University of Chicago Law School

Indiana University School of Business

1111 East 60th Street

10th Street & Fee Lane

Chicago, IL 60637

Bloomington, IN 47

405Tel: 312.702.7311

Tel: 812.855.3345

Fax: 312.702.0730

Fax: 812.855-8679

email: mark_ramseyer@law.uchicago.edu

email: erasmuse@indiana.edu

Abstract:

Because civil-law systems hire unproven jurists into career judiciaries, many maintain elaborate incentive structures to prevent their judges from shirking. We use personnel data (backgrounds, judicial decisions, job postings) on 275 Japanese judges to explore general determinants of career success and to test how extensively politicians manipulate career incentives for political ends. We find strong evidence that the judicial system rewards the smartest and most productive judges. Contrary to some observers, we find no evidence of on-going school cliques, and no evidence that the system favors judges who mediate over those who adjudicate. More controversially, we locate three politically driven phenomena. First, even as late as the 1980's, judges who joined a prominent leftist organization in the 1960's were receiving less attractive jobs. Second, judges who decided a high percentage of cases against the government early in their careers were still receiving less attractive jobs than their peers in 1980s. Finally, whenever a judge decided a case against the government, he incurred a significant risk that the government would soon punish him with a less attractive post.

-------------------------------------------------------------

* We gratefully acknowledge the helpful comments and suggestions of John Lott, Setsuo Miyazawa, Frances Rosenbluth, Tokuo Sakaguchi, Eiji Tsukahara, and participants in workshops at Georgetown (law), Harvard (government), and Indiana (business economics) Universities.

1. Introduction.

Because civil law courts hire unproven jurists into career judiciaries, many use elaborate incentive structures to prevent their judges from shirking. In Japan, the administrative office of the courts, the Secretariat, regularly monitors and evaluates each judge. On the basis of the evaluations, it then assigns judges to new posts every three years. Because not all posts are equal, this rotation system can be used to influence judges, for good or ill. For good, it induces them to exert effort in their job. For ill, it has the potential to enforce political orthodoxy.

In the article that follows, we use data from Japan to test both the determinants of judicial career success generally and the extent of political manipulation. Toward that end, we assembled career data on all of the 274 judges hired from 1961 to 1965. Within this data base, we find strong evidence that the judicial system rewards the smartest and most productive judges. Contrary to some observers, we find no evidence of on-going school cliques, and no evidence that the system favors judges who mediate over those who adjudicate.

More controversially, we also find three signs of politics influencing judicial careers. First, those judges who joined a prominent leftist organization in the 1960s were receiving less attractive jobs than their peers in the 1980s. Second, those judges who decided a high percentage of cases against the government early in their careers were still receiving less attractive jobs than their peers in 1980s. Finally, when a judge decided a case against the government, he incurred a significant risk that the government would soon punish him with a less attractive post.

We begin by outlining the theoretical and empirical literature on judicial independence (Section 2). We then explain the structure of Japanese courts (Section 3) and the anture of our data base (Section 4). Using an ordered probit model, we first explore the determinants both of a judge's initial job posting (Section 5.1) and of his posts later in his career (Section 5.2). With more extensive data for the class of 1965, we then test whether a judge who decides cases against the government receives less attractive posts (Section 6). Finally, we turn to the most common anecdotes of a link between the political content of a judge's decisions and his career success: cases involving the constitutionality of the ban on door-to-door canvassing (Section 7).

2. Civil-law systems and judicial independence.

2.1. Manipulability. To understand the potential manipulability of civil law courts, consider first -- by way of contrast -- the U.S. federal courts. To them, the President generally appoints only prominent middle-aged lawyers. Most have proven themselves both politically loyal and congenitally workaholic. He appoints these men and women to particular posts in particular towns. There, they hear cases for the rest of their working lives.

Sometimes, a federal judge will move from the District Court to the Court of Appeals, or other movement occurs within the system. A Stephen Breyer moves from the Court of Appeals to the Supreme Court. A William Webster moves from the courts to the FBI. Otherwise, a typical judge never moves out of town, never changes jobs, never earns a pay raise except in lockstep with every other federal judge. For most federal judges, how they do their jobs will have little effect on either tenure, advancement, or compensation.\footnote{ State court judges may face greater mobility. Yet even they seldom worry that an administrator might move them from Los Angeles to Fresno, or demote them from an appellate court to traffic court. More often, they are influenced not by fear of reprisals from politicians, but from voters, exactly the effect intended by the state constitution, whether wise policy or not.}

Not so in many civil-law regimes. There, judges face just such threats. Often, they join the courts immediately upon passing the bar. Because they are young and unproven, the government ahs relatively little information about them. It will, for starters, seldom know their political preferences. Yet given the apolitical nature of most of the docket, politics will be the least of its worries. More importantly, it will not know how hard or how fast they work. Instead, it will have access only to a few rough proxies like exam performance.

Accordingly, in civil law jurisdictions, the government will often use elaborate monitoring institutions and incentive systems to induce its judges not to shirk. Toward that end, it will maintain a judicial administrative office. Through that office, it will regularly grade their work and reward them appropriately. The better the work, the more attractive a job it may give the judge, and the more money it it may pay the judge.

Incentive structures, however, are manipulable. A government may introduce an institutional structure to induce effort, but -- hypothetically -- it can use it later to enforce political loyalty. At least in theory, in many civil-law systems it thus could use the structure to reward judges by the political complexion of the judgments they issue and the opinions they write. The loyal it could ply with prestigious posts in attractive cities and a quick climb up the pay scale. The heterodox it could let languish at low pay in branch offices in the outback. In most cases, politics will not matter, for most cases involve no political issues of moment. The question at stake, though, is whether it might yet manipulate the structure to shape decisions in those relatively unusual -- but sometimes vitally important -- cases about which it cares deeply.

2.2. Theory. Whether rational politicians will protect judicial independence is an issue on which theory can go either way.\footnote{Note that whether politicians are self-interested or public-interested is not a central issue here. It is by no means clear that an unselfish politician would prefer judicial independence; he does not want the judiciary to thwart the policies he proposes for the public good, whether because the judge disagrees or because that beneficial policy is truly unconstitutional.} On the one hand, independent courts potentially solve several vexing political problems. First, they add credibility to governmental promises. Whether to maximize the rents it extracts (Landes & Posner, 1975) or to lower the cost of its debt (North & Weingast, 1989), a government may want to make its commitments credible. Subjecting its promises to the judgment of an independent court sometimes does just that.

Second, independent courts help police the bureaucracy . By giving disaffected citizens the right to sue a bureaucracy, a government can potentially obtain access to information about how well its bureaucrats perform (McCubbins & Schwartz, 1984). Armed with that information, it can potentially improve bureaucratic performance and its party's electoral odds.

Third, independent courts minimize a party's losses while out of power. To the extent judges are independent, they do not necessarily serve those politicians in control. That they do not, in turn, will often comfort the rest (Ramseyer, 1994). If a party expects to sometimes be out of power, it will prefer a moderating judiciary.

On the other hand, independent courts introduce political problems of their own. Politicians do not maximize votes by promising desired policies but rather by delivering desired policies. Independent judiciaries can obstruct that delivery. Moreover, many politicians rationally take short-term perspectives. They care less about longterm credibility than about the next election. And there will always be a temptation to infringe on judicial independence in small ways, because the ideal for the ruling party is for the judiciary to have a reputation for independence (thus making the government's promises and contracts credible and keeping the bureaucracy cowed), while actually being subservient to the politicians.

Ultimately, therefore, the extent to which politicians keep judges independent will depend on factors external to the courts. The more readily politicians can make their promises credible, the more cheaply they can monitor their bureaucrats, and the less likely they are to revert to being a minority party, the smaller their incentive to keep judges independent. In all cases, the politicians will wish to pretend that the judges are independent, however, so any analysis of the actual outcome must depend on observing behavior rather than words.

2.3. Empirical studies. Existing empirical studies do not tell us whether politicians in civil law systems keep their courts independent. Over the last decade, several scholars have begun to publish sophisticated empirical analyses of judicial independence, but to date, however, they have studied only the comparatively hard-to-manipulate common-law systems (Spiller & Gely, 1992; Toma, 1991; Anderson, Shughart & Tollison, 1989; Cohen, 1991). Generally, they find some evidence, relatively weak, either that judicial institutional structures affect the political cast of what judges do, or that they respond -- as Spiller & Gelly (1992: 489) nicely put it -- "albeit quite indirectly, to interest-group and voter pressures" .

Civil law systems put less emphasis on the distinction between law and statute, and would seem to provide more opportunities for political intervention, but we know of no systematic econometric study of judicial independence in the more highly manipulable civil-law environments. Scholars have considered the role of judges in relation to politicians, but using historical rather than quantitative approaches, and with an emphasis on crisis situations rather than year-to-year activities (see, e.g., Muller, 1991 on the acquiescence of the German judiciary to Nazi policies in the 1930's.). Even in Japan, where the debate has taken a aggressively political tone, most scholars have produced only anecdotal accounts. The most common accounts involve the ban in Section 138 of the Public Offices Elections Act on door-to-door canvassing. Because incumbents obtain free media coverage but challengers do not, the ban disproportionately benefits incumbents. During the postwar period , the Liberal Democratic Party (the LDP) has had the most incumbents, so the ban disproportionately benefited the LDP.

From time to time, lower court judges insisted that the canvassing ban violated the constitutional guarantee of freedom of expression. According to the anecdotal accounts, those judges were penalized as a result (e.g., Ramseyer & Rosenbluth, 1993: ch. 9). Take Haruhiko Abe. He held Sectin 138 unconstitutional in 1968, and by 1990 had spent 11 years in branch offices of the courts, far more than normal. Or take Masato Hirayu. He held Section 138 unconstitutional in 1979, and by 1990 had spent 8 years in branch offices.

Yet anecdotal accounts are always subject to the criticism that they reflect special situations and not the norm, and different anecdotes will lead to different conclusions. John Haley (1995), dean of Japanese law scholars in the U.S., recently examined the evidence carefully and concluded that Japanese courts remained fundamentally autonomous. Premier Japanese law & society scholar Setsuo Miyazawa (1994) did the same and concluded the opposite. The question remains open.

3. The Japanese courts.

3.1. Appointment and reappointment. As in most civil law systems, Japanese judges begin their careers as judges immediately upon passing the bar. They then stay judges for most of their working lives, which end at the mandatory retirement age of 65 for lower-court judges. In 1990, the Japanese court system had the authority to employ about 2,800 judges. These judges decided some cases alone, but most as 3-judge panels.

Formally, lower-court judges work under a series of 10-year contracts. The Prime Minister, who from 1955 to 1993 was always a member of the modestly conservative LDP, legally has discretion over both the initial appointment and the later reappointments. In fact, he defers to the Secretariat on these personnel matters, reappointing all sitting judges until they either resign or reach retirement age.

3.2. Postings.

During their careers of thirty-odd years, Japanese judges move through a variety of posts, which vary along several dimensions. First, they vary by geography. The Secretariat can -- and does -- routinely move judges from city to city. Second, they vary along the judicial hierarchy. The Secretariat can -- and does -- bounce judges up and down the hierarchy from the High Courts (the courts of appeals) to the District Courts to the Family Courts to the branch offices of the District and Family Courts. Moreover, this is not an extraordinary occurrence. A judge who moves to a worse posting is not necessarily in disgrace, and has not necessarily been singled out for special treatment. Third, some posts involve prestigious administrative duties. The most successful judges become one of the eight High Court Presidents late in their career. Those who are modestly successful become a District or Family Court Chief Judge as the high point of their career. Some judges will work several years in the Secretariat or at the Ministry of Justice. And most will spend some time as a district judge with internal personnel responsibilities (a skatsu assignment). Prestige and influence do not overlap completely, however; a sokatsu assignment or staff position within the Secretariat can be highly influential, even if not as visible as a position as a High Court judge. By the Constitution, a judge's pay cannot be cut, but the Secretariat can promote judges along the pay scale at different speeds. Critics have accused it of doing this to penalize the politically heterodox, but the data is not available to us to test this.

Since posts vary in quality, and movement is normal, the LPD could potentially control judicial careers without public or obtrusive intervention. This is important, because no constitution forbids political intervention if the politicians are willing to be heavy-handed enough. If the U.S. Congress dislikes a judge's decisions, for example, it is free to impeach him on trumped-up charges, or to alter the law setting up the details of the federal court system to transfer him to an undesirable posting. The cost of doing so is high, however, because it is both time-consuming and destroys any illusions people may have about judicial independence.

For the Prime Minister to terminate the appointment of a sitting judge would have been legal, but obvious, and costly to his reputation. We do not even know of anecdotal evidence that elected or party officials intervened overtly to manipulate postings, a much easier intervention to hide, and thus more tempting, but one which, like bribery, would be hard to hide for forty years. A more subtle form of control may have been used, however: the stacking of key positions in the administrative positions with reliable judges who could discipline the rest.

This combination of potential control and formal non-intervention poses the question central to this study: did the LDP not intervene in the courts because it could not do so quietly, because the judges largely anticipated what the LDP wanted without waiting for overt intervention, or because the LDP wished to maintain a reputation for non-interference so as to credibly make commitments under the rule of law?

3.3. Judicial preferences. To determine the relative attractiveness of the various judicial appointments, we discussed the issue with a wide variety of Japanese observers, and looked at the careers of the most successful judges. Idiosyncratic preferences aside, most judges seem to prefer Tokyo posts to all others, and to prefer Osaka if they cannot be in Tokyo. They mildly prefer High and District Court posts to Family Court posts, and strongly prefer all such posts to lower court branch offices. Most aspire to some administrative responsibilities.

In Table 1 we display the percentage of a judge's career spent in various assignments for two groups of judges: the most successful judges (who eventually became Supreme Court justices or High Court Presidents), and the entire cohort of judges, successes and failures, who began their careers in 1965. Note that the most successful judges spend more time in Tokyo and Osaka, more time in the Secretariat and other non-judicial posts (e.g., the Ministry of Justice), and less time in branch offices. They do not spend much more time as sokatsu than other judges, but this is because they more quickly move to higher administrative roles such as chief judgeships.

Table 1: Exceptional Judges

and the Class of 1965

Exceptional Judges Class of 1965
	Mean	Range	Mean	Range
Personal data:
Starting Age	26.28	18-29	29.85	24-38
Sex	.96	0-1	.91	0-1
Tokyo U	.76	0-1	.2	0-1
Kyoto U	.12	0-1	.036	0-1
Chuo U	0	0-0	.2	0-1
No Univ	.04	0-1	.47	0-1
Opinions	4.02	0-16.5	1.75	.16-7.82
YJL	.08	0-1	.27	0-1
Observations:	25			55
Percentage of career in various posts:
Tokyo	.56	0-.89	.23	0-.88
Osaka	.13	0-.73	.068	0-.56
Sokatsu	.14	0-.33	.10	0-.4
Secretariat	.21	0-.58	.007	0-.12
Other non-judicial	.14	0-.78	.083	0-.56
Branch offices	.035	0-.19	.15	0-.58
Observations:	21		55

Notes:

The initial personal variables are as defined in the text.

"Exceptional judges" are those judges who were named eventually either to the Supreme Court or to the Presidency of a High Court, and whose career records appear in the ZSKS. This rules out those appointed to these positions early in the post-war era, as they would have begun their careers prior to the 1948 and thus would not appear in the ZSKS.

For purposes of deriving these figures, the time of appointment to the Supreme Court is treated as the time of retirement.

The percentage postings figures give the percentage of career, as of 1990, spent in the various positions. Only 21 of the exceptional judges had finished their careers by 1990, which is why the number of observations falls from 25 to 21. For definitions of the variables, see Section 4.3 below.

3.4. The Supreme Court and the Secretariat. The Japanese Supreme Court consists of fifteen justices, who are appointed by the Prime Minister and serve until mandatory retirement at age 70. Among the last 20 justices, the mean age at appointment has been 64. The Chief Justice supervises the Secretary General, the head of the Secretariat. Generally, at least one Supreme Court justice is himself a former Secretary General.

Note the political implications. For most of the post-War years, the Supreme Court included only LDP appointees, and only recent LDP appointees. By appointing them with only six years to go before retirement, the LDP could minimize the "Harry Blackmun problem": the risk that a politically loyal appointee would evolve over time into a very different beast who promoted his own agenda instead of his benefactor's. In contrast to U.S. Presidents, the LDP felt safe in appointing judges whose seats would soon become open again because it faced very high odds (though less than 1, as it turned out) of staying in power. Short terms also have the advantage that the LDP had more, if slightly less valuable, rewards to give to judges of whose performance they approved. Given the extent to which the miniscule probability of a Supreme Court appointment coming open at the right time and under the right administration is said to influence some American appellate judges, this carrot may have been quite useful to the LDP. Since the Supreme Court included at least one justice who recently had headed the Secretariat and knew how its workings intimately, the Court's control of the Secretariat was more than nominal, and talented judges in the Secretariat had an extra incentive to please the Prime Minister. Finally, the Secretariat controlled all judicial personnel matters, except for having to take the general pool of judges as a given. Indirectly, the LDP controlled judicial careers--- but was this effect so diluted as to be unimportant?

4. The data.

4.1. Sources. We collected data from three sources.

First, for information on judicial careers, we used the Zen saibankan keireki soran (ZSKS): a list of all postings for every judge hired after 1948.

Second, for data on judicial opinions, we used the TDK LEX/DB data base of judicial opinions. Available on eight CD-ROM disks, the data base works much like the Lexis and Westlaw systems. Unfortunately, the collection is still slightly incomplete. TDK began compiling the opinions only a few years ago, and since precedent is so much less important in a civil law system, it had nothing like the West national reporter system from which to work. Nonetheless, we have checked the compilation scheme, and have no reason to think the coverage is biased in any way relevant here.

Third, we obtained the membership roster for the leftish Young Jurists League (YJL) from Osorubeki saiban. The authors of the book took the roster -- current as of mid-1969 -- from the League's own newsletter.

4.2. Datasets. From this material, we produced four datasets.

(a) Exceptional judges. We collected data on the mostsuccessful of the post-War judges: all judges in the ZSKS who eventually obtained postings to either the Supreme Court or the Presidency of a High Court. As discussed above, we used this data to learn which posts constitute advantageous assignments (see Table 1).

(b) Judges who ruled on Section 138 cases. To explorewhether judges who decide politically sensitive cases in ways contrary to LDP interests receive unfavorable assignments, we investigated all judges who published opinions on the issue most commonly cited in this context: the constitutionality of the ban on door-to-door canvassing under Section 138 of the Elections Act.

(c) Judges from the classes of 1961 to 1965. We compiled career data on all judges, not just a sample, who entered the courts during 1961 to 1965. We then dropped those judges who had left the judiciary by April 1990, so that we could compare careers of equal length. Some critics accuse the Secretariat of pressing left-leaning judges into early retirement. To the extent that this happened, our findings understate the true scope of any political discrimination.

(d) Judges from the class of 1965. For judges in the class of 1965 (a subset of dataset (c)), we investigated every decision the judge published that involved the government as litigant in one of four fields (labor, administrative, tax, and criminal law). We included all opinions, whether written alone or by a three-judge panel. We coded an opinion as "anti-government" if the party fighting the government won a full or partial victory.

One may complain that many of these opinions will not involve distinctively LDP interests, which is quite correct. The government may even be pleased that it does not win every dispute. Using biased judges to win every case would create incentive problems for the effort level of government litigators. Those litigators may themselves not represent government interests always. The commitment problem Landes & Posner identified means that the government would not want to to obviously control the judiciary. And there is a selection effect to consider (of which more later): a government which thinks the judges favor it will behave with less restraint, simply moving the borderline of legality so that even pro-government judges will eventually balk.

Despite these objections, our coding scheme does contain useful information. First, the test of court attitude is objective. We considered coding opinions according to our subjective sense of whether they furthered LDP interests, but concluded that doing so would invite charges that we "cooked" the data. We therefore opted for a less precise coding, using government interests, to minimize any conscious or unconscious bias on our part. Even this coding was time-consuming, since who ``wins'' in a legal dispute is not always clear, but it is simpler to decide which litigant achieved more of his objectives than whether those objectives matched those of a political party. Second, caveats about incentive effects, promissory credibility, and agency slack notwithstanding, governments generally litigate disputes because they want to win them, even if they do not expect to win every time. To that straightforward -- and forthrightly simplistic -- extent, a decision that a government loses is an "anti-government" decision, and one which will disappoint the l eaders of the government.

4.3. The variables. We construct the following variables. STARTING AGE: the age at which a judge joined the judiciary. To become a judge (or lawyer or prosecutor) in Japan, one must graduate from the government-run two-year Legal Research & Training Institute (the LRTI). During most of the years at stake, the pass rate on the entrance exam to this Institute ranged from 1 to 4 percent. Would-be lawyers, prosecutors, and judges typically pass it only on their 3th, 4th or 5th tries. We hypothesize, therefore, that the lower the age at which a person graduates from the Institute, the higher his cognitive ability and the stronger his determination to succeed. To the extent that success depends on intelligence and drive, STARTING AGE should correlate

inversely with success.

SEX: 1 if a judge is male, and 0 if female. TOKYO U: 1 if a judge went to Tokyo University, and 0 otherwise. Because observers widely consider the Tokyo University Law Department the most selective, graduation there should positively correlate with intelligence and drive. Note, however, that many critics argue that Tokyo University alumni form a clique within the courts and help each other in their careers, independent of ability.

KYOTO U: 1 if a judge went to Kyoto University, and 0 otherwise. Traditionally, observers have considered the Kyoto University Law Department second only to Tokyo University. Critics have accused Kyoto University alumni of running a clique as well.

CHUO U: 1 if a judge went to Chuo University, and 0 otherwise. Chuo University operates a large and respectable but not first-tier law department. We include the variable because so many judges attended the school.

NO UNIV: 1 if the ZSKS lists no university for a judge, and 0 otherwise. A 1 could mean that the judge either attended the LRTI without graduating from a university, or chose not to disclose his educational background.

OPINIONS: the number of recorded decisions a judge published up to 1990, divided by the number of years he spent on the bench. We exclude those years during which he handled only administrative work. Note a potential problem here. The law reporters (there are both official and unofficial ones) do not publish all opinions. Instead, they publish an opinion only if the editors find it interesting or important. If a branch office judge hears less important cases, this could mean that he will not publish as much even if he works as hard, creating a simultaneity problem. Suppose OPINIONS were positively correlated with career success. That fact could mean either that judges receive inferior assignments because they publish less, or that they publish less because they receive inferior assignments. To resolve this problem, we used our Class of 1965 data to create another variable: productivity for all years in courts other than lower court branch offices or summary courts. Fortunately for our purposes, the correlation between that new variable and OPINIONS was .98, indicating that adjusting for poor assignments would make little difference.

1ST POST: the prestige of the first assignment a judge receives. The value is 3 if the posting involves an administrative assignment, 2 if it involves sokatsu duties, 1 if it is on a District or Family Court, and 0 if it involves a lower court branch office or Summary Court. (No judge went to a High Court for his first posting.) For the vast majority of judges, the value was 1.

1980S POST: the prestige of a judge's assignments during the 1980's. If he spent at least 3 years in an administrative assignment, it is 3; if he spent at least 3 years in either an administrative assignment or a sokatsu post (but not 3 years in an administrative assignment), it is 2; if he does not qualify for the categories above and spent at least 3 years in a lower court branch office, it is 0; otherwise, it is 1. For this and the other variables, however, we count time in the branch office only if the judge was not the official head of the branch office, and did not have sokatsu status.

1ST LOCATION: the location of a judge's initial assignment. This is 3 if the judge's first assignment was in Tokyo (including Hachioji), 2 if it was in Osaka, 1 if it was in another large metropolitan area (Yokohama, Nagoya, Sapporo, Kobe, Kyoto, Fukuoka, Kawasaki, Hiroshima, or Kitakyushu), and 0 otherwise. 1980S LOCATION: a judge's location during the 1980s. It is 3 if he spent at least 5 years in Tokyo, 2 if at least 5 years in Osaka or Tokyo (but not 5 years in Tokyo), 1 if at least 5 years in a major metropolitan area (but not 5 years in Tokyo or Osaka), and 0 otherwise. YJL: membership in the Young Jurists League (YJL). The YJL is an organization of lawyers, law professors, and judges that generally supports leftist causes and which its detractors consider a Japan Communist Party affiliate. The variable is 1 if the judge was a member in 1969, and 0 otherwise.

EARLY ANTI-GOVT: the number of anti-government decisions (defined at Section 4.2(d)) that a judge issued during 1965-74. LATE ANTI-GOVT: the number of anti-government decisions that a judge issued during 1975-84. %EARLY ANTI-GOVT: the percentage of anti-government decisions that a judge issued during 1965-74.

%LATE ANTI-GOVT: the percentage of anti-government decisions a judge issued during 1975-84. ANY EARLY ANTI-GOV: 1 if a judge issued any anti-government decisions during 1965-74, and 0 if otherwise.

ANY LATE ANTI-GOV: 1 if a judge issued any anti-government decisions during 1975-84, and 0 if otherwise.

Table 2: Selected Summary Statistics, Classes of 1961-65
	Mean	Range
Starting Age	28.73	24-38
Sex	.96	0-1
Tokyo U	.16	0-1
Kyoto U	.19	0-1
Chuo U	.14	0-1
No University	.43	0-1
Opinions	2.02 .04-10.42
1st Post	.91	0-3 1980s
Post	1.83	0-3
1st Location	1.00	0-3
1980s Location	1.06	0-3
YJL	.35	0-1
Early Anti-Govt%*	9.64	0-
100 Late Anti-Govt%*	2.56	0-100

Observations: 275 for all except asterisked items, which are for the Class of 1965 only (55 observations).

----------------------------------------------------------------

5. The results.

5.1. First assignments. We begin by investigating the factors that determine a judge's initial assignment. The best jobs, our regressions suggest, go to the smartest and hardest working judges. Table 3 reports the results: an ordered probit regression of the characteristics of a judge on two measures of the attractiveness of his first job. Consider each column separately. Column A: Recall that 1ST POST measures whether a judge receives administrative responsibilities, receives a routine District or Family Court assignment, or is stationed to a branch office or Summary Court. Because no judge begins his career with administrative responsibilities, column A effectively shows only that the worst initial jobs (primarily the branch office assignments) go to the oldest novice judges.

Because age at appointment roughly correlates with the number of times the judge failed the LRTI exam, it inversely correlates with intelligence and drive. The worst initial jobs, this suggests, go to the least smart and least hardworking judges.

Column B: The regression on 1ST LOCATION asks who receives the prized Tokyo and Osaka assignments. These jobs go to the judges (i) who are younger, and (ii) who attended the most selective universities. Once more, the regression suggests that the best jobs go to the smartest and hardest working judges. Note that the coefficient for YJL is insignificant in both regressions. Because the League's membership rolls did not become public until 1969, the Secretariat probably would not have known who was a member. Nonetheless, if the coefficient had been significant, it would have suggested that the Secretariat had access to other information about a judge's political beliefs, correlated with YJL membership, and used that information to discriminate by ideology. In fact, it seems not to have done so.

Table 3: Determinants of First Assignment

A. B. 1st Post 1st Location

Sex	-.56 (.97) [.33]	.33 (.92) [.36]
Starting Age	-.12 (3.17) [.00]	-.14 (4.81) [.00]
Tokyo U	-.19 (.40) [.69]	1.28 (4.15) [.00]
Kyoto U	-.73 (1.60) [.11]	.69 (2.34) [.02
Chuo U	-.54 (1.17) [.24]	.10 (.30) [.76]
No Univ	-.49 (1.16) [.25]	.18 (.64) [.53]
YJL	-.07 (.34) [.74]	-.03 (.17) [.86]

Pseudo R2: .08 .11

Observations: 274 274

Notes: Coefficients, followed by T-statistics in parenthesis, and confidence levels in brackets. Program: STATA, running ordered probit.

---------------------------------------------------------------

5.2. Late assignments. Turn now to Table 4, the determinants of late-career success. In the next set of regressions, we ask which judges received the most prized jobs in the 1980s, some 15 to 30 years after they joined the courts. First, in the location regression, STARTING AGE is significant, but in both regressions university affiliation is not. That STARTING AGE continues to be important decades later suggests that intelligence and drive matter, and in ways byond their effect on the judge's first job. That university affiliation loses significance (other than through its effect on 1ST LOCATION) suggests that critics exaggerate the importance of university cliques. If cliques mattered, university affiliation should affect later assignments, perhaps even more than the initial assignment, since over the course of time a judge's university classmates would rise to power in the Establishment. That the judge's university matters only through the initial assignment (while STARTING AGE has an independent continuing effect) implies that it is used as a proxy for ability in determining a judge's initial assignment, but becomes less useful as a proxy once he has a proven track record.

Second, 1ST LOCATION correlates with a judge's later assignments. This corroborates accounts suggesting that the Secretariat places some judges on fast and some one slow tracks, and that an initial assignment to the Tokyo District Court predicts later success. Third, OPINIONS matters: judges who write many publishable opinions do better than those who write few. Although this restates the importance of intelligence and hard work, its potential significance goes further. From time to time, observers suggest that Japanese society may reward judges who settle cases rather than decide them. Because of a cultural preference for negotiated settlements, they argue, the Japanese encourage their judges to settle cases when they an. Because settlements do not appear in our data, we do not know whether the most successful judges settle the lowest percentage of their disputes.

We do know that the most successful judges are the most prolific in writing published opinions for cases that fail to settle.

Last, independent of intelligence and hard work, political preferences matter: whether a judge was a YJL member inversely correlates with whether he received prestigious administrative responsibilities in the 1980's. Those judges named as part of a Marxist group in 1969 were still receiving less attractive jobs 10 to 20 years later. Curiously, YJL membership did not affect the location where the judge worked. Perhaps the Secretariat was willing to assign able leftists to the cities, but not with the highest positions within the judicial hierarchy.

-------------------------------------------------------------

Table 4: Determinants of Career Success

B. 1980s Post 1980s Location

Sex	.31 (.92) [.36]	.21 (.57) [.57]
Starting Age	-.029 (1.07) [.29]	-.070 (2.26) [.03]
Tokyo U	.092 (.30) [.77]	.076 (.22) [.82]
Kyoto U	.058 (.20) [.84]	.0089(.03) [.98]
Chuo U	.10 (.34) [.73]	.40 (1.20) [.23]
No Univ	-.15 (.56) [.58]	-.11 (.38) [.70]
1st location	.19 (2.67) [.01]	.31 (4.05) [.00]
Opinions	.18 (4.11) [.00]	.20 (4.35) [.00]
YJL	-.28 (1.94) [.05]	.13 (.83) [.41]

Pseudo R2: .08 .13

Observations: 274 274

Notes: Coefficients, followed by t-statistics in parenthesis, and confidence levels in brackets.

Program: STATA, running ordered probit.

---------------------------------------------------------------

6. The effect of anti-government opinions. 6.1. The method. We now turn to a more complex inquiry: whether the way a judge decides cases influences his career success. We find that it does. The phenomenon seems to have two aspects: (a) a punishment effect, where anti-government opinions translate immediately into less attractive job assignments in the short-term, and (b) an information effect, where anti-government preferences revealed in a judge's decisions translate into a career disadvantage in the long term. Our political variable so far has been YJL membership, which is cleanly defined and easily measured for the entire sample of 275 judges. We now wish to look at a more complicated characteristic: whether a judge rules for or against the government when the government came to court. This introduces difficulties in measurement, econometrics, and theory.

We have already discussed how we measured whether a given judge ruled in favor or against the government. Because it is so time-consuming, we examined the opinions only of the Class of 1965, a much smaller number of observations than for all the Classes between 1961 and 1965. The econometric problem comes in combining our 55 observations on judicial opinions with our 275 observations on all the other variables relevant to a judicial career. If we were willing to drop 220 observations, the econometrics would be simpler: we would repeat the probit regressions in Table 4, but adding opinion variables on the right-hand-side. This not only would discard information, however, but would raise doubts about the validity of the estimates and the standard errors, since probit is a nonlinear, asymptotic technique for which having a large sample is especially important. What we have done, therefore, is to take a different approach. We begin with the regressions of Table 4, which use all 275 observations to predict career success. These regressions do not explain all the variance in the data, and generate an unexplained residual for each judge. If we can explain this residual using judicial opinion variables, we will have shown that a judge's opinions matter to his career success, and ought to have been in the regressions in Table 4. Moreover, since the residual is a continuous variable, we can use ordinary least squares, which does not rely on asymptotics for its validity.

More specifically, we first turn to our Class of 1965 dataset and use our Table 4 regressions to generate a "residual" for each judge. For this process, we combine our estimated Table 4 coefficients with the the particular values the variables take for each judge. This generates a left-hand-side "score" for each judge. Ordered probit also generates estimated cutoff ranges to translate the scores to observed values, so we match each judge's "score" with the cut-off ranges to generate a predicted posting for each judge, an integer from 0 to 3. An example may help. We did not report the estimated cutoff scores above, but for the regression in Table 4.A., they were -1.37, -.90, and .78. Suppose judge X has a true score of -3.3. Because it falls below the bottom cutoff of -1.37, his predicted posting is 0. If his score were -1.20 intead, he would fall in the -1.37 to -.90 range and have a predicted posting of 1.

Because for each judge we have only an estimated score, our predictions do not take straightforward integer values. If we knew with certainty that X's score were -3.3, we could predict with certainty a posting of 0. Because -3.3 is just an estimate, however, his true score might be higher or lower--- there is some probability that his true score is +0.5, given an estimated score of -3.3, in which case his posting would be 3, not 0.

Accordingly, if X's score is -3.3, our best prediction is not a posting of 0 but a weighted average of 0, 1, 2, and 3.

Those weights will be our estimated probabilities of the true score lying in the four intervals of [-infinity, -1.37], [-1.37, -.90], [-.90, 0.78], and [0.78, +infinity]. We find the probabilities by using the standard error of our estimate. Our predicted career quality is the resulting weighted average. By now comparing this predicted career quality with a judge's actual posting (of 0, 1, 2, or 3), we calculate a residual -- a continuous variable that measures the judge X's unexplained career quality. If positive, it indicates that he did better than our regression predicted; if negative, it indicates he did worse.

Finally, we used a logit transformation to map the value of the residual, which lies between -3 and +3, to the entire real line, between positive and negative infinity. This transformation maps the raw residual u to log[(u + 3)/(3 - u)].

A variable representing the unexplained success of a judge's career having been thus constructed, we can now see whether a tendency to decide public-law cases against the government has a negative correlation with that variable. If judge X's decisions had no impact on his career, regressing his residual on a variable summarizing his decisions would yield an insignificant coefficient. If they did have an impact, then the coefficient might be significant.

6.2. The punishment effect. According to Table 5, judges who decided cases against the government soon receive less attractive jobs. In Part A of Table 5, the absolute number of anti-government opinions that a judge writes in 1975-84 inversely correlates with the odds of receiving a post in an attractive city in the 1980s. In Part B, whether a judge decides any anti-government opinions in 1975-84 (a 0-1 variable) inversely correlates with receiving high-status posts in the judicial hierarchy in the 1980s.

The simplest explanation is a straightforward punishment strategy: if you decide cases against the government, the expected value of your next job falls. The probability of punishment may well be less than one, since the government does not care equally about all cases, will not expect to win every case, and will not necessarily wish to punish every judge on a 3-judge panel. (Note that the published opinions do not identify dissenters, but the Secretariat probably has access to that information.) Notwithstanding these caveats, according to Table 5, anti-government opinions translate directly into less attractive posts in the near future. We find the haphazard confidence levels in Table 5 a puzzle. In Part A, only the location residual is significant, and in B, only the post residual. We suspect that this reflects the noise in the data discussed in the preceding paragraph and the relatively small sample size. Despite the large standard errors, however, the signs for the late opinions are negative in all four regressions.

Table 5: The Punishment Effect

A. Number of Anti-Government Decisions:

Post Residual Location Residual

Early Anti-Govt	.025 (.77) [.45]	.015 (.48) [.63]
Late Anti-Govt	-.097 (l.18) [.25]	-.20 (2.51) [.02]

R2: .03 .11

B. Any Anti-Government Decisions:

Post Residual Location Residual

Any Early Anti-Govt	.17 (1.13) [.26]	-.21 (1.40) [.17]
Any Late Anti-Govt	-.32 (1.75) [.09]	-.036 (.20) [.84]

R2: .06 .06

Observations: 54 54

Notes: Coefficients, followed by t-statistics in parentheses, and confidence levels in brackets.

Program: STATA, running ordinary least squares on a logistic conversion of the residual from the career regression.

-------------------------------------------------------------

6.3. The information effect. Potentially, a judge's opinions convey information about his political preferences. If litigated cases were a random sample of all disputes, the percentage of cases that the government won before a judge would say something about his biases. The Secretariat could then use that information to decide where to post its judges. a. The selection bias. Litigated cases are not a random sample of all disputes, however, and the percentage of cases won says as much about the types of cases that go to trial as about how the judge views disputes in general (Priest & Klein, 1984). In order to avoid the costs of trial, most disputants settle disputes whenever they agree about the likely litigated outcome. As a result, cases do not go to trial randomly. Instead, they go to trial when the judge's expected decision is unclear.

Just because 80 percent of judge Y's decisions are pro-plaintiff does not mean that he is necessarily pro-plaintiff: he may only be pro-plaintiff in the most complicated cases, when the litigants found his decisions hardest to predict. For the purposes of this analysis, settlement could have an even more bizarre effect: it may be that the government goes to trial with its most outrageous cases only when it knows the judge is particularly pro-government, and is only moderately displeased when the government's arguments are too weak even for that judge to swallow. Thus, the judges who rule against the government most often might be the pro-government judges.

Settlement will be most common where the parties know a judge's style and biases most precisely. If nothing is known about a judge in advance, he will hear cases that are randomly selected. Given that randomness, his verdict rate will indeed tend to disclose his biases. A judge with a shorter track record is one about whom litigants will have less information. Accordingly, we hypothesize that the selection bias will be strongest among judges at the end of their careers, and weakest at the start. Our data contains separate variables for early and late anti-government decisions. Lack of correlation between these two variables could be a sign either that no judges are biased for or against the government, or that the selection effect means that once a judge acquires a track record, he tries only a selected group of cases and his anti-government percentage says nothing about his biases. The correlation between %EARLY ANTI-GOVT and %LATE ANTI-GOVT is .12. Whether the size of this correlation is evidence for or against selection bias is unclear, but it is statistically insignificant ($t=.28$, significance level 78 percent). Second, the correlation between YJL and %EARLY ANTI-GOVT is .23, and between YJL and %LATE ANTI-GOVT is .06. If membership in the YJL provided information about whether a judge would decide cases against the government, we would worry more about a selection bias even for younger judges. It does not seem to provide such information, however, for the t-statistic for the regression of %EARLY ANTI-GOVT on YJL and a constant is only 1.16 (a significance level of 25 percent).

b. The results. According to Table 6, a judge who decides a high percentage of cases against the government early in his career does indeed receive geographically less attractive jobs later in his career. The percentage of anti-government opinions early in a judge's career matters because it reveals information about his true biases. The later career percentage does not, though the absolute number matters to the government's short-term punishment strategy, as was seen in Table 5, because it conveys no meaningful information. -----------------------------------------------------------------

Table 6: The Information Effect

Post Residual Location Residual

%Early Anti-Govt	-.15 (.48) [.63]	-.52 (1.67) [.10]
%Late Anti-Govt	.27 (.62) [.54]	.46 (1.05) [.30]

R2: .01 .06 Observations: 44 44

Notes: Coefficients, followed by t-statistics in parentheses, and confidence levels in brackets.

Program: STATA, running ordinary least squares on a logistic conversion of the residual from the career regression. Only 44 of the judges are used, because the others had no published opinions during the early time period, so their percentage anti-govt. is undefined.

-----------------------------------------------------------------

7. Electoral law decisions. Consider now the most common anecdotal accounts of political bias: the alleged discrimination against judges who held the #164# 138 canvassing ban unconstitutional. Among lower court judges, we located 37 who held the ban constitutional and 9 who held it unconstitutional. Using the data on these 46 judges, we test whether a judge's decision on the issue affected the assignments he received. Toward this end, we introduce several new variables:

PRIOR POSTS: the prestige of a judge's assignment before the #164# 138 decision. The variable is 3 if he spent at least 3 years in an administrative job during the 10 years before the decision, 2 if he spent at least 3 years in an administrative or sokatsu capacity, 0 if he or she did not meet those requirements but spent at least 3 years in a lower court branch office or Summary Court, and 1 if otherwise. LATER POSTS: the equivalent to PRIOR POSTS for the 10 years after the decision. It takes the values 3, 2, 1, or 0. PRIOR LOCATION: the desireability of the judge's location before the #164# 138 decision. The variable is 3 if the judge spent at least 5 of the previous 10 years in Tokyo, 2 if at least 5 years in Osaka, 1 if at least 5 years in metropolitan areas generally, and 0 if otherwise. LATER LOCATION: the equivalent to PRIOR LOCATION for the 10 years after the decision. It takes the values 3, 2, 1, or 0. PRIOR PUB OPS: the judge's productivity (published opinions per year on bench) for the 10 years before the #164# 138 decision. 138 DECISION: 0 if the judge held the canvassing ban constitutional and 1 if otherwise.

Table 7: Effect of #164# 138 Opinions on Careers

Later Posts Later Location

Prior Posts .11 (.52) [.61]

Prior Location .22 (1.40) [.17]

Prior Opinions -.17 (1.62) [.11] .12 (0.94) [.35] 138

Decision -.01 (.023) [.98] -.31 (.59) [.56]

Pseudo R2: .03 .04

Observations: 46 46

Notes: Coefficients, followed by T-statistics in parentheses, and confidence levels in brackets. Program: STATA, running ordered probit.

----------------------------------------------------------------

Note the absence of statistically significant results in Table 7. Notwithstanding the many stories of judges punished for writing heterodox opinions on #164# 138, we find no evidence of the phenomena. Given this absence of statistically significant effects (but the presence of a punishment effect generally -- see Table 5), the puzzle shifts to the anecdotes: why do the stories exist? In part, the reason for the divergence between the anecdotes and Table 7 lies in the numbers: nine observations is a small number with which to obtain significant results -- especially since ordered probit's reliability requires large samples. In part, however, the reason may also lie in the Secretariat's sophistication: it seems already to have realized that these judges were unreliable before they wrote their #164# 138 opinions. Table 8 shows that the judges who held the canvassing ban unconstitutional did indeed hold worse jobs after the opinions than the judges who held it constitutional. Disproportionately, however, they already had worse jobs before the decisions. This, in turn, might help explain why they did not fear making decisions hostile to the government.

-----------------------------------------------------------------

Table 8: Summary 138 Statistics

Constitutional Unconstitutional Mean Mean .

Prior Posts 1.43 .44

Later Posts 1.62 1.56 Prior Location .86 .00

Later Location 1.00 .44

Sex 1.00 1.00

YJL .14 .56

Observations 37 9

--------------------------------------------------------------

8. Conclusions.

Because civil-law systems hire unproven jurists into career judiciaries, many such systems maintain elaborate incentive structures to prevent their judges from shirking. In this article, we used career data from the Japanese courts both to explore the general determinants of career success, and to test how extensively the government manipulated those incentives toward political ends. We find considerable evidence that the government rewards the smartest and hardest working judges. Contrary to some observers, we find little evidence of ongoing school cliques. We also find no evidence that the Japanese system rewards judges who mediate over those who adjudicate. Rather, the judges who do best are those who publish the most opinions.

More controversially, we locate several politically driven phenomena. First, those judges who joined a prominent leftist organization in the 1960s were still receiving less attractive jobs than their peers in the 1980s. Second, those judges who decided cases against the government faced a straightforward short-term penalty: on average, they received less attractive assignments over the next several years. Last, those judges who decided a high percentage of cases against the government early in their careers were still receiving less attractive jobs than their peers in the 1980s.

References:

Anderson, Gary, William Shughart, & Robert Tollison. 1989. "On the Incentive of Judges to Enforce Legislative Wealth Transfers." 32 J. Law & Econ. 215-28.

Clark, David. 1988. "The Selection and Accountability of Judges in West Germany: Implementation of a Rechtstaadt." 61 S. Cal. L. Rev. 1797-1847.

Cohen, Mark A. 1991. "Explaining Judicial Behavior or What's 'Unconstitutional' about the Sentencing Commission." 7 J. Law, Econ. & Org. 183-99.

Easterbrook, Frank. 1982. "Ways of Criticizing the Court." 95 Harv. L. Rev. 802-32.

Haley, John O. 1995. "Judicial Independence in Japan Revisited." 25 Law in Japan 1.

Hayakawa, Takeo. 1971. "The Japanese Judiciary in the Whirlwind of Politics." 7 Kobe Univ. L. Rev. 17.

Hoso mondai kenkyu kai (ed.). 1983. Hosokai no sayoku seiryoku (zenteiban) [Leftist Influence in the Legal World (Rev. Ed.)]. Tokyo: Tachibana shoten.

Landes, William, & Richard A. Posner. 1975. "The Independent Judiciary in an Interest-Group Perspective." 18 J. Law & Econ. 875-901.

McCubbins, Mathew D., & Thomas Schwartz. 1984. "Congressional Oversight Overlooked: Police Patrols versus Fire Alarms." 28 Am. J. Pol. Sci. 165-79.

Merryman, John Henry. 1985. The Civil Law Tradition (2d ed.). Stanford: Stanford University Press.

Miyazawa, Setsuo. 1991. "Administrative Control of Japanese Judges." In Philip S.C. Lewis (ed.), Law and Technology in the Pacific Community. (Boulder: Westview Press).

Muller, Ingo. 1991. Hitler's Justice: The Courts of the Third Reich. Cambridge, Mass.: Harvard University Press. Nihon minshu horitsuka kyokai (ed.). 1990. Zen saibankan keireki soran: kaitei shinban [Biographical Information on All Judges: New Revised Edition]. Tokyo: Konin sha.

North, Douglass C., & Barry R. Weingast. 1989. "Constitutions and Commitment: The Evolution of Institutions Governing Public Choice in Seventeenth-Century England." 49 J. Econ. Hist. 803-32.

Priest, George L., & Benjamin Klein. 1984. "The Selection of Disputes for Litigation." 13 J. Legal Stud. 1.

Ramseyer, J. Mark. 1994. "The Puzzling (In)Dependence of Courts: A Comparative Approach." 23 J. Legal Stud. 721-47.

---------- & Frances McCall Rosenbluth. 1993. Japan's Political Marketplace. Cambridge: Harvard University Press.

Rasmusen, Eric. 1994. "Judicial Legitimacy as a Repeated Game." 10 J. Law, Econ. & Org. 62-83.

Shiso undo kenkyu sho (ed.). 1969. Osorubeki saiban [Fearsome Trials]. Tokyo: Zenbo sha.

Spiller, Pablo T. & Rafael Gely. 1992. "Congressional Control or Judicial Independence: The Determinants of U.S. Supreme Court Labor-Relations Decisions, 1949-1988." 23 RAND J. Econ. 463-92.

---------- & Matthew L. Spitzer. 1992. "Judicial Choice of Legal Doctrines." 8 J. Law, Econ. & Org. 8-46.

Toma, Eugenia Froedge. 1991. "Congressional Influence and the Supreme Court: The Budget as a Signalling Device." 20 J. Legal Stud. 131-46.

Tsukahara, Eiji. 1991. "Saibankan keireki to saiban kodo [The Relationship Between a Judge's Career and his Opinions]." 43 Ho shakai gaku 46.

Sakaguchi, Tokuo. 1988. "Saibankan ni taisuru ninchi yakushoku "sabetsu" ni tsuite [Regarding the "Discrimination" Against Judges in Location and Official Responsibilities." 5 [Osaka bengoshi kai,] Shiho mondai taisaku nyusu 4.

ZSKS. See Nihon minshu.