by Eric A. Posner and Miguel F. P. de Figueiredo
ERIC A. POSNER is Kirkland and Ellis Professor of Law, University of Chicago. MIGUEL F. P. DE FIGUEIREDO is a Ph.D. student in political science at the University of California at Berkeley. We thank Fay Booker, Jacob Cogan, Sean Farhang, Andrew Guzman, Larry Helfer, Richard Hynes, Matt Jarvis, Anup Malani, Andrew Martin, Simeon Nichter, Abdul Noury, Keith Smith, Duncan Snidal, Laura Stoker, Erik Voeten, Greg Wawro, John Yoo; participants at seminars at the University of Chicago, Washington University in St. Louis, Columbia University, and the University of California, Berkeley, Law School for helpful comments; and Bill Martin and Wayne Hsiung for excellent research assistance. We are especially grateful to John de Figueiredo, Rui de Figueiredo, Jr., and Tom Miles, who helped us resolve several difficult empirical problems. Posner thanks the Russell Baker Scholars Fund for financial support.
The International Court of Justice (ICJ) has jurisdiction over disputes between nations and has decided dozens of cases since it began operations in 1946. Its defenders argue that the ICJ decides cases impartially. Its critics argue that the members of the ICJ vote the interests of the states that appoint them. Prior empirical scholarship is ambiguous. We test the charge of bias using statistical methods. We find strong evidence that (1) judges favor the states that appoint them and that (2) judges favor states whose wealth level is close to that of the their own states, and weaker evidence that (3) judges favor states whose political system is similar to that of their own states and that (4) (more weakly) judges favor states whose culture (language and religion) is similar to that of their own states. We find weak or no evidence that judges are influenced by regional and military alignments.
The International Court of Justice (ICJ) is the principal judicial organ of the United Nations (UN) and the only international court that has general subject matter jurisdiction over disputes between all of the members of the UN, virtually every state in the world.1
The ICJ has considerable importance, both political and scholarly. Many of the ICJ’s judgments appear to have resolved real international disputes. And although in many other cases states have failed to comply with its judgments or to acknowledge its jurisdiction, the ICJ remains a potent symbol of the possibilities of an international legal system. For its defenders, the ICJ “plays the leading role in legitimating the [international legal] system by resolving its disputes in a principled manner” (Franck 1995, p. 346).2 Critics of the ICJ—mainly politicians and diplomats from states that have recently lost their cases—argue that the ICJ’s rulings are politically motivated (see also Robinson 2003; for some expressions of skepticism by international lawyers, see Reisman 1995 and Chayes and Chayes 1995). In the words of Jeane Kirkpatrick, the ICJ is a “semi-legal, semi-juridical, semi-political body which nations sometimes accept and sometimes don’t.”3
The ICJ is also of intrinsic scholarly interest for legal academics, even those who do not study international law. It is, after all, a court, and it resembles domestic courts in the United States and other countries. A large literature debates judicial voting in domestic courts, focusing on whether judges’ decisions reflect ideology or disinterested application of the conventions of legal reasoning (see Segal and Spaeth 1993; Epstein and Knight 1998; and Bergara, Richman, and Spiller 2003). The academic discussion has a parallel in the dispute about whether the voting of ICJ judges reflects national interests. A study of the voting patterns of ICJ judges might be of interest for those who study domestic judicial decision making.
This paper examines data on the voting patterns of ICJ judges. We test the claim of the critics that the judges vote the interest of the state of which they are a national (their “home state”) rather than enforce international law in a disinterested way. The null hypothesis then is that judges are unbiased. A judge votes in an unbiased way if he or she is influenced only by the relevant legal considerations—such as the proper interpretation of a treaty—and not by legally irrelevant considerations such as whether one party has a military alliance with the judge’s state. The ideal way to determine if a judge is unbiased is just to figure out the proper legal outcome of a dispute and then see if his or her vote matches that outcome, taking into account legitimate differences in the legal cultures in which judges are educated. The problem with this approach, however, is that the proper legal outcome is rarely obvious and, further, judges may make mistakes and vote the wrong way even though they are unbiased.
To avoid this problem, we can look at voting patterns alone and see if they are related to legally irrelevant factors. The null hypothesis implies that an unbiased judge from state X is no more likely to vote for state X than is an unbiased judge from state Y. The unbiased judge from state X is also no more likely to vote for state Z, where Z is an ally of X, than an unbiased judge from state Y, where Z is an enemy of Y. We are thus not assuming that unbiased judges always vote the same way—as there can be legitimate, legally relevant grounds for disagreeing on the outcome of a dispute—but only that their disagreements are random (or correlated with relevant legal factors) and not correlated with political factors.
The simplest way to test this claim is to examine whether judges vote in favor of their home states when that state appears as a party. Previous studies have found some support for this claim but have also disputed the significance of this finding.4 We use more sophisticated empirical tests, as well as more data, to show that, in fact, judges are significantly biased in favor of their home states when that state appears as a party. Whereas judges vote in favor of a party about 50 percent of the time when they have no relationship with it, that figure rises to 85–90 percent when the party is the judge’s home state.
This finding has limited importance, however, because it does not tell us anything about the voting behavior of judges when their home state is not a party. It is possible that only the judges whose home states are parties are biased, in which case their votes cancel out, leaving 13 or so other judges to resolve the case impartially. We hypothesize that even when a judge’s home state is not a party, his or her home state may have an interest in one party prevailing, and that the judge’s vote will reflect his or her state’s interest. Previous studies have found no evidence for this hypothesis. The most recent such study concluded, “[T]he record does not reveal significant [voting] alignments, either on a regional, political, or economic basis. There is a high degree of consensus among the judges on most decisions. The most that can be discerned is that some judges vote more frequently together during certain periods than do others, and that in rare instances, notably with the Soviet and Syrian judges, they have always voted the same way. But there have not been persistent voting alignments which have significantly affected the decisions of the Court” (Weiss 1987, p. 134). However, this study and the earlier studies all have flaws, chiefly, the failure to rely on statistical techniques that control for relevant factors.
To test our hypothesis, we classify states into blocs—on the basis of region, wealth, culture, military and political alliances, and similar factors—so that we can determine whether judges are biased in favor of state parties that belong to the same bloc as the judges’ home states. We find substantial evidence for this hypothesis.
The paper proceeds as follows. Section 2 provides some background, including the history of the ICJ and a brief discussion of the political and academic debates about the ICJ. Section 3 provides our hypotheses. Section 4 describes the data and tests the hypotheses.
The ICJ is not the first world court; it is the successor of the Permanent Court of International Justice (PCIJ). The PCIJ began operations in 1922, and at its peak in the late 1920s and early 1930s it issued about two judgments on contentious cases per year. However, it gradually lost relevance for governments beset by the problems created by the worldwide depression and the rise of fascism. By the late 1930s the PCIJ, like the League of Nations, had become irrelevant, and it was not used at all during World War II.
The founders of the UN resurrected the PCIJ, albeit with a new name, in the hope that a world court would operate more successfully if backed by the UN, which was designed to be a stronger institution than the League of Nations and enjoyed the participation and leadership of the United States.
The ICJ is based on the statute of the ICJ, which is independent of, but referenced by, the UN charter. All members of the UN charter are parties to the statute, so virtually every state has been subject to the jurisdiction of the ICJ. The statute of the ICJ is a vague document and has been supplemented over the years with other agreements, internal court orders, and customs.
The ICJ has jurisdiction over three types of cases: (1) cases by “special agreement,” in which the parties to a dispute agree to submit their case to the court, (2) cases authorized by a treaty that provides that future disputes arising under the treaty will be adjudicated by the ICJ, and (3) cases between states that have declared themselves subject to the “compulsory jurisdiction” of the court. Sixty-four states have accepted the compulsory jurisdiction of the court, albeit frequently with reservations, and numerous multilateral treaties provide for ICJ adjudication.
Fifteen judges sit on the ICJ. Each judge has a 9-year, renewable term. Their terms are staggered, so that the composition of the court shifts by one-third (not counting retirements and deaths) every 3 years. No two judges may share a nationality. Judges must have the standard qualifications, and typically they have significant experience as lawyers, academics, diplomats, or domestic judges. Judges are nominated by states and then voted on by the Security Council and the General Assembly. If a state appears before the court as a party and a national from that state is not currently a judge, the state may appoint an ad hoc judge who serves only for that case but otherwise has the same powers as the permanent judges.
If there are 15 slots but 191 states (by the end of our period), how are the states that receive representation determined? The slots are distributed by region, currently as follows: Africa, three; Latin America, two; Asia, three; Western Europe and “other” states (including Canada, the United States, Australia, and New Zealand), five; and Eastern Europe (including Russia), two. This distribution is the same as that of the Security Council, and the permanent members of the Security Council have, by custom, one slot each.5 Thus, the United States, Russia, Britain, and France nearly always have a judge on the court;6 an average term of about 9 years. In 79 proceedings, one or both of the parties used an ad hoc judge.
The history of the ICJ can be seen as a struggle between the inter- nationalist aspirations of the court’s supporters and the efforts of states to limit their international obligations. Consider the bases of jurisdiction. Jurisdiction by special agreement poses no threat to states because they can avoid it simply by refusing to consent to jurisdiction. The ICJ, in special agreement cases, serves as an elaborate arbitration device. To be sure, unlike traditional arbitration, the state parties that use the ICJ do not select most of the judges, so that the ICJ, unlike traditional arbitration panels, may be willing to decide cases in a way that reflects the interests of states other than the two parties. But for just this reason states may use traditional arbitration rather than the ICJ, if they wish.
Next we have treaty-based jurisdiction. Here state consent is also needed—at the time that the treaty is ratified—so in theory states have nothing to fear from treaty-based jurisdiction. But in practice states sometimes must agree to ICJ resolution of treaty disputes if they want any of the benefits of the treaty, and, as ICJ jurisdiction is always reciprocal, states agree to ICJ jurisdiction so that they have the power to bring other states to court. These states can then find themselves pulled before the ICJ against their will, often many years after the treaty was ratified.
Finally, we have compulsory jurisdiction. Again, states can avoid compulsory jurisdiction by not filing a declaration. But many states have filed this declaration, apparently because they believe the benefit—being able to pull another state before the ICJ—exceeds the costs—being pulled before the ICJ by another state. Note that the obligation is strictly reciprocal: a state can be pulled before the ICJ only by another state that has itself filed the declaration. In addition, most states have, through reservations, consented to compulsory jurisdiction only for a narrow range of cases. The U.S.’s declaration, for example, excluded cases involving national security. When the ICJ nonetheless found that this clause was satisfied in the Nicaragua case (discussed below), the United States pulled out of compulsory jurisdiction. France also withdrew from compulsory jurisdiction after the ICJ took a case without France’s consent in the early 1970s. No permanent member of the Security Council re- mains subject to compulsory jurisdiction except the United Kingdom.
One hundred four cases have been filed with the ICJ; about a quarter of these were dropped before the ICJ was able to make a substantive decision. In 76 cases, the ICJ judges voted on substantive questions.7 The most common type of case involved a border dispute (31), followed by use of force (22), property (14), and aerial incident (14) (Ginsburg and McAdams 2004).8
A few examples follow:
Corfu Channel (1947–49). This case was the ICJ’s first contentious case. In 1946 British warships struck mines in Albanian waters and were damaged. The United Kingdom filed an application with the ICJ charging that Albania was responsible either for laying the mines or for not clearing them. The ICJ held that Albania violated international law and awarded Britain damages of £844,000. The Albanian government re- fused to pay, and a settlement was not reached until 1992 (Rosenne 1995, p. 44).
Treatment in Hungary of Aircraft and Crew of the United States of America
(1954). This case is the first between the two superpowers; it also disappeared because the Soviet Union refused to participate. A few other cases in which the United States or other Western powers filed applications against the Soviet Union or its satellites also never advanced beyond preliminary stages. The Soviet Union and its satellites have never filed applications. For the most part, the ICJ was used during the Cold War (and after) only by Western powers and developing countries.
The Temple of Preah Vihear (1962). This case was one of many border disputes arising from decolonization. Cambodia filed an application against Thailand complaining that Thailand illegally occupied Cambodian territory around the Temple of Preah Vihear. The ICJ ruled in favor of Cambodia. Thailand accepted the judgment and relinquished its claim.
South West Africa (1966). South Africa controlled neighboring territory (now Namibia), claiming the right under a League of Nations mandate. Ethiopia, Liberia, and many other African countries objected to South Africa’s control and its policies and, after political efforts failed, filed an application with the ICJ. The ICJ took jurisdiction over the application on a close vote but then subsequently (after a change in the bench) ruled that it did not have jurisdiction. The case is significant because the outcome outraged the newly powerful bloc of former colonial countries, which resolved to boycott the ICJ (McWhinney et al. 1991, p. 158). The court repudiated its reasoning in a later case (the Namibia advisory opinion ), an event likened to the U.S. Supreme Court’s repudiation of the jurisprudence of Lochner (Lochner v. United States ).9
U.S. Diplomatic and Consular Staff in Tehran (1979–81). The United States filed an application against Iran after the Iranian government permitted militants to seize the U.S. embassy and take members of the embassy staff hostage. The ICJ ruled in favor of the United States, but the ruling did not appear to have any influence on Iran, which refused to participate in proceedings.
Nicaragua (1984). The United States had been supporting insurgents in Nicaragua, which was controlled by the Soviet-backed Sandinista government. The Central Intelligence Agency mined Nicaraguan ports and harbors in a secret operation; when Nicaragua found out, it filed an application in the ICJ, claiming that the United States had violated various treaties as well as general principles of international law. The United States argued that the ICJ did not have jurisdiction because (1) the treaties did not confer jurisdiction on the ICJ and (2) compulsory jurisdiction did not apply. When the ICJ held against the United States, the United States refused to comply with the ruling and withdrew its consent to compulsory jurisdiction.
Breard (1998). Paraguay brought proceedings that challenged the United States’ failure to advise a Paraguayan national of his rights under the Vienna Convention on Consular Relations at the time of his arrest. The ICJ tried to stop the United States from executing the Paraguayan national, but the United States refused to obey the ICJ’s order. The United States lost two subsequent cases (LaGrand , brought by Germany, and Avena , brought by Mexico) on similar facts, and in both those cases also refused to obey the ICJ’s orders.
Figure 1 shows the size of the ICJ’s docket by decade. The docket declined in the 1960s and 1970s and recovered somewhat in the late 1980s and 1990s. The reasons for these changes are complex. In part, the number of states increased; but at the same time, the court seems to have become less popular among the major Western states (for a discussion, see Posner 2004).10
Scholars have proposed a range of motives for judges of domestic courts: they may seek to maximize their wealth, their status, their leisure, attainment of their political goals, or the probability of elevation or other future position. They also may seek sincerely to rule in the manner dictated by law. Empirical studies so far have been suggestive but inconclusive. Numerous studies find that judicial votes are correlated with the ideology or party affiliation of the judge, but these studies are vulnerable to methodological objections. The international setting adds a new factor: national identity. National identity could affect decision making in three ways: psychologically, economically, or via selection effects.
Psychologically, if judges identify with their countries, they may find it difficult to maintain impartiality. International Court of Justice judges are not only nationals who would normally have strong emotional ties with their countries; they also have spent their careers in national service as diplomats, legal advisors, administrators, and politicians.11 Even with the best intentions, they may have trouble seeing the dispute from the perspective of any country but that of their native land. National and linguistic differences may also interfere with the establishment of collegiality on the court.
Economically, judges may be motivated by material incentives. Judges who defy the wills of their governments by holding against it may be penalized. The government may refuse to support them for reappointment and also refuse to give them any other desirable government position after the expiration of their term. These considerations are likely to weigh even more heavily in the calculations of judges from authoritarian states, as these judges do not necessarily have the option to take refuge in the private sector if they displease their government.
The selection effect works as follows. Because governments nominate judges, they can ensure that their judges are not too independent minded by drawing from the pool of officials who have shown reliability and the appropriate attitudes. There is evidence that the nomination of judges is a highly political process (see Rosenne 1995).
It is not the purpose of this paper to decide which of these explanations, if any, are true.12 We are concerned with the question whether these factors or some other factors cause judges to vote in favor of the interests of their home state or, in the case of ad hoc judges, the state that appoints them.13 The contrary view—the null hypothesis—is that judges take their legal role seriously because they are ideologically committed to the development of international law, or think that they are more likely to be rewarded for impartiality than for bias, or are not selected on the basis of national bias.
The simplest hypothesis is that ICJ judges vote in favor of their home state when that country is a party to the case. Thus, if the applicant is the United States and the judge is an American, then the judge will vote in favor of the applicant. If the respondent is Nigeria, and the judge is an ad hoc appointee of Nigeria (whether he or she is Nigerian or not), then the judge will vote in favor of the respondent.
This first hypothesis is simple and easily tested, but it does not resolve the main question, which is whether the ICJ, as a court, is biased. For the normal two-party case, only two of the judges are nationals of the parties. We expect that their votes will cancel each other out, and the question is, what about the other judges? Regarding these judges, we hypothesize that they will vote in favor of the state party whose strategic interest is more closely aligned with the strategic interest of the judge’s home state. We examine several such alignments:
Region. United Nations General Assembly voting often divides along regional lines, and the ICJ has region-based representation. Accordingly, we predict regional alignments. We will focus on continental alignments (North America, South America, Africa, Europe, Asia).14
Military. We predict that North Atlantic Treaty Organization (NATO) states and states within the Soviet sphere of influence voted as blocs during the Cold War (before 1989).
Wealth. Wealthier and poorer countries often form blocs in international conflicts, for example, over trade. Thus, we predict that judges from wealthier countries will favor wealthier parties and that judges from poor countries will favor poorer parties. States may also support members of trade alliances or organizations such as the European Union (EU) and the Organisation for Economic Co-operation and Development (OECD).
Democracy. Many scholars argue that democracies share interests and are more likely to cooperate in international relations (for examples, see Bueno de Mesquita et al. 1999; Lipson 2003; Lake 1992; and Schweller 1992). We thus test the hypothesis that judges from democracies are more likely to favor democracies; we also look at whether judges from nondemocracies are more likely to favor nondemocracies.
Culture. Judges might be biased in favor of states for which they have a cultural affinity. We focus on majority language and religion: we hypothesize that judges are more likely to vote for states with the same language and religion as the judge’s home state. Note that these variables might, in fact, be better proxies for political alignments, especially post- colonial alignments.
United Nations Organization. We look at whether judges from states that are permanent members of the Security Council are more likely to vote for permanent members of the Security Council.
One might argue that region, wealth, democracy, and culture do not reflect strategic alignments so much as national affinities. Perhaps judges from democracies vote in favor of democracies because they feel more sympathy for democracies or the types of actions taken by democracies. If so, rejection of the null hypothesis would not show that judges are motivated by the strategic interests of their home states but by psychological or perhaps philosophical factors. We will return to this possibility in the conclusion.
The case reports include a majority opinion plus concurring and dissenting opinions when they exist. The reports also show a vote tally for each issue that is decided. Earlier reports showed only the vote tally and not the identities of the judges who voted each way, but one can usually (though not always) determine each judge’s vote on each issue by reading all the opinions. Later reports give the vote tally and also reveal the way each judge voted. In general, we coded judges as voting in favor of the applicant if they joined the majority or concurrence, but we also would code a judge as voting in favor of the applicant if he or she dissented because the majority opinion did not give everything that the applicant sought.15
Let Vcj equal one if the vote (V) by a particular judge (j) from a particular state is in favor of the applicant in a particular case (c); otherwise, Vcj equals zero.16 The regression equation is
The first variable—applicant-judge nationality match—is a dummy variable equal to one if the applicant state and the judge’s state are the same; otherwise, the variable equals zero. Respondent-judge nationality match equals one if the respondent state and the judge’s state are the same. For example, if the case is United States versus Iran, then for the observation containing the U.S. judge, applicant match equals one and respondent match equals zero. For the observation containing the Iranian judge, the reverse is true. For the observations containing other judges, both variables equal zero. (See the Appendix for an explanation of the variables.)
Next we look at bloc voting. The NATO applicant match equals one if the applicant is a NATO country and the judge comes from a NATO country. The variable equals zero if the applicant is not a NATO country or it is and the judge does not come from a NATO country. Similarly, NATO respondent match equals one if the respondent is a NATO country and the judge comes from a NATO country. The variable equals zero if the respondent is not a NATO country or it is and the judge does not come from a NATO country. We use a separate interaction variable to capture cases in which the applicant, the respondent, and the judge are from NATO, in which case we predict no bias.17 Note that when a NATO country is an applicant, its own judge is a nationality applicant match as well as a NATO match. The nationality applicant match variable serves as a control in cases such as this.
These principles guide our tests of the other alliances and regional groups, including region, the OECD, EU, and Warsaw Pact18 and also of language and religion. The language and religion match variables equal one if the applicant (or respondent) has the same majority language or religion as the judge’s home state and zero otherwise.
In the case of democracy and wealth, we can use a single variable for each. The formula for the democracy measure is
where the democracy score ranges from 0 (authoritarian) to 10 (democracy). The formula for the wealth measure is
In each case, the variable takes a positive value when the judge’s state and the applicant state are similar along the relevant dimension and the respondent’s state is different. The variable takes a negative value when the judge’s state and the applicant state are different and the judge’s state is closer to the respondent state.
Finally, we use some controls, including controls for type of case (border dispute, use of force, and so forth), type of jurisdiction, the existence of multiple applicants or respondents, the existence of interveners, and so forth.19 Most important, we use fixed effects for cases and judges, in order to ensure that case-specific and judge-specific factors do not bias the results (Greene 1993, pp. 466–69). Suppose, for example, that bloc voting occurs only in hard cases or cases with certain attributes such as geopolitical salience, and does not occur in other cases. If we do not control for case-specific effects, our results will be inflated. A similar point can be made about judge-specific factors.
Before we turn to the data, we should discuss selection effects. We already mentioned one kind of selection effect: governments might nominate judges who are impartial—in the sense that they vote according to legal principles—but happen to hold an idiosyncratic view of the world that favors the legal principles that will end up helping the appointing state in any ICJ litigation. On this view, the ICJ may be biased as an institution even though the individual judges are unbiased. Because we are interested in the institution as a whole, and less interested in the motivations of the judges, this selection effect does not undermine our empirical analysis. We do note, however, that it is unlikely that jurisprudence could be so elastic that a judge could always vote in good faith in favor of his own country.
A more troublesome possible selection effect could occur at the filing stage rather than the appointment stage. Suppose that states file cases with the ICJ only when they predict that the judges will favor them. As a result, the pool of observations does not contain those cases in which (say) a judge votes against his or her home state, and our regression results will exaggerate the extent of the bias. But note that respondents do not have any control over whether they will be pulled into court, so there ought to be no selection bias with respect to judges whose home states are respondents. Yet judges who are nationals of respondent states vote in favor of respondents at roughly the same rate that judges who are nationals of applicant states vote in favor of their home states.20
A final type of selection bias is related to the type of cases that the ICJ is hearing. Suppose that states tend to settle easy cases and litigate hard cases, with the result that only hard cases make it into our data set. It is possible that the ICJ judges would resolve the easy cases in an impartial way (and thus the cases are settled against the expectation of impartial adjudication) and that their biases affect results only when the proper legal outcome is ambiguous. If so, then our results may exaggerate the overall bias of the ICJ.
This problem is more troubling than the others. To address it properly, we would need to have data about the entire universe of cases in which the possibility of litigation in the ICJ existed. This is clearly impossible, and thus our results demonstrate only bias in the cases that actually reach judgment. We should note, however, that it seems unlikely that governments would believe that the ICJ would resolve easy cases impartially if it does not appear to resolve hard cases impartially. If governments observe that hard cases are being resolved in a biased fashion, then one would expect that some government would find it in its interest to try bringing a somewhat easier case to the ICJ when it expects that the ICJ judges would be heavily biased in the government’s favor, as would sometimes be the case. If the ICJ resolves cases like these impartially, then this would be reflected in our results; if not, then there is no reason to doubt our results.
4.2. Description of Data
4.2.1. Who Litigates? Although ICJ judges come from all regions, litigants do not. The main litigants have been the United States (21 cases), Britain (15), France (11), Germany (7), Belgium (5), Iran (5), India (4), Spain (4), Australia (4), and the Netherlands (4). Two-thirds of states have never appeared before the ICJ, including China, Japan, Russia/ USSR, South Korea, Brazil, Argentina, and Poland. During the Cold War, no Soviet satellite appeared before the ICJ. And, until recently, most cases have pitted Western nations against each other or developing countries against each other. Thus, it is immediately clear that we are unlikely to have enough variation to test our hypothesis that general regional and military alignments predict voting patterns.
Table 1. Votes of Party and Nonparty Judges in Proceedings
|In Favor of Applicant||In Favor of Respondent|
4.2.2. Party Judges. By “party judges,” we mean (1) judges who are nationals of one of the state parties and (2) ad hoc judges appointed by one of the state parties because it does not have a national already on the court. Several earlier studies investigate whether party judges are biased. Most of these studies have concluded that they are somewhat but not very biased on the basis of an issue-by-issue comparison of their votes with the votes of nonparty judges (see Weiss 1987). Table 1 pro- vides our data.
Judges favor their home states. They vote for nonhome parties about half of the time; they vote for home states about 90 percent of the time. There is thus substantial evidence that party judges vote in favor of their home state. However, the votes of party judges may cancel each other out, and it is possible that the nonparty judges are unbiased and that therefore the ICJ as a whole renders impartial decisions.
4.2.3. Nonparty Judges. We attempt to measure the biases of nonparty judges by looking for links between their states and the state parties. We hypothesize that nonparty judges are more likely to vote in favor of states that belong to a geopolitical bloc shared by their own states. Table 2 reports results for voting by bloc or alignment.
The table provides some support for the hypothesis of bloc voting. When the judge’s state and one party match—both are members of OECD or they share language or religion—the judge is more likely to vote for the matching state than the nonmatching state. These results are especially strong for language and religion. Regional alignments and Security Council membership seem to be irrelevant, however, and NATO and EU matches are as predicted and significant only for the respondent case. The last two rows provide a basis for comparison. They show that for observations in which the judge does not match either state along any of these dimensions, and for all observations (except when the judge’s home state is a party), the percentage of votes for the applicant is around 50, as one would expect. The only troublesome result is that judges favor respondents when their home state’s region matches that of the applicant, at a statistically significant level.
Table 2. Bloc Voting When Judge’s State Is Not a Party
|Judge-Applicant Match||Judge-Respondent Match|
|%||N||t-Test of Difference||%||N||t-Test of Difference|
Note. Percentage of votes for applicant by a judge from a state that matches the applicant or respondent along the relevant dimension. We exclude cases in which the judge’s home state is a party (or the judge is an ad hoc appointee). NATO = North Atlantic Treaty Organization; EU = European Union; OECD = Organisation for Economic Co-operation and Development.
Figure 2 shows the relationship between wealth alignment and the probability of a judge favoring the applicant.21 The y-axis shows the probability of voting for the applicant. The x-axis shows the extent of the match between the wealth of the judge’s state and the wealth of the applicant: 5 means that the judge’s state and the applicant have high gross domestic products (GDPs) while the respondent has a low GDP or that the judge’s state and the applicant have low GDPs while the respondent has a high GDP. Lower numbers mean that the judge’s state’s GDP is closer to the respondent’s (whether high or low) and farther from the applicant’s. In short, higher values mean that the judge’s state is closer to the applicant’s, and lower values mean that the judge’s state is closer to the respondent’s. The observations are divided evenly among each value on the x-axis (about 120 per value).
Figure 2 shows the predicted relationship. A judge is more likely to vote in favor of wealthy states (per capita GDP) when the judge’s state is wealthy than when the judge’s state is poor. A judge is more likely to vote in favor of a poor state when the judge’s state is poor than when the judge’s state is wealthy.
Figure 3 shows a similar relationship between regime type and the likelihood of a judge favoring the applicant. A low value on the x-axis means that the democracy score for the judge’s state (whether high or low) is close to the democracy score of the respondent (data are from Polity IV; see Marshall and Jaggers 2002). Again, the observations are divided evenly among values (about 182 per value). A high value means that the democracy score for the judge’s state is close to the democracy score of the applicant.
The figure shows the predicted relationship. Judges from democratic states favor democracies over nondemocracies, and judges from non- democratic states favor nondemocracies over democracies.
So far we have limited ourselves to the raw data. The raw data are suggestive but of limited value. In this section, we report the results of several regressions. The main obstacle for our regressions is multicollinearity: wealth, democracy, language, religion, and the various regional groupings are all, to some extent, related—in some cases, with correlations as high as .522 To address this problem, we run several regressions with different groups of independent variables.
We use a series of probit models with and without fixed effects (for examples of papers that have relied on variations of this model or that justify its use, see Hausman and Wise 1978; Beck, Katz, and Tucker 1998; and Laisney and Lechter 2002)23 Tables 3 and 4 report two sets of these regressions. Table 3 contains the results of regressions without fixed effects; Table 4 contains the results of regressions with judge and case fixed effects. The tables contain the standardized coefficients (the marginal probability calculated at the mean) and the value of the z– statistic for each of the variables (in parentheses). The dependent variable is one if the judge votes in favor of granting the applicant relief. Typically, this means that the judge joined the majority or filed a concurrence. We do not report the results of regressions with judge fixed effects only and with case fixed effects only. These results are largely consistent with the reported results.
The first two rows in both tables show the results for the applicant-judge match and respondent-judge match variables. The coefficients are consistent with our hypothesis and highly significant across almost all of the regressions.
The results for wealth are as predicted and significant, as well as robust against alternative specifications.24 The results for democracy are also as predicted and significant at the 99 percent level of confidence when tested alone with the judge-party matches; democracy maintains its statistical significance and is positively signed in accordance with our hypothesis when controls are introduced into the regression.25
The results for religion and language are more mixed but still strong. When a judge’s home state and the applicant have a common religion, the judge displays a strong pattern of voting in favor of the applicant at a high level of statistical significance; the situation is the same when the judge’s home state and the respondent have a common religion (Table 4, regression ). We have a similar result for language, although it is weaker for respondent matches.
Table 3. Probit Estimates with No Fixed Effects
|Applicant||.40 (6.40) **||.38 (5.07)**||.38 (5.98)**||.41 (6.63)**||.40 (6.30)**||.31 (4.20)**||.34 (4.80)**||.26 (2.74)**|
|Respondent||-.42 (6.63) **||-.35 (4.77)**||-.40 (5.93)**||-.42 (6.44)**||-.39 (5.85)**||-.40 (5.76)**||-.37 (5.11)**||-.30 (3.32)**|
|Democracy||.02 (5.95)**||.01 (3.52)**|
|In Per Capita GDPa||.11 (4.12)**||.07 (2.49)*|
|App-region||-.07 (1.38)||-.03 (.49)|
|Res-region||-.06 (1.05)||-.07 (1.03)|
|Inter-region||.10 (1.33)||.07 (.69)|
|App-NATO||-.03 (.42)||-.14 (1.81)+|
|Res-NATO||-.16 (3.72)**||-.10 (1.63)|
|Inter-NATO||.11 (1.16)||.12 (1.06)|
|App-language||.24 (4.48)**||.26 (3.97)**|
|Res-language||-.06 (1.35)||-.01 (.12)|
|Inter-language||-.11 (1.03)||-.09 (.69)|
|App-religion||.15 (3.48)**||.11 (2.11)*|
|Res-religion||-.13 (2.97)**||.01 (.20)|
|Inter-religion||-.04 (.56)||-.13 (1.54)|
Note. Robust z-statistics are in parentheses. The dependent variable is pwin, which is equal to one if the judge rules in favor of the applicant and zero otherwise. Standardized coefficients are reported.
a We rely on a procedure for dealing with missing GDP data described in note 25. Results for the coefficients in model 3 were .00, with a .07 z-statistic, and in model 8 they were -.01, with a .39 z-statistic. Neither result was statistically significant.
z-statistic. Neither result was statistically significant.
+ Significant at the 10% level.
* Significant at the 5% level.
** Significant at the 1% level.
Table 4. Probit Estimates with Judge and Case Fixed Effects
|Applicant||.52 (4.67)**||.57 (3.73)**||.52 (4.47)**||.52 (4.65)**||.52 (4.60)**||.51 (3.46)**||.50 (3.44)**||.41 (1.71)+|
|Respondent||-.55 (4.22)**||-.48 (3.91)**||-.55 (3.87)**||-.55 (4.26)**||-.55 (3.94)**||-.56 (3.89)**||-.55 (3.93)**||-.46 (3.04)**|
|Democracy||.02 (3.22)**||.01 (1.80)+|
|In Per Capita GDPa||.13 (2.86)**||.11 (1.99)*|
|App-region||-.01 (.09)||-.04 (.33)|
|Res-region||.00 (.05)||-.01 (.09)|
|Inter-region||.01 (.09)||.05 (.26)|
|App-NATO||-.14 (1.08)||-.31 (2.19)*|
|Res-NATO||-.18 (1.70)+||-.08 (.56)|
|Inter-NATO||.13 (.72)||.34 (1.71)+|
|App-language||.39 (3.93)**||.48 (3.78)**|
|Res-language||-.10 (1.04)||-.12 (1.09)|
|Inter-language||-.45 (2.65)**||-.36 (1.37)|
|App-religion||.33 (4.41)**||.23 (2.25)*|
|Res-religion||-.16 (2.40)*||-.03 (.34)|
|Inter-religion||-.11 (.84)||-15 (.87)|
Note. Robust z-statistics are in parentheses. The dependent variable is pwin, which is equal to one if the judge rules in favor of the applicant and zero otherwise. Standardized coefficients are reported.
a We rely on a procedure for dealing with missing gross domestic product data described in note 25. Results for the coefficients in model 3 were .07, with a .59 z- statistic, and in model 8 were .01, with a .04 z-statistic. Neither result was statistically significant.
+ Significant at the 10% level.
* Significant at the 5% level.
** Significant at the 1% level.
The results for regional matches and NATO matches are weak. In many regressions, we do not get significant results, and in some regressions the signs are the opposite of what was predicted (regression  in both tables for applicant-NATO match). We do not have a satisfactory explanation for this anomaly, but the problems here are lack of variation, high levels of multicollinearity,26 and, in a few cases, not enough observations. We think the safest conclusion is that we cannot reject the null hypothesis that judges are not biased by NATO and regional matches.27
There are a number of other possible control variables that one might want to use.
Cold War. Perhaps judges were more biased during the Cold War when the world was more polarized than it is today.
Jurisdiction. Perhaps judges are less biased when the parties appear voluntarily (special agreement) than when one party is a respondent.
Type of Case. Perhaps certain types of cases like border disputes are less polarizing than others (like use of force) and judges are less biased in the former types of cases.
Judge Country. Perhaps judges from certain countries (for example, liberal democracies) are less biased than judges from other countries.
Applicant and Respondent Countries. Perhaps some countries are more or less likely to be the subject of bias because they are generally considered good (Sweden) or bad (Libya) world citizens.
We tested all these possibilities and will not burden the reader with our results.28 It is sufficient to say that our main results—for applicant and respondent match, for democracy, for wealth, and (somewhat weaker) for language and religion—are robust. The controls themselves do not appear to be important.29
What do these numbers mean? Are the biases we identify trivial or important?
As we have seen, a judge whose home state does not share a relevant characteristic with either the applicant or respondent votes in favor of the applicant with a probability of around .5. Holding all of the other independent variables at their means, the probability of a judge voting in favor of the applicant increases by 26 percentage points when the judge and applicant are from the same country; when the judge’s country matches the respondent’s country, the likelihood of his or her voting for the applicant decreases by about the same amount. As the democracy variable increases from its minimum to its maximum, the likelihood of a judge favoring the applicant increases by 24 percentage points. As it increases 1 standard deviation around the median, the likelihood of favoring the applicant increases by 7 percentage points. As the GDP per capita variable increases from its minimum to its maximum, the prob- ability that the judge favors the applicant increases by 29 percentage points. As it increases 1 standard deviation around the median, the likelihood increases by 5 percentage points. The probability of a judge voting in favor of the applicant increases by 26 percentage points when the judge’s home state’s language is the same as that of the applicant, compared with the case of no match. But the probability is virtually unchanged when the language match is with the respondent. Religion is similar to language: the judge’s probability of favoring the applicant increases by 11 percentage points when the judge’s state’s majority religion matches the applicant’s, and the probability is also virtually un- changed when the judge’s state’s religion matches the respondent’s religion.
The bottom line on the regressions is clear. Judges vote in favor of their own countries and in favor of countries that match the economic, political, and (somewhat more weakly) cultural attributes of their own countries. As for regional and military groupings—whether economic or strategic—we are hampered by multicollinearity and lack of variation.30
The data suggest that national bias has an important influence on the decision making of the ICJ. Judges vote for their home states about 90 percent of the time. When their home states are not involved, judges vote for states that are similar to their home states—along the dimensions of wealth, culture, and political regime. Judges also may favor the stra- tegic partners of their home states, but here the evidence is weaker because of multicollinearity; if they do, the magnitude of the bias is probably low.
We have not shown in a straightforward way that judges are consciously biased. All that we have shown is that the judges, on the margin, do not vote impartially in the manner prescribed by the null hypothesis. The motivation for their votes may be psychological or cultural; a judge does not necessarily consciously choose to favor a state that is similar to his or her own states.
We also have not shown that judges—consciously or unconsciously— vote in a manner that promotes that strategic interests of their home states; it is possible that the judges vote in a manner that reflects their own psychological or philosophical biases. For example, suppose that the United States seeks closer relations with a poor dictatorship because of that country’s useful strategic location in the war against terror. In an ICJ dispute between that state and, say, Canada, the U.S. judge may vote in favor of Canada because of political, economic, and cultural biases, even though the vote in favor of Canada may violate the U.S. national interest. We do not have enough data to reject this possibility, although the fact remains that whatever motivates the judge in this example, it is not the law.
The evidence also does not prove that the ICJ is dysfunctional, although it gives one pause. For one thing, judges may vote dispassionately when the applicant and respondent are both very similar to their own state; they may also vote dispassionately when the applicant and respondent are both very different from their own states. In these cases, there is no reason for the judges to be biased, although they may be outvoted by judges who are biased. How often such cases arise is hard to say.
In addition, even biased judges may sometimes swallow their biases and vote in an unbiased manner. Judges who vote 90 percent in favor of their home states vote 10 percent against their home states, and so in this small fraction of cases their votes may carry special weight. Of course, it is also possible that they are masking their bias—they may vote against their own states on occasion in order to help maintain the appearance of impartiality. But the possibility of unbiased voting in some cases cannot be dismissed on the basis of our data set.
Whether this level of bias matters depends on what the ICJ is sup- posed to accomplish. According to one study, compliance with ICJ judgments hovers around the 60 percent level (Ginsburg and McAdams 2004; but for doubts, see Posner 2004). It may be that states are aware that the ICJ judges are sometimes but not always biased and that the states are more likely to use the ICJ and comply with judgments when they believe that the judgments are not biased. When a state’s own judge votes against his or her home state, or when judges from a given bloc vote against a party from that bloc, the state may take the judgment more seriously than otherwise and may be more inclined to comply with it. If so, the ICJ may play a useful role, albeit under narrow conditions and for limited purposes.
The founders of the ICJ did anticipate the problem of judicial bias. Some people thought that judges should not be allowed to hear cases involving their home states, precisely because they feared that such judges could not decide the cases impartially; for the same reason, the ad hoc system was anathema. Our evidence vindicates the premise of these critics but not their remedy. For our evidence suggests that even nonparty judges would be influenced by legally irrelevant factors.
The designers themselves appeared to think that party judges would ensure that each state would get a fair hearing during deliberations. It is certainly possible that judges could not be made to understand the claims of a state whose perspective they do not share unless one of their number was a national or representative of that state. Our evidence does not reveal whether the cases were decided more impartially than they would have been if party judges had been prohibited.
APPENDIX: VARIABLE DEFINITIONS AND SOURCES
Our data set consists of all cases for which there was an opinion on a preliminary objection, the merits, or a similarly substantive issue, from the beginning of the ICJ’s operations in 1946 through March 1, 2004. For each case, we determine the vote of each judge on each issue, whether the judge ultimately sided with one party or the other, and whether the judge was a part of the majority, con- currence, or dissent. Records improve over time, in a few earlier cases, we could not always answer these questions for a particular judge in a particular case, in which case the observation was dropped.
Some cases involved multiple proceedings with separate votes (for example, on a preliminary objection and then on the merits). In the reported regressions, we generally used the latest proceeding unless it seemed minor (like an interpretive case), but we reran our regressions using all the proceedings, and the results differ only trivially.
Table A1 contains our main variables with coding and sources, and Table A2 contains the summary statistics for variables that were used in the regression analysis.
Table A1. Variable Coding and Sources
|Variable||Coding and Sources|
|pwin||The dependent variable; equal to one if the judge rules in favor of the applicant and zero otherwise|
|Applicant||Equal to one if the applicant’s country and judge’s country match and zero otherwise|
|Respondent||Equal to one if the respondent’s country and judge’s country match and zero otherwise|
|App-region||Equal to one if the region of the applicant’s country and the judge’s country are the same and zero otherwise; regions are defined by individual continents|
|Res-region||Equal to one if the region of the respondent’s country and the judge’s country are the same and zero otherwise; regions are defined by individual continents|
|Inter-region App-NATO||App-region multiplied by res-region|
Equal to one if the respondent’s country and the judge’s country are both members of NATO and zero otherwise.
|Res-NATO||Equal to one if the respondent’s country and the judge’s country are both members of NATO and zero otherwise.|
|Inter-NATO App-language||App-NATO multiplied by res-NATO|
Equal to one if applicant and judge’s state have same majority language and zero otherwise (Robertson 2004)
|Res-language||Equal to one if respondent and judge’s state have same majority language and zero otherwise (Robertson 2004)|
|Inter-language App-religion||App-language multiplied by res-language|
Equal to one if applicant and judge’s state have same majority religion and zero otherwise (World Facts and Figures 2001)
|Res-religion||Equal to one if respondent and judge’s state have same majority religion and zero otherwise (World Facts and Figures 2001)|
|Inter-religion Democracy||App-religion multiplied by res-religion|
Absolute value of the difference between the judge’s state’s democracy score and the applicant’s democracy score minus the difference between the judge’s state’s democracy score and the respondent’s democracy score; democracy scores are from Polity IV
|ln Per capita GDP||Absolute value of the difference between the judge’s state’s logged per capita GDP and the applicant state’s logged per capita GDP minus the difference between the judge’s state’s logged per capita GDP and the respondent’s logged per capita GDP; equal to zero if data are missing; data come from the Penn World Tables; the figures are adjusted to a 1996 base year and adjust for purchasing power parity|
|Missing GDP data||Equal to one if applicant, respondent, or judge country does not have missing GDP data and zero if the applicant, respondent, or judge country has missing GDP data|
Note. NATO = North Atlantic Treaty Organization; GDP = gross domestic product.
Table A2. Summary Statistics for Variables Used in Regression Analysis
|In GDP per capita||1,560||.02||.56||-2||2|
|Missing GDP data||1,560||.49||.5||0||1|
- Alter, Karen J. 1998. Who Are the “Masters of the Treaty”? European Governments and the European Court of Justice. International Organization 52: 121–47.
- Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker. 1998. Taking Time Seriously: Time-Series–Cross-Section Analysis with a Binary Dependent Variable. American Journal of Political Science 42:1260–88.
- Bergara, Mario, Barak D. Richman, and Pablo T. Spiller. 2003. Modeling Supreme Court Strategic Decision Making: The Congressional Constraint. Legislative Studies Quarterly 28:247–80.
- Bueno de Mesquita, Bruce, James D. Morrow, Randolph M. Siverson, and Alastair Smith. 1999. An Institutional Explanation of the Democratic Peace. American Political Science Review 93:791–807.
- Chamberlain, Gary A. 1980. Analysis of Covariance with Qualitative Data. Review of Economic Studies 47:225–38.
- Chayes, Abram, and Antonia Handler Chayes. 1995. The New Sovereignty: Compliance with International Regulatory Agreements. Cambridge, Mass.: Harvard University Press.
- Cohen, Jacob, and Patricia Cohen. 1983. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences. 2d ed. Hillsdale, N.J.: Erlbaum Press.
- Epstein, Lee, and Jack Knight. 1998. The Choices Justices Make. Washington, D.C.: Congressional Quarterly Press.
- Feenstra, Robert C. 2002. U.S. Import Data. University of California at Davis, Department of Economics and the Center for International Data. http://data.econ.ucdavis.edu.
- Franck, Thomas. 1995. Fairness in International Law and Institutions. New York: Oxford University Press.
- Gibson, James L., and Gregory A. Caldeira. 1995. The Legitimacy of the Court of Justice in the European Union: Models of Institutional Support. American Political Science Review 89:356–76.
- Ginsburg, Tom, and Richard H. McAdams. 2004. Adjudicating in Anarchy: An Expressive Theory of International Dispute Resolution. William and Mary Law Review 45:1229–1339.
- Greene, William. 1993. Econometric Analysis. 2d ed. New York: Macmillan.
- Hausman, Jerry A., and David A. Wise. 1978. A Conditional Probit Model for Qualitative Choice: Discrete Decisions Recognizing Interdependence and Heterogeneous Preferences. Econometrica 46:403–26.
- Hensley, Thomas R. 1968. National Bias and the International Court of Justice. Midwest Journal of Political Science 12:568–86.
- Heston, Alan, Robert Summers, and Bettina Aten. 2002. Penn World Table Version 6.1. Philadelphia: University of Pennsylvania, Center for International Comparisons. http://pwt.econ.upenn.edu/php_site/pwt_index.php.
- Howell, David C. 2002. Treatment of Missing Data. University of Vermont, Department of Psychology. http://www.uvm.edu/∼dhowell/StatPages/More_Stuff/Missing_Data/Missing.html.
- Laisney, Francois, and Michael Lechter. 2002. Almost Consistent Estimation of Panel Probit Models with “Small” Fixed Effects. ZEW Discussion Paper 02-64. Center for European Economic Research, Mannheim.
- Lake, David. 1992. Powerful Pacifists: Democratic States and War. American Political Science Review 86:24–37.
- Lipson, Charles. 2003. Reliable Partners: How Democracies Have Made a Separate Peace. Princeton, N.J.: Princeton University Press.
- Marshall, Monty G., and Keith Jaggers. 2002. Polity IV Dataset (computer file). College Park: University of Maryland, Center for International Development and Conflict Management.
- McWhinney, Edward, Douglas Ross, Grigory Tunkin, and Vladen Vereshchetin, eds. 1991. From Coexistence to Cooperation: International Law and Organization in the Post–Cold War Era. Boston: Martinus Nijhoff.
- Posner, Eric A. 2004. The Decline of the International Court of Justice. John M. Olin Program in Law and Economics Working Paper No. 233. University
of Chicago Law School, Chicago
- Reisman, Michael. 1995. Review of Metamorphoses: Judge Shigeru Oda and the International Court of Justice. Canadian Yearbook of International Law 33:185–221.
- Robertson, Raymond. 2004. International Language Data. http://www .macalester.edu/research/economics/PAGE/HAVEMAN/Trade.Resources/Data/ Gravity/language.txt.
- Robinson, Davis R. 2003. The Role of Politics in the Election and the Work of Judges of the International Court of Justice. Proceedings of the American So- ciety of International Law 97:277-93.
- Rosenne, Shabtai. 1995. The World Court: What It Is and How It Works. Boston: Martinus Nijhoff.
- Samore, William. 1956. National Origins v. Impartial Decisions: A Study of World Court Holding. Chicago-Kent Law Review 34:193–221.
- Schweller, Randall. 1992. Domestic Structure and Preventive War: Are Democracies More Pacific? World Politics 44:235–69.
- Segal, Jeffrey A., and Harold J. Spaeth. 1993. The Supreme Court and the Atti- tudinal Model. New York: Cambridge University Press.
- Suh, Il Ro. 1969. Voting Behavior of National Judges in International Courts. American Journal of International Law 63:224–36.
- Voeten, Erik. 2000. Clashes in the Assembly. International Organization 54: 185–215.
- Weiss, Edith Brown. 1987. Judicial Independence and Impartiality: A Preliminary Inquiry. Pp. 123–54 in The International Court of Justice at a Crossroads, edited by Lori F. Damrosch. Dobbs Ferry, N.Y.: Transnational.
- Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross-Section and Panel Data. Cambridge, Mass.: MIT Press.
- World Facts and Figures. 2001. Religion Statistics by Country. http://www .worldfactsandfigures.com/religion.php.
[Journal of Legal Studies, vol. 34 (June 2005)]
© 2005 by The University of Chicago. All rights reserved. 0047-2530/2005/3402-0018$01.50
References [ + ]
|1.||↩||The International Court of Justice (ICJ) has two other functions as well: to provide advisory opinions to certain international organizations and to appoint arbitrators to other international tribunals; these functions are outside the scope of this paper.|
|2.||↩||Franck describes criticisms of the ICJ as “remarkably toothless.”|
|3.||↩||See Freepedia, “Nicaragua v. United States” (http://en.freepedia.org/Nicaragua_v. _United_States.html).|
|4.||↩||For example, Suh (1969, p. 230) found that judges vote in favor of their government in 82 percent of the cases but concluded that the data do “not support the theoretical contention that the system of national judges must necessarily be out of harmony with international justice.” See also Hensley (1968), Samore (1956), and Weiss (1987).|
|5.||↩||The distribution is not formally recorded but is the custom (Rosenne 1995). There is no official list of the countries in each region, which is a problem for our coding.|
|6.||↩||China did not have a judge from 1967 to 1985.|
|7.||↩||In our data analysis, we consider only these 76 cases and ignore the others. The majority of the 28 remaining cases consist of pending cases or of rulings on administrative and procedural matters, which do not reveal much about the ideological decision making of judges. In addition, we also have some cases that were dropped prior to a merits decision. Note also that a case may have several of what we classify as “proceedings,” that is, an opportunity for the judges to vote. Most cases have one (a ruling on preliminary objections) or two (a ruling also on the merits); a few have three (an interpretive ruling, a ruling on remedy, or a ruling on a request for provisional measures). On average, each case had a bit less than two proceedings.|
|8.||↩||Our numbers are slightly different from Ginsburg and McAdams’s because we count cases in a different way.|
|9.||↩||Lochner-era jurisprudence favored property rights and interfered with the implementation of the New Deal; it was repudiated in 1937, possibly as a result of concerns about Roosevelt’s court-packing plan, and New Deal legislation was allowed to proceed.|
|10.||↩||Part of the spike in the late 1990s is a result of Serbia’s filing of 10 separate cases against the members of the North Atlantic Treaty Organization (NATO). The cases all involved the use of force against Yugoslavia by the NATO powers.|
|11.||↩||See the biographies of the current court at http://www.icj-cij.org/icjwww/igeneralinformation/icvjudge/tomka.htm. Biographies of members of earlier courts can be found in the ICJ Yearbooks (1947–2004). Some of the judges were former professors, albeit ones who have been heavily involved in international legal affairs on behalf of their country.|
|12.||↩||Our data set is not rich enough to allow us to do this; for some speculation about these issues, see Posner (2004).|
|13.||↩||The literature on domestic courts makes a distinction between ideological and sophisticated voting—where an ideological judge votes his or her ideology and a sophisticated judge takes account of the possible responses of Congress and so may suppress his or her ideological instincts when doing so would elicit a negative reaction from Congress. We do not take into account the possibility that states or international institutions might respond to ICJ judgments by overturning them or ignoring them, although the latter does happen. Our working hypothesis is that judges care more about their own government’s and state’s attitudes toward them than the attitudes of other states or international organizations. Thus, the judges are sophisticated, but their incentives are national, not international. Future research might consider the possibility that judges seek to strengthen and legitimate the court and for that reason would sometimes vote against the interests of their own states. A stream of literature on the European Court of Justice (ECJ) (see Alter 1998; Gibson and Caldeira 1995) argues that judges of the ECJ vote impartially in order to strengthen the legitimacy of the court. Unfortunately, ECJ votes are not public, so we cannot test this hypothesis using the method advanced in this paper.|
|14.||↩||The regional representation on the ICJ is not quite the same, but alternative coding does not produce results that are appreciably different.|
|15.||↩||An alternative approach would be to treat the vote on each issue as an observation—some cases had several distinct issues, which the judges voted on separately. The problem is that many issues are of no importance, and often only a single issue matters, so an issue-by-issue approach would overweight trivial issues at the expense of important issues.|
|16.||↩||Note that the special agreement cases do not technically involve an applicant and respondent because they are brought jointly by the two parties. In these cases, the words “applicant” and “respondent” are just placeholders and should be read as “one party or the other.” Nothing in the analysis turns on the identity of a party as an applicant rather than as a respondent.|
|17.||↩||We also test to see if non-NATO judges favor non-NATO litigants but find no such relationship at a statistically significant level.|
|18.||↩||We eventually decided not to include these variables in the reported regressions because of multicollinearity problems, but we provide some data related to them below.|
|19.||↩||We do not report regressions using these controls either because they do not have a substantial impact on the results or because they eliminate too many degrees of freedom.|
|20.||↩||We do find that for some variables (language, religion), the coefficients and level of statistical significance are higher for applicant matches than for respondent matches, which suggests that this selection effect may exist to some extent.|
|21.||↩||We used purchasing power parity-adjusted GDP figures; see Heston, Summers, and Aten (2002).|
|22.||↩||Another problem is that we run probit regressions and almost all of our independent variables are categorical variables. This creates statistical problems that we acknowledge but have no remedy for. We do note that two of our independent variables—the wealth measure and the democracy measure—are continuous and significant in most of the regressions.|
|23.||↩||Wooldridge (2002) suggests that the fixed-effects probit model has attractive features but that its main practical drawbacks include obtaining maximum likelihood for more than about five alternatives and the difficulty in obtaining partial effects on the response probabilities, which involve complex calculation. Chamberlain (1980) suggests that a probit with two-way fixed effects can present a number of statistical irregularities. To address these concerns, we test our data with conditional logit models and find that the results do not substantively vary. These results are available from the authors.|
|24.||↩||Because we have a significant amount of missing data for GDP, we follow a technique suggested by Cohen and Cohen (1983) in which we code all missing GDP data with a zero. We then create a dichotomous missing GDP variable where missing data are coded with a zero and all other values are coded with a one. For a detailed justification for the technique, see Howell (2002).|
|25.||↩||We also test democracy scores dichotomously at a democracy score of 6 and a democracy score of 7; the signs remain the same, but we lose statistical significance.|
|26.||↩||When tested jointly, they are significant at the 95 percent level for applicants but not significant for respondents.|
|27.||↩||Another possible measure of strategic alignment is trade: it is possible that a judge from a state with good trading relations with a party would be more likely to vote for that party. We tried to test this hypothesis using data on trade flows (taken from Feenstra 2002), but because the data are relatively recent (post-1962) and partial (excluding many states), we do not have much confidence in regressions. Simple correlations suggest a positive relationship between a vote for the applicant and variables that measure the relative strength of the trading relationship between the judge’s state and the applicant. There is a positive relationship, for example, between the probability of voting for the applicant and the sum of exports and imports between the judge’s state and the applicant, a negative relationship between the probability of voting for the applicant and the sum of exports and imports between the judge’s state and the respondent, and a positive relationship between the probability of voting for the applicant and the ratio of exports and imports with applicant over exports and imports with respondent (both weighted for judge’s state’s GDP and not). Most of these relationships are significant at the 10 percent level, but not all of them are.|
|28.||↩||Available from the authors. We also ran the regressions after dropping all cases involving interveners, multiple applicants, and multiple respondents; doing this changes our results only trivially. And we ran regressions with fixed effects for applicant, respondent, and judge country; for year; and for type of case and found only minor changes in our results. We also held the sample size constant, dropped interaction terms, tested a number of dummy variables for jurisdiction; tried various alternative measures for GDP and democracy; dropped cases that were unanimous or had only one dissent; and so forth. These variations did not substantially change our results, but some of them reduced degrees of freedom by so much that statistical significance was lost or anomalous results occurred.|
|29.||↩||Controlling for the Cold War period does not have a significant impact on our results, except that we lose significance for the defendant match variable, apparently because after the Cold War we have several observations in which the respondent was not represented by a judge or the U.S. judge voted three times against the United States, or, in two other cases, the judge from the respondent state voted against the respondent. Putting aside the defendant match variable, this is consistent with Voeten (2000, p. 213), which found that post–Cold War voting in the General Assembly shows an East-West cleavage similar to that which prevailed during the Cold War.|
|30.||↩||As an additional test of our results, we did an in-sample prediction on our probit regression that included every variable. We found that our regression coefficients accurately predicted case outcomes 69 percent of the time.|