The most dangerous justice rides into the sunset.

• Author: Edelman, Paul H.
• Position: Symposium: The Rehnquist Court in Empirical and Statistical Retrospective

1. INTRODUCTION

   Spatial models of voting behavior suggest that the preferences of the median voter, under majority rule and with a single issue dimension, will determine policy outcomes. (1) This theoretical insight has been applied to numerous policymaking institutions, and the Supreme Court of the United States is no exception. (2) Identifying and analyzing the median voter on the Court has generated something of a cottage industry, both in scholarly research and in the popular press. (3) Yet identification of the "power center" on the Court may depend on the methodology employed and on the assumptions that underlie evaluators' methodological choices.

   One key assumption often made is that the Justices' voting patterns in most cases can be arrayed along a single ideological dimension. Although the Justices' voting behavior in nonunanimous cases will often fall along predictable ideological lines, a substantial minority of cases exhibit coalitions that are not so predictable. Consider, for example, the 2007 decision in Philip Morris v. Williams (4)--in which the Court invalidated a punitive damage award to the widow of a deceased smoker because the jury had improperly calculated that award based on harm to smokers other than the widow's husband. Voting in the majority were Justices Roberts, Alito, Breyer, Kennedy, and Souter. In dissent were Justices Ginsberg, Stevens, Thomas, and Scalia. These odd coalitions clearly fail to conform to expectations concerning the Justices' shared policy preferences. In short, such an outcome is inconsistent with the notion that the Justices' votes are best described in all cases as the product of a single ideological dimension.

   The presence of unpredictable voting coalitions suggests that Supreme Court Justices' decisions may in some cases be structured along divergent or cross-cutting issue dimensions. These alternative issue dimensions can complicate the identification of the Court's pivotal justice. To account for and accommodate these alternative dimensions, our earlier research (5) constructed a method for identifying the most powerful Justice without relying on the assumption of unidimensional policy preferences. Instead, earlier efforts focused on the unique policy coalitions formed by the Justices in non-unanimous cases. We ranked the Justices in terms of their individual ability to alter or shape Court outcomes. Rather than focusing solely on the identification of the Court's median Justice, we calculated a power index that allowed us to rank all of the Justices in terms of voting power. In this paper, we identify the most powerful Justice for each term of the Rehnquist Natural Court (1994 to 2004), completing our analysis of the longest natural 9-member Court in history.

2. METHODOLOGY

   1. THE MADNESS OF OUR METHOD

      Our previous articles have explained our methods at length. For the sake of brevity we will give just the barest outline of our techniques and refer readers to our earlier work for further discussion and justification.

      We begin by assembling the collection of unique coalitions that formed during the relevant Terms of the Court. By a coalition we mean a subset of the Justices who agreed on a legal proposition and whose complementary set of Justices did not agree. These legal propositions could be a complete opinion, or an agreement in part of an opinion. In particular, if a Justice concurs in a judgment but does not join the opinion, then he or she is not part of the coalition associated with that opinion.

      It is particularly important to note that we do not keep track of the number of times a particular coalition forms. While this decision is certainly subject to criticism, we believe that it is justifiable. Since the Justices have substantial control over their docket, a single coalition can agree to hear multiple cases. For the purposes of assessing the Justices' ability to form multiple coalitions, counting the coalitions by the number of times they appear overstates the power of the individual Justices involved.

      From this collection of data we compute three different indexes: the Sophisticated Index, the Naive Index, and the Modified Median Index. The Sophisticated Index is a variant of the Banzhaf index, and is computed only over cumulative data. The Naive Index accounts solely for the 5-4 decisions of the Court, and we compute it both Term-by-Term and cumulatively. The Modified Median Index is our variant of an index proposed by Lynn Baker, (6) which we also compute Term-by-Term and cumulatively. (7)

      The Naive Index is computed simply by counting the number of times that a Justice appears in a 5-Justice coalition and then normalizing so that the numbers add to 100 percent. It succinctly captures the notion that power is related to the number of times that a Justice's vote is decisive with respect to the outcome. It is easy to compute and meaningful both on a Term-by-Term basis as well as cumulatively.

      If all the Justices are equally powerful, each Justice will be assigned a power of 11.1%. To facilitate our analysis we compute the Judicial Quotient for each judge by rescaling the numbers by a factor of 9, so a 11.1% index results in a quotient of 100. We follow this convention for all of our indices.

      The Sophisticated Index expresses the intuition that not every Justice in a 5-Justice coalition has a credible threat of defecting. Those Justices without such a threat should not be deemed powerful in the coalition. To identify those Justices that do have a credible threat, we examine whether the other four Justices have ever themselves formed a coalition. If they have, then there is evidence that the Justice in question has indeed defected and so we assign him power in that coalition. On the other hand, if the other four Justices have never formed a coalition then we assume that the Justice in question does not have a credible threat to defect. We therefore assign that Justice no power in that coalition. For each Justice, we total the number of 5-Justice coalitions in which he or she has power, and then normalize those scores so they add to 100. The result is the Sophisticated Index for each Justice. Because of the very stringent requirement of assigning power only for a credible threat of defection, the Sophisticated Index works poorly on Term-by-Term data. The number of cases heard by the Court does not allow enough coalition formation to manifest itself in a single term. We thus only compute the Sophisticated Index for cumulative data.

      Finally, we compute the Modified Median Index. This index is our variant of a computation suggested by Lynn Baker. (8) It attempts to capture the intuition that the power of a Justice is measured by how close that Justice is to the median. To compute this index, we tally the number of winning combinations (i.e. coalitions of size 5 or greater) that contain a particular Justice, and then normalize so the scores add to 100%. Though we believe that this statistic does capture certain aspects of power, we remain skeptical of the notion that the median Justice is the Court's most powerful member.

   2. THE MEDIAN IS NOT THE MESSAGE REDUX (9)

      In response to the first of our articles, Lynn Baker proposed that a better approach to assessing voting power was to seek to identify the median Justice. (10) The intuition was that the closer a Justice was to the median, the more power that Justice would have. In a purely one-dimensional spatial voting model, of course, the median Justice has all of the power. We further challenged Professor Baker's assertion by questioning whether the Median Voter Theorem really applies to the Court. (11) In particular we argued that the cases before the Court were not predominately one dimensional (12) and that the quest for a median Justice would therefore be fruitless.

      Since that exchange there have been a number of methodological advances in the search for a median Justice. (13) One paper in particular challenged our assertion specifically: "We too can identify particular cases that violate the condition of a single-dimension issue space but, as it turns out, the great majority of disputes before the Supreme Court do not. E.g., of the 8,889 case in which the Court heard oral arguments and decided between the 1953 and 2002 terms, only 3.79 percent (n=337) contained more than one issue ... See Harold J. Spaeth, United States Supreme Court Database, ..." (14) We remain unpersuaded.

      This appeal to Harold Spaeth's database leaves unanswered the question of whether the one-dimensional spatial voting model realistically and usefully tracks the behavior of the Supreme Court. While the unidimensional model may indeed be adequate for many situations, we doubt whether it is adequate for capturing the very fine-grained nature of locating the median Justice. To illustrate this concern, we will show that even a very sophisticated method of finding the median Justice leads to problematic results that cast doubt on the consistency of the results with the observed coalitional structure of the Court.

      There are two reasons why we reject the above-cited numerical rejoinder and appeal to the Spaeth database. The first is that the 3.79% figure expresses very little about the number of cases with a multi-dimensional issue space. The second is that the coding used to construct the Spaeth database is not a reliable indicator of the...