JUDICIAL INGROUP BIAS IN THE SHADOW OF TERRORISM
Moses Shayo Asaf Zussman
October 11, 2010
Abstract
We study ingroup bias ñthe preferential treatment of members of oneís group ñin naturally occurring data, where economically signiÖcant allocation decisions are made under a strong non-discriminatory norm. Data come from Israeli small claims courts during 2000-2004, where the assignment of a case to an Arab or Jewish judge is e§ectively random. We Önd robust evidence for judicial ingroup bias. Furthermore, this bias is strongly associated with terrorism intensity in the vicinity of the court in the year preceding the ruling. The results are consistent with theory and lab evidence according to which salience of group membership enhances social identiÖcation.
JEL classiÖcation codes: D03, D71, J15, K4, Z13
Keywords: judicial decisions, social identity, ingroup bias, terrorism, ethnicity, discrim- ination.
Wethanktheeditor,LarryKatz,threeanonymousreferees,MayaBar-Hillel,DanBenjamin,E¢ Benm- elech, Yoav Dotan, Haggay Etkes, David Genesove, Eric Gould, Christine Jolls, Dean Karlan, Esteban Klor, Saul Lach, Victor Lavy, Barak Medina, Omer Moav, Daniele Paserman, Michael Rabin, Andrei Shleifer, Raanan Sulitzeanu-Kenan, Noam Zussman and participants of seminars at the Hebrew University, Yale University, the Jerusalem Conference on Behavioral Economics (2010), and the NBER Summer Institute (2010) for valuable comments. Excellent research assistance was provided by Eli Bing, Yifat Ferder, Moran Kaganovski, Dani Kariv, Shelly Mana, Ofer Menachem, Tal Orbach, Adi Raíanan, Reut Rosenthal, Rotem Shacham, Yaíarit Shani, and especially Ittai Shacham. The Maurice Falk Institute for Economic Research in Israel provided generous Önancial support.
1
1 Introduction
Traditional economic models assume that people care only about their self interest. However, people may also care about groups to which they belong. We refer to this phenomenon as social identiÖcation. One of the most extensively studied manifestations of social identiÖca- tion is ingroup bias: the preferential treatment of members of oneís own group. A crucial observation here is that social identiÖcation is endogenous: people do not automatically identify with any group they belong to. In particular, social identiÖcation has been shown to be a§ected by the salience of group-speciÖc attributes.
Evidence for the existence of ingroup bias and for its sensitivity to group salience comes mostly from experiments. In this paper we study ingroup bias and saliency e§ects in naturally occurring data, where professional decision makers make economically signiÖcant allocation decisions under a strong non-discriminatory norm. SpeciÖcally, we analyze judicial decisions in Israeli small claims courts during 2000-04. These courts handle civil cases between private litigants.
Several features make this setting especially suitable for investigating ingroup bias. First, when making a decision, a judge in these courts allocates resources between two individuals who may or may not belong to her social group. This feature resembles standard lab experi- ments which measure ingroup bias. However, unlike allocation decisions in lab experiments, the decisions we study are made by professional judges who are expected to apply the law blindly. Second, the Israeli setting allows us to study social identiÖcation with naturally occurring, ìreal-lifeî, groups: Arabs and Jews. Third, the assignment of cases to judges within a given court is e§ectively random. This facilitates credible estimation of the extent of ingroup bias. Finally, the period studied is characterized by intense ethnically-based ter- rorist attacks. Since these attacks are plausibly exogenous to the legal procedure, they allow us to study the e§ects of ethnic salience on ingroup bias.1
The main source of data used in our analysis is transcripts of decisions made by judges in the small claims courts. From these documents we extract information on the court, litigants, subject of the claim, timing of decision, and claim outcome. The ethnicity of judges and litigants is deduced from their names. Our dataset covers the universe of documents available for 2000-04 where a plainti§ of one ethnicity faces a defendant of a di§erent ethnicity. Our main analysis focuses on 1,748 judicial decisions, 31% of which were made by Arab judges and the rest by their Jewish colleagues.
We Önd robust evidence for the existence of judicial ingroup bias in this period. A claim
1The procedure for allocating cases to judges is described in section 2.2. Randomization tests are in section 4. The exogeneity of terror intensity with respect to case allocation is examined in section 5.1.
2
is between 17% and 20% more likely to be accepted if assigned to a judge of the same ethnicity as the plainti§. In monetary terms the estimated bias translates to over $200 per case. Interestingly, the extent of the bias does not seem to vary with judge experience or with other observed judge characteristics such as gender and education.
The above estimates represent a level of bias that is characteristic of the period as a whole. We next ask whether this bias is an exogenously given feature of inter-ethnic relations in Israel or if, alternatively, it varies with the salience of ethnic cleavages. In particular, we examine whether judicial ingroup bias is related to the intensity of Palestinian politically motivated fatal attacks inside Israel. Results suggest that judicial ingroup bias is positively and signiÖcantly associated with terrorism intensity as measured by the number of fatalities per capita in the area surrounding the court in the year preceding the judicial decision. Furthermore, the data seem to indicate that terrorism a§ects judges of both ethnicities, leading Arab judges to favor Arab plainti§s and Jewish judges to favor Jewish plainti§s.
We interpret these Öndings in terms of a general framework for modeling social identity developed in Shayo (2009). This framework ñoutlined in the online appendix ñattempts to capture both the behavioral e§ects of social identiÖcation and the endogenous determination of the groups people identify with. The basic structure of the model is as follows. A society may have many social groups ñìIsraeliî, ìArabî, ìmiddle classîand so on ñbut in any given situation individuals ìidentifyîwith only some of these. Given their social identities, individuals choose courses of action, which determine the aggregate outcome. That outcome forms the social environment that in turn a§ects the pattern of social identities. A Social Identity Equilibrium is a steady state where (i) each individualís behavior is optimal given her social identity; (ii) social identities are optimal given the social environment; and (iii) the social environment is determined by the behavior of the individuals. The present paper seeks to shed light on two of the major components of the model. First, the e§ect of social identiÖcation on behavior (where we focus on ingroup bias); and second, the e§ect of the social environment on identiÖcation patterns (where we focus on the e§ect of the salience of group-speciÖc attributes).
We note, however, that since this is not a controlled experiment, we cannot completely rule out two alternative interpretations of our Öndings. First, in our setting the person making the allocation decision (the judge) communicates with the individuals receiving the allocation (the litigants). It is thus possible that what underlies the results is not a preferen- tial treatment of members of oneís own group but rather better transmission of information between the litigant and the judge when they belong to the same ethnic group. While we cannot dismiss this possibility, it seems implausible that terrorism intensity should a§ect such di§erence in the ability to communicate. Second, the results might be driven by lit-
3
igant behavior rather than by judge behavior. This does not seem to be a major concern in our setting, since from a legal perspective the relevant actions take place before the legal procedure starts, and certainly before the parties involved know the identity of the judge that will rule in their case. We discuss the interpretation of our results in section 6.
The paper relates to two major literatures. The Örst is the literature on social identity and ingroup bias and the second is the literature on ethnic and racial bias in economic and legal contexts. Ingroup bias has been studied extensively using the experimental setting known as the Minimal Group Paradigm. In these experiments an individual allocates some resource between two other individuals, where the only thing she knows about them is whether they belong to her group or not. Starting with Tajfel et al. (1971), this literature has demonstrated that ingroup bias can emerge even in artiÖcially created groups, and has examined various factors which facilitate its emergence. Another prominent line of research looks at how the salience of group membership a§ects contributions to public goods (e.g. Orbell, van de Kragt, and Dawes 1988; Bornstein 2003; Eckel and Grossman 2005). Other settings where ingroup bias has been studied are reported in Bernhard, Fischbacher, and Fehr (2006), Chen and Li (2009), Fong and Luttmer (2009) and Klor and Shayo (2010). See Shayo (2009) for a review of this literature.
Beyond the literature on social identity, our paper is closely associated with the extensive literature on discrimination. The economic literature identiÖes two major types of discrim- ination: taste-based (Becker 1957) and statistical (Phelps 1972; Arrow 1973). Our paper is more closely related to the former, but rather than treating the taste for discrimination as exogenously given, we seek to study its determinants. Methodologically, the approach we take in identifying the e§ect of terrorism on judicial bias is similar to that of earlier stud- ies which examine the e§ects of shocks to tastes associated with political events. A recent example is Michaels and Zhi (2010) who examine the e§ect of a deterioration in relations be- tween the USA and France in 2002-03 on trade between the two countries. In a similar vein, Moser (2010) argues that WWI adversely a§ected German Americansíprospects of gaining membership in the New York Stock Exchange. In the Israeli context, Miaari, Zussman, and Zussman (2009) examine how the outbreak of the second Palestinian Intifada (uprising) in September 2000 a§ected labor market outcomes of Arabs relative to those of Jewish Israelis.
Finally, a large literature studies possible bias against Blacks and Hispanics in the Ameri- can criminal justice system.2 A major methodological obstacle in this context is the di¢ culty of ruling out potential correlation between race and ethnicity on the one hand and unob- served case characteristics on the other. Several innovative strategies have been used to
2In Israel, Gazal-Ayal and Sulitzeanu-Kenan (2010) study ethnic ingroup bias in detention decisions in criminal courts.
4
tackle this problem. For example, Abrams, Bertrand, and Mullainathan (2007) rely on the random assignment of cases to judges to examine the between-judge variation in incarcer- ation rates of Blacks relative to Whites. They Önd large inter-judge disparity, suggesting that at least some judges di§erentially treat defendants based on their race. Alesina and La Ferrara (2010) use discrepancies in decisions made in lower versus higher courts to provide evidence of bias against minority defendants in capital sentencing. Glaeser and Sacerdote (2003) examine data on vehicular homicides, where the identity of the victim is arguably random, and Önd that drivers who kill Blacks receive signiÖcantly shorter sentences. Finally, McConnell (2010) analyzes judicial decisions in federal courts following 9/11 and Önds no change in sentencing outcomes for any ethnic group other than Hispanics. A novel feature of the identiÖcation strategy we use to study judicial bias is the combined use of random assignment of judges to cases with exogenous variation in the salience of ethnicity. This allows us not only to estimate the extent of ingroup bias but also to explore its sources.
We proceed as follows. The next section describes the historical and institutional setting in which our empirical investigation takes place. In section 3 we explain how the dataset was constructed and provide descriptive statistics. Section 4 estimates the overall level of judicial ingroup bias in the period under investigation while section 5 studies the e§ect of terrorism on the extent of the bias. Section 6 discusses alternative interpretations of the Öndings and section 7 concludes.
2 The setting 2.1 Time and place
We analyze decisions involving Arabs and Jews in Israel in 2000-2004. In this period Arab citizens of Israel numbered roughly 1.25 million, or about 20% of the countryís population. Arab-Jewish relations inside Israel are strongly associated with developments in the Arab- Israeli conáict and in particular with relations between Israel and the Palestinians in the Occupied Territories.
In late September 2000, following a period of relatively calm relations between Israel and the Palestinians, the second Intifada erupted. The ensuing years saw an intense wave of violence between Israelis and Palestinians, claiming the lives of thousands. Suicide bombings resumed in late 2000, peaked in 2002 and subsided in 2004.
In the Örst days of October 2000 there were mass demonstrations and clashes between Arab Israelis and the police which left twelve Arab Israelis dead. These ìOctober Eventsî are widely considered a turning point in Arab-Jewish relations in Israel, contributing to a
5
rise in ethnocentric views among both Arabs and Jews.3
2.2 Small claims courts
Small claims courts operate in many countries around the world, including Australia, Canada, England and the USA. These courts handle civil cases between private litigants. The amount of monetary judgments they can award is capped: in Israel during the period under investi- gation the cap was set at 17,800 New Israeli Shekels (NIS), roughly equal to $US 4,000.
The rules of civil procedure and of evidence in the Israeli small claims courts are relatively simple. The procedure starts when the plainti§ Öles a claim at the court, provides supporting documentation, and pays a small fee. Claims can only be submitted to the court where either: (1) the relevant transaction took place or was supposed to take place; and/or (2) the defendant lives or works. Immediately following the Öling of the claim, the defendant is notiÖed and is instructed to provide a defense statement within Öfteen days. The defendant has the right to submit a counter-claim to which the original plainti§ needs to respond within seven days.
Once a claim was Öled ñor, in Öve courts, after the defendant has responded ñthe case is assigned a trial date and a judge. Due to a backlog in the system, trials are scheduled several months in advance. Each case is assigned to the Örst available slot (this procedure is used in all courts). This means that the assignment of judges to cases within a given court is in principle orthogonal to characteristics of the case.4
The judge receives the case materials no earlier than a week before the trial. Importantly, the plainti§ and the defendant represent themselves in the trial, i.e. the litigants appear without lawyers. During the trial, which typically lasts only a few minutes, the judge sees the litigants for the Örst time and hears their arguments.5 The judge has to issue a ruling in the case within seven days of the trial. Litigants who wish to appeal a ruling need to Örst request approval from the relevant district court.
Three features of these small claims courts make them especially appealing for analyzing ingroup bias. First, unlike courts which handle criminal cases, in small claims courts the judge decides on monetary transfers between two individuals. Our investigation is restricted to cases where litigants belong to two di§erent ethnic groups. Since the judge belongs to one
3It is important to note, however, that participation of Arab Israelis in acts of politically motivated violence (either in concert with Palestinians from the Occupied Territories or independently) has remained negligible in scale.
4Judges cannot normally decline to rule in a case unless they are personally acquainted with one of the litigants. In such cases they need to notify the court management of the circumstances.
5Litigants have a right to ask for an interpreter to be present in the court if they are not proÖcient in Hebrew.
6
of these groups, this generates a situation resembling standard (Minimal Group) experiments measuring ingroup bias. However, there are three crucial di§erences between our setting and typical experimental settings: (a) decision makers are professional and operate under a strong non-discriminatory norm (equality before the law); (b) monetary stakes are quite signiÖcant ñthe average compensation requested by plainti§s in our sample is roughly $1,460; and (c) the groups in our setting are natural, i.e. they are not formed by the researcher.
A second important feature of small claims courts is that judges receive the case materials at most a week before the trial, meet the litigants only once, and are forced to produce decisions within a week. This means that the proximate timing of the decision is known (as opposed to, for example, criminal cases involving protracted procedures, where it is hard to tell at what stage of the trial the judicial decisions are actually made). Another possible implication of the need to make decisions quickly is that it can make judges more susceptible to stereotyping and bias (on implicit bias see e.g. Bertrand, Chugh, and Mullainathan [2005] and Jolls and Sunstein [2006]).
Finally, since the ability of litigants to appeal decisions is limited and since the decisions do not attract public attention ñthese are after all small claims ñjudges in these courts enjoy almost complete discretion. The fact that the decision is made by a single judge (rather than by a jury or a panel of judges) reinforces the judgeís discretionary power.
3 Data
Our main source of data is online transcripts of judicial decisions (rulings). These documents Örst became available online in late 2000 in a handful of courts and coverage widened over time. The documents record the names of the judge and each of the litigants and typically include several paragraphs which sketch the arguments made by the litigants and the ruling of the judge. We cover the universe of available decision documents until December 31, 2004 (N=26,444). For each document, we code whether each of the litigants is a private citizen, a business or a government agency. If the litigant is private, we code his or her ethnicity (Arab or Jewish) using a procedure detailed in the online appendix. Coding ethnicity employs data from the Israel Population Registry which allow us to compute the likelihood of any given name being associated with an Arab or Jewish citizen. The accuracy of this procedure follows from the fact (apparent in the Registry data) that there is very little overlap between Jewish and Arab names.6
Having coded litigantsíethnicities for all available documents, we keep only ìmixed
6Note that the data only allows us to distinguish between Jews and Arabs, and not between subgroups (e.g. Moslem and non-Moslem Arabs).
7
casesî: those where at least one private plainti§ and one private defendant are of di§er- ent ethnicities (N=2,027). For these cases we conduct a comprehensive analysis of the documents.7 Focusing on mixed cases allows us to examine the situations that are of prime interest and that most resemble standard lab experiments which study ingroup bias.8
For the mixed cases, we extract data on the following:
Court.
Judgeís name (which we later link with biographical information).
Litigants: in addition to information about type (private, business or a government agency) and ethnicity, we use the wording of the decision document and litigantsí names to code gender.
Claim subject (e.g. breach of contract, tra¢ c accident etc.).
Timing of decision (trial dates are not reported in the decision documents but, as
mentioned above, the decision is made within seven days of the trial).
Monetary compensation requested by the plainti§ and whether a counter claim has been Öled.
Claim outcome: whether the claim was accepted (partly or fully), rejected, settled outside the court or withdrawn; the monetary transfers; and the legal expenses awarded (if any).
The main analysis in this paper excludes cases that were settled outside the court (121 cases) or withdrawn (58).9 We also exclude cases with multiple plainti§s (defendants) such that one plainti§ (defendant) is Jewish and another is Arab (107). Finally, we exclude cases where the court is located in the Occupied Territories (1). This leaves us with 1,748 cases.
Table I shows, for each of the 25 courts, the percentage of cases by the ethnicity of the judge, plainti§ and defendant. Most of the cases are in the two northernmost districts
7 Each document is coded independently by two di§erent coders (law students at the Hebrew University). A third (senior) coder veriÖes the coding and adjudicates cases where there is an incompatibility across coders in any of the Öelds (this happened in 14% of the cases).
8A comprehensive analysis of the universe of cases would have been prohibitively costly and would not drastically alter our ability to address the questions at hand.
9 In section 4 we examine the possibility that litigants strategically decide to settle cases outside the court or withdraw them. It might be interesting to note that the share of cases settled outside of court (6%) seems rather low. If this is indeed the case, it could suggest a di¢ culty in reaching settlement when the litigants are from di§erent ethnic groups (we thank Andrew Daughety, Jennifer Reinganum and Kathryn Spier for this observation).
8
(Northern and Haifa). This is largely due to the combination of two factors. First, the Israeli Arab population is concentrated in the north of the country. Second, online coverage of cases began earlier in the north than in other parts of the country. Overall, 31% of the cases in our data were ruled by Arab judges. Arabs make up 44% of the plainti§s and 56% of the defendants. In several courts there are no Arab judges while in others most of the cases are ruled by Arab judges.
[Table I]
We use Öve di§erent measures of trial outcome. The main measure is a binary variable which takes the value of one if the claim was accepted and zero otherwise. A second outcome variable attempts to distinguish between claims that were partly or fully accepted. This distinction is not straightforward: while in all cases we have information on the monetary compensation awarded by the judge, in more than 60% of the cases we do not know the sum requested. Nonetheless, we can sometimes deduce from the wording of the decision that the claim was ìfully accepted.îThis yields an ordered categorical variable that takes three values: rejected (coded 0), partly accepted (1), or fully accepted (2).
A third measure of trial outcome is the monetary compensation awarded by the judge to the plainti§ net of the compensation awarded to the defendant (in case there was a counter claim). A fourth measure is the legal expenses awarded to the plainti§ net of the expenses awarded to the defendant. The last two variables can take positive as well as zero and negative values. Finally, we look at the ratio between the net monetary compensation awarded by the judge to the plainti§ (inclusive of legal expenses) and the sum requested by the plainti§.
Additional information on judges is obtained from their biographies. Most biographies are available online. The rest were obtained from the court system using freedom of information procedures. Overall, we have 132 judges, Öfteen of whom are Arab.
Table II provides summary statistics for cases (panel A) and judges (panel B). With respect to outcomes, 73% of the claims were accepted (53% partly). Mean net monetary compensation is NIS 3,079 (roughly $700) and mean net legal expenses is NIS 189. On average, plainti§s receive 80% of the amount they request. Turning to case characteristics, we see that tra¢ c accidents account for almost seventy percent of the claims in our data, while thirteen percent have to do with a breach of contract.10 In about Öfteen percent of the cases the subject of the claim cannot be deduced from the decision document. Some documents note that the ruling was given under a condition of ìno defense.îThis means
10The high percentage of tra¢ c accidents might be a sign of ethnic segregation: Arabs interact with Jews during this period mostly on the road.
9
either that no defense statement was submitted or that the defendant(s) failed to appear in the trial (it is not possible to distinguish between these two possibilities). This happened in thirteen percent of the cases (with the others coded ìdefense presentî). A counter claim was Öled by the defense in nine percent of the cases. There is usually only one plainti§ in a case, but often more than one defendant. Almost all cases were Öled by private plainti§s while the share of private litigants out of the total number of defendants is 74% on average. The vast majority of litigants are male. Finally, for the 660 cases for which we have information on the compensation requested by the plainti§, the average amount is NIS 6,424 ($1,460). In terms of timing, there are relatively few cases in 2000-01, as online coverage of decision documents was still limited. Cases are uniformly distributed across months with very few cases decided on a weekend (not reported).
[Table II]
Panel B shows judge characteristics separately for Arab and Jewish judges. Arab judges are on average 5 years younger than their Jewish colleagues, with the average age of Arab judges being 44. Average tenure as a judge is four years for Arab judges and Öve years for Jewish ones. Two thirds of the Arab judges are male while Jewish judges are equally split across genders. Twenty three percent of the Jewish judges, and none of the Arab judges, were born outside of Israel. There are also some di§erences in the academic institution from which judges received their LLB degree, with the Hebrew University having the largest share overall. Less than 20% of these judges hold a degree higher than LLB.
4 Judicial ingroup bias
In this section we estimate the extent of judicial ingroup bias for the entire period under study (2000-04). Based on the court procedures described in section 2.2, our identiÖcation strategy assumes that within each court judge assignment is orthogonal to case characteristics. We start by assessing the validity of this assumption.
Table III examines di§erences in observed characteristics of cases assigned to Arab and Jewish judges. The Örst two columns show mean characteristics for each set of cases, while the third column presents the di§erence between the two means. The fourth column reports this di§erence controlling for court Öxed e§ects. Comparing raw overall means, we Önd some statistically signiÖcant di§erences (column (3)). For example, cases assigned to Arab judges are 11 percentage points more likely to have an Arab plainti§.11 Similarly, the share
11The Arab population is geographically concentrated in certain regions, and Arab judges mostly serve in courts in these regions (see Table I). If plainti§s tend to Öle claims in the area in which they live, courts where there is a high concentration of Arab judges will also have a high concentration of Arab plainti§s.
10
of private defendants and the share of male defendants are lower, while the share of male plainti§s is higher, in cases assigned to Arab judges. However, consistent with the assignment procedure described above, within courts there is no signiÖcant di§erence across Arab and Jewish judges in any observable case characteristic (column (4)). This lends support to our identiÖcation strategy: there is little evidence to suggest that ñ in a given court ñ cases assigned to Arab judges are systematically di§erent from those assigned to Jewish ones.
[Table III]
The results in Table III notwithstanding, a potential concern arises from the fact that since in principle plainti§s can Önd out the identity of the judge prior to the trial, they may withdraw their claim if they were assigned a judge of the opposite ethnicity. This could a§ect our estimates of judicial bias. If ñas may seem reasonable ñweaker claims are more likely to be withdrawn, then claims handled by a judge of a di§erent ethnicity from that of the plainti§ would be stronger on average than claims handled by a judge of the same ethnicity as the plainti§. This would produce a downward bias in the estimated judicial ingroup bias. In Table C.1 of the online appendix we test whether cases are more likely to be withdrawn when assigned to a judge of the opposite ethnicity to that of the plainti§. We Önd no evidence for such an association over the period as a whole. We similarly Önd no evidence that cases are more (or less) likely to be settled outside of court when assigned to a judge of the opposite ethnicity to that of the plainti§.
A Önal potential concern is that, under certain circumstances, plainti§s may have the opportunity to choose in which court to submit their claim (see section 2.2). Since we generally do not know where litigants reside, we cannot tell in which cases plainti§s had such a choice. To the extent that such opportunities were present, one might worry that plainti§s would tend to Öle their claims in courts where there is a relatively high proportion of judges from their own ethnic group (call this proportion p). This may bias our estimate of judicial ingroup bias if there is an association between p and the strength of the claim. However, it is again reasonable to assume that such an association ñif it exists ñwill produce a downward biased estimate, as plainti§s are more likely to choose courts strategically when their claims are weaker.12
12 To see this more clearly, consider a plainti§ with a choice between two courts, one closer to her place of residence than the other (call these courts Home and Away). Suppose that the plainti§ knows the p in each court and believes that her chances of winning are higher the higher is p. This plainti§ would incur the cost of submitting her claim in the Away court if and only if this su¢ ciently improves her chances of winning the case. There are two possibilities. If p is at least as high in the Home as in the Away court, she would submit at Home. In contrast, if p is higher in the Away court she may Öle there. Now, if the case is ìairtightî(i.e. the probability of winning at Home is close to 1) there is little reason to Öle in the Away court since this cannot signiÖcantly improve the chances of winning. However, if the case is su¢ ciently weak,
11
4.1 Results
We start by presenting general patterns of judicial decisions in the raw data. Figure I displays the share of claims accepted by judge and plainti§ ethnicity. The left pair of bars pertains to cases where the plainti§ is Jewish and the defendant is Arab. Seventy nine percent of these claims are accepted when the judge is Jewish while only 72% are accepted when the judge is Arab. This in itself is not necessarily evidence for ingroup bias: for example, it may be the case that compared to their Arab colleagues, Jewish judges are somewhat more inclined towards plainti§s. However, if this was the only reason for the di§erence, we would expect to observe a similar pattern regardless of plainti§ ethnicity. In fact, the right pair of bars shows that when the plainti§ is Arab, the pattern is reversed: Jewish judges accept 65% of these claims while Arab judges accept 75%.
[Figure I]
Table IV presents a di§erences-in-di§erences analysis of the raw data. As the top row shows, Arab judges are 3.7 percentage points more likely to accept a claim when the plainti§ is Arab rather than Jewish. Again, in itself this is no evidence for ingroup bias: Arab plainti§s might on average Öle stronger claims than Jewish plainti§s. However, Jewish judges (second row) are 14.4 percentage points less likely to accept a claim when the plainti§ is Arab rather than Jewish. The di§erence in these di§erences ñ18% ñprovides an indication of the extent of ingroup bias (i.e. by how much are Arab judges more likely than their Jewish colleagues to accept a claim Öled by an Arab plainti§ rather than by a Jewish one). It should be emphasized that, absent an ethnicity-free benchmark, it is impossible to speculate on whether and to what extent Jewish judges favor Jewish litigants and Arab judges favor Arab litigants. We revisit this issue in section 5.2 below.
[Table IV]
We now turn to an econometric investigation. Our baseline speciÖcation is of the form: yijct = 0 + 1ArabPlainti i + 2ArabJudgei + 3ArabPlainti ArabJudgei (1)
+c + ijct
and the expected gain from Öling at the Away court is independent of the strength of the case, there may be su¢ cient incentive to Öle at the Away (high-p) court. This would mean that claims Öled in courts with a high proportion of judges from the same ethnic group as the plainti§ would be weaker on average. The same conclusion obtains if the expected gain from Öling at the Away court decreases with the strength of the claim (results are ambiguous if the gain increases with the strength of the claim).
12
where yijct is the outcome of case i; assigned to judge j; in court c; at time t; c is a court Öxed e§ect; and ijct is an error term clustered within judge.13 ArabPlainti§, ArabJudge and the interaction term ArabPlainti§ ArabJudge are indicator variables.
Equation (1) allows for two possible di§erences across ethnic groups which, as mentioned above, do not necessarily indicate ingroup bias. First, it is possible that claims submitted by Arab plainti§s have di§erent unobserved characteristics than those submitted by Jewish plainti§s. Thus, 1 may be nonzero even in the absence of ingroup bias. Second, it is possible that Arab judges are di§erently inclined towards plainti§s than their Jewish colleagues. In other words, 2 may be nonzero even in the absence of ingroup bias. Our interest is in 3, which captures ingroup bias.14
Column (1) in Table V presents the results using the binary outcome measure, i.e. whether the claim was accepted or rejected. The estimates suggest that Arab plainti§s are Öfteen percentage points less likely than Jewish plainti§s to win a case and that Arab judges are eight percentage points less likely than Jewish judges to accept claims. The main result is in the third row, which shows a positive and highly statistically signiÖcant degree of ingroup bias. A claim is seventeen percentage points more likely to be accepted if assigned to a judge of the same ethnicity as the plainti§.
[Table V]
We next augment the baseline speciÖcation with additional controls. SpeciÖcally, we estimate:
yijct = 0 + 1ArabPlainti i + 3ArabPlainti ArabJudgei (2) + j +tenurejt +Xi0 +t +c +ijct
13Notice that while the judge is the relevant treatment and we allow for clustering at this level, the cluster- ing problem is not very central in our setting since the main explanatory variable ñArabPlainti§ ArabJudge ñvaries within the treatment group. Nonetheless, we allow for clustering at the judge level to address possible within-judge correlations (which might exist even with the judge Öxed e§ects in equation (2) below). This yields slightly higher standard errors than either uncorrected or heteroskedasticity-robust standard errors.
14This coe¢ cient captures a di§erence-in-di§erences controlling for court Öxed e§ects (and potentially other factors). To see this note that
E(yjArabJudge, JewishPlainti§, controls) E(yjJewishJudge, JewishPlainti§, controls) = 2 E(yjArabJudge, ArabPlainti§, controls) E(yjJewishJudge, ArabPlainti§, controls) = 2 + 3:
13
where j is a judge Öxed e§ect and tenurejt is judgeís tenure at the job.15 The vector Xi is a list of case-speciÖc controls that includes: the number of plainti§s; the number of defendants; the share of private plainti§s; the share of private defendants; the share of male plainti§s; the share of male defendants; the amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for ìdefense presentî; and an indicator for cases where the defendant Öled a counter-claim. t is a vector of year, month and day of week dummies.
In columns (2)-(4) of Table V we progressively add these sets of controls. This sig- niÖcantly increases the explanatory power of the regression. The degree of ingroup bias is, however, robust to the inclusion of the additional variables and has a point estimate of 19.2% in the full speciÖcation. This is a slightly stronger e§ect than that of Öling a counter claim (included in the unreported case controls and estimated at 0:154 with a 0:046 standard error).
To check the extent to which these results are driven by the decisions of a single judge, we repeatedly estimate the full regression (column (4) of Table V), each time removing from the sample cases ruled by a di§erent judge. The point estimate of 3 in these 132 regressions ranges from 0.158 to 0.207 (and is always highly statistically signiÖcant).
As we have seen in Panel B of Table II, Arab and Jewish judges di§er in some of their characteristics. It is possible that such di§erences, rather than ethnic ingroup bias, are behind the above results. For example, the share of judges holding only an LLB degree is higher among Arabs than among Jews. If educated judges are more sympathetic to ñor better understand ñeducated litigants and Jewish litigants are more educated than Arab ones, this will generate an association between judge-litigant ethnicity combinations and claim outcomes.16 To examine this possibility, in Table VI we augment equation (2) with interactions between judge characteristics and the Arab Plainti§ indicator. While some of these interactions seem to be associated with case outcomes, the estimated judicial ingroup bias (second row) is robust to their inclusion.
[Table VI]
In the online appendix we report results from examining two sub-samples. First, as Glaeser and Sacerdote (2003) argue, tra¢ c accidents often take place without each of the parties involved being aware of the identity of the other party. This contrasts with other disputes that involve face-to-face interactions, potentially generating di§erences in the extent
15The judge Öxed e§ect picks up any time invariant judge characteristics which may a§ect her rulings. Note that adding the judge Öxed e§ect implies dropping the ArabJudge indicator from the model. Note also that we keep the court Öxed e§ect c as a few judges rule in two (and in one case three) nearby courts.
16We thank Larry Katz for this observation.
14
of ingroup bias across di§erent types of cases. However, the estimated judicial ingroup bias (using equation (2)) for the sub-sample of tra¢ c-related cases (N=1,205) is 0.185, which is very similar to the estimate of 0.192 obtained using the full sample (see Table C.2 in the online appendix). Second, as mentioned in section 2.2, some courts allocate cases to judges only after the defense statement is received. Excluding these courts from the analysis (leaving 1,190 observations) yields an estimated ingroup bias of 0.214 (see Table C.3 in the online appendix).17
In Table VII we estimate the extent of ingroup bias using the four alternative outcome measures described above. In the Örst two columns the dependant variable takes three values according to whether the claim was rejected (0), partly accepted (1), or fully accepted (2). The qualitative results, using either OLS or Ordered Probit, are the same as those obtained in Table V using the binary outcome measure.
[Table VII]
In column (3) the dependent variable is the net monetary compensation awarded by the judge to the plainti§ (compensation awarded to plainti§ minus compensation awarded to defendant). The results indicate that a plainti§ facing a judge of the same ethnicity receives on average NIS 926 (roughly $210) more than a similar plainti§ facing a judge of the opposite ethnicity. To put this Ögure in perspective, recall that the maximum compensation that can be requested in these small claims courts is NIS 17,800 while ñin the 660 cases where we have this information ñthe average compensation requested by the plainti§s is NIS 6,424.
We next examine the net legal expenses awarded by the judge to the plainti§ (expenses awarded to plainti§ minus expenses awarded to defendant). Legal expenses were awarded in 76% of the cases. The decision on legal expenses is plausibly even more discretionary than the decision to accept or reject the claim and the decision on the amount of compensation to award. The decision to accept a claim is in principle grounded in the judgeís reading of the facts of the matter while the compensation awarded is based on the documents (e.g. a car damage assessment) submitted to the court. In contrast, it is hard to establish the appropri- ate legal expenses, e.g. the amount and value of time expended on the legal procedure (recall no lawyers are allowed in small claims courts). The results (column (4)) indicate an ingroup bias of NIS 224 ($50) in legal expenses. This is roughly 0.45 of the standard deviation of net legal expenses (see Table II, Panel A) whereas the bias in net monetary compensation reported in column (3) is 0.24 of the standard deviation of this variable.
Finally, in column (5) the dependent variable is the monetary yield of the claim, deÖned as the ratio between the net monetary compensation (including legal expenses) awarded
17We are grateful to two anonymous referees for suggesting these tests. 15
to the plainti§ and the compensation requested by the plainti§. As mentioned above, the denominator in this ratio is available for only 660 cases. Consequently, the bias is not estimated very precisely. Nonetheless, the point estimate suggests that a plainti§ receives on average 10% more of the amount requested when facing a judge of the same ethnicity.
An important remaining question is whether the extent of ingroup bias varies with judge characteristics. The relationship between agent characteristics and discriminatory behavior has received relatively little attention in the economics literature. A notable exception is List (2004) who Önds that experienced dealers are better able to statistically discriminate against minorities. A crucial di§erence between our setting and that studied by List is that we an- alyze public-sector decision makers who, while operating under a strong non-discriminatory norm, have no direct monetary stakes in trial outcomes. In Table VIII we conduct an ex- ploratory investigation of the relation between judicial bias and judge characteristics. We do so by interacting observable judge characteristics with the ArabPlainti§ *ArabJudge variable (controlling for the judge characteristics and their interactions with the ArabPlainti§ and ArabJudge variables). The data exhibit no systematic association between the extent of the bias and any of the judge characteristics.
[Table VIII]
5 The shadow of terrorism
The previous section establishes the existence of judicial ingroup bias in Israeli small claims courts during 2000-04. An interesting and important question is whether and to what extent this bias is a§ected by the social environment. In particular, the period under study is characterized by intense levels of ethnically-based violence, which may well lead to stronger ethnic identiÖcation. In this section we examine whether variations in terrorism intensity across space and time ñwhich are plausibly exogenous to the legal procedure ña§ect the extent of judicial ingroup bias. Such a result would be consistent with the extensive literature on the e§ects of group salience on ingroup bias.
5.1 Data
We use data on all Palestinian politically motivated fatal attacks inside Israel (i.e. excluding the Occupied Territories). For each attack we have information about date, location, and number of civilian and security forces fatalities.18 We merge these data with the judicial
18The dataset combines information from several sources: BíTselem, the Israeli Information Center for Human Rights in the Occupied Territories; The Israeli Ministry of Foreign A§airs; the Israeli National
16
decision data used above.19
Table IX reports the number of fatalities from terrorist attacks by district and year.
Panel A reports civilian fatalities only, while panel B reports total fatalities (civilian and security forces). These Ögures are normalized by the population in each district and year. The table reveals substantial variation across districts with the most severely hit districts being Jerusalem and Haifa. The intensity of violence increased until 2002 and subsided in the following years. Overall, there were 615 fatalities, 514 of them civilian.
[Table IX]
In the analysis below, our measure of terrorism intensity is the (population adjusted) number of fatalities from attacks that occurred in a given geographical area around the court during the year preceding the judicial decision. We examine three alternative geographical areas around the court. Natural area is the smallest geographic unit examined, followed by sub-district and district. Our data span 24 natural areas, 15 sub-districts, and 6 districts. Descriptive statistics on fatalities are at the bottom of Panel A of Table II.
The use of within-country temporal and spatial variation in terrorism fatalities to identify the e§ects of terrorism follows a long list of previous studies. A key advantage of this strategy is that it controls for any developments at the national level which are correlated with the country-wide intensity of terrorism and may a§ect the outcome of interest (see Gould and Klor [forthcoming] for a recent discussion). In our setting, one might imagine that terrorist attacks lead to (or follow) various actions and statements by government o¢ cials which could a§ect judicial decision making country-wide.
Before turning to the results, Table C.4 in the online appendix examines whether cases assigned to Arab judges become di§erent from cases assigned to Jewish judges as the number of fatalities in the vicinity of the court increases. Overall, there is not much evidence of di§erential e§ects of terrorism intensity on the characteristics of cases assigned to an Arab versus a Jewish judge.20 Nonetheless, there seems to be some indication that terrorism intensity is associated with an increased probability that a judge is assigned cases with plainti§s from her ethnic group. This might suggest that plainti§s understand that ethnic identiÖcation is stronger during periods of intense conáict and, consequently, withdraw their
Insurance Institute; and the Israeli Ministry of Defense. See Romanov, Zussman, and Zussman (forthcoming) for details.
19Our identiÖcation strategy relies on variation in the intensity of ethnic violence in the vicinity of the courts. Hence we cannot use data on (predominantly Palestinian) fatalities in the Occupied Territories. As mentioned above, there was only one case in this period handled in a court located in the Occupied Territories which involved litigants of opposite ethnicities. This case is dropped from our analyses.
20The only case characteristic that is consistently associated with a di§erential e§ect is whether or not the claim was related to a private conáict. There are only 23 (1.3%) such cases in our data.
17
claims or attempt to settle them outside of court if assigned to an outgroup judge. Further results reported in the online appendix lend some empirical support to this hypothesis: a case is somewhat more likely to be withdrawn when assigned to a judge of the opposite ethnicity in periods of intense terrorist activity around the court (Table C.5). There is no evidence of a similar e§ect with respect to reaching a settlement outside the court (Table C.6). As discussed above (section 4), selective withdrawals in times of intense conáict would tend to generate a downward bias in our estimate of ingroup bias if weaker claims are more likely to be withdrawn.
5.2 Results
Figure II compares case outcomes (share of claims accepted) when there are no civilian fatal- ities in the close vicinity (natural area) of the court to outcomes obtained when the number of civilian fatalities is positive. As in Figure I, both panels indicate the existence of ingroup bias. However, the extent of the bias is signiÖcantly smaller in the ìNo fatalitiesî cases, which make up 41% of the total (left panel). A simple di§erence-in-di§erence calculation suggests an ingroup bias of only 6% in these cases.21 By contrast, the bias in the ìPositive number of fatalitiesîcases is 25%.22
[Figure II]
Another way to examine the e§ect of fatalities is to augment equation (2) with measures of terrorism intensity interacted with the ethnicity variables. SpeciÖcally, we estimate an equation of the form:
yijct = 0 + 1ArabPlainti i + 3ArabPlainti ArabJudgei (3) +0Fatalitiesct + 1Fatalitiesct ArabPlainti i + 2Fatalitiesct ArabJudgei +3Fatalitiesct ArabPlainti ArabJudgei
+ j +tenurejt +Xi0 +t +c +ijct
where Fatalitiesct is the number of fatalities (per 10,000 population) in the vicinity of court c in the year preceding the judicial decision. Our main interest is in 3: the e§ect of terrorism intensity on judicial ingroup bias.23
2 1 (0:789 0:764) (0:720 0:757) = 0:062 :
2 2 (0:792 0:674) (0:617 0:748) = 0:249 :
23The estimated 3 captures the marginal e§ect of an additional fatality (per 10,000 population in the
vicinity of the court) on the di§erence-in-di§erence estimator described in footnote 14. The easiest way to 18
Table X reports the results. The Örst column replicates the results from the full regression without controlling for terrorism (column (4) of Table V). Columns (2)-(7) report results of the augmented regressions for the di§erent measures of Fatalitiesct. We Önd strong evidence that terrorism intensity is associated with higher levels of judicial ethnic ingroup bias. The estimated 3 in columns (2)-(4) imply that an additional civilian fatality per 100,000 popu- lation in the vicinity of the court is associated with a 2.6-3.9 percentage points larger bias. Somewhat smaller coe¢ cients are obtained when examining total rather than only civilian fatalities (columns (5)-(7)). Using any of the three geographical measures, and either civilian or total fatalities, a one standard deviation increase in the number of fatalities is associated with roughly a nine percentage points increase in the bias.24
[Table X]
Note also that the estimated 3 in the second row captures the expected judicial ingroup bias when the number of fatalities is zero (under the econometric speciÖcation in equation
(3) which assumes linearity of the bias in the number of fatalities). A comparison of the estimates in columns (2)-(7) with that reported in column (1) indicates that in the absence of terrorism the extent of judicial bias is substantially lower.
An alternative way to estimate ingroup bias in the absence of terrorism is to use an indicator variable for positive number of fatalities (rather than the number of fatalities) as the measure of terrorism intensity. Estimating equation (3) using binary versions of the fatality variables used in Table X yields estimates of 3 which are for the most part smaller than 0.1 and statistically indistinguishable from zero.25 Together with Figure II, these results can be interpreted as suggesting that there is rather little ethnic ingroup bias in the Israeli courts except during periods in which political violence intensiÖes ethnic identiÖcation. In other words, by heightening identiÖcation, ethnic conáict can dramatically undermine the proper functioning of an ostensibly impartial institution like the court system.
see this is to think of Fatalitiesct as an indicator which takes the value of 1 if the number of fatalities is strictly positive. Denoting by F, AP, AJ, JP and JJ indicators for Fatalities, Arab Plainti§, Arab Judge, Jewish Plainti§ and Jewish Judge, respectively, we have (conditional on the control variables):
[E(yjAJ, AP, F=0 ) E(yjJJ, AP, F=0 )] [E(yjAJ, JP,F=0 ) E(yjJJ, JP, F=0 )] = 3 [E(yjAJ, AP, F=1 ) E(yjJJ, AP, F=1 )] [E(yjAJ, JP,F=1 ) E(yjJJ, JP, F=1 )] = 3 + 3
The coe¢ cient 3 thus captures a di§erence-in-di§erence-in-di§erence.
2 4 Similar to the results in Table VI, interacting judge characteristics, the Arab plainti§ indicator, and our
fatalities variables does not signiÖcantly alter the estimated e§ect of terrorism intensity on judicial ingroup bias (see Tables C.7-C.13 in the online appendix).
25The only exception is when using civilian fatalities at the district level, in which case the estimated 3 is 0.18 with p-value=0.09. The full results are in Table C.14 of the online appendix.
3
19
Finally, the variation in terrorism intensity allows us to address an issue that could not be resolved when estimating the overall judicial bias in section 4. As noted there, in the absence of an ethnicity-free benchmark, one cannot establish whether and to what extent Jewish judges favor Jewish litigants and Arab judges favor Arab litigants. However, if terrorism increases the salience of ethnicity and thereby strengthens ethnic identiÖcation, we can use variations in terrorism intensity to estimate the marginal e§ect of ethnic identiÖcation on judicial decisions.26 Crucially, we can do this separately for Arab and Jewish judges. In Table XI we hence estimate an equation of the following form, separately for judges of each ethnicity:
yijct = 0 + 1ArabPlainti i + 0Fatalitiesct + 1Fatalitiesct ArabPlainti i (4) + j +tenurejt +Xi0 +t +c +ijct
where all the variables are deÖned as before. Our interest is in 1: the e§ect of terrorism intensity on the di§erential treatment of Arab versus Jewish plainti§s.
[Table XI]
Columns (1) and (5) show benchmark results (without controlling for the e§ect of terror- ism). Jewish judges are 11 percentage points less likely, and Arab judges are 7 percentage points more likely, to accept a claim Öled by an Arab rather than by a Jewish plainti§. As emphasized above, due to possible unobserved case characteristics we cannot tell whether these estimated e§ects represent bias on the part of Jewish judges, Arab judges or both. Columns (2)-(4) examine the e§ect of terrorism on Jewish judges. While the coe¢ cients are imprecisely estimated, they suggest that terrorism makes Jewish judges less likely to accept claims Öled by Arab plainti§s. Similarly, columns (6)-(8) indicate that terrorism makes Arab judges more likely to accept claims Öled by Arab plainti§s.27 Thus, while we cannot separately measure the extent of ingroup bias among Arab and Jewish judges, the fact that decisions of judges of both ethnicities co-move with fatalities suggests that ingroup bias exists on both sides.
26This approach is similar to that taken by Benjamin, Choi, and Strickland (2010) who identify the marginal behavioral e§ects of social identities by manipulating the salience of ethnic identities of laboratory subjects.
27Table XI shows the results using civilian fatalities. Very similar results are obtained using total fatalities instead.
20
6 Alternative interpretations
A Önal issue concerns the interpretation of our Öndings. We believe that judicial ingroup bias is the most plausible interpretation. It is consistent with the experimental literature in social psychology (as well as recent results in experimental economics) and follows naturally from the particular features of the institutional context we study. However, our setting deviates in two ways from typical experiments which have been used to establish ingroup bias and saliency e§ects. First, in the lab the experimenter can control (and often completely ban) any communication between the person making the allocation decision and the receivers. Second, in controlled experiments the allocation decision can be divorced from any actions taken by the receivers (who may be made completely passive). Accordingly, we cannot rule out two alternative interpretations of our results.
The Örst is that while judicial decisions vary with litigant ethnicity, this variation is not due to a preferential treatment of members of oneís own group but rather to di§erences in the information available to the judge. The judge may pay equal attention to arguments made by both sides, yet understand better those made by members of her own ethnic group (e.g. due to linguistic subtleties). While possible, we think that technical di§erences in the quality of ingroup-ingroup versus ingroup-outgroup communication are unlikely to drive our results. Cases are typically very simple (e.g. ìfender-benderîaccidents) and the decision is essentially about whose version of the events to accept and not about sophisticated lines of argumentation. More importantly, it is hard to see why technical di§erences in the quality of communication between Arabs and Jews should respond to the intensity of terrorism in the vicinity of the court.
The second alternative interpretation is that it is not judge behavior but rather litigant behavior that is driving the results. Such an alternative interpretation is perhaps best illustrated by a recent study of NBA refereeing decisions (Price and Wolfers, forthcoming). The study Önds that more personal fouls are called against players when they are o¢ ciated by an opposite-race refereeing crew than when o¢ ciated by an own-race crew. An important di¢ culty that arises in the NBA context is that player behavior may depend on the racial composition of the refereeing crew. For example, white players may play more aggressively when there are more blacks in the refereeing crew. Notice, however, that in the NBA setting, when a player makes a foul he already knows the racial composition of the refereeing crew. This may conceivably a§ect his behavior. In contrast, in our setting, when the defendant hits the plainti§ís car she cannot possibly know the ethnicity of the judge who will rule in the case. In other words, the legally relevant actions over which the judge decides cannot be a§ected by the ethnicity of the judge.
21
Nonetheless, one might still worry that litigant behavior could a§ect the judicial decision if litigants are more e§ective in presenting their case when facing a judge of the same ethnicity, and if this di§erence is accentuated during periods of ethnic tension. Since litigant behavior in the courtroom is unobserved (we only have access to the decision document written by the judge), we cannot rule out this possibility. However, several indirect tests may shed some light on this issue. We have information on three types of decisions taken by litigants. The Örst is the plainti§ís decision to withdraw the claim. The second is the defendantís decision to appear in court. The third is the litigantsídecision to reach a settlement outside of court. To some extent, these variables could indicate expected litigant courtroom behavior. SpeciÖcally, if litigants are more e§ective in presenting their case when facing a judge of the same ethnicity, they may be more likely to appear in court in this situation compared to when facing a judge of the opposite ethnicity. It is important to stress, however, that such behavior (if it exists) could also reáect litigantsíreaction to expected discriminatory behavior by the judge.28
As discussed in section 4, we Önd no evidence for selective withdrawals or selective set- tlement outside of court in the period as a whole. We Önd some evidence for selective withdrawals ñbut not for settlements outside of court ñwhen the intensity of terrorism is high (section 5.1). We do not have perfect information regarding defendantsídecision to appear in court. What we have instead is information on situations of ìno defenseî: these are cases where either the defendant failed to show up to the trial or did not submit a defense statement. We can therefore check whether defendants are less likely to appear in court (or provide a defense statement) when assigned to a judge of the opposite ethnicity. Similarly, we can examine whether such a response is a§ected by terrorism intensity. We Önd no evidence for such e§ects (see Table C.16 in the online appendix).
Overall, while we cannot entirely rule out alternative interpretations, there is not much evidence to support them. The most plausible interpretation of the results in our view is that they reáect judicial ingroup bias.
7 Conclusion
The voluminous literature on ingroup bias and its determinants has largely relied on lab experiments and (to a lesser extent) on structured Öeld experiments. Two of the major results in this literature relate to (1) the e§ect of group membership on individual behavior toward ingroup and outgroup members, and in particular the display of ingroup bias; and (2) the sensitivity of this e§ect to the salience of group membership. While these results
28We thank Christine Jolls for making this point.
22
are quite robust, concerns regarding external validity of experimental studies, especially if conducted in the lab, are widespread (e.g. Levitt and List 2007). This paper contributes to our understanding of ingroup bias by examining behavior in naturally occurring data. Using a unique dataset of judicial decisions in Israeli courts, we Önd support for both of the above experimental results.
Our identiÖcation strategy, which relies on plausibly exogenous variation both in the assignment of judges and in the salience of ethnicity, allows us to overcome a major challenge facing the literature on ethnic and racial bias in judicial decisions, namely the potential correlation between ethnicity and unobserved case characteristics.
The overall level of judicial ingroup bias we uncover in the period studied is arguably quite substantial. A claim is 17% to 20% more likely to be accepted if assigned to a judge of the same ethnicity as the plainti§. While this poses a challenge to the Israeli judicial system, perhaps more important is the fact that the bias is hardly an exogenously given fact. In areas which experienced relatively little ethnic strife in the recent past, the bias is substantially lower. This may suggest the feasibility of debiasing e§orts.
From the perspective of the literature on ethnic conáicts, our results shed light on the poorly understood e§ect of such conáicts on institutions and social norms (see Blattman and Miguel 2010 for a discussion). SpeciÖcally, it highlights a possibly important e§ect of ethnic conáicts, often overlooked in the conáict literature. Indeed, as we have seen, even if the conáict does not directly involve the domestic ethnic groups, by intensifying ethnic identities it can produce distortions in judicial decisions, thus potentially eroding property rights and public trust in the rule of law.
The Hebrew University of Jerusalem The Hebrew University of Jerusalem
REFERENCES
Abrams, David S., Marianne Bertrand, and Sendhil Mullainathan, ìDo Judges Vary in Their Treatment of Race?îUnpublished manuscript (2007).
Alesina, Alberto, and Eliana La Ferrara, ìA Test of Racial Bias in Capital Sentencing,î Unpublished manuscript (2010).
Arrow, Kenneth J., ìThe Theory of Discrimination,îin Discrimination in Labor Markets, Orley Ashenfelter and Albert Rees, eds. (Princeton: Princeton University Press, 1973).
Becker, Gary S, The Economics of Discrimination, (Chicago: The University of Chicago Press, 1957).
Benjamin, Daniel J., James J. Choi, and A. Joshua Strickland, ìSocial Identity and Prefer- ences,îAmerican Economic Review, 100 (2010), 1913-1928.
23
Bernhard, Helen, Urs Fischbacher, and Ernst Fehr, ìParochial Altruism in Humans,îNature, 442 (2006), 912-915.
Bertrand, Marianne, Dolly Chugh, and Sendhil Mullainathan, ìImplicit Discrimination,î American Economic Review, 95 (2005), 94-98.
Blattman, Christopher, and Edward Miguel, ìCivil War,îJournal of Economic Literature, 48 (2010), 3ñ57.
Bornstein, Gary, ìIntergroup Conáict: Individual, Group, and Collective Interests,î Per- sonality and Social Psychology Review, 7 (2003), 129-145.
Chen, Yan, and Sherry Xin Li, ìGroup Identity and Social Preferences,îAmerican Economic Review, 99 (2009), 431ñ457.
Eckel, Catherine C., and Philip J. Grossman, ìManaging Diversity by Creating Team Iden- tity,îJournal of Economic Behavior and Organization, 58 (2005), 371-392.
Fong, Christina M., and Erzo F. P. Luttmer, ìWhat Determines Giving to Hurricane Katrina Victims? Experimental Evidence on Racial Group Loyalty,îAmerican Economic Journal: Applied Economics, 1 (2009), 64-87.
Gazal-Ayal, Oren, and Raanan Sulitzeanu-Kenan, ìLet My People Go: Ethnic In-Group Bias in Judicial Decisions ñEvidence from a Randomized Natural Experiment,îJournal of Empirical Legal Studies, 7 (2010), 403-428.
Glaeser, Edward L., and Bruce Sacerdote, ìSentencing in Homicide Cases and the Roles of Vengeance,îJournal of Legal Studies, 32 (2003), 363-382.
Gould, Eric D., and Esteban F. Klor, ìDoes Terrorism Work?îQuarterly Journal of Eco- nomics, forthcoming, 2010.
Jolls, Christine, and Cass R. Sunstein, ìThe Law of Implicit Bias,îCalifornia Law Review, 94 (2006), 969-996.
Klor, Esteban F., and Moses Shayo, ìSocial Identity and Preferences over Redistribution,î Journal of Public Economics, 94 (2010), 269ñ278.
Levitt, Steven D., and John A. List, ìWhat Do Laboratory Experiments Measuring Social Preferences Reveal About the Real World?îJournal of Economic Perspectives, 21 (2007), 153ñ174.
List, John A., ìThe Nature and Extent of Discrimination in the Marketplace: Evidence from the Field,îQuarterly Journal of Economics, 119 (2004), 49-89.
McConnell, Brendon, ìEthnicity, Sentencing and 9/11.îUnpublished manuscript (2010). Miaari, Sami, Asaf Zussman, and Noam Zussman, ìEthnic Conáict and Job Separations,î
Unpublished manuscript (2009).
Michaels, Guy, and Xiaojia Zhi, ìFreedom Fries,î American Economic Journal: Applied Economics, 2 (2010), 256-281.
24
Moser, Petra, ìTaste-Based Discrimination at the NYSE – Empirical Evidence from a Shock to Preferences after WWI,îUnpublished manuscript (2010).
Orbell, John M., Alphons J.C. van de Kragt, and Robyn M. Dawes, ìExplaining Discussion- Induced Cooperation,îJournal of Personality and Social Psychology, 54 (1988), 811- 819.
Phelps, Edmund S., ìThe Statistical Theory of Racism and Sexism,î American Economic Review, 62 (1972), 659-61.
Price, Joseph, and Justin Wolfers, ìRacial Discrimination among NBA Referees,îQuarterly Journal of Economics, forthcoming, 2010.
Romanov, Dmitri, Asaf Zussman, and Noam Zussman, ìDoes Terrorism Demoralize? Evi- dence from Israel,îEconomica, forthcoming.
Shayo, Moses, ìA Model of Social Identity with an Application to Political Economy: Nation, Class and Redistribution,îAmerican Political Science Review, 103 (2009), 147-174.
Tajfel, Henri, Michael G. Billig, Robert P. Bundy, and Claude Flament, ìSocial Categoriza- tion and Intergroup Behavior,îEuropean Journal of Social Psychology, 1 (1971), 149ñ178.
25
TABLE I
CASES BY ETHNICITY OF THE JUDGE, PLAINTIFF AND DEFENDANT percent in each category by district and court
District Jerusalem
Northern
Haifa Central
Tel Aviv-Yafo Southern
Total
Court
Bet Shemesh Jerusalem
Afula
Akko
Bet She’an Nazareth
Qazrin
Qiryat Shemona Tiberias
Zefat
Hadera
Haifa
Krayot
Netanya
Petah Tiqwa Ramla
Rehovot
Rishon Leziyyon Tel Aviv-Yafo Ashdod Ashqelon
Be’er Sheva Dimona
Elat
Qiryat Gat
Jewish Arab Arab
Arab Jewish Jewish Cases
Judge: Jewish Arab Jewish Arab
Plaintiff: Jewish Defendant: Arab 100.0
51.89
31.19
42.34
83.33
3.88
100.0
67.65 0.00 80.00 0.00 83.33 16.67 63.48 0.00 34.72 26.39 42.86 14.29 65.63 0.00 75.00 0.00 44.90 0.00 83.33 0.00 100.0 0.00 48.89 6.67 100.0 0.00
0.00 0.00 14.29 30.16 50.00 0.00 42.86 0.00 83.33 0.00 41.08 15.16
0.00 0.00
0.00 1
0.00 106 23.85 109 6.57 137 0.00 12 55.04 258
0.00 3 0.00 34 0.00 5 0.00 12 0.00 178 16.67 216 10.71 336 0.00 32 0.00 56 0.00 49 0.00 12 0.00 11 3.33 90 0.00 4 0.00 1
44.44 63 0.00 4 0.00 7 0.00 12
16.02 1,748
0.00 48.11 31.19 13.76 16.06 35.04
0.00 16.67 29.84 11.24 0.00 0.00
32.35 20.00 0.00 36.52 22.22 32.14 34.38 25.00 55.10 16.67 0.00 41.11 0.00 100.0 11.11 50.00 57.14 16.67 27.75
26
TABLE II SUMMARY STATISTICS PANEL A: CASES (N=1,748)
Variable category Claim outcome
Case characteristics
Claim subject
Defense
Variable
Claim accepted
– partly accepted
Net monetary compensation Netlegalexpenses
Monetary yield1
Breach of sales contract Breach of service contract Housing related
Private conflict
Traffic accident Miscellaneous
Missing
Defense present
Defense made a counter claim Plaintiffs
Defendants
Plaintiffs
Defendants
Plaintiffs
Defendants
by plaintiff/s 1
2000 2001 2002 2003 2004 Civilian Total Civilian Total Civilian Total
Mean Std. Dev. 0.7340
0.5297
3079.3 3923.6
188.8 497.1 0.7993 0.4268 0.0320
0.0950
0.0109
0.0132
0.6894
0.0126
0.1470
0.8661
0.0881
1.1127 0.3181 1.7243 0.7134 0.9982 0.0311 0.7369 0.2584 0.8212 0.3643 0.8747 0.3128 6423.9 5085.4 0.0023
0.0864
0.2294
0.3450
0.3370
0.2817 0.3518 0.3231 0.4034 0.2510 0.2580 0.3068 0.3374 0.2299 0.2206 0.2900 0.2789
Number of
litigants
Private litigants
(share of total)
Male litigants
(share of private) Compensation requested
Timing of judicial decision
Terrorism fatalities2
Y ear
Natural area Sub district District
Notes:
1 Data on compensation requested by plaintiff/s and monetary yield are available for 660 cases.
2 “Terrorism fatalities” = civilian or total (civilian and security forces) fatalities from terrorist attacks in the natural area/sub-district/district of the court in the year preceding the judicial decision per 10,000 population; data on lagged fatalities are available for 1,744 cases.
27
TABLE II SUMMARY STATISTICS PANEL B: JUDGES (N=132)
Mean
Arab judges Jewish judges
Difference (1) (2) (3)
Age 43.84648.947-5.102**
Tenure at job
Male
Immigrant (Jewish)
LLB degree – Hebrew U. – Tel Aviv U.
– Bar Ilan U.
– other institutions
Highest degree – LLB – master
– doctoral
(7.445)
3.779 (3.973)
(9.580) [2.571] 5.110 -1.331
0.000 0.231
0.733 0.410
0.200 0.410
0.000 0.145
0.067 0.034
0.933 0.803 0.000 0.171
0.067 0.026
-0.231** [0.110]
0.323** [0.134]
-0.210 [0.133]
-0.145 [0.092]
0.032 [0.053]
0.130 [0.106]
-0.171* [0.098]
0.041 [0.047]
(6.680)
[1.767]
0.6670.5210.145 [0.137]
N 15117
Notes: Standard deviations in parentheses in columns (1)-(2). Standard errors in brackets in column (3).
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
28
TABLE III
BALANCING TESTS FOR THE ASSIGNMENT OF JUDGES
Mean
Difference in means Arab vs. Jewish judge
Obs.
(5) 1,748
1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 1,748 660
Arab plaintiff
Number of plaintiffs
Number of defendants
Private plaintiffs (share of total )
Private defendants (share of total)
Male plaintiffs (share of private plaintiffs) Male defendants (share of private defendants) Claim subject – Breach of sales contract Claim subject – Breach of service contract Claim subject – Housing related
Claim subject – Private conflict
Claim subject – Traffic accident
Claim subject – Miscellaneous
Claim subject – Missing
Defense present
Defense made a counter claim
Compensation requested (NIS)
Arab judge (1)
0.514
1.112 (0.316) 1.756 (0.708) 0.998 (0.030) 0.719 (0.255) 0.850 (0.341) 0.846 (0.344) 0.024
0.095 0.004 0.011 0.692 0.009 0.165 0.848 0.077
6,481 (5,260)
Jewish judge (2)
0.403
1.113 (0.319) 1.710 (0.716) 0.998 (0.031) 0.745 (0.259) 0.808 (0.374) 0.888 (0.297) 0.036
0.095 0.014 0.014 0.688 0.014 0.139 0.874 0.093
6,401 (5,018)
Without court FEs (3)
0.111*** [0.025] -0.001 [0.016] 0.046 [0.037] 0.000 [0.002] -0.026** [0.013] 0.041** [0.019] -0.042*** [0.016] -0.012 [0.009] 0.001 [0.015] -0.010* [0.005] -0.003 [0.006] 0.003 [0.024] -0.005 [0.006] 0.026 [0.018] -0.027 [0.018] -0.016 [0.015] 80 [437]
With court FEs (4)
-0.013 [0.032] -0.010 [0.021] -0.033 [0.047] 0.001 [0.002] -0.006 [0.017] 0.036 [0.024] -0.031 [0.021] -0.003 [0.012] 0.032 [0.020] -0.009 [0.007] 0.007 [0.008] -0.043 [0.029] -0.005 [0.008] 0.022 [0.022] -0.031 [0.021] -0.001 [0.019] 942 [619]
Notes: Standard deviations in parentheses in columns (1)-(2). Standard errors in brackets in columns (3)-(4). Each entry in columns (3)-(4) is derived from a separate OLS regression where the explanatory variable is an indicator for Arab judge.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
29
TABLE IV DIFFERENCES IN DIFFERENCES share of claims accepted
Arab judge Jewish judge
Difference
Arab plaintiff
Jewish plaintiff Difference
0.037 (0.038) [N=545]
-0.144*** (0.026) [N=1,203]
0.180*** (0.046) [N=1,748]
0.754 (0.026) [N=280]
0.647 (0.022) [N=485]
0.717 (0.028) [N=265]
0.791 (0.015) [N=718]
0.106*** (0.035) [N=765]
-0.074** (0.030) [N=983]
Notes: Standard errors in parentheses.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
30
TABLE V JUDICIAL INGROUP BIAS
Dependent variable: claim accepted
Arab plaintiff
Arab judge
Arab plaintiff * Arab judge
Court fixed effects
Judge fixed effects and tenure Case characteristics
Time controls
Observations
R-squared
(1)
-0.151*** (0.026)
-0.077* (0.044)
0.170*** (0.054)
Yes No No No 1,748 0.0439
(2)
-0.150*** (0.029)
0.166*** (0.056)
Yes Yes No No 1,748 0.1383
(3)
-0.121*** (0.030)
0.199*** (0.049)
Yes Yes Yes No 1,748 0.2377
(4)
-0.117*** (0.031)
0.192*** (0.049)
Yes Yes Yes Yes 1,748 0.2479
Notes: Regressions are estimated by OLS. Standard errors, clustered by judge, are reported in parentheses. Case characteristics include: number of plaintiffs; number of defendants; share of private plaintiffs; share of private defendants; share of male plaintiffs; share of male defendants; amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for “defense present”; and an indicator for cases where the defendant filed a counter-claim. Time controls include indicators for year, month and day of week.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
31
TABLE VI
JUDICIAL BIAS AND OTHER JUDGE CHARACTERISTICS
Dependent variable: claim accepted
Arab plaintiff Arab plaintiff
Arab plaintiff
Arab plaintiff
Arab plaintiff
Arab plaintiff
Arab plaintiff
* Arab judge
* Judge age
* Judge tenure * Male judge * Judge HU
* Judge>LLB
(1)
-0.117*** (0.031)
0.192*** (0.049)
(2)
-0.017 (0.105)
0.171*** (0.058)
-0.002 (0.002)
(3)
-0.098** (0.041)
0.182*** (0.051)
-0.003 (0.002)
Yes Yes Yes Yes 1,748 0.2483
(4)
-0.062 (0.038)
0.175*** (0.049)
-0.102** (0.049)
Yes Yes Yes Yes 1,748 0.2506
(5)
-0.104*** (0.040)
0.196*** (0.049)
(6)
-0.107*** (0.032)
0.193*** (0.047)
(7)
-0.032 (0.211)
0.174** (0.067)
-0.000 (0.005)
-0.000 (0.007)
-0.085 (0.059)
-0.012 (0.051)
-0.082 (0.072)
Yes Yes Yes Yes 1,748 0.2512
Court fixed effects
Judge fixed effects and tenure Case characteristics
Time controls
Observations
Yes Yes Yes Yes 1,748 0.2479
Yes Yes Yes Yes 1,748 0.2484
-0.027 (0.046)
Yes Yes Yes Yes 1,748 0.2481
-0.120* (0.061)
Yes Yes Yes Yes 1,748 0.2491
R-squared
Notes: Regressions are estimated by OLS. Standard errors, clustered by judge, are reported in parentheses. Case characteristics include: number of plaintiffs; number of defendants; share of private plaintiffs; share of private defendants; share of male plaintiffs; share of male defendants; amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for “defense present”; and an indicator for cases where the defendant filed a counter-claim. Time controls include indicators for year, month and day of week. “Judge HU” and “Judge>LLB” are indicators for whether judge attained LLB at the Hebrew University of Jerusalem and whether judge has a master or PhD degree, respectively.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
32
TABLE VII
JUDICIAL BIAS – ALTERNATIVE OUTCOME MEASURES
Dependent variable Estimation methodology
Arab plaintiff
Arab plaintiff * Arab judge
OLS (1)
-0.138*** (0.040)
0.208*** (0.057)
Yes Yes Yes Yes
Ordered Probit (2)
-0.369*** (0.005)
0.587*** (0.009)
Yes Yes Yes Yes
-135.3** (53.7)
224.3*** (84.5)
-0.052 (0.041)
0.101* (0.059)
Claim outcome {0,1,2}
Net monetary compensation
OLS (3)
-662.3*** (248.8)
925.7** (448.3)
Yes Yes Yes Yes
Net legal
expenses yield
OLS OLS (4) (5)
Monetary
Court fixed effects
Judge fixed effects and tenure Case characteristics
Time controls
Observations R-squared/Pseudo R-squared
Yes Yes Yes Yes Yes Yes Yes Yes
1,748 1,748 1,748 1,748 660
0.5473 0.4014 0.4298 0.2291
Notes: In columns (1)-(2) the dependent variable takes the value of 0 if the claim was rejected, 1 if the claim was partly accepted, and 2 if the claim was fully accepted. In column (3) the dependent variable is the net monetary compensation awarded by the judge to the plaintiff (compensation awarded to plaintiff minus compensation awarded to defendant). In column (4) the dependent variable is the net legal expenses awarded by the judge to the plaintiff (expenses awarded to plaintiff minus expenses awarded to defendant). In column (5) the dependent variable is the ratio between the net monetary compensation (including legal expenses) awarded by the judge to the plaintiff and the compensation requested by the plaintiff. Case characteristics include: number of plaintiffs; number of defendants; share of private plaintiffs; share of private defendants; share of male plaintiffs; share of male defendants; amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for “defense present”; and an indicator for cases where the defendant filed a counter-claim. Time controls include indicators for year, month and day of week. In column (5) the monetary compensation requested by the plaintiff is not included in the case characteristics. Standard errors,
clustered by judge, are reported in parentheses.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
33
0.5683
TABLE VIII
EXPLORING HETEROGENEITY IN INGROUP BIAS
Dependent variable: claim accepted
(1)
-0.022 (0.107)
0.229 (0.308)
* Judge age -0.001 (0.007)
(2)
-0.096** (0.042)
0.172*** (0.063)
0.003 (0.008)
Yes Yes Yes Yes Yes 1,748
include: number of plaintiffs; number of defendants; share of private plaintiffs; share of private defendants; share of male plaintiffs; share of male defendants; amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for “defense present”; and an indicator for cases where the defendant filed a counter-claim. Time controls include indicators for year, month and day of week. For each judge characteristic z, “Additional interactions” include Arab plaintiff*z and (for time varying z’s) Arab judge*z. “Judge HU” and “Judge>LLB” are indicators for whether judge attained LLB at the Hebrew University of Jerusalem and whether judge has a master or PhD degree, respectively.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
Arab plaintiff Arab plaintiff
Arab plaintiff
Arab plaintiff
Arab plaintiff
Arab plaintiff
Arab plaintiff
* Arab judge
* Arab judge * Arab judge * Arab judge * Arab judge * Arab judge
(3)
-0.056 (0.040)
0.161*** (0.056)
0.034 (0.102)
Yes Yes Yes Yes Yes 1,748
(4)
-0.096** (0.043)
0.165* (0.097)
(5)
-0.105*** (0.032)
0.188*** (0.050)
(6)
-0.045 (0.215)
0.424 (0.709)
-0.007 (0.018)
0.006 (0.027)
0.040 (0.137)
-0.023 (0.132)
0.049 (0.143)
Court fixed effects
Judge fixed effects and tenure Yes Case characteristics Yes Time controls Yes Observations 1,748 R-squared 0.2490
* Judge tenure * Male judge * Judge HU
* Judge>LLB
Yes
0.056 (0.112)
Yes
0.058 (0.097)
Additional interactions
Yes Yes Yes Yes Yes 1,748 0.2483
Yes Yes Yes Yes Yes 1,748 0.2491
Yes Yes Yes Yes Yes 1,748 0.2519
0.2490
Notes: Regressions are estimated by OLS. Standard errors, clustered by judge, are reported in parentheses. Case characteristics
0.2507
34
TABLE IX
FATALITIES FROM TERRORIST ATTACKS per 10,000 population
PANEL A: CIVILIAN FATALITIES
2000 2001 0.04 0.39 0.00 0.09 0.04 0.34 0.00 0.10 0.00 0.19 0.00 0.00
0.01 0.17
2002 2003 2004
1.06 0.71 0.23 2.44 0.16 0.05 0.00 0.30 0.49 0.44 0.00 1.30 0.36 0.03 0.00 0.49 0.14 0.23 0.03 0.58 0.00 0.01 0.33 0.36
0.33 0.21 0.08 0.80
2000-04
Jerusalem Northern Haifa Central TelAviv Southern
Total
Jerusalem Northern Haifa Central TelAviv Southern
Total
Notes: Total fatalities
fatalities. In the last
fatalities in 2000-04 is divided by the average population in that period. Fatality data from Romanov, Zussman and Zussman (forthcoming). Population data from Israel Central Bureau of Statistics.
PANEL B: TOTAL FATALITIES
2000 2001 0.04 0.40 0.04 0.14 0.04 0.39 0.00 0.11 0.00 0.26 0.00 0.00
0.02 0.20
2002 2003 2004
1.11 0.73 0.26 2.55 0.20 0.07 0.03 0.47 0.88 0.46 0.00 1.77 0.37 0.09 0.00 0.57 0.15 0.27 0.03 0.70 0.07 0.02 0.33 0.44
0.41 0.23 0.10 0.96
2000-04
refer to the sum of civilian and security forces column of both panels the cumulative number of
35
TABLE X TERRORISM AND JUDICIAL BIAS
Dependent variable: claim accepted
Civilian fatalities Natural Sub-
Total fatalities
Natural Area (5)
-0.084** (0.036)
0.125** (0.052)
0.052 (0.039)
-0.080* (0.048)
-0.123 (0.104)
0.256*** 0.353***
(0.089) (0.118) (0.142) (0.077) (0.101)
Court fixed effects
Judge fixed effects and tenure
Case characteristics
Time controls
Observations
R-squared
Notes: “Fatalities” is the number of civilian/total (civilian and security forces) fatalities from terrorist attacks in the natural area/sub-district/district of the court in the year preceding the judicial decision per 10,000 population. Regressions are estimated by OLS. Case characteristics include: number of plaintiffs; number of defendants; share of private plaintiffs; share of private defendants; share of male plaintiffs; share of male defendants; amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for “defense present”; and an indicator for cases where the defendant filed a counter-claim. Time controls include indicators for year, month and day of week. Standard errors, clustered by judge, are reported in parentheses.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
Arab plaintiff
Arab plaintiff * Arab judge
Fatalities
Fatalities * Arab plaintiff
Fatalities * Arab judge
Fatalities * Arab plaintiff * Arab judge
(1)
-0.117*** (0.031)
0.192*** (0.049)
Area District
(2) (3)
-0.085** -0.089** (0.035) (0.037)
0.128** 0.109* (0.050) (0.057)
0.075* 0.022 (0.044) (0.067)
-0.088 -0.090 (0.059) (0.086)
-0.152 -0.113 (0.123) (0.145)
District
(4)
-0.084** (0.041)
0.112* (0.058)
0.040 (0.105)
-0.112 (0.106)
-0.216 (0.163)
Sub- District (6)
-0.095** (0.039)
0.115* (0.061)
0.012 (0.040)
-0.057 (0.057)
-0.081 (0.116)
District
(7)
-0.093** (0.041)
0.135** (0.063)
0.005 (0.069)
-0.064 (0.070)
-0.110 (0.081)
0.213* (0.110)
0.387***
0.240***
0.288***
Yes Yes Yes Yes 1,748 0.2479
Yes Yes Yes Yes 1,744 0.2468
Yes Yes Yes Yes Yes Yes Yes Yes 1,744 1,744 0.2465 0.2464
Yes Yes Yes Yes Yes Yes Yes Yes 1,744 1,744 0.2467 0.2463
Yes Yes Yes Yes 1,744 0.2458
36
Arab plaintiff
0.032
Fatalities
Fatalities * Arab plaintiff
0.073 0.011 (0.047) (0.073)
0.038 (0.119)
(0.105)
(0.132) (0.188)
Court fixed effects
Judge fixed effects and tenure Casecharacteristics
Time controls
Observations
R-squared
Yes
Yes
Yes Yes
Yes Yes Yes Yes Yes Yes Yes Yes 545 545 0.2241 0.2253
Yes Yes Yes Yes Yes Yes Yes Yes 545 545 0.2261 0.2260
TABLE XI
TERRORISM AND JUDICIAL BIAS BY JUDGE ETHNICITY
Dependent variable: claim accepted
Jewish Judge Natural Sub-
Arab Judge
Area District District (1) (2) (3) (4)
(5)
Natural Area (6)
Sub- District (7)
District (8)
-0.113*** -0.082** -0.086** -0.078*
(0.034) (0.036) (0.037)
Yes 1,203 0.2857
Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
-0.085 -0.086 (0.059) (0.085)
-0.116 (0.103)
0.125 (0.074)
0.207** 0.234*** (0.073) (0.065)
Yes 1,199 0.2834
1,199 1,199 0.2827 0.2828
Notes: “Fatalities” is the number of civilian fatalities from terrorist attacks in the natural area/sub-district/district of the court in the year preceding the judicial decision per 10,000 population. Regressions are estimated by OLS. Case characteristics include: number of plaintiffs; number of defendants; share of private plaintiffs; share of private defendants; share of male plaintiffs; share of male defendants; amount of compensation requested (and an indicator for missing values); indicators for claim subjects; an indicator for “defense present”; and an indicator for cases where the defendant filed a counter-claim. Time controls include indicators for year, month and day of week. Standard errors, clustered by judge, are reported in parentheses.
*, **, *** represent statistical significance at the 10, 5, and 1 percent levels.
37
0.072*
(0.041) (0.039) (0.033)
0.029
(0.043) (0.038)
0.048
-0.071 -0.095 -0.193
0.79
FIGURE I BASELINE INGROUP BIAS
Share of claims accepted
0.60 0.65 0.70 0.75 0.80
Jewish plaintiff
Jewish Judge
0.65
Arab plaintiff
Arab Judge
0.72
Based on 1,748 mixed cases. Capped ranges indicate 95% confidence intervals.
38
0.75
FIGURE II TERRORISM AND INGROUP BIAS
by number of civilian fatalities in natural-area in the preceding year
0.79
0.79
Jewish plaintiff
Jewish Judge
Jewish plaintiff
Jewish Judge
Arab plaintiff
Arab Judge
No fatalities
0.76
Positive number of fatalities
0.76 0.72
Arab plaintiff
Arab Judge
0.75
Share of claims accepted
0.50 0.60 0.70 0.80 0.90
0.50 0.60 0.70 0.80 0.90
Based on 1,744 mixed cases, 711 with no fatalities in the natural area in the year preceding the judicial decision. Capped ranges indicate 95% confidence intervals.
39
0.67
0.62