• 1 Introduction

  The International Court of Justice (ICJ) is one of the principal organs of the United Nations (UN) and was established in accordance with Article 7 of the Charter of the UN in 1945.¹x The UN Charter, signed on 26 June 1945, entered into force on 24 October 1945. The ICJ Statute is the constitutional document for the ICJ, which is an integral part of the UN Charter. The ICJ accepts cases that only involve States as applicant and respondent.²x Art. 34(1) of the ICJ Statute. All member States of the UN, covering almost all States in the world, are ipso facto parties to the ICJ Statue.³x Art. 93(1) of the UN Charter. The ICJ possesses a broad jurisdiction covering all matters and all cases that State parties refer to it.⁴x Art. 36(1) of the ICJ Statute. Most common types of cases before the ICJ include boundary dispute, use of force, property rights and aerial incidents.⁵x T. Ginsburg and R.H. McAdams, ‘Adjudicating in Anarchy: An Expressive Theory of International Dispute Resolution’, 45 William & Mary Law Review 1229, at 1229-1340 (2004).
  The ICJ consists of fifteen permanent judges, each of whom shall come from a different State.⁶x Art. 3(1) of the ICJ Statute. Furthermore, pursuant to Article 3(2) of the ICJ Statute, in case of a judge having more than one nationality, he or she shall be deemed to be a national of the one in which he or she ordinarily exercises civil and political rights. Usually all these fifteen judges shall sit before a particular case.⁷x Art. 25(1) of the ICJ Statute. But there are two exceptions. One is applied when a judge is on leave (judicial vocations or periodic leaves) or prevented from attending due to illness⁸x Art. 23 of the ICJ Statute. or because that he or she should not take part in the decision of a particular case for some special reason.⁹x Art. 24 of the ICJ Statute. Another exception is that the ICJ may form a chamber composed of three or more judges to deal with particular categories of cases, such as labour cases and cases concerning transit and communications.¹⁰x Art. 26 of the ICJ Statute. Notably, the form of chambers has not been commonly used by the ICJ to date. Between 2005 and 2016, only one case among thirty-two contentious cases has utilised the form of a chamber.¹¹x The case using a chamber is the 2004 Frontier Dispute (Benin/Niger) case. The ICJ declared a Chamber to deal with this case composed of five judges. See Frontier Dispute (Benin/Niger), Order of 16 February 2005 (Composition of Chamber), available at: <www.icj-cij.org/docket/files/125/11047.pdf> (last visited 25 November 2017). Among these five judges, one is from the application State Benin, and one is from the respondent State Niger. As a result, this special chamber form does not affect our final calculations. A State party can appoint an ad hoc judge if no judge of its nationality sits among the permanent judges.¹²x Art. 31 of the ICJ Statute. As a result, under normal circumstances, there will be fifteen to seventeen judges sitting for an ICJ case.
  It is stated in the ICJ Statute that the ICJ judges, either permanent or ad hoc, shall exercise his or her powers impartially.¹³x Art. 20 of the ICJ Statute. This paper is interested in whether this statement is in line with the practice. In this light, the research question of this paper is whether the voting pattern of the ICJ judges is biased. For this purpose, an empirical study is conducted. As to the structure of this paper, the second part summarises the previous literature on the same topic. On this basis, the third part introduces the methodology for this study. Relevant data are described in the fourth part and results of the empirical analysis are presented in the fifth part. In the sixth part, we discuss some possible explanations for the results that we have observed. Several conclusions are drawn in the final part.

• 2 Literature Review

  A considerable amount of literature addresses the voting behaviour of international judges from the perspective of normative analysis. Some literature explores the historical development of the legal regime of national judges, especially ad hoc judges. Article 31 of the ICJ Statute sets forth the regime of ad hoc judges, as mentioned above. In fact, this provision mirrors Article 31 of the Statute of the Permanent Court of International Justice (PCIJ), which was the predecessor of the ICJ.¹⁴x A. Zimmermann et al., The Statute of the International Court of justice: A commentary (2012), at 533. In accordance with the preparatory work regarding the drafting process of Article 31 of the PCIJ Statute, the ad hoc judges’ regime was established for the purpose of addressing the inequality between parties which might be caused when there was “upon the bench a judge of the nationality of only one of the parties to the case”.¹⁵x Ibid., at 531. In this light, Lord Phillimore, a member of the Advisory Committee, stated:

  it would be preferable to give a national representative to both parties, not only to protect their interests, but to enable the Court to understand certain questions which require highly specialised knowledge and relate to the differences between the various legal systems.¹⁶x League of Nations. Advisory Committee of Jurists, Procès-verbaux of the proceedings of the Committee, June 16th-July 24th, 1920, with Annexes, at 528-9, available at: <https://archive.org/details/procsverbauxof00leaguoft> (last visited 25 November 2017).

  In a similar vein, several scholars have addressed the raison d’être of the national judges, especially ad hoc judges, by underlining the political necessity behind this regime.¹⁷x S. Rosenne, The International Court of Justice: An Essay in Political and Legal Theory (1957), at 147; M. Kuijer, ‘Voting Behaviour and National Bias in the European Court of Human Rights and the International Court of Justice’, 10 Leiden Journal of International Law 49, at 51-2, 54-6 (1997); S. Nāgendra, The Role and Record of the International Court of Justice (1989), at 193-4. In particular, it is pointed out that:

  If during the deliberations it becomes apparent that, despite the best efforts of counsel for the party, the case presented by that party has not been clearly understood, the national judge is best placed to appreciate this, and…can ensure that any rejection of that approach – whether he himself favours such rejection or not – is reasoned in such a way as to reassure the party that its point of view has been understood, and that justice has been done.¹⁸x Nāgendra, above n. 17, at 193-4.

  It is widely accepted that the above described function of the national judges, namely to ensure the full understanding of the case presented by a State party, can increase the confidence of that State party in participating in relevant international dispute settlement procedures.¹⁹x Kuijer, above n. 17, at 66-7.
  Furthermore, some literature aims at identifying the factors influencing the impartiality of international judges. Inter alia, it is suggested that the ICJ is a “semi-legal, semi-judicial, semi-political body which nations sometimes accept and sometimes not”.²⁰x D.R. Robinson, ‘The Role of Politics in the Election and the Work of Judges of the International Court of Justice’, 97 Proceedings of the Annual Meeting (American Society of International Law) 277, at 277-82 (2003). Also see M. Reisman, ‘Metamorphoses: Judge Shigeru Oda and the International Court of Justice’, 33 Canadian Yearbook of International Law/Annuaire canadien de droit international 185, at 185-221 (1996); A. Chayes and A.H. Chayes, The New Sovereignty: Compliance with International Regulatory Agreements (1996). Put differently, the decisions made by the ICJ are not only based on legal considerations but also on political considerations. Besides, influential factors on judicial behaviour are summarised as including legal education and training, views on the function of law, conceptions of institutional propriety and belonging, wider external community expectations, and national or political bias.²¹x G.I. Hernández, ‘Impartiality and Bias at the International Court of Justice’, 1 Cambridge J. Int’l & Comp. L. 183, at 183-207 (2012).
  In addition to normative analysis, some scholars also conduct empirical researches on this topic. Nevertheless, most empirical researches only focus on the voting behaviour of the national judges, including judges elected or appointed ad hoc, in cases where their home or appointing States are involved, namely “the extent to which national judges reach their decisions consistently with the position of their national governments”.²²x Kuijer, above n. 17, at 50; I.R. Suh, ‘Voting Behavior of National Judges in International Courts’, 63 American Journal of International Law 224, at 224-36 (1969). These researches show that the national judges have voted in favour of his/her own government in more than 90% of the disputes.²³x “In almost 94% of the disputes, the national judge voted in favour of his own government…In only four instances did a judge vote against the position of his government.” See Kuijer, above n. 17, at 63. In accordance with Il Ro Suh’s research, about 91% (108 out of 119) of the ad hoc judges voted in favour of their governments’ positions in the contentious cases or advisory opinions of the PCIJ and the ICJ between 1922 and 1967. See Suh, above n. 22, at 228, 230. As observed by William Samore, “except in the most isolated of instances, it would be unrealistic to expect that a judge of an international tribunal would ever vote against the country of his origin or nationality”.²⁴x W. Samore, ‘National Origins v. Impartial Decisions: A Study of World Court Holdings’, 34 Chicago-Kent Law Review 193, at 193-221 (1955). In concreto, some scholars conclude that ad hoc judges are more likely to vote in favour of their appointing States than permanent judges.²⁵x In accordance with Il Ro Suh’s research, about 91% of the ad hoc judges voted in favour of their governments’ positions while only about 70% of the permanent judges voted for their home States. See Suh, above n. 22, at 228, 230. ‘It is true that a regular judge should possess a greater sense of responsibility toward his judicial duties than a judge who is specially appointed. For this reason, the case against ad hoc judges is stronger’. See Samore, above n. 24, at 204.
  One of the most systematic empirical researches on this topic was done by Posner and de Figueiredo who studied the ICJ cases from 1946, the beginning year of the ICJ’s operations, to 2004, one year before the publication of their research.²⁶x E.A. Posner and M.F.P. de Figueiredo, ‘Is the International Court of Justice Biased?’, 34 The Journal of Legal Studies 599, at 599-630 (2005). Unlike the above referred researches, this research analysed not only the voting behaviour of national judges but also non-national judges. Their research showed that national bias played an important role in influencing the decision-making of the ICJ.²⁷x Ibid., at 624. Detailed findings of this research included that the ICJ judges voted for their home States or appointing States about 90% of the time. Moreover, when their home States or appointing States were not involved, the ICJ judges were more likely to vote for States that were similar to their home States or appointing States along the dimensions of wealth, culture and political regime.²⁸x Ibid. Posner and de Figueiredo also examined other variables but they did not find significant results. These variables include region, NATO, cold war, jurisdiction, type of cases, judge country, applicant and respondent countries.

• 3 Methodology

  Inspired by the above referred literature, our research intends to conduct a continued study on the ICJ cases between 2005 and 2016. The research purpose is to examine whether the aforesaid findings based on the data between 1946 and 2004 made by Posner and de Figueiredo are still applicable to the data between 2005 and 2016.
  Our null hypothesis is that the voting pattern of the ICJ judges is unbiased. Assuming the null hypothesis is true, decisions made by the ICJ judges are influenced only by relevant legal considerations, such as the proper interpretation or application of an international rule, but not by legally irrelevant considerations, such as whether the applicant or respondent is the judge’s home State (in the case of permanent judges) or appointing State (in the case of ad hoc judges).²⁹x Ibid., at 600-1.
  On the premise of the null hypothesis, our first hypothesis can be stated as below: no difference exists in the voting pattern of an ICJ judge no matter whether his or her home State or appointing State is an applicant or respondent to a particular case or not. A judge whose home State or appointing State is a party to the case can be called a ‘party judge’, and otherwise a ‘nonparty judge’.
  The second hypothesis is that no difference exists in the voting pattern of an ICJ judge no matter whether the applicant or respondent to a particular case shares the same or similar political, economic or cultural levels with the judge’s home State or appointing State. As inspired by the referred literature, we use democracy, wealth and language to respectively illustrate or measure the political, economic or cultural levels of relevant States.³⁰x Ibid., at 607-10. We code the variables, including one dependent variable and six explanatory variables, as follows (see Table 1.

  3.1 Judge-Vote

  The dependent variable. In accordance with Article 95(1) of the 1978 Rules of Court, the ICJ judgements shall contain the number and names of the judges constituting the majority.³¹x Art. 95(1) of the 1978 Rules of Court, adopted on 14 April 1978 and entered into force on 1 July 1978. Consequently, these judgements are able to show clearly each judge’s vote on each issue.³²x Data of the ICJ judgements, available at: <www.icj-cij.org/en/cases> (last visited 25 November 2017). The value of this variable equals to one if the judge rules in favour of the applicant and zero otherwise. Usually judges are asked to decide on more than one substantive issue in one case. Under such circumstances, we code judges as voting in favour of the applicant if they join the majority or concurrence which upholds everything or significant requests that the applicant seek, or if they dissent when the majority opinion does not support the significant requests that the applicant seeks.³³x For example, in the 2014 Whaling in the Antarctic (Australia v. Japan: New Zealand intervening) case, the majority opinion upheld the applicant’s claims. However, four judges (Judges Owada (Japan), Abraham (France), Bennouna (Morocco), Yusuf (Somalia)) dissented the majority opinion. Hence, twelve judges joining the majority opinion voted in favour of the applicant while the foresaid four judges voted for the respondent. Likewise, in the 2015 Application of the Convention on the Prevention and Punishment of the Crime of Genocide (Croatia v. Serbia) case, the majority opinion rejected the applicant’s claims. However, two judges (Judge Cançado Trindade (Brazil); Judge ad hoc Vukas (appointed by the applicant)) dissented the majority opinion. Hence, fifteen judges joining the majority opinion voted in favour of the respondent while the foresaid two judges voted for the applicant.
  

  Table 1 Variable Coding and Data Sources

  Variables	Coding & Sources
  Judge-vote	The dependent variable; equal to one if the judge rules in favour of the applicant and zero otherwise; source from the ICJ website
  App-nationality	Equal to one if the applicant’s country and judge’s country match and zero otherwise; source from the ICJ website
  Res-nationality	Equal to one if the respondent’s country and judge’s country match and zero otherwise; source from the ICJ website
  Democracy	|(judge’s country’s democracy score – respondent’s democracy score)| – |(judge’s country’s democracy score – applicant’s democracy score)|; source from Polity IV
  Wealth	|(judge’s country’s logged per capita GDP – respondent’s logged per capita GDP )| – |(judge’s country’s logged per capita GDP – applicant’s logged per capita GDP )|; source from IMF
  App-language	Equal to one if applicant and judge’s country have same majority language and zero otherwise; source from www.macalester.edu
  Res-language	Equal to one if respondent and judge’s country have same majority language and zero otherwise; source is www.macalester.edu

  Note: Similar methodology of coding variables can be seen in Table A1 of Posner and de Figueiredo’s paper (above n. 26, at 627).

  3.2 App-Nationality

  An explanatory variable. The value of this variable equals to one if the applicant and judge’s home State or appointing State match and zero otherwise. The data on nationality of applicants and judges can be accessed from the ICJ judgements.

  3.3 Res-Nationality

  An explanatory variable. The value of this variable equals to one if the respondent and judge’s home State or appointing State match and zero otherwise. The data on nationality of respondents and judges can be accessed from the ICJ judgements.

  3.4 Democracy

  An explanatory variable. We follow the formula used by Posner and de Figueiredo for the democracy measure³⁴x Ibid., at 612. as below:

  |(judge’s State’s democracy score – respondent’s democracy score)| – |(judge’s State’s democracy score – applicant’s democracy score)|

  We obtain the data on democracy score for each State from the database of Polity IV³⁵x We use the Polity IV Annual Time-Series (1800-2016) database developed by the Center for Systemic Peace as the reference for coding of democracy, available at: <www.systemicpeace.org/inscrdata.html> (last visited 25 November 2011). The latest version was published on 25 July 2017. An earlier version was published in 2014, from 1946-2013. See Polity IV Individual Country Regime Trends, 1946-2013, available at: <www.systemicpeace.org/polity/polity4.htm> (last visited 25 November 2017). which ranges from –10 (authoritarian) to 10 (democracy).³⁶x It is noted that in Posner and de Figueiredo’s paper, though it also used the data from Polity IV, the democracy score range that it gave was from 0 (authoritarian) to 10 (democracy). This is because the indicator we use in this research is POLITY, other than DEMOC used by Poser and de Figueiredo. The reason for using the indicator POLITY is that ‘the POLITY score was added to the POLITY IV data series in recognition of its common usage by users in quantitative research and in the overriding interest of miantaining uniformity among users in this application’. In this sense, the POLITY score is a better way of conducting empirical research. See M. Marshall, T.R. Gurr & K. Jaggers, ‘Polity IV Project: Dataset Users’ Manual’, available at: <www.systemicpeace.org/inscr/p4manualv2016.pdf> (last visited 25 November 2011). The democracy score for each country corresponds to the Polity IV score for the year when the judgement was made.³⁷x In the 2007 Bosnia and Herzegovina v. Serbia and Montenegro case, the score for Bosnia and Herzegovina is represented by Bosnia of 5. As indicated by the 2017 database, Bosnia experienced interruption in the year of this ICJ judgement. Thus, its democracy score for the year 2007 was -66, which represented ‘a period of interruption’. Due to mathematic calculation consideration, we use the 2013 database in which Bosnia got the score of 5, available at: <www.systemicpeace.org/polity/bos2.htm> (last visited 25 November 2011). Likewise, the democracy score for Serbia and Montenegro is based on the score for Serbia and Montenegro from 2003-2006, which stands at 6, excluding the impact of 2006 referendum. In a similar vein, Somalia was experiencing interregnum periods in 2009-2011 and thus its democracy score for this period was -77, which represented ‘periods of interregnum during which there is a complete collapse of central political authority’. Due to calculation reasons, the score for Somalia in the 2009-2011 was -7 as indicated in the 2013 database, available at: <www.systemicpeace.org/polity/som2.htm> (last visited 25 November 2017). In addition, the score for the former Yugoslav Republic of Macedonia is represented by the score for Macedonia, which remains at 9 during the research period. See Marshall et al., above n. 36. In accordance with this formula, if the value of the democracy measure is bigger than zero (positive value), it means that the judge’s State has a closer democracy score to the applicant. When the value is smaller than zero (negative value), then the judge’s State has a closer democracy score to the respondent. Otherwise (zero value) the democracy score of the judge’s State is neither closer to the applicant nor the respondent.

  3.5 Wealth

  An explanatory variable. Like the democracy measure, we also use a single variable for the wealth measure and adopt the formula used by Posner and de Figueiredo³⁸x Posner and de Figueiredo, above n. 26, at 612. as follows:

  |(judge’s State’s logged per capita GDP – respondent’s logged per capita GDP )| – |(judge’s State’s logged per capita GDP – applicant’s logged per capita GDP )|

  We obtain the data on GDP based on PPP per capita for each State for the year when the judgement was made from the database of the International Monetary Fund (IMF).³⁹x Data on GDP based on PPP per capita from IMF, available at: <www.imf.org/external/datamapper/PPPPC@WEO/OEMDC/ADVEC/WEOWORLD> (last visited 25 November 2017). Pursuant to this formula, if the value of the wealth measure is bigger than zero (positive value), it means that the judge’s State has a closer wealth level to the applicant. When the value is smaller than zero (negative value), then the judge’s State has a closer wealth level to the respondent. Otherwise (zero value) the wealth level of the judge’s State is neither closer to the applicant nor the respondent.⁴⁰x Somalia data are missing, thus it is coded as zero as Posner and de Figueiredo did in their research. Also, the data for Serbia and Montenegro are calculated by ½ (Serbia + Montenegro).

  3.6 App-Language

  An explanatory variable. The value of this variable equals to one if the applicant and judge’s home State or appointing State have the same majority language and zero otherwise. We obtain the data on the majority language for each State from the database of Macalester.edu.⁴¹x Data on majority language for each State, available at: <https://www.macalester.edu/research/economics/PAGE/HAVEMAN/Trade.Resources/Data/Gravity/language.txt> (last visited 25 November 2017).

  3.7 Res-Language

  An explanatory variable. The value of this variable equals to one if the respondent and judge’s home State or appointing State have the same majority language and zero otherwise.⁴²x There are several states that are not listed in the MACALESTER coding system. Thus, the authors check the main language and code each state. These States are as follows: (1) Bosnia and Herzegovina, Serbo-Croatian; (2) Serbia and Montenegro, Serbo-Croatian; (3) Ukraine, Ukrainian; (4) Croatia, Serbo-Croatian.
  We mainly adopt two methods in this study. One is the difference tool. Assuming the first hypothesis is true, the ratios of party judges and nonparty judges voting in favour of applicants (or respondents) shall be close. Hence, we use the difference between the aforesaid ratios to examine whether a judge is voting in a biased way under the influence of the variable nationality. Normally, in a case, only two of the judges are party judges and the rest (normally thirteen to fifteen judges) are nonparty judges. If both party judges are biased, we expect that their votes can cancel each other out and thus their bias will not have an important impact on the final outcome. Therefore, it is also of vital importance to examine whether nonparty judges are biased.
  With respect to the second hypothesis, we divide the nonparty judges into two subgroups: nonparty judges whose home States or appointing States share the same or similar political, economic or cultural levels with applicants (or respondents), and other nonparty judges. Assuming the second hypothesis is true, the ratios of these two subgroups of nonparty judges voting in favour of applicants (or respondents) shall be close. The differences between the aforesaid ratios can be used to examine whether a nonparty judge is voting in a biased way under the influence of the variables democracy, wealth or language.
  The other method is the regression tool. We use a similar regression equation as Posner and de Figueiredo’s.⁴³x Posner and de Figueiredo, above n. 26, at 611. Our regression equation is as below:

  Judge-Vote = a + b * [App-Nationality] + c * [Res-Nationality]
  + d * [Democracy] + e * [Wealth]
  + f * [App-Language] + g * [Res-Language]

  We conduct several regressions on the basis of the above referred regression equation. The data that we use for regressions include votes by all judges, either party or nonparty. At the outset, we run a series of single-variable regressions with the data of the dependent variable and each of the explanatory variables in the order of app-nationality and res-nationality, democracy, app-language and res-language, and wealth. Furthermore, we run a series of double-variable regressions. The variable nationality (i.e. app-nationality and res-nationality) is involved in all these double-variable regressions, while the other variable rotates in the order of democracy, app-language and res-language, and wealth. In the end, we run a multiple-variable regression involving all the aforesaid explanatory variables.

• 4 Description of Data

  The ICJ has adjudicated thirty-two cases in total between 2005 and 2016.⁴⁴x Data of the ICJ judgements, available at: <www.icj-cij.org/en/cases> (last visited 25 November 2017). However, only in eighteen cases the ICJ judges had voted on substantive issues.⁴⁵x The list of these 18 cases is as follows: the 2005 Frontier Dispute (Benin/Niger) case; the 2007 Application of the Convention on the Prevention and Punishment of the Crime of Genocide (Bosnia and Herzegovina v. Serbia and Montenegro) case; the 2007 Territorial and Maritime Dispute between Nicaragua and Honduras in the Caribbean Sea (Nicaragua v. Honduras) case; the 2008 Sovereignty over Pedra Branca/Pulau Batu Puteh, Middle Rocks and South Ledge (Malaysia/Singapore) case; the 2008 Certain Questions of Mutual Assistance in Criminal Matters (Djibouti v. France) case; the 2009 Maritime Delimitation in the Black Sea (Romania v. Ukraine) case; the 2009 Dispute regarding Navigational and Related Rights (Costa Rica v. Nicaragua) case; the 2010 Pulp Mills on the River Uruguay (Argentina v. Uruguay) case; the 2010 Ahmadou Sadio Diallo (Republic of Guinea v. Democratic Republic of the Congo) case; the 2011 Application of the Interim Accord of 13 September 1995 (the former Yugoslav Republic of Macedonia v. Greece) case; the 2012 Jurisdictional Immunities of the State (Germany v. Italy: Greece intervening) case; the 2012 Questions relating to the Obligation to Prosecute or Extradite (Belgium v. Senegal) case; the 2012 Territorial and Maritime Dispute (Nicaragua v. Colombia) case; the 2013 Frontier Dispute (Burkina Faso/Niger) case; the 2014 Maritime Dispute (Peru v. Chile) case; the 2014 Whaling in the Antarctic (Australia v. Japan: New Zealand intervening) case; the 2015 Application of the Convention on the Prevention and Punishment of the Crime of Genocide (Croatia v. Serbia) case; and the 2015 Construction of a Road in Costa Rica along the San Juan River (Nicaragua v. Costa Rica) case. Thus, we select these eighteen cases as our study subject for the purpose of researching the voting pattern of ICJ judges.

  4.1 Party Judges

  As mentioned above, ‘party judges’ are those who are either nationals of (in the case of permanent judges) or appointed by (in the case of ad hoc judges) applicants or respondents. In eighteen cases, all applicants and respondents have their party judges except in the 2009 Dispute regarding Navigational and Related Rights (Costa Rica v. Nicaragua) case, the applicant, Costa Rica, did not have its party judge.⁴⁶x At the beginning of the proceedings, Costa Rica chose Mr. Antônio Cançado Trindade (Brazilian) as its ad hoc judge. But Mr. Cançado Trindade was subsequently elected as a permanent judge of the ICJ. Costa Rica informed the ICJ that it had decided not to choose a new ad hoc judge. See the Dispute regarding Navigational and Related Rights (Costa Rica v. Nicaragua) case, Judgement of 13 July 2009, 10, para. 4. As a result, there are thirty-five party judges in total, with seventeen of them nationals of or appointed by applicants and eighteen nationals of or appointed by respondents. In concreto, only two of the thirty-five party judges are permanent judges of the ICJ,⁴⁷x These two permanent judges are: the German judge, Judge Bruno Simma, in the 2012 Jurisdictional Immunities of the State (Germany v. Italy: Greece intervening) case; and the Japanese judge, Judge Hisashi Owada, in the 2014 Whaling in the Antarctic (Australia v. Japan: New Zealand intervening) case. while the remaining thirty-three party judges are ad hoc judges. In this sense, in terms of party judges, this research has more explanatory power on the voting pattern of ad hoc judges rather than permanent party judges. Our data on the voting pattern of party and nonparty judges are provided in Table 2.
  As shown in Table 2, nonparty judges vote for applicants or respondents about 50% of the time. By contrast, party judges vote for applicants that appoint them about 88% of the time. Hence, the probability of a judge voting in favour of the applicant increases by 37 percentage points when the applicant State appoints the judge. Likewise, the probability of a judge voting in favour of the respondent increases by 40 percentage points when the respondent State appoints the judge. As a result, there is substantial evidence that party judges of the ICJ vote in favour of their home States or appointing States.
  Notably, among the thirty-five party judges, two permanent judges, namely two out of two (100%), vote for parties that are their home States. A majority of ad hoc judges, namely twenty-nine out of thirty-three (88%), vote in favour of their appointing States. There are only four exceptions: Firstly, in the 2008 Certain Questions of Mutual Assistance in Criminal Matters (Djibouti v. France) case, all sixteen judges, including the ad hoc judge appointed by France, Judge Gilbert Guillaume, unanimously voted in favour of the applicant’s key claim, finding that “the French Republic…failed to comply with its international obligation under Article 17 of the Convention on Mutual Assistance in Criminal Matters between the two Parties”.⁴⁸x Certain Questions of Mutual Assistance in Criminal Matters (Djibouti v. France), Judgement of 4 June 2008, para. 205(2)(a). Secondly, in the 2009 Maritime Delimitation in the Black Sea (Romania v. Ukraine) case, all fifteen judges, including the ad hoc judge appointed by Ukraine, namely Judge Jean-Pierre Cot, unanimously voted in favour of the applicant’s key claim on drawing a single maritime boundary.⁴⁹x Maritime Delimitation in the Black Sea (Romania v. Ukraine), Judgement of 3 February 2009, para. 219. Thirdly, in the 2012 Territorial and Maritime Dispute (Nicaragua v. Colombia) case, all fifteen judges, including the ad hoc judge appointed by Nicaragua, namely Judge Mohammed Bedjaoui, unanimously rejected the applicant’s key claims.⁵⁰x Territorial and Maritime Dispute (Nicaragua v. Colombia), Judgement of 19 November 2012, para. 251(3)(4). Fourthly, in the 2013 Frontier Dispute (Burkina Faso/Niger) case, all seventeen judges, including the ad hoc judge appointed by Burkina Faso, namely Judge Yves Daudet, unanimously rejected the applicant’s key claims.⁵¹x Frontier Dispute (Burkina Faso/Niger), Judgement of 16 April 2013, para. 114.

  Table 2 Votes of party and nonparty judges

  	In Favour of Applicant		In Favour of Respondent
  Judges (N=270)	Ratio	%	Ratio	%
  Party Judges	15/17	88	16/18	89
  Nonparty Judges	119/235	51	116/235	49
  Difference (percentage points)		37		40

  Source: the ICJ judgements and the authors’ own calculations.

  4.2 Nonparty Judges

  We intend to study whether the voting pattern of nonparty judges is biased. For this purpose, we examine the difference of the voting behaviour between two subgroups of nonparty judges: nonparty judges whose home States or appointing States share the same or similar political, economic or cultural levels with applicants (or respondents), and other nonparty judges. On the basis of the null hypothesis, we hypothesise that the voting pattern of nonparty judges would not differ regardless of whether their States have same majority language as, or share closer political level or economic level to the applicant (respondent) or not. The raw data of the variable language (app-language and res-language) are binary and thus can be presented in the form of a descriptive table. By contrast, the raw data of the variables democracy and wealth are a series of discrete values and thus will be presented in a different form.
  Table 3 reports results for the voting pattern of nonparty judges whose States speak the same majority languages as applicants (respondents) and other judges. Judges whose States have different majority languages from applicants (respondents) vote in favour of applicants (respondents) about 50% of the time. When judges’ States speak the same majority languages as respondents, the probability of these judges voting in favour of respondents increases by 32 percentage points. However, the change is different when it comes to the situation when judges’ States speak same majority languages as applicants. As indicated by Table 3, there are sixteen judges whose States speak same majority languages as applicants, and only seven out of these sixteen judges vote for applicants. Under this circumstance, the probability of judges voting for applicants decreased by 9 percentage points. In order to figure out the explanation for this result, we have checked the voting details of the remaining nine votes, and found that six of them actually happened when States of judges, applicants and respondents all speak the same majority languages.⁵²x This is the difference between the variable language and the variable nationality. The State of a judge cannot be the same State as the applicant and the respondent at the same time. But the State of a judge, the applicant and the respondent can speak the same majority language. Apparently in accordance with the collected data at present, most judges favour respondents when such circumstance occurs. But it cannot be interpreted to undermine the conclusion that judges of the ICJ vote in favour of party States that speak the same majority languages as their States.
  

  Table 3 Votes of judges whose states speak same majority languages as applicants (respondents) and other judges

  	In Favour of Applicant		In Favour of Respondent
  Judges (N=235)	Ratio	%	Ratio	%
  Judges whose States speak same majority languages as applicants (respondents)	7/16	44	11/14	79
  Other Judges	109/205	53	96/205	47
  Difference (percentage points)		–9		32

  Source: the ICJ judgements, the database of Macalester.edu and the authors’ own calculations.

  Figure 1 shows the relationship between democracy alignment and the probability of a judge favouring the applicant. The y-axis shows the probability of a judge voting in favour of an applicant. The x-axis shows the extent of the match between the democracy score of the judge’s State with the applicant: the bigger the number, the closer the extent of the match. The observations are divided evenly among each value (0-5) on the x-axis (about 39 per value).⁵³x For this purpose, we first sort the entries of the variable democracy_measure from the smallest to the largest. Then we code the first group of thirty-nine entries as value 0, the second, third, fourth and fifth group of thirty-nine entries as values 1, 2, 3 and 4, and in the end the sixth or last group of forty entries as value 5. Figure 1 is in line with the predicted relationship described in our null hypothesis. A judge is not more likely to vote for an applicant when the judge’s State has a closer democracy score to that applicant.
  Likewise, Figure 2 shows the relationship between wealth alignment and the likelihood of a judge voting in favour of the applicant. The y-axis shows the probability of a judge voting in favour of an applicant. The x-axis shows the extent of the match between the wealth level of the judge’s State with the applicant: the bigger the number, the closer the extent of the match. The observations are divided evenly among each value (0-5) on the x-axis (about 39 per value).⁵⁴x For this purpose, we first sort the entries of the variable wealth_measure from the smallest to the largest. Then we code the first group of thirty-nine entries as value 0, the second, third, fourth and fifth group of thirty-nine entries as values 1, 2, 3 and 4, and in the end the sixth or last group of forty entries as value 5. Figure 2 endorses the predicted relationship described in our null hypothesis. A judge is not more likely to vote for an applicant when the judge’s State has a closer wealth level to that applicant.
  

  Relationship between Judge’s Votes and Matching Democracy Level
  Source: the ICJ judgements, the database of Polity IV and the authors’ own calculations.
  Relationship between Judge’s Votes and Matching Wealth Level
  Source: the ICJ judgements, the database of IMF and the authors’ own calculations.

• 5 Results

  In this section, the result of several regressions will be presented. As indicated by Posner and de Figueiredo, the main obstacle for the regression is multicollinearity: the variables, namely democracy, language and wealth, are to some extent related.⁵⁵x Posner and de Figueiredo, above n. 26, at 618. To address this problem, we also run several regressions with different groups of independent variables. Table 4 contains the results of these regressions, describing each variable’s standardised coefficients, with P-values in parentheses.⁵⁶x There are eight regression results of (1) nationality; (2) democracy; (3) nationality and democracy; (4) language; (5) nationality and language; (6) wealth; (7) nationality and wealth; and (8) nationality, democracy, language and wealth. Among them, (1), (3), (4) and (5) are calculated on each variable independently, with the purpose of finding each variable’s coefficient and significance; (2), (4) and (6) are calculated with the nationality variable since nationality is the most significant variable according to the previous research and is the basis of democracy, language and wealth as well; and (8) is the final overall result.
  As can be seen from the regression results, the first row, namely app-nationality variable, is highly significant in all different regressions, and its coefficients are consistent with the above observation that a judge who is the national of or appointed by the applicant is more likely to vote for the applicant. By contrast, the second variable, i.e. res-nationality, is significant in all regressions except when the variable language (including variables app-language and res-language) are involved (regressions (5) and (8) in Table 4).
  In terms of the variable democracy, it has significance when considered alone (regression (2) in Table 4). However, when this variable is combined with the nationality variable (including variables app-nationality and res-nationality), the democracy variable loses its significance. And when the multiple-variable regression is conducted, the democracy variable lacks any significance (regression (8) in Table 4).
  Similarly, with respect to the variable language (including variables app-language and res-language), it also has significance when considered alone (regression (4) in Table 4), but when this variable is combined with the variable nationality, the variable app-language loses its significance and the significance of the variable res-language is weakened (regression (5) in Table 4). A similar situation also exists in the multiple-variable regression (regression (8) in Table 4).
  As to the variable wealth, no significance is shown in all regressions, no matter calculated alone in regression (6), or taken into account other variables in regressions (7) and (8).

  Table 4 Regression Results of Nationality, Democracy, Language and Wealth

  	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
  App-nationality	0.376*** (0.002)		0.363*** (0.004)		0.484*** (0.002)		0.377*** (0.002)	0.468*** (0.005)
  Res-nationality	–0.395*** (0.001)		–0.384*** (0.002)		–0.185 (0.253)		–0.396*** (0.001)	–0.175 (0.298)
  Democracy		0.167* (0.087)	0.006 (0.569)					–0.004 (0.694)
  App-language				0.231** (0.026)	0.017 (0.889)			0.019 (0.874)
  Res-language				–0.346*** (0.001)	–0.233* (0.064)			–0.232* (0.069)
  Wealth						0.101 (0.152)	–0.001 (0.991)	0.010 (0.892)
  Adjusted R2	0.070	0.007	0.067	0.032	0.077	0.004	0.067	0.070

  Notes: N=270. Standardised regression coefficients are reported. P-values are in parentheses. The dependent variable is judge-vote, which equals to one if the judge rules in favour of the applicant and zero otherwise.
  * Significant at the 10% level; ** Significant at the 5% level; *** Significant at the 1% level.
  Source: the ICJ judgements, the databases of Polity IV, Macalester.edu & IMF, and the authors’ own calculations.

• 6 Discussion

  On the basis of the above results, we feel confident to conclude that the variable nationality does play a role in ICJ judges’ decision-makings. This finding regarding cases between 2005 and 2016 is in line with Posner and de Figueiredo’s finding for cases between 1946 and 2004. In their paper, both the variables app-nationality and res-nationality are significant across all regressions.⁵⁷x Posner and de Figueiredo, above n. 26, at 619. On this basis, we can further conclude that the significance of the variable nationality is consistent across all time, from 1946 to 2016. In concreto, the probability of party judges favouring applicants or respondents increase by around 40 percentage points across all time when compared with nonparty judges (Table 2).
  It is worth noting that 33 out of 35 party judges (94%) involved in this empirical research are ad hoc judges. Hence, the above conclusion about the voting behaviour of party judges has more explanatory power on ad hoc judges rather than permanent judges. Put differently, it can be inferred from the results of this research that nationality does play an important role in influencing the voting pattern of ad hoc judges. Besides, as mentioned above, some scholars conclude that ad hoc judges are more likely to favour the appointing States than permanent judges. Considering most of the party judges involved in this research are ad hoc judges, the data of this research are insufficient to confirm or negate the above conclusion.
  With respect to the variable democracy, it does not show any significance in Figure 1 when assessing the voting pattern of nonparty judges alone. By contrast, when taking into account both party and nonparty judges together, this variable shows significance in Table 4 (regression (2)). The latter analysis differs from the former in the sense that the latter adds the voting data of party judges into discussion. Accordingly, it is suggested that the significance in regression mainly results from party judges. As concluded above, the variable nationality does influence the voting pattern of party judges. Considering that the variables nationality and democracy are co-related, the difference in significance between these two results is understandable. This argument is also supported by regressions (3) and (4) in Table 4, in which democracy loses significance when combined with the nationality variable. To sum up, the democracy variable has little impact on the voting pattern of nonparty judges. Nonparty judges from democratic States do not necessarily favour democracies over non-democracies, and vice versa. This is contrary to the results of Posner and de Figueiredo’s research, which found that the democracy variable had significance in the voting behaviour of nonparty judges.⁵⁸x Ibid., at 617-18, 620.
  The variable language, on the other hand, also exerts influence on judges’ decision-making. As indicated by regression (4) in Table 4, both the variables app-language and res-language show a significant connection with the dependent variable judge-vote. Inter alia, the variable res-language has a higher correlation. This finding is contrary to that concluded by Posner and de Figueiredo. In their research, the variable app-language showed stronger significance.⁵⁹x Ibid. Such contrast may result from different cases selected. Another interesting point deserving attention is regression (5) in Table 4. The regression results show that when both variables nationality and language are taken into account, the variables res-nationality and app-language lose their significance, and the significance of the variable res-language becomes weaker. A possible explanation for this finding may be that the variables nationality and language are highly co-related. For example, when a judge’ State is the same as the applicant (or respondent), that judge’s State, for sure, speaks the same majority language as the applicant (or respondent).
  Regarding the remaining variable wealth, the results shown in Figure 2 as well as in Table 4 are totally different from Posner and de Figueiredo’s findings.⁶⁰x Ibid. We do not find any significant correlation between the variable wealth with judges’ voting pattern. Put differently, a judge appointed by a poor State is not necessarily more likely to vote in favour of a poor state than a wealthy State, and vice versa. Different findings between cases in the period 1946-2004 and cases during 2005-2016 in this regard may result from the difference in numbers and substances of the cases selected.⁶¹x It is worthy of note that Posner and de Figueiredo’s data set is about four to five times larger than ours. Unfortunately, we cannot obtain their raw data in the public domain. Therefore, we cannot conduct further comparison between their raw data and ours.
  Notably, the results of this empirical research are not exactly the same as those of Posner and de Figueiredo’s, especially with respect to the influence exerted by the democracy and wealth alignments. These two variables were significant in cases between 1946 and 2004, but not any more in cases between 2005 and 2016. It is noted that the world has experienced significant changes around the Millennium, including the dissolution of the Soviet Union, the end of the cold war, the emergence of multi-polar world economy, globalisation and so on. The diluted influence exerted by the democracy and wealth alignments on the judges’ voting pattern may not be an isolated or surprising event. Just as predicted by Samuel P. Huntington in his classic work, The Clash of Civilizations and the Remaking of World Order, “[w]orld politics is entering a new phase…the fundamental source of conflict in this new world will not be primarily ideological or primarily economic. The great divisions among humankind and the dominating source of conflict will be cultural”.⁶²x S.P. Huntington, The Clash of Civilizations and the Remaking of World Order (1997), at 22. This prediction also echoes with our finding on the variable language that is shown to be significant in influencing the ICJ judges’ voting pattern.
  Various recommendations are raised to tackle the concern that the voting behaviour of ad hoc judges is very likely to be influenced by the political factor or national interests. For example, it is suggested to deprive a judge ad hoc of the right to vote.⁶³x Kuijer, above n. 17, at 67. The rationale behind this suggestion is that since “the function of the judge ad hoc is one of understanding and interpretation”, “the right to vote does not seem to be a necessity”.⁶⁴x Ibid. However, practically speaking, in order to make his/her voice or the additional information he/she provides heard and considered by other judges, it is very important that the judge ad hoc acts as a full member of the Bench.⁶⁵x Ibid. Moreover, as shown by the results of this empirical research, the voting behaviour of nonparty judges is also influenced by other factors such as culture. Following the same rationale of the above suggestion, these nonparty judges should also be deprived of the right to vote. Consequently, none of the judges, including party and nonparty judges, should be entitled to vote, which is apparently ridiculous and impractical. Hence, the suggestion of depriving a judge ad hoc of the right to vote is untenable.
  Meanwhile, the current appointing mechanism of ad hoc judges should be reformed to lessen the interference of the States concerned.⁶⁶x Ibid. The fifteen permanent judges of the ICJ are recommended by State parties to the ICJ Statute and elected for nine-year terms by the concurrent action of the General Assembly and the Security Council of the UN.⁶⁷x Arts. 4 and 5 of the ICJ Statute. By contrast, ad hoc judges are simply designated by a State to the dispute.⁶⁸x Art. 31 of the ICJ Statute. Though the State concerned is recommended to give preference to those who have been nominated as candidates for permanent judges of the ICJ, this preference is not obligatory and in practice, the State has often bypassed the candidate list.⁶⁹x Samore, above n. 24, at 196. In this light, some scholar proposes to reinforce the appointing mechanism of ad hoc judges through the following reforms:

  each member State would be obliged to file a list if three candidates for the position of judge ad hoc at the Registry of the Court before a dispute has risen. In case a judge ad hoc is subsequently needed, the Registry can contact the first candidate without further interference of the state concerned. Should the first candidate be unavailable, the registrar would be able to contact the second and third candidates. In addition, member states could be urged not to appoint a judge ad hoc when neither party to a dispute is represented on the Bench by a regular judge.⁷⁰x Kuijer, above n. 17, at 67.

  Regardless of whether this proposal will be accepted in practice by States parties to the ICJ Statute, this proposal tends to weaken the function of ad hoc judges, which, as mentioned above, is for the purpose of ensuring the full understanding of the case presented by a State party and further increasing the confidence of that State in participating in the judicial proceedings.
  Taking a step back, a normative question raised here is whether the biased voting behaviour of the judges, including party and nonparty, will lead to the unfair interpretation or application of law. The key to answering this question is to determine whether such voting behaviour is a proper exercise of the discretion power of a judge. It is widely accepted that the judges’ discretion is inevitable. Both legal positivists and legal naturalists acknowledge that judges are entitled to exercise their discretion power during judicial procedures.⁷¹x Regarding the opinions of legal positivists, see H.L.A. Hart, The Concept of Law, 2nd ed. (1994). Ronald Dworkin objected to the discretion theory as proposed by legal positivists. In his view, positivists proposed a strong disrection that judges might reach decisions free from legal standards. Nevertheless, he acknowleged the existence of two types of weaker discretion, namely judgement discretion (judges can reach decisions with their own judgement when the standards cannot be applied mechanically) and finality discretion (judges can make final decisions without being reviewed or reversed by other authorities). See R. Dworkin, ‘Judicial Discretion’, 60 The Journal of Philosophy 624, at 624-38 (1963); R. Dworkin, Taking Rights Seriously (1978), at 31-9. Criticism regarding Dworkin’s position, see J. Raz, ‘Legal Principles and the Limits of Law’, 81 The Yale Law Journal 823, at 823-54 (1972).
  Nevertheless, a judge’s discretion is definitely not “free to reach any result he pleases as long as he is able to give some legally plausible argument for it”.⁷²x T.D. Perry, ‘Judicial Method and the Concept of Reasoning’, 80 Ethics 1, at 9 (1969). Instead, a judge “must give the decision that he honestly thinks is best, and reasons which he sincerely regards as the best reasons he is able to give”.⁷³x Ibid. A question naturally arises as to which factors should be taken into consideration by a judge when deciding what constitutes the best decision and the best reasons. In this regard, Thomas M. Franck proposed as below:

  since [judgements are] made by men who, in their attitudes, proclivities, and intellectual tendencies, are to a significant degree products of the environments that related them to local and national systems of social values, there can be no man impartial in disputes between States.⁷⁴x T.M. Franck, ‘Some Psychological Factors in International Third-Party Decision-Making’, 19 Stanford Law Review 1217, at 1220 (1966).

  As shown by this empirical research, the voting pattern of party or nonparty judges is influenced by the political or cultural factors. In this sense, these two factors can be regarded as the influential factors that a judge takes into account when exercising his/her own discretion power. It is understandable that judges are trained to best exercise their discretion, and that such training is inevitably associated with certain political and cultural contexts. This research cannot draw any conclusion regarding whether judges consciously vote in a biased pattern. It is possible that a judge votes for a State party to the dispute that happens to share similar politics or culture with his/her home State without any intention. Consequently, given that discretion is allowed in all legal systems including the system of international law, the influence exerted by the political or cultural factors on the voting behaviour of the judges does not necessarily lead to the unfair interpretation or application of law. Instead, the coming-into-play of these factors may be considered as a proper exercise of the discretion power of a judge.

• 7 Conclusion

  The data provide strong evidence that judges of the ICJ favour applicants that appoint them, and that judges favour respondents that speak same majority languages as their States, and weaker evidence that judges favour respondents that appoint them, and more weakly or no evidence that judges favour applicants that speak same majority languages as their States or that judges are influenced by democracy and wealth alignments. These results suggest that political and cultural factors do play a role in influencing the decision-making of the ICJ judges.
  Nevertheless, to avoid doubt, this research has not shown that judges are consciously biased. All it has shown is that their voting pattern is not in line with the manners described by the null hypothesis. A judge does not necessarily consciously decide to vote in favour of a State that is the same or similar to his or her own State. It is also possible that when a State party needs to appoint a person as a judge in the ICJ, the State may tend to appoint a person with legal opinions in favour of its part at the beginning. All in all, this research aims at providing an empirical perspective on the voting pattern of ICJ judges. It does not reveal that the cases would have been decided more impartially if all the validated bias triggers had been prohibited. In the end, this research does not suggest that the biased voting pattern of the ICJ judges will necessarily lead to unfair interpretation or application of law.

Noten

• * The authors would like to thank Prof. Koen Caminada and the anonymous reviewers for their precious comments during writing this paper and the Chinese Scholarship Council for funding their PhD research at Leiden University, Netherlands. The views expressed are the personal observations of the authors, and the usual disclaimer applies.

• 1 The UN Charter, signed on 26 June 1945, entered into force on 24 October 1945.

• 2 Art. 34(1) of the ICJ Statute.

• 3 Art. 93(1) of the UN Charter.

• 4 Art. 36(1) of the ICJ Statute.

• 5 T. Ginsburg and R.H. McAdams, ‘Adjudicating in Anarchy: An Expressive Theory of International Dispute Resolution’, 45 William & Mary Law Review 1229, at 1229-1340 (2004).

• 6 Art. 3(1) of the ICJ Statute. Furthermore, pursuant to Article 3(2) of the ICJ Statute, in case of a judge having more than one nationality, he or she shall be deemed to be a national of the one in which he or she ordinarily exercises civil and political rights.

• 7 Art. 25(1) of the ICJ Statute.

• 8 Art. 23 of the ICJ Statute.

• 9 Art. 24 of the ICJ Statute.

• 10 Art. 26 of the ICJ Statute.

• 11 The case using a chamber is the 2004 Frontier Dispute (Benin/Niger) case. The ICJ declared a Chamber to deal with this case composed of five judges. See Frontier Dispute (Benin/Niger), Order of 16 February 2005 (Composition of Chamber), available at: <www.icj-cij.org/docket/files/125/11047.pdf> (last visited 25 November 2017). Among these five judges, one is from the application State Benin, and one is from the respondent State Niger. As a result, this special chamber form does not affect our final calculations.

• 12 Art. 31 of the ICJ Statute.

• 13 Art. 20 of the ICJ Statute.

• 14 A. Zimmermann et al., The Statute of the International Court of justice: A commentary (2012), at 533.

• 15 Ibid., at 531.

• 16 League of Nations. Advisory Committee of Jurists, Procès-verbaux of the proceedings of the Committee, June 16th-July 24th, 1920, with Annexes, at 528-9, available at: <https://archive.org/details/procsverbauxof00leaguoft> (last visited 25 November 2017).

• 17 S. Rosenne, The International Court of Justice: An Essay in Political and Legal Theory (1957), at 147; M. Kuijer, ‘Voting Behaviour and National Bias in the European Court of Human Rights and the International Court of Justice’, 10 Leiden Journal of International Law 49, at 51-2, 54-6 (1997); S. Nāgendra, The Role and Record of the International Court of Justice (1989), at 193-4.

• 18 Nāgendra, above n. 17, at 193-4.

• 19 Kuijer, above n. 17, at 66-7.

• 20 D.R. Robinson, ‘The Role of Politics in the Election and the Work of Judges of the International Court of Justice’, 97 Proceedings of the Annual Meeting (American Society of International Law) 277, at 277-82 (2003). Also see M. Reisman, ‘Metamorphoses: Judge Shigeru Oda and the International Court of Justice’, 33 Canadian Yearbook of International Law/Annuaire canadien de droit international 185, at 185-221 (1996); A. Chayes and A.H. Chayes, The New Sovereignty: Compliance with International Regulatory Agreements (1996).

• 21 G.I. Hernández, ‘Impartiality and Bias at the International Court of Justice’, 1 Cambridge J. Int’l & Comp. L. 183, at 183-207 (2012).

• 22 Kuijer, above n. 17, at 50; I.R. Suh, ‘Voting Behavior of National Judges in International Courts’, 63 American Journal of International Law 224, at 224-36 (1969).

• 23 “In almost 94% of the disputes, the national judge voted in favour of his own government…In only four instances did a judge vote against the position of his government.” See Kuijer, above n. 17, at 63. In accordance with Il Ro Suh’s research, about 91% (108 out of 119) of the ad hoc judges voted in favour of their governments’ positions in the contentious cases or advisory opinions of the PCIJ and the ICJ between 1922 and 1967. See Suh, above n. 22, at 228, 230.

• 24 W. Samore, ‘National Origins v. Impartial Decisions: A Study of World Court Holdings’, 34 Chicago-Kent Law Review 193, at 193-221 (1955).

• 25 In accordance with Il Ro Suh’s research, about 91% of the ad hoc judges voted in favour of their governments’ positions while only about 70% of the permanent judges voted for their home States. See Suh, above n. 22, at 228, 230. ‘It is true that a regular judge should possess a greater sense of responsibility toward his judicial duties than a judge who is specially appointed. For this reason, the case against ad hoc judges is stronger’. See Samore, above n. 24, at 204.

• 26 E.A. Posner and M.F.P. de Figueiredo, ‘Is the International Court of Justice Biased?’, 34 The Journal of Legal Studies 599, at 599-630 (2005).

• 27 Ibid., at 624.

• 28 Ibid. Posner and de Figueiredo also examined other variables but they did not find significant results. These variables include region, NATO, cold war, jurisdiction, type of cases, judge country, applicant and respondent countries.

• 29 Ibid., at 600-1.

• 30 Ibid., at 607-10.

• 31 Art. 95(1) of the 1978 Rules of Court, adopted on 14 April 1978 and entered into force on 1 July 1978.

• 32 Data of the ICJ judgements, available at: <www.icj-cij.org/en/cases> (last visited 25 November 2017).

• 33 For example, in the 2014 Whaling in the Antarctic (Australia v. Japan: New Zealand intervening) case, the majority opinion upheld the applicant’s claims. However, four judges (Judges Owada (Japan), Abraham (France), Bennouna (Morocco), Yusuf (Somalia)) dissented the majority opinion. Hence, twelve judges joining the majority opinion voted in favour of the applicant while the foresaid four judges voted for the respondent. Likewise, in the 2015 Application of the Convention on the Prevention and Punishment of the Crime of Genocide (Croatia v. Serbia) case, the majority opinion rejected the applicant’s claims. However, two judges (Judge Cançado Trindade (Brazil); Judge ad hoc Vukas (appointed by the applicant)) dissented the majority opinion. Hence, fifteen judges joining the majority opinion voted in favour of the respondent while the foresaid two judges voted for the applicant.

• 34 Ibid., at 612.

• 35 We use the Polity IV Annual Time-Series (1800-2016) database developed by the Center for Systemic Peace as the reference for coding of democracy, available at: <www.systemicpeace.org/inscrdata.html> (last visited 25 November 2011). The latest version was published on 25 July 2017. An earlier version was published in 2014, from 1946-2013. See Polity IV Individual Country Regime Trends, 1946-2013, available at: <www.systemicpeace.org/polity/polity4.htm> (last visited 25 November 2017).

• 36 It is noted that in Posner and de Figueiredo’s paper, though it also used the data from Polity IV, the democracy score range that it gave was from 0 (authoritarian) to 10 (democracy). This is because the indicator we use in this research is POLITY, other than DEMOC used by Poser and de Figueiredo. The reason for using the indicator POLITY is that ‘the POLITY score was added to the POLITY IV data series in recognition of its common usage by users in quantitative research and in the overriding interest of miantaining uniformity among users in this application’. In this sense, the POLITY score is a better way of conducting empirical research. See M. Marshall, T.R. Gurr & K. Jaggers, ‘Polity IV Project: Dataset Users’ Manual’, available at: <www.systemicpeace.org/inscr/p4manualv2016.pdf> (last visited 25 November 2011).

• 37 In the 2007 Bosnia and Herzegovina v. Serbia and Montenegro case, the score for Bosnia and Herzegovina is represented by Bosnia of 5. As indicated by the 2017 database, Bosnia experienced interruption in the year of this ICJ judgement. Thus, its democracy score for the year 2007 was -66, which represented ‘a period of interruption’. Due to mathematic calculation consideration, we use the 2013 database in which Bosnia got the score of 5, available at: <www.systemicpeace.org/polity/bos2.htm> (last visited 25 November 2011). Likewise, the democracy score for Serbia and Montenegro is based on the score for Serbia and Montenegro from 2003-2006, which stands at 6, excluding the impact of 2006 referendum. In a similar vein, Somalia was experiencing interregnum periods in 2009-2011 and thus its democracy score for this period was -77, which represented ‘periods of interregnum during which there is a complete collapse of central political authority’. Due to calculation reasons, the score for Somalia in the 2009-2011 was -7 as indicated in the 2013 database, available at: <www.systemicpeace.org/polity/som2.htm> (last visited 25 November 2017). In addition, the score for the former Yugoslav Republic of Macedonia is represented by the score for Macedonia, which remains at 9 during the research period. See Marshall et al., above n. 36.

• 38 Posner and de Figueiredo, above n. 26, at 612.

• 39 Data on GDP based on PPP per capita from IMF, available at: <www.imf.org/external/datamapper/PPPPC@WEO/OEMDC/ADVEC/WEOWORLD> (last visited 25 November 2017).

• 40 Somalia data are missing, thus it is coded as zero as Posner and de Figueiredo did in their research. Also, the data for Serbia and Montenegro are calculated by ½ (Serbia + Montenegro).

• 41 Data on majority language for each State, available at: <https://www.macalester.edu/research/economics/PAGE/HAVEMAN/Trade.Resources/Data/Gravity/language.txt> (last visited 25 November 2017).

• 42 There are several states that are not listed in the MACALESTER coding system. Thus, the authors check the main language and code each state. These States are as follows: (1) Bosnia and Herzegovina, Serbo-Croatian; (2) Serbia and Montenegro, Serbo-Croatian; (3) Ukraine, Ukrainian; (4) Croatia, Serbo-Croatian.

• 43 Posner and de Figueiredo, above n. 26, at 611.

• 44 Data of the ICJ judgements, available at: <www.icj-cij.org/en/cases> (last visited 25 November 2017).

• 45 The list of these 18 cases is as follows: the 2005 Frontier Dispute (Benin/Niger) case; the 2007 Application of the Convention on the Prevention and Punishment of the Crime of Genocide (Bosnia and Herzegovina v. Serbia and Montenegro) case; the 2007 Territorial and Maritime Dispute between Nicaragua and Honduras in the Caribbean Sea (Nicaragua v. Honduras) case; the 2008 Sovereignty over Pedra Branca/Pulau Batu Puteh, Middle Rocks and South Ledge (Malaysia/Singapore) case; the 2008 Certain Questions of Mutual Assistance in Criminal Matters (Djibouti v. France) case; the 2009 Maritime Delimitation in the Black Sea (Romania v. Ukraine) case; the 2009 Dispute regarding Navigational and Related Rights (Costa Rica v. Nicaragua) case; the 2010 Pulp Mills on the River Uruguay (Argentina v. Uruguay) case; the 2010 Ahmadou Sadio Diallo (Republic of Guinea v. Democratic Republic of the Congo) case; the 2011 Application of the Interim Accord of 13 September 1995 (the former Yugoslav Republic of Macedonia v. Greece) case; the 2012 Jurisdictional Immunities of the State (Germany v. Italy: Greece intervening) case; the 2012 Questions relating to the Obligation to Prosecute or Extradite (Belgium v. Senegal) case; the 2012 Territorial and Maritime Dispute (Nicaragua v. Colombia) case; the 2013 Frontier Dispute (Burkina Faso/Niger) case; the 2014 Maritime Dispute (Peru v. Chile) case; the 2014 Whaling in the Antarctic (Australia v. Japan: New Zealand intervening) case; the 2015 Application of the Convention on the Prevention and Punishment of the Crime of Genocide (Croatia v. Serbia) case; and the 2015 Construction of a Road in Costa Rica along the San Juan River (Nicaragua v. Costa Rica) case.

• 46 At the beginning of the proceedings, Costa Rica chose Mr. Antônio Cançado Trindade (Brazilian) as its ad hoc judge. But Mr. Cançado Trindade was subsequently elected as a permanent judge of the ICJ. Costa Rica informed the ICJ that it had decided not to choose a new ad hoc judge. See the Dispute regarding Navigational and Related Rights (Costa Rica v. Nicaragua) case, Judgement of 13 July 2009, 10, para. 4.

• 47 These two permanent judges are: the German judge, Judge Bruno Simma, in the 2012 Jurisdictional Immunities of the State (Germany v. Italy: Greece intervening) case; and the Japanese judge, Judge Hisashi Owada, in the 2014 Whaling in the Antarctic (Australia v. Japan: New Zealand intervening) case.

• 48 Certain Questions of Mutual Assistance in Criminal Matters (Djibouti v. France), Judgement of 4 June 2008, para. 205(2)(a).

• 49 Maritime Delimitation in the Black Sea (Romania v. Ukraine), Judgement of 3 February 2009, para. 219.

• 50 Territorial and Maritime Dispute (Nicaragua v. Colombia), Judgement of 19 November 2012, para. 251(3)(4).

• 51 Frontier Dispute (Burkina Faso/Niger), Judgement of 16 April 2013, para. 114.

• 52 This is the difference between the variable language and the variable nationality. The State of a judge cannot be the same State as the applicant and the respondent at the same time. But the State of a judge, the applicant and the respondent can speak the same majority language.

• 53 For this purpose, we first sort the entries of the variable democracy_measure from the smallest to the largest. Then we code the first group of thirty-nine entries as value 0, the second, third, fourth and fifth group of thirty-nine entries as values 1, 2, 3 and 4, and in the end the sixth or last group of forty entries as value 5.

• 54 For this purpose, we first sort the entries of the variable wealth_measure from the smallest to the largest. Then we code the first group of thirty-nine entries as value 0, the second, third, fourth and fifth group of thirty-nine entries as values 1, 2, 3 and 4, and in the end the sixth or last group of forty entries as value 5.

• 55 Posner and de Figueiredo, above n. 26, at 618.

• 56 There are eight regression results of (1) nationality; (2) democracy; (3) nationality and democracy; (4) language; (5) nationality and language; (6) wealth; (7) nationality and wealth; and (8) nationality, democracy, language and wealth. Among them, (1), (3), (4) and (5) are calculated on each variable independently, with the purpose of finding each variable’s coefficient and significance; (2), (4) and (6) are calculated with the nationality variable since nationality is the most significant variable according to the previous research and is the basis of democracy, language and wealth as well; and (8) is the final overall result.

• 57 Posner and de Figueiredo, above n. 26, at 619.

• 58 Ibid., at 617-18, 620.

• 59 Ibid.

• 60 Ibid.

• 61 It is worthy of note that Posner and de Figueiredo’s data set is about four to five times larger than ours. Unfortunately, we cannot obtain their raw data in the public domain. Therefore, we cannot conduct further comparison between their raw data and ours.

• 62 S.P. Huntington, The Clash of Civilizations and the Remaking of World Order (1997), at 22.

• 63 Kuijer, above n. 17, at 67.

• 64 Ibid.

• 65 Ibid.

• 66 Ibid.

• 67 Arts. 4 and 5 of the ICJ Statute.

• 68 Art. 31 of the ICJ Statute.

• 69 Samore, above n. 24, at 196.

• 70 Kuijer, above n. 17, at 67.

• 71 Regarding the opinions of legal positivists, see H.L.A. Hart, The Concept of Law, 2nd ed. (1994). Ronald Dworkin objected to the discretion theory as proposed by legal positivists. In his view, positivists proposed a strong disrection that judges might reach decisions free from legal standards. Nevertheless, he acknowleged the existence of two types of weaker discretion, namely judgement discretion (judges can reach decisions with their own judgement when the standards cannot be applied mechanically) and finality discretion (judges can make final decisions without being reviewed or reversed by other authorities). See R. Dworkin, ‘Judicial Discretion’, 60 The Journal of Philosophy 624, at 624-38 (1963); R. Dworkin, Taking Rights Seriously (1978), at 31-9. Criticism regarding Dworkin’s position, see J. Raz, ‘Legal Principles and the Limits of Law’, 81 The Yale Law Journal 823, at 823-54 (1972).

• 72 T.D. Perry, ‘Judicial Method and the Concept of Reasoning’, 80 Ethics 1, at 9 (1969).

• 73 Ibid.

• 74 T.M. Franck, ‘Some Psychological Factors in International Third-Party Decision-Making’, 19 Stanford Law Review 1217, at 1220 (1966).