: We used the individual justices' votes on all relevant first amendment cases from the Warren court to the most recently available Rehnquist Court decisions. 0 corresponds to a liberal vote, and 1 indicates a conservative vote. A liberal vote supports free speech; a conservative vote supports the government. Positive estimates for independent variables indicate a support for the government rather than the speaker. A total of 47 cases were involved, producing a total of 421 votes for analysis. Votes on the Brandenburg case itself were excluded from the analysis.

Independent variables

Most of the independent variables take the form of sets of dummy variables. In the description that follows, we have listed all of the categories of each set, but shown the base (omitted) category in a smaller type to set it off.

Attitudes of Justices: The attitude of each justice is measured using the scale developed by Segal, Cover, and Cameron (1988-89), and updated by Segal, Epstein, Cameron and Spaeth (1995). The scale is based on content analysis of newspaper editorials at the time of the confirmation hearings of each justice. The scale runs from -1.00 (very conservative) to 1.00, (very liberal). We expect the parameter to be negative.

Brandenburg: This is a dummy variable indicating whether a case was decided before or after Brandenburg. 0 indicates the case was decided before Brandenburg, and 1 indicates a decision after Brandenburg. We hypothesize that the parameter will be negative.

Lawless Action (a set of dummy variables):

Imminent Risk of Lawless Action: If the speech is directed toward producing imminent lawless action, the value is 1; otherwise it is 0 (hypothesized to be positive before and after Brandenburg).

Likely Risk of Lawless Action: If the speech is likely to produce or incite lawless action, the value is 1; otherwise it is 0 (hypothesized to be positive before and after Brandenburg). Imminent and Likely risk of lawless action were coded identically, so we use one variable to represent both of them.

Abstract advocacy: If the speech is abstractly directed toward producing lawless action, the value is 1; otherwise it is 0 (hypothesized to be negative before and after Brandenburg). This is the baseline category for our analysis of this set of variables.

Not an issue: If these factors were not an issue in the case, the value is 1; otherwise it is 0.

Setting: This set of dummies covers the public forum doctrine, which is not specifically changed by Brandenburg. One category covers first amendment cases which did not involve a public forum question. This includes compelled speech before a legislative investigatory committee or before government interrogators. It also includes first amendment cases involving speech in private, and denial of benefits or official actions against individuals for their membership in subversive or communist organizations. The hypotheses listed below show what we expect to observe regardless of doctrinal change.

Public forum: If the speech occurs in a traditional or designated public forum, the value is 1; otherwise it is 0. The estimate is expected to be negative.

Semi-public forum: If the speech occurs in a semi-public forum such as a library or school, the value is 1; otherwise it is 0. The estimate is expected to be positive.

Non-public forum: If the speech occurs in a non-public forum such as a jail or military base, the value is 1; otherwise it is 0. The estimate is expected to be positive. This is the baseline category for the setting variables.

Not an issue: If the type of forum is not an issue in the case, the value is 1; otherwise it is 0.

Political speech: If the speech is political, the value is 1; otherwise it is 0. The estimate is expected to be negative regardless of doctrinal change). As it was coded, there turned out to be no variation in political speech in cases decided before Brandenburg, so we are unable to compare effects before and after Brandenburg.

Government party: This set of variables considers the level of government acting against the speaker. One category is comprised of private lawsuits against speakers. The hypotheses listed below show what we expect to observe regardless of doctrinal change.

University: When the level of government is a school, university or college, the value is 1; otherwise it is 0. The estimate is expected to be negative. The lack of variation in this category after Brandenburg caused it to be dropped from the analysis.

Local: When the government is below the state level, for example a town, city or county, the value is 1; otherwise it is 0. The estimate is expected to be negative. This is the baseline variable for the group of government variables.

State: When level of government is a state, the value is 1; otherwise it is 0. The estimate is expected to be negative.

Federal: When the federal government is acting against the speaker, the value is 1; otherwise it is 0. The estimate is expected to be positive.

Non-government: When no government is acting against the speaker, the value is 1; otherwise it is 0. The estimate is expected to be negative. The lack of variation in this category after Brandenburg caused it to be dropped from the analysis.

Identity of Speaking Party: The "other" category includes speakers who are members of religious or other groups.⁽¹²⁾ The hypotheses listed below show what we expect to observe without taking into account doctrinal change, and then in the set of cases decided before Brandenburg; after Brandenburg we expect that any relationships will disappear.

Communist: When the speaker is speaking as a communist, the value is 1; otherwise it is 0. The estimate is expected to be positive.

War/military protester: When the speaker is speaking as a war or military protester, the value is 1; otherwise it is 0. The estimate is expected to be positive. This is the baseline variable for the speaker group of dummy variables.

Racial minority: When the speaker is speaking as a racial minority, the value is 1; otherwise it is 0. The estimate is expected to be positive.

Summary statistics for all of the independent variables are shown in Table 12.

RESULTS

Table 13 reports the results of three models: the overall doctrine-constrained attitudinal model, and the two conditional models for cases decided before and after Brandenburg. The overall model model predicts 76.01 of the cases correctly, with a proportional reduction in error of 34.85, and a psuedo R-squared of .240. The set of doctrine variables is signficant at the .0002 level. The set of government variables is borderline in statistical significance (p= .1064), and the set of speaker variables is significant (p< .005). Turning to individual variables, it is not a surprise that attitudes remain a strong predictor of justices votes in cases involving questions of speech which may produce lawless action. The variable which captures influence of the Brandenburg doctrine is significant (p < .01). As hypothesized, Brandenburg influences the justices to vote for the rights of the speakers. In addition, we observe a significant, positive influence of the variable capturing the presence of speech which is likely to produce an imminent risk of lawless action, as compared to the baseline variable, which captures whether the speech merely advocated lawless action in the abstract.

Examining the other variables in the model which are not directly related to Brandenburg, we see that the justices are inclined to give greater first amendment protection to political speech. This result is statistically significant. The school or university government variable influences the justices votes in a significant, negative direction, as compared to the baseline variable of local governments as the party. Looking to speaker's identity, we see that as compared to war or draft protestors, the variables for communists and racial minorities are significantly associated with greater support for the rights of the speakers.

Examining the before and after Brandenburg conditional models in Table 13, the most important result is the Chow test of whether the sets of model parameters differ before and after Brandenburg produces a highly significant chi square of 87.98 (13 d.f., p<.0001). It is noteworthy that a number of variables that were not significant before Brandenburg are significant aftewards, although the sets of variables are significant overall both before and after. The before model predicts 75.78 percent of the cases correctly, with a proportional reduction in error of 34.73 percent and a psuedo R-squared of .249; the after model predicts 83.64 percent of the cases correctly, with a proportional reduction in error of 55.01 percent and a psuedo R-squared of .492.

The attitudes of the justices are a significant, negative predictor of the justices' votes both before and after, and the value of the coefficient changes little. For the after model, the set of government variables is significant (p=.0138), as is the set of speaker variables (p=.0168); neither is significant before Brandenburg. After Brandenburg, the federal and state government variables significantly influence the justices to vote in support of the government. Prior to Brandenburg, the directions of coefficients for these variables were negative (but not significant). One possible explanation is that before Brandenburg clarified the Court's doctrinal support for free expression, federal and state officials were not clear as to whether the action they were taking against speakers actually violated the first amendment, so they were more likely to take unconstitutional action against the speakers. After Brandenburg, federal and state officials may have limited themselves to prosecuting speakers who were clearly engaging in speech which was not constitutionally protected under Brandenburg. Perhaps this also explains why, as compared to war protestors, communists, racial minorities and others were likely to have their speech rights limited after Brandenburg. The coefficients of the variables for these speakers are significant and positive; before Brandenburg, the coefficients were negative but not significant.

Table 14 repeats the above analysis with only the justices who were on the Court at the time of the Brandenburg decision (this is to be sure that the before/after differences are not simply an artifact of personnel change). This group includes justices Black, Brennan, Douglas, Harlan, Marshall, Stewart and White. We excluded Chief Justice Warren and justice Fortas because they left the Court very close to the time of the Brandenburg decision. This model enables us to compare this group of justices to the larger population of justices. As in Table 13, we report three models, an overall model and the two conditional models of cases decided before and after Brandenburg. The Brandenburg justices doctrine-constrained attitudinal model predicts 80.78 percent of the cases correctly with a 35.80 percent reduction in error and a psuedo R-squared of .247.

For the overall model, three of the four of the sets of variables are significant: the doctrine set of variables (p=.0165); the government set (p=.0531); and the speaker set (p=.0173). The setting set of variables is not statistically significant (p=.1573). Looking at individual variables, rather than sets, we observe three significant, negative influences on the votes of the justices: attitudes, Brandenburg, and political speech. The coefficient for speech which is likely to produce imminent lawless action is significant and positive as compared to the abstract advocacy of lawless action. This is as hypothesized. According to Brandenburg, mere advocacy of lawless action should not make speech unprotected. Only when the speech is likely to produce imminent lawless action should it be curtailed. As compared to war protestors, racial minorities and communists are significantly more likely to have their speech protected. As compared to local government, schools and universities are significantly more likely to influence the justices to vote in favor of the rights of speakers; non-government organizations are significantly more likely to have the opposite effect.

Table 14 reports the results of the conditional before and after Brandenburg models for justices who participated in Brandenburg. Again, the Chow test shows significant differences for the before and after model (chi square 57.41, 13 d.f., p<.0001). One striking aspect of the estimates for the before/after conditional model is the sharp differences in individual coefficients; this in part undoubtedly reflects the reduced number votes used in the estimation. The before model predicts 78.05 percent of the cases correctly, with a 38.99 percent proportional reduction in error and a psuedo R-squared of .257. The after model predicts 92.31 percent of the cases correctly, with a 58.83 percent proportional reduction in error and a psuedo R-squared of .553. None of the sets of variables are significant for either model (again probably reflecting in substantial part the reduced sample sizes), but the Chow test shows that there are differences between the two models. The ideologies of the justices remain as a significant predictor of the justices' votes. The other individual variables in the two models are not significant, with the exception of the non-government variable in the before model, which has a positive coefficient.

CONCLUSION

It is perfectly plausible that Justices are political actors who seek to make policy, and in so doing, to maximize their policy preferences. In this sense, the Supreme Court is similar to Congress or the Presidency. However, it is rare that political actors can or do act with no constraints placed on their ideological, or policy, preferences. For example, members of congress are constrained by their constituencies, parties, career motivations, the desire to make "good" public policy and the need for information, and thus their votes cannot be explained by ideology alone (see Fenno 1973; Kingdon 1992; Kingdon 1989; Mayhew 1974). The nature of the constraints must surely be different, since the nature of the Court is different from that of Congress. Our task has been to identify constraints on Supreme Court Justices and to eventually build a more complete and powerful model of Supreme Court decision making.

This paper begins to show ways in which law systematically impacts on the decisions of Supreme Court justices. We have structured our analysis around the concept of the law-constrained attitudinal model of Supreme Court decision-making. Our primary empirical analyses have focused some specific legal areas (search and seizure and free speech). Clearly, our analyses constitute only a first step. Our approach needs to be extended into other areas of Supreme Court decision-making. We plan to do precisely this for what we have labeled the doctrinal model in a paper at next fall's Southern Political Science Association. In addition to updating our analysis of the impact of Brandenburg, we will look at areas such as obscenity, free exercise, the establishment clause, and possibly search and seizure.

Finally, before concluding, it is worth noting that using the "standard" measure of votes of Supreme Court justices, as we have done, actually biases the analysis toward acceptance of the attitudinal model and rejection of the legal model. In the standard data sets, votes (and decisions) are coded as "liberal" and "conservative" according to a set of coding rules that equate the former with positions that are consistent with contemporary American "liberalism" or "conservatism" (this approach dates at least to the work of Glendon Schubert (1962; 1965) The attitudinal model then seeks to explain those votes in terms of measures of judicial attitudes that fall along the same ideological continuum (see, for example, Songer, Segal and Cameron 1994). It would be very surprising if votes measured along this continuum were not strongly related to judicial attitudes measured along the same continuum. However, it is easy to imagine other measures of judicial votes (e.g., "Congressional support" by justices for cases challenging acts of Congress) that might be explained by other variables. Just as students of Congress have formulated a wide variety of roll call voting indices to measure different aspects of roll call voting (e.g., presidential support score, party unity score, etc.), students of the Supreme Court could develop measures of decision-making that were not in themselves explicitly ideological. That was not our goal in the analysis presented above. However, the fact that we were able to explain votes measured in ideological terms with nonideological/nonattitudinal factors, analyses focused on voting measures that are not ideological in nature should be even more successful in applying factors such as those we have discussed.

REFERENCES

Aliotta, Jilda. 1995. "The Impact of Case Characteristics and Judicial Ideologies: A Comparison Across Constitutional Issues." Paper presented at Law & Society Association, Toronto, June 1-4, 1995,

Ashenfelter, Orley, Theodore Eisenberg, and Stewart J. Schwab. 1995. "Politics and the Judiciary: The Influence of Judicial Background on Case Outcome." Journal of Legal Studies 24:257-281.

Baum, Lawrence. 1992. "Membership Change and Collective Voting Change in the United States Supreme Court." Journal of Politics 54:3-24.

Brams, Steven J. , and Douglas Muzzio. 1977. "Unanimity in the Supreme Court: A Game-Theoretic Explanation in the White House Tapes Case." Public Choice 32:67-83.

Brenner, Saul, and Theodore S. Arrington. 1987. "Unanimous Decision Making on the Supreme Court: Case Stimuli and Judicial Attitudes." Political Behavior 9:75-86.

Brenner, Saul, and Marc Stier. 1996. "Retesting Segal and Spaeth's Stare Decisis Model." American Journal of Political Science 40:1036-1048.

Brisbin, Richard A. 1996. "Slaying the Dragon: Segal, Spaeth and the Function of Law in Supreme Court Decision Making." American Journal of Political Science 40:1004-1017.

Dworkin, Ronald M. 1996. Freedom's Law: The Moral Reading of the American Constitution.

Epstein, Lee, and Jack Knight. 1998. The Choices Justices Make. Washington: CQ Press.

Epstein, Lee, and Joseph F. Kobylka. 1992. The Supreme Court and Legal Change: Abortion and the Death Penalty. Chapel Hill: University of North Carolina Press.

Eskridge, William N. 1991. "Overriding Supreme Court Statutory Interpretation Decisions." Yale Law Journal 101:331-456.

Fenno, Richard F., Jr. 1973. Congressmen in Committees. Boston: Little, Brown.

George, Tracy E., and Lee Epstein. 1992. "On the Nature of Supreme Court Decision Making." American Political Science Review 86:323-337.

Hagle, Timothy M. 1991. "But Do they Have To See It To Know It? The Supreme Court's Obscenity and Pornography Decisions." Western Political Quarterly 44:1039-1054.

Hanushek, Eric A., and John E. Jackson. 1977. Statistical Methods for Social Scientists. New York: Academic Press.

Ignagni, Joseph A. 1993. "U.S. Supreme Court Decision-making and the Free Exercise Clause." Review of Politics 55:511-529.

Ignagni, Joseph A. 1994. "Explaining and Predicting Supreme Court Decision Making: The Burger Court's Establishment Clause Decisions." Journal of Church and State 36:301-327.

Kairys, David. 1990. "Freedom of Speech." Pp. 237-272 in The Politics of Law: A Progressive Critique, edited by David Kairys. New York: Pantheon Books.

Kingdon, John E. Jackson and John W. 1992. "Ideology, Interest Group Scores, and Legislative Votes." American Journal of Political Science 36:805-823.

Kingdon, John W. 1989. Congressmen's Voting Decisions. Ann Arbor: University of Michigan Press.

Knight, Jack, and Lee Epstein. 1996. "The Norm of Stare Decisis." American Journal of Political Science 40:1018-1035.

Kort, Fred. 1957. "Predicting Supreme Court Cases Mathematically: A Quantitative Analysis of the 'right to Counsel' Cases." American Political Science Review 51.

Lawrence, Susan E. 1994. "Review Symposium: The Supreme Court and the Attitudinal Model." Law and Courts 4:3-12.

Lockhart, William B., Yale Kamisar, Jesse H. Choper, Steven H. Shiffrin, and Richard H. Fallon, Jr. 1996. Constitutional Law. St. Paul: West Publishing Co.

Matsuda, Mari, and et al. 1993. Words that Wound. Boulder, CO: Westview Press.

Mayhew, David R. 1974. Congress: The Electoral Connection. New Haven: Yale University Press.

McGuire, Kevin T. 1990. "Obscenity, Libertarian Values, and Decision Making in the Supreme Court." American Politics Quarterly 18:47-67.

Nelson, Lisa. 1998. "Toward an Understanding of  Stare Decisis in Supreme Court Decision Making." Unpublished Ph.D. Dissertation, Department of Political Science, University of Wisconsin.

Perry, H.W. 1991. Deciding to Decide: Agenda Setting in the United States Supreme Court. Cambridge: Harvard University Press.

Pickerill, J. Mitchell. 1996. "The Unanimously Ignored Issue of Unanimity and the Attitudinal Model." Paper presented at Annual Meeting of the Midwest Political Science Association, Chicago, IL, April, 18-20.

Pritchett, C. Herman. 1948. The Roosevelt Court: A Study in Judicial Politics and Values, 1937-1947. New York: Macmillan.

Pritchett, Herman C. 1941. "Voting Behavior on the United States Supreme Court." American Political Science Review 35:890-898.

Rabban, David M. 1983. "The Emergence of Modern First Amendment Doctrine." University of Chicago Law Review 50:1207-1263.

Rohde, David W., and Harold J. Spaeth. 1976. Supreme Court Decision Making. San Francisco: W.H. Freeman and Co.

Schubert, Glendon A. 1962. "The 1960-61 Term of the Supreme Court: A Psychological Analysis." American Political Science Review 56:90-107.

Schubert, Glendon A. 1965. The Judicial Mind: Attitudes and Ideologies Os Supreme Court Justices. Evanston: Northwestern University Press.

Segal, Jeffrey. 1984. "Predicting Supreme Court Cases Probabilistically: The Search and Seizure Cases, 1962-1981." American Political Science Review 78:891-900.

Segal, Jeffrey. 1986. "Supreme Court Justices as Human Decision Makers: An Individual-Level Analysis of the Search and Seizure Cases." Journal of Politics 48:938-955.

Segal, Jeffrey A., and Albert D. Cover. 1989. "Ideological Values and the Votes of U.S. Supreme Court Justices." American Political Science Review 83:557-566.

Segal, Jeffrey A., Albert D. Cover, and Charles M. Cameron. 1988-89. "The Role of Ideology in Senate Confirmation of Supreme Court Justices." Kentucky Law Review 77:485-508.

Segal, Jeffrey A., Lee Epstein, Charles M. Cameron, and Harold J. Spaeth. 1995. "Ideological Values and the Votes of U.S. Supreme Court Justices Revisited." Journal of Politics 57:812-823.

Segal, Jeffrey A., and Harold J. Spaeth. 1993. The Supreme Court and the Attitudinal Model. New York: Cambridge University Press.

Segal, Jeffrey A., and Harold J. Spaeth. 1994. "The Authors Respond." Law and Courts 4:10-12.

Segal, Jeffrey A., and Harold J. Spaeth. 1996. "The Influence of Stare Decisis on the Votes of United States Supreme Court Justices." American Journal of Political Science 40:971-1003.

Siegel, Paul. 1987. "Protecting Political Speech: Brandenburg v. Ohio Updated." Pp. 136-153 in Perspectives on Freedom of Speech: Selected Essays from the Journals of the Speech Communication Association, edited by Thomas L. Tedford, John J. Makay, and David L. Jamison. Carbondale: Southern Illinois University Press.

Songer, Donald. 1987. "The Impact of the Supreme Court on Trends in Economic Policy Making in the United States Courts of Appeals." Journal of Politics 49:830-844.

Songer, Donald R., and Stefanie Lindquist. 1996. "Not the Whole Story: The Impact of Justices' Values on Supreme Court Decision Making." American Journal of Political Science 40:1049-1063.

Songer, Donald R., Jeffrey A. Segal, and Charles M. Cameron. 1994. "The Hierarchy of Justice: Testing a Principal-Agent Model of Supreme Court-Circuit Court Interactions." American Journal of Political Science 38:673-696.

Songer, Donald R., and Reginald S. Sheehan. 1990. "Supreme Court Impact on Compliance and Outcomes: Miranda and New York Times in the United States Courts of Appeals." Western Political Quarterly 43:297-316.

Spaeth, Harold J. 1989. "Consensus in the Unanimous Decisions of the U.S. Supreme Court." Judicature 92:274-281.

Spaeth, Harold J. 1995. "The Attitudinal Model." Pp. 296-314 in Contemplating Courts, edited by Lee Epstein. Washington: CQ Press.

Tate, C. Neal, and Roger Handberg. 1991. "Time Binding and Theory Building in Personal Attribute Models of Supreme Court Voting Behavior, 1916-88." American Journal of Political Science 35:460-480.

Wolpert, Robin M. 1991. "Explaining and Predicting Supreme Court Decision-Making: The Gender Discrimination Cases, 1971-1987." Paper presented at Annual Meeting of the Midwest Political Science Association, Chicago, Illinois, April 18-20, 1991.

Wright, Gerald C. 1976. "Conditional Models for Evaluating Conditional Relationships." American Journal of Political Science 20:349-373.

NOTES

1. The Oncale case just happened to come down this term while we were working on this paper. It is not an anomoly, however. There are ample unanimous cases from which to choose. The following is merely a sample of recent unanimous cases involving civil liberties and civil rights issuse: Washington v. Glucksberg 117 S.Ct.. 2258 (1997) and Vacco, Attorney General of New York 117 S.Ct. 2293 (1997) (the Court ruled unanimously in both cases that there is no constitutionally protected right to physician assisted suicide); Richards v. Wisconsin 117 S.Ct. 1416 (1997) (The Court unanimously refused to allow a blanket exception to the general Fourth Amendment Search and Seizure requirement that police knock and announce their identity before executing a search warrant at a residence for cases involving illegal drugs); Bracy v. Gramley 117 S.Ct. 1793 (1997) (The Court unanimously ruled in favor of a convicted triple murderer in Illinois who filed a federal habeus corpus petition requesting a new trial because of alleged bias by a corrupt judge who presided over his trial); Whren v. United States 116 S.Ct. 1769 (1996) (The Court unanimously ruled that police may temporarily detain a motorist under Fourth Amendment Search and Seizure law if they have probable cause to believe a traffic violation occurred even if their primary prupose is to look for drugs); 44 Liquormart, Inc. v. Rhode Island 517 U.S. 484, 116 S.Ct. 1495 (1996) (The Court unanimously struck down a state law banning liquor price advertising as a violation of the First Amendment Free Speech protection); O'Connor v. Consolidated Coin Caterers Corp. 517 U.S. 308, 116 S.Ct. 1307 (1996) (The Court unanimously ruled that a plaintiff can win an age discrimination suit without proving he was replaced by someone under forty years old, the threshold age for coverage under the federal Age Discrimination in Employment Act); McKennon v. Nashville Banner Publishing Co. 513 U.S. 352, 115 S.Ct. 879 (1995) (The Court unanimously ruled that an employee does not lose her right to sue for job discrimination because an employer discovers evidence that would have justified her dismissal of known earlier); United States v. Gaudin 515 U.S. 506, 115 S.Ct. 2310 (1995) (The Court unanimously held that a trial jusge violated the defendant's right to a jury trial under a federal false statement statute by refusing to allow jurors to decide whether the statements made were "material" as required for conviction); Rubin v. Coors Brewing Co. 514 U.S. 476, 114 S.Ct. 1585 (1995) (The Court unanimously struck down the 60 year old federal law which prohibited breweries from including the alcohol content of their malt beverages on their labels for violating the First Amendment protection of commercial speech). These are just a few examples of unanimous cases involving civil liberties and civil rights handed down by the Supreme Court over the past few years. A casual perusal of CQ's annual Supreme Court Yearbook and of the United States Supreme Court Database reveals abundant unanimous cases in all issue areas in every term of the Court.

2. Percentages from the U.S. Supreme Court Judicial Database. 1995 is the most recent term included in the database.

3. Both data sets were obtained from the Interuniversity Consortium for Political and Social Research. We, of course, bear full responsibility for the analyses reported herein.

4. Brenner and Arrington claim that this split in direction of decision does not lend any support to the two situation model. While this may be so, it is not clear why conservative Justices would join liberal unanimous decisions much more frequently than liberals join conservative unanimous decisions. This is particularly true in affirm cases, where there is no reason for the court to have chosen to hear the case, as I briefly mentioned earlier.

5. As it turns out, the percentages of affirmed and reversed decisions which are unanimous are very similar, whether all oral argument cases or only cert cases are used. We believe that it is correct to use only cert. cases since the Justices have total discretion over whether or not to hear them. If indeed the Justices are unanimous in cases because they are correcting ideological outliers, we would expect them to be voluntarily choosing and then unanimously affirming lower court decisions.

6. Patterns of unanimity also vary depending upon the authority for review (see, Pickerill 1996). The Court is more likely to reach unanimous decisions in cases involving statutory interpretation than in those involving judicial review; this is especially true of those cases selected on cert. There are two likely explanations. First, the language of statutes is generally less ambiguous and more detailed than the language of the Constitution, and statutes are frequently more modern than the Constitution. It is, simply, easier to figure out plain meaning and legislative intent when interpreting statutes, and when the meaning is clear, Justices are more constrained from acting purely on their policy preferences. Second, Justices have strategic motivation to hand down unanimous statutory interpretation decisions: their decision is less likely to be overturned by Congress (see Eskridge 1991). As City of Boerne v. Flores 117 S.Ct. 2157 (1997) demonstrated, Congress cannot overturn the Court's interpretation of the Constitution without amending the Constitution. But as the Civil Rights Act of 1991 also demonstrated, Congress and the President can overturn the Court's interpretation of statutory language. As Eskridge has demonstrated, Congress is less likely to overturn statutory interpretations the more unanimous the Court decision.

7. Some scholars have suggested that strategic considerations might account for unanimous decisions (see, Brams and Muzzio 1977; Rohde and Spaeth 1976, 193-207); we believe that such factors can account for only a small fraction of unanimous decisions.

8. "[C]onstitutional guarantees of free speech and free press do not permit a state to forbid or proscribe advocacy of the use of force or of law violation except where such advocacy is directed to inciting or producing imminent lawless action and is likely to incite or produce such action" (399 U.S 444, 447 [1969]).

9. This is a conservative estimate based on the Spaeth data set. We have counted only cases coded as issues 250, reapportionment, and only cases coded as orally argued with full opinions. The data set includes 64 such entries, but there are only 44 unique docket numbers.

10. In another article, Segal (1986) carried out a vote-level analysis quite similar to that which we present below.

11. We would like to thank Jeff Segal for making his search and seizure data available to us.

12. We also considered coding for racial supremacists, but the only Supreme Court case involving racial supremacist speakers in addition to Brandenburg was R.A.V. v. City of St. Paul, 505 U.S. 377 (1992).