Revision of July 12, 2000 Why Is the Japanese Conviction Rate So High? J. Mark Ramseyer Harvard Law School Eric B. Rasmusen Indiana University, Kelley School of Business Abstract Conviction rates in Japan exceed 99 percent. Why? One explanation involves judicial incentives. Because Japanese judges can be penalized by a personnel office if they rule in ways the office dislikes, perhaps they face biased incentives to convict. Using data on the careers and opinions of 321 Japanese judges, we find that judges who acquit do indeed have worse careers following the acquittal. On closer examination, though, we find that the punished judges are not those who acquit on the ground that the prosecutors charged the wrong person. Rather, they are judges who acquit for reasons of statutory or constitutional interpretation, often in politically charged cases. Thus, the apparent punishment of acquitting judges seems unrelated to any pro-conviction bias at the judicial administrative offices. We suggest an alternative explanation: that the high conviction rates reflect case selection and low prosecutorial budgets; understaffed prosecutors present judges only with the most obviously guilty defendants. Ramseyer: Harvard Law School, Cambridge, MA 02138. Office: 617-496-4878. Fax: 617-495-1110. Ramseyer@law.harvard.edu. Rasmusen: Indiana University, Kelley School of Business, Dept. of Business Economics and Public Policy, BU 456, 1309 E. 10th Street, Bloomington, IN 47405-1701. Office: 812-855-9219. Fax: 812-855-3354. Erasmuse@indiana.edu;. Php.indiana.edu/~erasmuse. We thank John Coates, John Lott, Setsuo Miyazawa, Eric Posner, Richard Posner, Roberta Romano, Arthur Rosett, Detlev Vagts, Mark West, an anonymous referee, and participants at workshops at the American Law and Economics Association, the University of Chicago, Harvard University, Kyushu University, Washington University St. Louis, the University of Wisconsin, and Yale University for helpful comments. We received generous financial assistance from the Lynde and Harry Bradley Foundation, the Sarah Scaife Foundation, the John M. Olin Foundation, and the John M. Olin Center for Law, Economics & Business at the Harvard Law School. Virtually every American law professor knows at least one fact about Japanese law: Japanese criminal courts never acquit anyone. The puzzle is what to make of this. Perhaps Japanese judges convict because Japanese prosecutors only bring true criminals into the courtrooms. If so, the high conviction rates represent a phenomenon many observers would envy. Or perhaps Japanese judges convict because they further their careers by convicting. Even so, most defendants might be guilty -- but then again, they might not. That Japanese prosecutors might do very well at keeping innocent defendants out of the court system follows straightforwardly from budget constraints. Chronically understaffed, Japanese prosecutors lack the time to prosecute any but a small fraction of the suspects forwarded by the police. Rather than waste their time with dubious cases (other than the occasional politically driven corruption case), perhaps rational prosecutors prosecute only the most obviously and gruesomely guilty. That Japanese judges might face skewed incentives follows from the bureaucratic structure of the Japanese courts. Ordinarily, judges in Japan take their jobs straight out of the national law school. They then stay judges until they retire. During their career, they work as judicial bureaucrats. Where they work and what they earn depends on the reputation they maintain with their judicial betters at the courts' administrative office. Elsewhere, we have noted how the best jobs in the system go to the judges who are smartest, hardest working, and politically most reliable. Perhaps they also go to the judges who most readily convict. Using data on the careers and published opinions of 321 judges -- every judge who published an opinion in a criminal case in 1976 or 1979 -- we will here explore the effect criminal opinions have on Japanese judicial careers. We find mixed results. On the one hand, Japanese judges indeed do receive better posts if they convict than if they acquit. A judge sitting alone who acquits will spend an extra year and a half in branch offices over the next decade. Also, a judge who finds a conviction reversed on appeal will suffer no significant penalty, but one who finds an acquittal reversed will spend an extra three years in branch offices. On the other hand, this punishment generally occurs only in highly unusual cases -- those involving politically sensitive crimes, for example, or where the acquittal results from statutory interpretation such as the definition of mental capacity or the timing of the statute of limitations. Almost never does a judge find himself punished for deciding that prosecutors nailed the wrong man. For acquitting defendants in run-of-the-mill crimes, judges seem not to suffer a career hit. Accordingly, the high conviction rates in routine cases result more plausibly from budget constraints than from skewed judicial incentives. We begin by comparing conviction rates in Japan and the U.S. (Section I). We then discuss why the severe budget constraints in Japanese prosecutorial offices could result in a high ratio of guilty to innocent prosecutions (Section II). Finally, we explore empirically the pressure Japanese judges face to convict (Section III). I. Comparative Conviction Rates: A. Introduction. Conviction rates are high in Japan. They are high in most countries, of course, but they are particularly high in Japan. In U.S. federal courts, prosecutors win 85 percent of all criminal cases (46,773 out of 54,980 in 1995) and convict 83 percent of murder defendants (265 out of 313). In U.S. state courts, they win roughly 87 percent of their felony cases and 88 percent of their misdemeanors. Japanese district court judges convict 99.9 percent of all defendants (49,598 out of 49,643 in 1994). Of the defendants up on murder charges, they convict 99.7 percent (587 out of 589). Conviction rates like 83.3 or 99.7 percent are not the odds a defendant who contests his guilt actually faces, either in the U.S. or in Japan. The data for both countries include cases where defendants decided not to contest the charges. In Japan, to be sure, the law does not allow plea bargains. Instead, all defendants prosecuted face trial, and courts convict them only if prosecutors prove their guilt beyond a reasonable doubt. Yet the lack of formal plea bargains in Japan does not mean the parties ignore analogous calculations. Defendants need not contest their guilt. Instead, they can freely confess. In exchange, prosecutors can freely suggest a sentence lower than the sentence the defendant would receive at a contested trial. The suggestion matters, because courts routinely accept the recommended sentence. Such an implicit bargain would not be enforceable formally. Still, both sides will generally find it binding in fact. If only to encourage defendants to confess, the prosecutor's office will want to maintain its reputation for playing by customary norms. Having already confessed, the defendant would find it hard effectively to renege even if he wanted. B. Confessions in Japan: For both the prosecutor and the defendant, uncontested proceedings create gains from trade. By agreeing not to contest his guilt, the defendant saves the prosecutor resources he can use to prosecute other people. So long as the deterrent effect of these additional convictions outweighs the savings to the settling defendant, the arrangement increases the deterrent value of the law. Perhaps more important, it increases the prosecutor's observable work product: the productivity he can show his present bureaucratic supervisors and future private-sector employers. The gains to the defendant are equally obvious: he gains if the plea-bargained sentence is less costly to him than the sum of (a) the expected risk-adjusted value of the sentence from a contested trial and (b) his expected litigation costs. Procedural protections for defendants in Japan loosely track those in America. Although courts hesitate to mandate a blanket exclusionary rule, they do exclude coerced confessions on reliability grounds, they impose a presumption of innocence, and they demand proof at levels close to the U.S. "reasonable doubt" standard. They enforce a right to counsel at trial (with state- appointed counsel for the poor), a right to remain silent, and a right to interrogate witnesses, and they require warrants for searches and seizures. Absent an explicit or implicit bargain, a rational defendant gains little by confessing. In Japan, as in the United States, some defendants do confess or plead guilty without a promise of a lower penalty. Exceptions do not prove rules, however, and we have little reason to think defendants anywhere systematically act against their self-interest. If Japanese defendants routinely file confessions, that very fact suggests that they routinely receive lower sentences. Japanese defendants certainly do confess. Of the 49,856 criminal cases in Japanese district court in 1994, defendants contested prosecution in only 7.3 percent. If a confession generally represents a functional analogue to a plea bargain (obviously, they are not exact equivalents), then Japanese defendants cut a bargain nearly 93 percent of the time. By contrast, of the 54,980 criminal cases in U.S. federal district courts in 1995, defendants pleaded innocent in 22 percent. In the state courts, they pleaded innocent in 11 percent. Notwithstanding the formal absence of plea bargains, Japanese defendants are modestly more likely not to contest prosecution than Americans. C. The Logic to the Comparative Rates: The somewhat lower frequency of contested prosecutions in Japan probably reflects the greater predictability of trials there. The logic follows from the well-known model of litigation and settlement. A prosecutor faces lower costs if he can simply meet with the defendant's lawyer and bargain than if he must prove his case in a contested trial. A defendant similarly saves time and expense by avoiding trial. In bargaining, moreover, both sides reduce risk, exchanging a gamble between conviction with a high sentence and acquittal with no sentence for a sentence that is definite. If both sides similarly estimate the odds of the various trial outcomes, then unless one of them enjoys risk they will plea bargain. Even if they do not agree completely on the odds, so long as their opinions are close enough and the cost and risk of trial are high enough they will cut a deal. In Japan, two factors dramatically increase the predictability of criminal litigation relative to the U.S. First, trials are discontinuous. The prosecutor and defendant begin the trial by assembling before a judge. They discuss the issues at stake and then recess to gather evidence. Having collected evidence on the first issue, they reconvene, litigate that issue, and recess again to gather evidence on the next. In the process, the judge has considerable chance to disclose how he leans. As he discloses his inclinations, the parties obtain increasingly accurate estimates of the outcome at trial. Second, all trials are bench trials -- Japanese courts do not use juries. Unlike juries, judges have "seen it all before," and the lawyers have seen them seeing it. Judges thus have histories that lawyers can investigate to find out how they approach cases. Whether because they take professional pride in uniformity, or because (like most mortals) they economize on effort, judges tend to decide similar cases similarly. This predictability reduces the variance in the prosecutor's and defendant's estimates of success. Plea bargains fail and litigation ensues when both parties are optimistic -- where the difference in their estimates of the contested outcome is so great that they cannot agree to meet in the middle and save their trial costs. To the extent that the predictability of the judicial system reduces variance, it reduces the frequency with which prosecutors and defendants will arrive at dramatically different estimates. If their estimates are close, they will cut a deal. Evidence from civil suits corroborates this explanation. As observers often note, disputants in Japan settle a larger fraction of civil disputes than in the U.S. They settle them, however, in the shadow of the law. Out of court, defendants pay and plaintiffs collect amounts that track the amounts a court would award. They settle because the trial outcome is usually clear, and so neither side has reason to bother litigating. Rather than sue, they settle, along the lines a judge would rule. D. Residual Conviction Rates: Given the high rate of uncontested proceedings in Japan, what is the residual conviction rate in contested cases? The contrast with the United States is stark. In 1994 Japanese defendants refused to confess in 3,648 cases. Courts acquitted them in 45, yielding a contested conviction rate of 98.8 percent. In 1995, U.S. federal defendants pleaded innocent in 11,877 cases. Courts acquitted them (or dismissed the charges) in 8,207, yielding a contested conviction rate of 30.9 percent. In state courts, the contested conviction rate apparently stands even lower. II. Prosecutorial Incentives: A. Introduction: For a scholar, high conviction rates are a bedeviling puzzle. They might be high because prosecutors only prosecute the guilty, and judges then duly convict. Or they might be high because judges dutifully convict everyone prosecuted, guilty or no. To determine the truth -- which could include both reasons -- we would seem to need independent evidence of the guilt of the accused. That, of course, is information we rarely have. As a simple heuristic, consider Figure 1. In this summary, prosecutors can screen suspects carefully or cursorily. The courts, in turn, can decide the cases fairly (carefully and without a pro- conviction bias) or unfairly (either cursorily or with bias). [Insert Figure 1 about here.] If a prosecutor screens suspects carefully, then whether a court reviews cases fairly (row (a)) or unfairly (row (b)), conviction rates will be high and few innocent defendants will be convicted. If a prosecutor screens only cursorily, then if a court reviews cases fairly it will periodically acquit (row (c)). Conviction rates will be lower, but here too few innocent defendants will stand convicted. If a court merely rubber-stamps prosecutions, then conviction rates will remain high and innocent defendants may indeed comprise a substantial fraction of those convicted (row (d)). The puzzle is which case best describes Japan. Given the high conviction rates, it cannot be row (c). Given that prosecutors will not have much incentive to be careful if judges rubberstamp all prosecutions, row (b) does not seem a sub-game perfect equilibrium. But which of the other two might it be? Is it the intuitively troubling row (d), or the apparently less problematic row (a)? Lacking independent evidence of the guilt of the suspects, we cannot test the issue directly. To explore it indirectly, we instead proceed in two steps. We first explore the likely effect of the resource constraints on prosecutorial screening (this Section II). As a way to ask whether we are in row (a) (with unbiased adjudication) or row (d) (biased adjudication), we then test empirically whether judges in Japan face unbiased incentives (Section III). B. The Prosecution Budget Story: 1. The intuition. One need not posit prosecutorial clairvoyance to suggest that there might be a difference in the ratio of innocent to guilty defendants in Japanese and American courts. Instead, one need only show a difference in prosecution budgets. Suppose, as seems likely, a prosecutor promotes his career by convicting defendants. A win helps, while a loss is at best neutral, but also is time wasted that might have used to obtain more convictions. How then does a prosecutor win cases? Obviously, he could try to be brilliant, experienced, and prepared. He could also, however, just prosecute easy cases. By only filing cases with overwhelming evidence, he could avoid ever losing except by accident. If averse to even a small chance of losing, he could offer such generous plea bargains that no rational defendant would ever plead innocent. Things are not so simple, of course. Given the usual bureaucratic structures, a prosecutor cannot maximize his current conviction rate by prosecuting one easy larceny case each year and dropping all the other cases brought to his desk. He must at least keep busy, and he faces some pressure from police and public to convict as many criminals as possible. In Japan, this pressure is formalized in "prosecution review commissions,” commissions of 11 randomly chosen citizens who hold hearings on non-indictments and issue non-binding recommendations to prosecutors. Even given these pressures, a prosecutor will have strong incentives to prosecute mostly strong cases. Any prosecutor, in any system, has more cases than he has time to handle. Rationally, he will usually turn to his strongest cases first. In exceptional circumstances, he will try evidentially weak cases involving politically visible or particularly heinous crimes. Yet he will usually start with the most straightforward cases and move to the more problematic ones only as time permits. Japanese prosecutors face particularly stark incentives to win cases. Like the judges we discuss below, prosecutors are part of a national bureaucracy. Within this bureaucracy, some posts are prestigious and some are not. Yet every two years, prosecutors find themselves moved from post to post on the basis of their performance. Whether they move to a better or worse post depends on that performance, and, according to many observers, losing a case is a sure path to demotion -- as one Tokyo prosecutor put it, “[p]rosecutors regard acquittal as a very serious problem. … Each prosecutor examines each case very, very carefully, and if he doubts that it’s a strong case – or if there are extenuating circumstances – he doesn’t indict.” Sociologist David Ted Johnson made the connection between career failure and acquittals more explicitly: “[p]rosecutors found to be ‘negligently’ responsible for acquittals pay heavy fines in this currency of status” within the Ministry of Justice. This is not because the prosecutor's career depends on people who only have time to look at a few summary statistics, as with elected state prosecutors in the United States who worry about conviction rates and high-profile cases at election time. Rather, in Japan the Ministry of Justice apparently will evaluate a prosecutor both on the abilities he demonstrates in particular cases, and on the number of people he convicts. As a result, a Japanese prosecutor has less reason to care about his success rate per se than the U.S. elected prosecutor. Instead, his position more closely resembles that of an assistant U.S. district attorney, whose career depends on evaluation by professionals in his field. Given this dynamic, if prosecutorial offices try to maximize the number of convictions, then one with a lower budget (Japan, by our argument) may have a higher conviction rate than one with higher funding (the U.S.). Because of the severe resource constraints, the leaner office will have time to prosecute only the most egregious cases. In the extreme, a resource-starved office would only be able to prosecute one case a year, but have a conviction rate of 100 percent. Given more generous resources, a fatter office would move to prosecute riskier cases too, and its conviction rate would drop. This is not because the prosecutor is trying to maximize his conviction rate -- the fatter office would always have the higher conviction rate if that were true -- but because trying to put more criminals behind bars will result in more acquittals as well as more convictions. 2. Comparative statistics. Consider a few back-of-the-envelope comparisons. To deal with its legal affairs, the U.S. federal government employs 27,985 lawyers and the various state governments 38,242 (of which 24,700 are state prosecutors). The entire Japanese government employs 2,000. Although Japan has fewer inhabitants and lower crime, this vast difference in personnel still generates significantly higher prosecutor workloads in Japan. Each year, police in the U.S. make about 12 million arrests, excluding traffic offenses. Given that there are 25,000 state prosecutors, that comes to 480 arrests per year per prosecutor. Because many of the 28,000 federal lawyers also prosecute crimes, this figure even overstates the prosecutorial work load. Of the 12 million arrests, the FBI catalogs 2.4 million as “serious crimes.” Even if only state prosecutors handled those cases, they would still have just 96 serious criminal arrests apiece. By contrast, Japanese police make about 1.4 million arrests per year for Criminal Code violations. They make another 1.1 million for other (mostly traffic) crimes. This comes to 1,250 total arrests per prosecutor. If we exclude traffic offenses, it comes to 700. Differences in clearance rates reflect these differences in workload. In 1994, state courts in the U.S. convicted 870,000 people of a felony, and the federal courts another 44,000. Given the conviction rates cited earlier, prosecutors must have brought felony charges against slightly more than 1 million defendants. If we use the FBI’s definition of “serious crime” arrests as a rough proxy for felony arrests, we obtain a 42 percent prosecution/arrest ratio for felonies. In Japan, by contrast, of the 919,000 people arrested for Criminal Code violations in 1995, prosecutors filed charges against a scant 17.5 percent. Statistical categories can be hard to compare between countries, but consider murder, perhaps the most consistently defined and best measured crime. In 1994, U.S. police arrested 19,000 people for roughly 26,000 murders. Courts convicted 12,000. Again using conviction rates to infer prosecutions from convictions (at an 85 percent conviction rate, this suggests about 14,000 prosecutions), we can deduce that prosecutors prosecuted roughly 75 percent (14,000/19,000) of all people arrested on murder charges. In Japan, of the approximately 1,800 people arrested for 1,300 murders, prosecutors tried only 43 percent. Could the comparative prosecution rates be a function of police sloppiness, of a tendency in Japan for police to arrest irresponsibly? We suspect not -- the numbers just cited also imply that in the U.S. 54 percent of murders result in trial and 46 percent in conviction (under the false but simplifying assumption that each murder involves one murderer), whereas in Japan 60 percent of murders result in trial and conviction. And although we have little evidence on police “sloppiness,” American police do face more public complaints about heavy-handed behavior than their Japanese peers. Or could the cross-national variation reflect differences in police budgets?. The number of police officers per 10,000 population is similar: 18 in Japan (227,000 police) compared to 24 in the U.S. (604,000 police). That comes, however, to 175 police per murder in Japan but just 23 police per murder in the United States. In its murder rates the U.S. is an aberration among industrialized societies, so perhaps one cannot read too much into this. Yet it does seem that Japanese policeman have as fat budgets as Japanese prosecutors have lean. Do Japanese police officers make weaker arrests, just as U.S. prosecutors bring weaker cases? The analogy is unreliable. When the average prosecution has an 85 percent probability of conviction but the average arrest has only a 43 percent probability of prosecution, the marginal arrest has much less value than the marginal prosecution. On the other hand, with larger budgets Japanese police officers could potentially increase the quality of the arrest -- i.e., they could spend more effort making sure the arrest can generate a conviction. This in turn suggests that Japanese police are more likely both to arrest someone for a murder and to arrest the right person. If so, the low Japanese prosecution rate among those arrested is all the more telling. C. A Formal Model of Prosecutorial Incentives Our argument that an increased budget will reduce conviction rates does not apply under all circumstances, something formal modelling helps to show. Let us assume that the cost of any one case is trivial compared to the entire prosecutorial budget, so that we can model cases as points on a continuum. Potential cases will be indexed by ?, where ? varies from 0 to 1 in order of increasing ease of prosecution, so that the universe of possible cases has size 1. If the prosecutor decides to prosecute the easiest 1/3 of cases, he takes all the cases in the interval [2/3,1]. We assume that prosecutors maximize the number of convictions they obtain, and that all cases involve the same crime. As we discuss below (Section E.2.), even if, as seems likely, prosecutors also try to keep their conviction rates high, conviction rates can still decline with increasing prosecutorial budgets. Obviously, all cases do not involve the same crime, and in general prosecutors will face pressure to focus disproportionately on the more serious crimes. So long as they try to maximize convictions within each category of crime, however, conviction rates for each category will still decline with budget increases. We denote the prosecutor's budget by B, the fixed cost of each case prosecuted by F (which we will assume is the same for each case), and the variable cost by e. A case of a given type will have a probability of conviction of P(e,? ). Let us assume that dP/de ? 0, d2P/de2 ? 0, dP/d? ? 0 and d2P/d?2 ? 0. This says that the value of a case of a given type increases with prosecutorial effort, subject to diminishing returns, but it also allows for extra effort being unproductive (P’ = 0) or having neither increasing nor diminishing returns (P’’= 0). If two cases have equal expenditure, the easier case (with higher ?) has at least as great a chance of success, and perhaps a greater one, and also has at least as great a marginal benefit of effort. Figure 2 shows how P might vary with e and ? . [Insert Figure 2 about here.] We now come to the prosecutor's problem. He has two kinds of choices. First, he must decide which cases to prosecute. He will want to prosecute the easier cases first, which amounts to choosing a lower cutoff ?? for the interval of types [??,1] that he prosecutes. Second, he needs to pick the e(?) function which determines how much budget e he spends on each type ? of case. If the prosecutor chooses to equalize the marginal product of effort in two cases with ? = .8 and ? = 1, as in Figure 2, then he would choose e(.8) and e(1) to equalize the slopes of the P(e;?) functions, which means that e(1)> e(.8); i.e., he will spend more on the stronger case. Figure 3 shows one e(?) function he might pick and the two dimensions in which he can change it – the extensive margin of reducing the cutoff (horizontal) and the intensive margin of increasing spending on existing cases (vertical). [Insert Figure 3 about here.] The payoff function for the prosecutor is 1 (2) ? V(e(?);?)d?. ?? The budget constraint is 1 (3) ? [e(?)+ F]d? ? B. ?? The shaded area in Figure 3 represents the first part of this integral, the variable costs of prosecution. The prosecutor's solution needs two kinds of optimality, both simple enough to be described without further mathematics. First, the e(?) function must be chosen so that the marginal product of effort must be the same for each case prosecuted. This is the optimal intensive margin of prosecution. Second, the marginal product of effort for any case prosecuted must equal the average product of effort for the marginal case. This is the optimal extensive margin of prosecution. Three things affect whether increasing the budget leads to a rise or a fall in the average probability of conviction: the size of the fixed cost of prosecuting a case, the rate at which returns to effort diminish, and the variability in ease of prosecution dP/d?. First, size of the fixed cost matters. If it is big, then the extra budget would best go to existing cases, and the average conviction rate will rise. If it is very small, then the prosecutor will prosecute all cases -- another corner solution -- and an increase in the budget will again increase the average probability of success for sure. Thus, the average conviction rate increases with the budget if the fixed cost per case is either very large or very small. Second, the speed at which the returns to effort diminish matters. If the marginal return to effort does not change much as effort increases then as the budget increases, existing cases continue to provide almost as much marginal return to extra effort as they did before. Accordingly, most of the increased spending will go to those existing cases. The marginal case will not change much, but the effort on each existing case will increase, so the average conviction rate will go up. On the other hand, if there are diminishing returns to spending on existing cases, then if new budget dollars went to old cases, the marginal return to effort would fall sharply, to the point where it would be well below the average return to effort for new cases. Therefore, more of the new budget will go to new cases. The identity of the marginal case,?? , will decline and the average conviction rate will decline with it. Thus the average conviction rate will decline with extra budget if the diminishing returns are large. Third, it matters how rapidly the quality of cases declines, which is captured by the value of dP/d?. If dP/d? is large, then as the prosecutor selects better cases his probabilities of conviction rise dramatically. Correspondingly, dipping deeper into the case pool will reduce his success rate. This means that the prosecutor will be reluctant to add new cases as his budget increases, and will use more of the extra money to improve his chances in the old cases. Consider two extreme cases that illustrate these considerations. Situation Alpha. Let all cases be identical in their P(e, ?) functions, so dP/d? = 0, with the fixed cost low relative to the budget. All cases are taken. Additional budget leads to increased spending on existing cases, and higher average success. Situation Beta. Let all cases be identical in their P(e, ?) functions, so dP/d? = 0, with the fixed cost high. The marginal benefit of effort per case falls sharply, so P'' is very negative. Then, additional budget will lead to more cases, and a lower average success. E. The Argument Applied: 1. Implications for Japan. We believe Japan is more like Situation Beta: if a prosecutor were given a greater budget, he would tend to spend it on prosecuting more cases rather than in prosecuting his existing cases more intensely. First, as explained above, in Situation Alpha the prosecutor takes all cases. We know, however, that in Japan prosecutors take only a very small fraction of cases. Most criminal cases they simply drop, unprosecuted. Second, because prosecutors care about their reputations before judges, whether in Japan or in the U.S., they avoid irritating judges with sloppy work. As a result, the fixed cost F will be nontrivial, and prosecutors will not file cases unless they can invest enough resources to make a plausible showing of guilt. Third, in bench trial jurisdictions like Japan, the investments a prosecutor makes in a case should earn high returns initially (quickly raising the expected conviction rate from near zero to a comfortable probability), but run into sharply diminishing marginal returns thereafter. After investing enough effort in a case to make the basic showing, the U.S. prosecutor devotes much of the rest of his time to such tasks as voi dire in jury selection or showmanship before the jury. By contrast, Japanese prosecutors argue before professional judges. It is important for the prosecutor to present the facts and the law, but how he presents them -- a task with enormous potential for polishing and artistry -- matters less. That initial investment in presenting the facts, however, is all the more important, since before a jury a small investment in dramatic performance can sometimes cure a sloppy case with poor advance research. 2. Qualifications. What of the well-known Priest-Klein hypothesis that because of settlement and plea bargaining, litigated verdict rates do not reflect the distribution of liability or guilt in the underlying population of cases? The point is simply irrelevant to our discussion. The effect of prosecutorial poverty in Japan is to remove the credibility from the prosecutor’s threats to try hard cases. If the threat is not credible, then only the obviously guilty will plead guilty. Accordingly, the prosecutor’s small budget will cause both the plea-bargained and the contested cases to be disproportionately selected from defendants who are clearly guilty. The rest of the cases will be neither settled, nor tried, but simply dropped. Nor does it matter whether the average success rate at trial enters directly into the prosecutor’s utility function as a separate argument from the expected number of convictions. This is plausible. Particularly when running for office, American prosecutors do like to boast about the percentage of their cases that result in convictions. In a typical press release for a local prosecutor's race, for example, we read: Attorney Thomas Broderick Jr. will formally announce Tuesday that he is a candidate for the Democratic nomination for Madison County prosecutor. … He was chief deputy for Prosecutor William F. Lawler Jr. for more than seven years, claiming a 98 percent personal conviction rate on cases ranging from drugs to murder. That the conviction rate enters directly in a prosecutor’s utility functions, however, still will not eliminate the effects of budget increases that we have described. In allocating a budget of any size, a prosecutor would slant his allocation towards improving his percentage won, but if he were given an extra yen, he would split it between his two objectives of improving his percentage and improving his number of cases won. Increasing the number of cases would be a better use of the money under the extreme cases we have described above (e.g. P'' very negative), even if it lowered his conviction rate. An interesting implication is that criminal conviction rates should be much higher than civil plaintiff victory rates. The reason has nothing to do with jury unanimity or burden of proof. Such requirements will not affect the percentage of trials that end in convictions, since prosecutors will take them into account in deciding whether to pursue a case. Rather, criminal conviction rates are higher than civil plaintiff win rates because prosecutors who care about conviction rates-- elected U.S. prosecutors, more than appointed Japanese or U.S. ones-- have more reason than plaintiffs to avoid losing at trial. III. Empirics: A. Introduction: If prosecutors in Japan prosecute a higher percentage of guilty defendants than in the U.S., higher conviction rates will result under unbiased adjudication. In this section, we ask whether the Japanese judicial bureaucracy does reward unbiased accuracy, or instead rewards convictions. We first outline the general bureaucratic organization of the courts (Section B). We then explain our systematic tests, our data, and the variables involved (Section C), and turn to our regression results (Section D). We conclude by exploring the details of our data (Section E), and speculating about the political economy of the U.S.-Japan contrast (Section F). B. The Judicial Bureaucracy: 1. Typical careers. A would-be judge in Japan begins his legal education while still a college undergraduate. There, he majors in law. During his last year, he takes the entrance exam for the one national law school, the Legal Research & Training Institute (LRTI). If extraordinarily skilled or lucky, he passes on his first try. More likely, he takes it once a year until he passes four or five years later, or he despairs and chooses another vocation. Students stay at the LRTI for two years (recently shortened to 1.5 years). There, they attend lectures on legal practice and serve clinical assignments at various private and public offices. Toward the end of their stay, they select a career. Most will become lawyers in private practice. The rest will become judges or prosecutors. Formally, the Cabinet chooses the judges to appoint, but usually it defers to the Supreme Court Secretariat, the administrative office of the courts. When the new judges join the court, the Secretariat will name them to specific courts. The Secretariat itself is staffed by career judges, and a posting there at some point during a judge’s career signals recognized talent. By law, all judges serve a succession of 10-year terms. In practice, the Cabinet generally reappoints them until they retire in their early 60's. All judges face mandatory retirement at age 65. The Japanese government maintains eight high courts and a widely dispersed array of district and family courts. Virtually all of the judges (other than those in small claims court, some 2000) are part of the career judiciary. Some of the courts also maintain branch offices in smaller cities. Like most Japanese professionals, judges generally prefer to work in Tokyo or one of the larger metropolitan areas. Almost uniformly, they consider branch office assignments the worst posts in the system. The Secretariat regularly moves judges around the country and up and down the judicial hierarchy. Some cities are more attractive than others, and some responsibilities more prestigious. As a result, by controlling these periodic assignments (usually made at 3-year intervals), it can reward and punish judges. Also, although it cannot lower a judge's salary, it can vary the pace at which it promotes judges up the salary scale. For any given judge, the earliest signs of talent are graduation from a selective university and admission to the LRTI within a very few years of graduation (the latter indicates he passed the entrance examination on one of his first tries). Generally, the most talented judges begin at the Tokyo District Court. During their career, they will rotate out of Tokyo and serve some time in less desirable posts. In general, however, they will stay on a faster track than their less talented peers. As part of this faster track, they will spend more time in desirable locations, less time in branch offices, and more time with important administrative responsibilities. Only the Supreme Court stands apart from this system. The Cabinet and Prime Minister appoint Supreme Court justices to serve until mandatory retirement at age 70. The justices are not subject to 10-year terms or to reassignment by the Secretariat. Nominally, voters can remove them at any national election; in fact, voters never have. Supreme Court justices typically take the job in their early 60's. A plurality of them have served as career lower-court judges before their appointment. As in the U.S., appointment to the Supreme Court depends not just on talent but also on luck. The Supreme Court contains 15 justices, one of whom is the Chief Justice. The Chief Justice also serves as the administrative head of the entire lower-court system. In that capacity, he supervises the head of the Secretariat. Usually, at least one of the justices headed the Secretariat shortly before joining the Supreme Court. 2. Manipulation. In several contexts, the Secretariat has indeed manipulated job assignments to reward and punish judges either for their politics or for the complexion of the opinions they write. In the 1960s, for example, many young judges joined the Young Jurists' League -- a left-wing organization analogous to the National Lawyers Guild. Under pressure from the government (controlled by the conservative Liberal Democratic Party [LDP]), the Secretariat instructed those judges to resign their membership. Most did. Throughout their careers, however, those who had once been members received less attractive posts than their peers, and climbed the pay scale more slowly. Similarly, consider politically sensitive cases. Japanese election law bans door-to-door canvassing. Like most other restraints on electioneering, the ban helps incumbents and hurts challengers. Because the LDP had many incumbents where the Communist Party had few, the ban helped the LDP and hurt the Communists. When Communist candidates challenged it, some lower court judges struck it down. When they did, they suffered in their careers. Similarly, judges who held the Japanese military unconstitutional, judges who more readily enjoined the government in administrative cases, and judges who held the electoral apportionment schemes unconstitutional in the early years when the LDP relied heavily on the overrepresented agricultural vote -- all these judges suffered in their careers. C. The Data 1. The test. In exploring whether Japanese judges face biased incentives to convict, the gist of our test is simple. We took all 455 district court criminal case opinions published in 1976 or 1979 (our reference years), and identified the deciding judges. We chose years in the late 1970s because we need 10 years of post-decision judicial career data and at the time we collected our data our information on judicial careers ran only to 1991. We chose 1976 and 1979 rather than two adjacent years because doing so generated a larger sample (321 judges). Recall that the Secretariat generally reassigns judges every three years. Because judges sometimes specialize by subject matter during a given a 3-year assignment, by choosing years three years apart we could collect data on two almost entirely different judicial cohorts. We divided our judges into two groups according to whether they acquitted a defendant in one of the reference years. We further divided those who acquitted a defendant into those who did so as part of a three-judge panel and those who did so as a single judge. Concurrently, we also divided all judges who had a case appealed according to whether the appeal resulted in a reversal or an affirmance. For each of our judges, we measured the attractiveness of the jobs he held during the 10 years after the relevant reference year. Holding constant a variety of factors found pertinent in related research (primarily seniority, and several proxies for ideology, effort, and intelligence), we then tested whether an acquittal led to a worse job posting. The incentives are clear enough. Judges in Japan probably do not have high incomes as their first priority, since most could earn more in the private sector. Most probably do, however, value prestige, a job in a city where their children can obtain a good education, proximity to the center of intellectual life, and judicial influence. Indeed, if they care less about money than their peers, they presumably care more about these other things -- and a posting to a branch office hurts in all these ways. Even if judges did not respond to threats of poor postings, however, the Secretariat would still have reason to place troublesome judges in branch offices. It would post them to branch offices simply to remove them from positions of influence. Given the differential location of case filings, judges sent to branch offices would have have less opportunity to make important rulings. Given the smaller number of colleagues there, they would also have less chance to persuade other judges to share their views. 2. The data. We obtained records of published opinions from the Hanrei taikei database. The database is functionally equivalent to Lexis and Westlaw, and is available on 9 or 10 CD-ROMs, depending on the year. It covers nearly all the Japanese public and private case reporters. We obtained the records of judicial careers from the Zen saibankan keireki soran. The book, published in 1990, includes every judicial post held by every judge educated since World War II. We located membership in the Young Jurists' League (the leftist legal organization mentioned earlier) in Osorubeki saiban. A right-wing group published this book to attack what it saw as the increasingly leftist tendencies in the courts. In it, the group included the 1969 Young Jurists' League membership roster, which it took from the Young Jurists' League's own newsletter. 3. The variables. We use the sources just discussed to construct the following variables. They fall into three general categories: job quality variables; control variables indirectly measuring effort, intelligence, or ideology (or some combination of the three); and variables related to published opinions. Summary statistics appear in Table 1. [Insert Table 1 about here.] a. Job quality. Good Jobs Before: The percentage of time during the 10 years preceding the reference year, in which a judge had a high status appointment: primarily, the time he held sokatsu responsibilities (a modestly prestigious position), worked in the Secretariat or the Ministry of Justice, or served as a Chief Judge. Good Jobs After: Analogous to Good Jobs Before, but for the 10 years after the reference year. Bad Jobs Before: The percentage of time during the 10 years preceding the reference year, in which a judge served in a branch office (we exclude appointments to the relatively desireable Hachioji branch office in suburban Tokyo; we include appointments to summary courts) -- a low status appointment. Bad Jobs After: Analogous to Bad Jobs Before, but for the 10 years after the reference year. As explained below, we use Good Jobs After and Bad Jobs After as our dependent variables. Because the Secretariat tends to track judges onto relatively faster and slower tracks, in general Good Jobs Before should be positively correlated with Good Jobs After, and Bad Jobs Before with Bad Jobs After. b. Control variables. Seniority: The number of years since the judge graduated from the LRTI. Exams Failed: The estimated number of years (using the judge's birth year) between a judge's graduation from university and his entrance to the LRTI. Because the LRTI had an entrance exam with a pass rate between 1 and 4 percent during this period, most people passed it only after failing several times first. As a result, Exams Failed approximates the number of times the judge failed the LRTI entrance exam, and will correlate inversely with demonstrated ability. Tokyo or Kyoto University?: 1 if a judge graduated from either of the two most prestigious universities, the University of Tokyo and the University of Kyoto, 0 otherwise. Tokyo Start?: 1 if a judge’s first assignment was in a Tokyo court, 0 otherwise. In general, the Secretariat identifies the most promising young judges at the time it hires them, and for their first job assigns them to the Tokyo District Court. Opinions/Year: The number of published opinions solely or jointly authored by a judge during the ten years preceding the reference year, divided by the number of years the judge was on the bench during that time (we exclude years when a judge held an administrative post that precluded writing opinions). This is an indicator of how hard a judge worked and how newsworthy the judicial reporters (some of which are official government publications, and some of which are private) considered his opinions to be. The combination of a maximum value of 14.1 and a mean of 1.3 indicates the presence of a few highly productive outliers, something that appears not just in other samples of Japanese judges but among U.S. judges as well. Young Jurists' League: Membership in the Young Jurists' League in 1969: 1 if a member, 0 otherwise. Each of these variables will plausibly influence the careers of the judges in our sample. According to our 1997 study, Exams Failed and Young Jurists' League are negatively correlated with career success; Seniority, Tokyo Start?, Opinions/Year and Tokyo or Kyoto University? are positively correlated with career success. c. Published opinions. Any Acquittals?: 1 if a judge participated in an opinion acquitting a defendant (or accepting part of a defendant's argument), 0 otherwise. Solo Acquittal?: The same as Any Acquittals?, but excluding acquittals by 3-judge panels. Acquittal Reversed?: 1 if a judge participated in an opinion acquitting a defendant (as defined above) that was reversed by a High Court (or where the High Court increased the penalty on appeal), 0 otherwise. Acquittal Affirmed?: 1 if a judge participated in an opinion acquitting a defendant that was affirmed by a High Court, 0 otherwise. Conviction Reversed?: 1 if a judge participated in an opinion convicting a defendant (or rejecting all of the defendant’s claims) that was reversed by a High Court (or where the High Court reduced the penalty on appeal), 0 otherwise. Conviction Affirmed?: 1 if a judge participated in an opinion convicting a defendant that was affirmed by a High Court and has values of zero for Acquittal Reversed?, Acquittal Affirmed?, and Conviction Reversed?, 0 otherwise. [Insert Table 2 about here.] D. The Results: 1. Introduction. In Table 2, we report the results of two sets of regressions. Throughout, we use tobit, as this enables us to correct for the way our dependent variables are censored by 0 and 1. The results displayed in columns 2.1 and 2.2 of Table 2 use Any Acquittals? as the key explanatory variable. The results displayed in columns 2.3 and 2.4 use Solo Acquittal? as the key explanatory variable. Within each set, the first column shows the regressions with Good Jobs After as the dependent variable, and the second with Bad Jobs After. Within each column, we first show the coefficients, and display the absolute value of the t-statistics below them in parentheses. 2. Job quality. The job quality variables show that the Secretariat does seem to track judges into faster and slower career trajectories. Good Jobs Before is positively correlated with Good Jobs After, though it does not quite reach statistical significance, and Bad Jobs Before is significantly correlated with Bad Jobs After. Those judges who spend more time with prestigious administrative responsibilities in the 1970s spent more time with such responsibilities in the 1980s. Those who spent more time in branch offices in the 1970s spent still more time in branch offices in the 1980s. 3. Control variables. The variables proxying for effort, intelligence, and ideology also have most of the predicted effects. First, Seniority is positively and significantly correlated with Good Jobs After, and negatively and significantly correlated with Bad Jobs After. The more senior the judge, the more time he spends with administrative responsibilities and the less time in branch offices. Second, the fewer the times a judge failed the LRTI entrance exam, the more time he spent with administrative responsibilities. Exams Failed is negatively and significantly correlated with Good Jobs After. Third, judges who began their careers at the Tokyo District court spent less time in branch offices in the 1980s. Tokyo Start? is negatively and significantly correlated with Bad Jobs After. The Secretariat begins its career tracking at the very outset, in other words, and those judges it identifies as most promising at the start continue to receive the best jobs several decades later. Last, leftist judges suffered. Young Jurists' League is positively and significantly correlated with Bad Jobs After. Those judges affiliated with the left-wing Young Jurists' League in 1969 were still spending more time in branch offices in the 1980s. 4. Opinions. a. Acquittals in General. To test whether Japanese judges face biased career incentives in criminal cases, consider Regressions 2.1 and 2.2. Here, we test whether a judge involved in any published opinion acquitting a defendant incurred a career penalty over the next decade. The results are inconclusive. The coefficient on Any Acquittals? is positive in the Bad Jobs After regression. With a t-statistic of 1.42, however, it is not significant at the 10 percent level. In the Good Jobs After regressions, the coefficient is not even in the predicted direction. b. Solo Acquittals. Importantly, the data set includes convictions and acquittals authored both by 3-judge panels (107 cases) and single judge panels (317). This loosely mirrors the way courts decide criminal cases more generally. Of all cases (published and unpublished) decided in 1994, 3-judge panels decided 7.6 percent (3,751 cases), and single judges decided 92.4 percent (45,529). By law, 3-judge panels must decide the most serious criminal cases. Single judges generally decide the rest. Although the assigning mechanism depends on the local court, commonly a minor incoming criminal case will go directly to a single judge, who will then decide it himself. If it is unusually hard, he will route it to a 3-judge panel. In general, if prosecutors seldom prosecuted obviously innocent defendants (as suggested in Part II), if judges were fair and unbiased, and if single judges handled the most straightforward cases themselves -– if all this were true, then cases decided by single judges would show lower acquittal rates than cases handled by three-judge panels. The data are consistent with this hypothesis. Among our published opinions, 33 percent of the panel opinions were acquittals, but only 7 percent of the single-judge opinions. Of all 1994 cases published and unpublished, 1.2 percent of all panel opinions were acquittals, but only 0.09 percent of all single-judge opinions. Yet, one should not be so sanguine. Consider Regressions 2.3 and 2.4, in which we test whether a single judge who acquits a defendant incurs a career penalty. According to Regression 2.4, a judge who acquits alone does indeed suffer. With a t-statistic of 2.27, the result is significant at better than the 5 percent level. And consider the size of the punishment. Because the dependent variable is censored by 0 and 1, we use tobit rather than OLS regressions here, and interpreting the value of tobit coefficients requires considerable mathematical manipulation. The conclusions, however, are stark. A median judge (with Solo Acquittal? = 0) can expect to spend .23 of the next decade in branch offices; an otherwise median judge with Solo Acquittal? = 1 can expect to spend .37 of the next decade there. All else equal, in other words, an otherwise median judge who, sitting alone, acquits a defendant will spend an average of 1.4 extra years in branch offices. The puzzle is what to make of this. On the one hand, the difference between the not-quite- significant results for Any Acquittals? and the strongly significant ones for Solo Acquittal? could reflect the fog involved in collegial panels. Many of the acquitting panels might have been split. Because Japanese lower courts never publish dissents (though the Supreme Court does), we would not know which of the judges favored acquittal and which opposed. Because the judges themselves knew, however, the internal personnel reports might well reflect that information. If so, then the Secretariat might well punish only the two judges who favored acquittal. Accordingly, perhaps the contrast between the two sets of regressions simply results from the fog in the Any Acquittals? variable. Alternatively, the difference between Regressions 2.2 and 2.4 might reflect a policy mandating that judges forward hard cases to a multi-judge panel. If so, then perhaps the Secretariat had no intent to punish judges who acquit after group deliberation. Instead, it wanted to punish those who flouted court policy and refused to forward ambiguous cases to three-judge panels. Perhaps, in short, the Secretariat did not punish judges for acquitting; it punished them for ignoring case-routing policy. [Insert Table 3 about here.] c. Reversals. The Solo Acquittal? results thus leave a puzzle: whether the career penalties paid by judges who acquit on single-judge panels reflect (i) penalties for acquitting or (ii) penalties for not forwarding hard cases to collegial panels. Because even a Secretariat with a pro-conviction bias might see some acquittals as proper and not want to reward all convictions, we explore the careers of judges who find their cases reversed. If the Secretariat punishes only for case-routing violations rather than acquittals, then judges who find acquittals reversed should do no worse than those who find convictions reversed – holding constant the effect of Solo Acquittal? To explore these issues, we divide all appealed cases into four groups: acquittals reversed, acquittals affirmed, convictions reversed, and convictions affirmed. We then rerun the regressions with Good Jobs After and Bad Jobs After as the dependent variables. As the coefficients on the opinion variables were insignificant in the Good Jobs After regressions, we focus on the Bad Jobs After regressions. In Regression 3.1 we show the result a judge faces if he finds an acquittal overturned: over the next 10 years, he will spend substantially more time in branch offices. With a t-statistic of 2.52, the result is significant at the 1.2 percent level. By itself, Regression 3.1 does not necessarily show a pro-conviction bias. Instead, it could reflect a general Secretariat policy of punishing judges who make mistakes – and a reversal indicates that the trial judge was wrong. Accordingly, in Regression 3.2 we use all four appeal- related variables. The results show a stark bias: the Secretariat only punishes judges who find their acquittals reversed. Only the coefficient on Acquittal Reversed? is either large or significant. In terms of judicial bias, the relevant question is whether the coefficient on Acquittal Reversed? is greater than that on Conviction Reversed?. Indeed, it is -- .329 for Acquittal Reversed? compared to .002 for Conviction Reversed? It remains possible that the penalty paid by judges who find their acquittals reversed merely reflects the penalty paid by single-panel judges who (to continue the earlier argument) failed to follow the policy of forwarding hard cases to three-judge panels. Given that the correlation between Acquittal Reversed? and Solo Acquittal? is only .11, we were skeptical. Yet to test for that phenomenon directly, in Regression 3.3 we include both the four appeal-related variables and Solo Acquittal? The results in Regression 3.3 confirm the independent bias against those who write reversed acquittals. Both Acquittal Reversed? and Solo Acquittal? are independently significant. Holding constant the effect of Solo Acquittal?, judges who find their acquittals reversed still suffer larger penalties over the next decade. The median judge had no acquittals reversed or solo acquittals. The coefficients from Regression 3.3 imply (after calculations omitted here) that such a median judge could expect to spend .22 of the next decade in branch offices. For a judge with a three- judge panel acquittal reversed, the prediction rises to .42, an extra two years in branch offices over the decade. For a judge with a solo acquittal that was later appealed and affirmed, the prediction is .44. This is surpsing -- a judge is punished heaviliy for a solo acquittal even if his judgment is later confirmed by an appellate court. Rather less surprising, a judge who both acquits in a solo trial and is later reversed has a predicted percentage of time in branch offices of .59 -- over three-and-a-half years more than the median judge. E. The Acquittals in Detail: If, as Tables 2 and 3 imply, Japanese judges face skewed incentives to convict then we have no assurance that the high conviction rate reflects a high percentage of guilty defendants (row (a) in Figure 1). Instead, since judges would be rubber-stamping prosecutorial choices, prosecutors would have less incentive to screen cases, and Japanese judges may well be convicting innocent defendants regularly (row (d) in Figure 1). Unfortunately, we have very few acquittals, and the cases ending in acquittals could well raise other characteristics that the Secretariat sought to punish. To explore this potential omitted variable problem, we examined the Solo Acquittal? and Acquittal Reversed? cases in more detail. More specifically, we took all judges with Solo Acquittal? = 1 or Acquittal Reversed? = 1, and compared (i) the time they actually spent in branch offices in the decade after the opinion (observed Bad Jobs After) with (ii) the time they would have spent (given their other personal characteristics) had they not acquitted (predicted Bad Jobs After). If the latter (predicted Bad Jobs After) exceeds the former (observed Bad Jobs After), then by definition the judge suffered no punishment. Tables 4 and 5 summarize our results, and confirm our concerns. The Secretariat, they imply, punishes acquitting judges primarily when they acquit in politically sensitive cases or misconstrue statutes in ways that raise questions of judicial ability. Of the judges punished in Table 4, Hirayu and Ogawa acquitted defendants for violating the statutory ban on door-to-door canvassing. As noted in Section III.C., these were highly visible politically charged cases that regularly resulted in judicial punishments. Fujita acquitted leftists for beating a police officer during a riot on the grounds that the police were making illegal arrests. Torai let off a violent mobster by interpreting broadly the scope of an earlier summary proceeding that generated a trivial fine. Most of the other punished judges acquitted defendants on grounds of statutory interpretation: what constitutes mental capacity, can consensual sex be statutory rape, has the statute of limitations run, or when does negligence rise to manslaughter. In none of the cases where judges’ careers seemed to suffer did a judge rule: “the defendant didn’t do it.” [Insert Tables 4 and 5 about here.] Table 5 suggests much the same phenomenon. Again, one judge (Hirayu) held the door-to- door canvassing ban unconstitutional. Koike, Yamazaki and Noguchi acquitted defendants of crimes involving labor unrest. Given the role that socialists and communists played in Japanese labor unions (especially public sector unions), union cases can be very political and sometimes did result in punishments imposed on pro-labor judges. Ikeda and Omasa invoked the exclusionary rule -- which the Japanese Supreme Court has long viewed skeptically. Only two other punished judges remain, and both acquitted on grounds that turned on the interpretation of the statute of limitations. Here too, none of the judges acquitted on the grounds that the goverment prosecuted the wrong man. More basically, the point is this: despite the economically and statistically significant coefficients in Tables 2 and 3, the regressions provide no evidence that judges who second-guess whether the police and prosecutor found the right defendant suffer. Instead, the only evidence of punishment appears in non-routine cases -- cases where the judges arguably interpreted statutes wrongly or favored the leftist opposition. Given the extensive evidence elsewhere both that judges who promote leftist causes do worse than the mainstream, and that more able judges do better than the less talented, we find it hard to believe that Tables 2 or 3 support our biased-incentives hypothesis in run-of-the-mill criminal cases. F. The Political Economy of Convictions: One remaining puzzle is why Japan and the U.S. use such different criminal justice mechanisms. In both countries, politicians face competitive elections. Yet Japanese politicians hire so few prosecutors that they leave a majority of people arrested for murder unprosecuted, whereas U.S. politicians hire so many that they prosecute even many cases they cannot win. The Japanese system is the easier to explain. Assume, as seems plausible, that citizens in both countries want politicians to minimize crime subject to a budget constraint. By this metric, Japanese politicians have done extraordinarily well. On the one hand, they have kept violent crime low. Murders run 1.6 per 100,000 population, compared to 10.1 in the U.S. The Japanese rate is lower than the U.S. rate even correcting for the fraction of young males in the population and excluding murders in the urban ghettos. On the other, they have kept down law enforcement costs too. Not only do they use fewer prosecutors, they use fewer police. As noted earlier, in the mid-1990s, Japan had 18 police officers per 10,000 population, where the U.S. had 24 per 10,000. So the low budgets of Japanese prosecutors seem quite appropriate; the system is doing a good job of satisfying the voters' demand for crime control. The U.S. system presents the harder puzzle. Why do U.S. politicians win elections even though crime rates stay high and prosecutors waste resources bringing losing cases? In part, perhaps the answer lies in a constraint politicians cannot correct: the constitutional jury trial right. This right presents two consequences relevant here. First, it raises the cost of deterrence. As novices, juries tend not to realize what all professionals know: that most defendants prosecuted in modern democracies are guilty. Probably, they are guilty of the crime charged but, if not, then at least of some similar crime. Indeed, for just that reason U.S. criminal lawyers routinely demand jury trials. And precisely because convictions are harder to obtain from juries, deterrence in the U.S. will always come at a higher price. High prosecution budgets in the United States cannot keep crime low, but reducing the budgets would let crime increase even further. Second, the jury trial right gives prosecutors a private incentive to try more cases than socially optimal. As explained in Section I.C., the jury system increases the variance in the parties’ estimates of the expected litigated verdict. In the process, it increases the fraction of civil disputes that go to trial, and that increased probability of trial raises the returns to trial expertise. Those increased returns, in turn, present prosecutors with an incentive to use their job to obtain the trial experience they cannot find in the large modern law firms dominated by massive cases. The jury trial right creates, in other words, the agency slack that induces U.S. prosecutors to litigate more cases than U.S. citizens would like. The agency slack is costly, but it is a constitutionalized cost politicians can do little to reduce. The large increases in crime in the U.S. from 1960 to 2000 are clearly relevant to the politics of all this (though too large a topic to explore here). If we take it as given that the U.S. has a higher propensity towards crime(for whatever reason), then vote-maximizing politicians might reasonably decide to spend more on prosecution. Because crime rates are high enough to create severe social disruption, they may even reasonably decide to spend more on prosecutions per crime. By contrast, because crime reduction is probably subject to diminishing returns and increasing cost, Japanese politicians may conclude that cutting crime further from the already very low level is not worth the expense. The puzzle remains, however, of why the Japanese have so many police (almost the same number per capita as the U.S., as discussed in Section II.B.2) when they have so few prosecutors and so little crime. IV. Conclusions Japanese courts convict. Courts convict in America too, of course, but in Japan they convict with a vengeance: over 99 percent of the time. Even in cases where the defendant contests his guilt, they convict over 98 percent of the time. Are courts convicting the guilty and innocent alike, or are prosecutors merely choosing the guiltiest defendants to try? Absent independent evidence of the guilt of the accused, one cannot directly tell. In this article, we pursue indirect evidence on point. First, Japanese prosecutors are woefully understaffed. Tied as they are to a severe budget constraint, one might expect them to try only the most obviously guilty. Unbiased courts would then convict. The conviction rate would approach 100 percent, but only because most of the defendants were guilty. To return to our heuristic in Figure 1, Japan would be at the best-case row (a). Are Japanese courts unbiased? Initially, Japanese judges seem to face significantly skewed incentives: judges who acquit seem more likely to suffer a career penalty than those who convict. Yet a closer look at the judges punished for their acquittals suggests a classic omitted-variable problem -- and returns us to our hypothesis about prosecutorial resources. The acquittals that generate apparent punishment are sometimes cases where judges sided with opposition parties in politically charged cases, and otherwise cases where the judge may simply have interpreted the law wrongly. Never are they cases where the judge decided that the prosecutors simply brought charges against the wrong man. Instead, we know from other studies that the Japanese courts generally reward political reliability and intelligence -- and the observed punishment seems simply to reflect that broader phenomenon. Table 1: Sample Statistics Minimum Median Mean Maximum Good Jobs Before 0 0 .078 .970 Good Jobs After 0 .27 .341 1 Bad Jobs Before 0 .20 .204 1 Bad Jobs After 0 .20 .234 1 Seniority 2 13 13.5 29 Exam Failures 0 3 4.22 17 Tokyo or Kyoto Univ.? 0 0 .321 1 Tokyo Start? 0 0 .112 1 Opinions/Yr 0 .85 1.34 14.11 YJL 0 0 .109 1 Any Acquittals? 0 0 .271 1 Solo Acquittal? 0 0 .053 1 Acquittal Reversed? 0 0 .034 1 Acquittal Affirmed? 0 0 .040 1 Conviction Reversed? 0 0 .050 1 Conviction Affirmed? 0 0 .150 1 n: 321 Table 2: The Effect of Acquittals on Judicial Careers (2.1) (2.2) (2.3) (2.4) LHS: Good Jobs After Bad Jobs After Good Jobs After Bad Jobs After Intercept -.501*** .342*** -.487*** .354*** (5.47) (5.24) (5.39) (5.57) Good Jobs Before .229 ----------------------------- .224 ------------------------------ (1.50) ----------------------------- (1.46) ------------------------------ Bad Jobs Before ----------------------------- .329*** ------------------------------- .305*** ----------------------------- (2.91) ------------------------------- (2.72) Seniority .053*** -.023*** .053*** -.023*** (11.74) (6.85) (11.72) (6.92) Exam Failures -.018* .005 -.017** .004 (1.93) (0.62) (1.85) (0.59) Tokyo or Kyoto Univ.? -.002 -.063 -.010 -.050 (0.04) (1.18) (0.16) (0.96) Tokyo Start? .138 -.151* .071 -.168** (1.50) (1.90) (1.61) (2.12) Opinions/Yr? .007 -.006 .007 -.004 (0.38) (0.39) (0.36) (0.23) YJL -.042 .144** -.011 .142** (0.12) (2.00) (0.13) (1.99) Any Acquittals? .023 .081 ------------------------------ ------------------------------ (0.66) (1.55) ------------------------------ ------------------------------ Solo Acquittal? ---------------------------- ---------------------------- -.108 .224** ---------------------------- ---------------------------- (0.82) (2.27) Pseudo R2 .31 .16 .31 .17 Stan. Error .44 .37 .44 .37 # Censored (y<0, y>1) (132,28) (134, 5) (132, 28) (134, 5) Notes: Coefficients, followed by the absolute value of the t-statistic in parenthesis on row below. ***, **, *-- significant at more than the 1 percent level, 5 percent, and 10 percent levels. The regression uses tobit and there are 321 observations. Table 3: The Effect of Appeals on Judicial Careers (3.1) (3.2) (3.3) LHS: Bad Jobs After Bad Jobs After Bad Jobs After Intercept .358*** (5.64) .340*** (5.21) .330*** (5.11) Bad Jobs Before .333*** (2.98) .336*** (3.01) .312** (2.82) Seniority -.024*** (7.14) -.024*** (7.16) -.023*** (7.08) Exam Failures .006 (0.74) .005 (0.70) .004 (0.58) Tokyo Kyoto Univ.? -.064 (1.21) -.072 (1.35) -.072 (1.38) Tokyo Start? -.159** (2.01) -.161** (2.03) -.182** (2.31) Opinions/Yr -.005 (0.28) -.003 (0.16) .001 (0.05) YJL .153** (2.13) .157** (2.19) .155** (2.19) Acquittal Rev’d? .306** (2.52) .329*** (2.69) .301** (2.48) Acquittal Aff’d? ------- ------- .108 (0.96) .108 (0.98) Conviction Rev’d ------- ------- .002 (0.02) .014 (0.13) Conviction Aff’d ------- ------- .088 (1.40) .102 (1.63) Solo Acquittal? ------- ------- ------- ------- .215** (2.18) Stan.Error .37 .37 .36 Pseudo R2 .17 .17 .19 Notes: Coefficient, followed by the absolute value of the t-statistic. ***, **, *-- significant at more than the 1 percent level, 5 percent, and 10 percent levels. The regressions use tobit and there are 321 observations. In each regression, 134 observations are censored because of the bound of 0 and 5 because of the bound of 1 on the left-hand-side variable. Table 4: Judges with Solo Acquittals (1) (2) Judge Observed Predicted Bad Jobs After if the judge Name Bad Jobs After had not had a solo acquittal Kunio Ogawa 1.00 .33 Politician acquitted of violating campaign limits, on grounds that limits are unconstitutional. Kazumichi Hirai .90 .30 Defendant acquitted, on grounds of mental capacity. Yoshito Morita .87 .40 Defendant construction supervisor acquitted of manslaughter charge for negligently leaving construction site, where negligence not shown. Masato Hirayu .83 .27 Politician acquitted of violating campaign limits, on grounds that limits are unconstitutional. Yasuhiro Igaki .80 .39 Prosecution dismissed, on basis of interpretation of statute of limitations. Narishige Futakami .70 .26 Allegedly largely rehabilitated juvenile ordered released. Kenji Fujita .60 .27 Rioters acquitted of battery of riot police, on grounds that police were acting illegally in making arrests. Shigeo Yamamori .60 .28 Defendant acquitted of injuring a minor, where defendant had consensual sex with 16-year-old. Osamu Okushi .43 .33 Juvenile released from correctional facility, where reliable guardian petitioned for custody. Yasuo Torai .40 .26 Indictment of violent yakuza member dismissed, where series of violent batteries (perhaps mob discipline) were covered by summary proceeding on related charge leading to 100,000 yen ($330) fine. Satoshi Kataoka .04 .02 Partial acquittal on grounds that crime B is lesser included offense to crime A. Kenichi Hiruma .03 .17 Administrative traffic fine of juvenile vacated. Niro Shimada .00 .04 Prosecution for fraud in commercial loan application dismissed, on grounds of inadequately specific indictment. Kazunobu Araya .00 .06 Real estate broker acquitted of corporate tax fraud on factual grounds. Tetsuo Hirai .00 .28 Construction firm acquitted of tax fraud on factual grounds. Ryujiro Sugiura .00 .13 Upon petition of suspect, freer access to attorney ordered. Yasushi Sato .00 .27 Juvenile who confessed to burglary sent to juvenile detention instead of criminal system. Masaru Miyamoto .00 .28 Prosecution dismissed in tax case (facts not clear). Note: Column (1) shows the proportion of bad jobs the judge received in the decade after his decision. Column (2) shows the prediction of the OLS equivalent to Regression (2.4) if the judge had no solo acquittals. Table 5: Judges with Reversed Acquittals (1) (2) Judge Observed Predicted Bad Jobs After in the absence Name Bad Jobs After of Reversals and Solo Aquittals Masato Hirayu .83 .00 Politician acquitted of violating campaign limits, on grounds that limits are unconstitutional. (See also Table 4.) Yasuhiro Igaki .80 .07 Prosecution dismissed, on basis of interpretation of statute of limitations. (See also Table 4.) Yokichi Koike .57 .13 Labor union activists acquitted of destroying property by posting signs on company building; held, property not damaged. Katsuyuki Ikeda .48 .10 a. Defendant acquitted partially, based on exclusion of evidence obtained through unreliable confession. b. Defendant convicted of customs duty violations on primary counts; supplementary trial on additional counts held unconstitutional as cruel punishment. Ushio Yamazaki .40 .00 See Koike case, supra. Kizo Noguchi .37 .00 See Koike case, supra. Seichi Omasa .35 .00 See Ikeda cases a. and b., supra. Yasuhiro Morioka .30 .02 Fraud prosecution dismissed on statutes of limitations grounds. Reiji Noma .00 .00 See Morioka case, supra. Kiyoshi Inoue .00 .00 See Ikeda cases a. and b., supra. Tetsuo Hirai .00 .00 Defendant acquitted, on grounds of mental capacity. (See also Table 4.) Note: Column (1) shows the proportion of bad jobs the judge received in the decade after his decision. Column (2) shows the prediction of the OLS equivalent to Regression (3.3) if the judge had no acquittals that were solo or reversed. Negative numbers (possible because OLS is used here instead of tobit) are replaced by 0. Figure 1 – Prosecutorial Screening and Judicial Review Prosecutorial Judicial Conviction Innocents Screening Review Rates Convicted (a) High Yes Higher Fewer (b) High No Higher Fewer (c) Low Yes Lower Fewer (d) Low No Higher More Figures 2 and 3 are in separate files, japcon2.doc and japcon3.doc. J. Mark Ramseyer & Eric B. Rasmusen, Judicial Independence in a Civil Law Regime: The Evidence from Japan, 13 J. Law, Econ. & Org 259 (1997); J. Mark Ramseyer & Eric B. Rasmusen, Why the Japanese Taxpayer Always Loses, 72 S. Cal. L. Rev. 571 (1999); J. Mark Ramseyer & Eric B. Rasmusen, Why are Japanese Courts so Conservative in Politically Charged Cases, Discussion Paper # 268 (November 1999, Havard Law School John M. Olin Center for Law, Economics & Business). In a majority of Japanese murder cases, the charges against the suspect are dropped, as we discuss in Section II.B.2., below. For Japanese figures, see Homusho, Hanzei hakusho [White Paper on Crime] 122 (Tokyo Okura sho, 1996). The Saiko saibansho jimusokyoku, Shiho tokei nempo, Heisei 6-nen [Annual Report of Judicial Statistics, 1994] (Tokyo: Hosokai, 1994) gives slightly different figures. For U.S. federal courts, see Administrative Office of the U.S. Courts, Judicial Business of the U.S. Courts tab. D-4 (Washington, D.C.: Administrative Office of the U.S. Courts, 1995) (murder includes 1st and 2d degree; figures -- 10/94-9/95). For state courts, see U.S. Dept. of Justice, Bulletin – Prosecutors in State Courts, 1994 (1996, NCJ-151656). Daniel H. Foote, The Benevolent Paternalism of Japanese Criminal Justice, 80 Cal. L. Rev. 317, 352 (1992). The parallel between Japanese confessions and U.S. plea bargains is made explicitly in David Ted Johnson, The Japanese Way of Justice: Prosecuting Crime in Japan ch. 7 (Ph.D. Dissertation, Jurisprudence & Social Policy, University of California, Berkeley, 1996). For an analogous debate in the German context, see John Langbein, Land Without Plea Bargaining: How the Germans Do It, 78 Mich. L. Rev. 204, 214-15 (1979). Where lack of enforceability matters is in plea bargains in which the defendant wishes to promise cooperation as a witness in other cases. In such bargains, the cooperation sometimes comes after the prosecutor has recommended a sentence, and U.S. prosecutors, keenly aware of the potential for defendant breach, use all the formal legal tools at their disposal to prevent it. See, for example, Eric Rasmusen, Mezzanatto and the Economics of Self Incrimination, 19 Cardozo L. Rev. 1541 (1998). Lack of enforceable plea agreements, like low budgets, is an obstacle in the path of the Japanese prosecutor. This approach was pioneered by Frank Easterbrook, Criminal Justice as a Market System, 12 J. Leg. Stud. 289 (1983). The Japanese constitution guarantees counsel to indigent defendants, but about half the defendants use privately retained lawyers. See J. Mark Ramseyer & Minoru Nakazato, Japanese Law: An Economic Approach 172 (Chicago: University of Chicago Press, 1999). See generally J.M. Ramseyer & M. Nakazato, supra note 6, at 168-75. See Saiko saibansho, supra note 2, at vol. 2, tab. 31-4 (excluding traffic crimes); Administrative Office, supra note 2, at tab. D-4; U.S. Dept. of Justice, Bulletin – Felony Sentences in State Courts (1997, NCJ 163391). The earliest discussions of the model include William M. Landes, An Economic Analysis of the Courts, 14 J. L. & Econ. 61 (1971) and Richard A. Posner, An Economic Approach to Legal Procedure and Judicial Administration, 2 J. Legal Stud. 399 (1973). See J. Mark Ramseyer & Minoru Nakazato, The Rational Litigant: Settlement Amounts and Verdict Rates in Japan, 18 J. Legal Stud. 263 (1989). See Ramseyer & Nakazato, supra note 10. Saiko saibansho, supra note 2, at tabs. 31-4, 36-3; Administrative Office, supra note 2, at tab. D-4. If we ignore dismissals-- which might be part of negotiating ploys by prosecutors-- defendants were acquitted in 1,095 of the 4,765 trials in the U.S. , a conviction rate of 77.0%. Somewhat analogous calculations appear in Johnson, supra note 3, at ch. 7. A point implied by federal surveys placing both the aggregate conviction rate and the plea bargaining rate in the high 80 percent range. See U.S. Dept. of Justice, supra note 2 and note 8. Hypothetically, of course, courts could also have a pro-acquittal bias. Given that no one has suggested such a phenomenon in Japan, we ignore the possibility here. Though one can overstate the point. Prosecutors ultimately answer to politicians, and politicians who do not keep their prosecutors in line will -- all else equal -- do worse at the ballot box. There are 207 such commissions in Japan, serving for 6-month terms. Between 1949 and 1989 they held 77,000 hearings for 22 million non-indictments, a rate of .3 percent. Mark West, Note: Prosecution Review Commissions: Japan's Answer to the Problem of Prosecutorial Discretion, 92 Columbia L. Rev. 684, 697-98 (1992). The commissions recommended indictment in only about 7 percent of cases they heard, and only about 20 percent of these resulted in changed decisions by prosecutors. West, supra, at 702. The commissions are similar to German “motions for judicial decisions.” Hans-Heinrich Jescheck, The Discretionary Powers of the Prosecuting Attorney in West Germany, 18 Am. J. Comp. L. 508 (1970). The U.S. has only a weak equivalent, the statutes that in a few states allow private persons to challenge prosecutorial inaction. Stuart P. Green, Note, Private Challenges to Prosecutorial Inaction: A Model Declaratory Judgment Statute, 97 Yale L.J. 488 (1988). Perhaps Japan, with its lower prosecution rates, has more need of a formal check on prosecutorial discretion. See also Marcia E. Goodman, The Exercise and Control of Prosecutorial Discretion in Japan, 5 UCLA Pac. Basin L.J. 16 (1986). This is not accidental. Any justice system in which prosecutors have the resources to prosecute every possible case is grossly overfunded. It is more efficient to prosecute only a fraction of cases, and usually only the strongest cases. If a criminal cannot determine in advance whether he will be one of those prosecuted, such a system preserves deterrence and reduces costs. This is one corollary of the large-punishment-with-low-probability theory of Gary S. Becker, Crime and Punishment: An Economic Approach, 76 J. Pol. Econ. 169-217 (1968). There is, of course, a tradeoff between ease of prosecution and seriousness of the offense, as explored more fully in Richard A. Posner, Economic Analysis of Law, 5th ed. Sec. 22.3 (New York: Aspen, 1998). The prosecutor who just wants to win will not prosecute murders (which elicit heavy spending on defense) or fraud (which has complicated elements of proof). Even prosecutors who focus on the most serious classes of crimes, however, will tend to start with the easiest cases in each class and move to harder ones as budget and time permit. Quoted in Cynthia Mayer, Japan: Behind the Myth of Japanese Justice, Am. Law. July/Aug. 1984, at 113, 119. Johnson, supra note 3, at 499. See generally R. Posner, supra note 18, at Sec. 22.3; Richard A. Posner, The Problems of Jurisprudence 216 (Cambridge: Harvard University Press, 1990). Take the special independent prosecutors in the U.S. such as Lawrence Walsh and Kenneth Starr. With enormous budgets, they indict people who may be guilty but are far less obviously guilty than the typical defendant tried on burglary or auto theft charges. The calculations that follow rely on the following sources. On the U.S., see U.S. Dept. of Commerce, ed., Statistical Abstract of the United States tabs. 323, 333 (Washington, D.C.: U.S. Department of Commerce, 1996); U.S. Dept. of Justice, Bulletin – Prosecutors in State Courts, 1994 (1996, NCJ – 151656). On Japan, see Yasuo Watanabe, Setsuo Miyazawa, Shigeo Kisa, Shosaburo Yoshino & Tetsu Sato, Tekisuto bukku gendai shiho [Textbook: The Modern Judiciary] 110 (Tokyo: Nihon hyoron sha, 3d ed., 1997); Homusho, ed., Hanzai hakusho [White Paper on Crime] 402-03, 411-13, 430-31 (Tokyo: Okura sho, 1996). The resource shortage in Japan is rightly noted in John Owen Haley, Authority Without Power: Law and the Japanese Paradox ch. 6 (New York: Oxford University Press, 1991). Johnson claims that Japanese prosecutors are not over-worked, but primarily because he calculates indictments per prosecutor rather than arrests per prosecutor. See Johnson, supra note 3, at 84. The clearance rate discussion is based on: Homusho, supra note 22, at 432; U.S. Dept. of Commerce, supra note 22, at tabs. 316, 323; U.S. Dept. of Justice, supra note 2 and supra note 6. Similar calculations yield prosecution/arrest ratios in the U.S. of: drug trafficking - .6; robbery - .45; burglary - .44; and aggravated assault - .16. Note that the lower prosecution budget in Japan is not an indication that either the United States or Japan is behaving suboptimally. Lower prosecution rates have the cost of less crime deterrence, but the benefits of lower cost and fewer innocent people prosecuted. We discuss this issue more fully in Section III.F below Japanese prosecutors can choose to “suspend prosecution” in selected cases, by which they hope to signal to the public (how credibly being another question) that they believe the suspect is guilty even though they decided to forego prosecution. Daniel H. Foote, Prosecutorial Discretion in Japan: A Response, 5 UCLA Pac. Basin L.J. 96 (1986); West, supra note 14, at 693. In about two-thirds of the murder cases dropped, according to the official Ministry of Justice white paper, prosecutors simply lacked adequate suspicion against the person arrested. That is just another way of saying that the prosecutors tried the strongest cases first and ran out of time before they got to the rest. Homusho, supra note 22, at 260 (1995 data). Keisatsu cho, Keisatsu hakusho [Police White Paper] 325 (Tokyo: Okura sho, 1996) (1996 data for Japan); Bureau of the Census, Statistical Abstract of the United States tab. 334 (Washington, D.C.: Bureau of the Census, 1996) (1993 data for the U.S.). Surprisingly little work has been done on the economic analysis of prosecutor behavior. One exception is the recent working paper, Edward L. Glaeser, Daniel P. Kessler, Anne Morrison Piehl, What Do Prosecutors Maximize? An Analysis of Drug Offenders and Concurrent Jurisdiction, NBER Working Paper No. W6602 (June 1998), which focuses on the interaction between choices of federal and state prosecutors. More generally, we could replace the probability of conviction P in the model with a value V that incorporates the probability of a conviction or a plea bargain, the length of the sentence, the importance of punishing the particular type of crime, and the chance of mistakenly prosecuting someone innocent. For present purposes, we are most interested in conviction rates; our model can be considered as a model of the conviction rate for a case with given sentence length, importance, and probability of mistake. These assumptions are plausible, but not invariably true. It might be, for example, that the prosecutor can secure a high probability of conviction in a certain case with very little effort, but that further effort is useless, whereas a different case requires more effort in general but has higher marginal product of effort too. Our assumptions rule out that kind of situation, in which the term “easier case” becomes highly ambiguous. For a mathematical exposition, see J. Mark Ramseyer & Eric B. Rasmusen, Why Is the Japanese Conviction Rate so High, Discussion Paper # 240 (Oct. 1998, Harvard Law School John M. Olin Center for Law, Economics & Business). George L. Priest & Benjamin Klein, The Selection of Disputes for Litigation, 13 J. Legal Stud. 1 (1984). Attorney says he'll be candidate for prosecutor, The Indianapolis Star, January 17, 1998, at N02. Fuller discussions of the material in this summary, together with citation to the relevant statutes when appropriate, appear in Ramseyer & Rasmusen, Judicial, supra note 1; J. Mark Ramseyer & Frances McCall Rosenbluth, Japan’s Political Marketplace chs. 8-9 (Cambridge: Harvard University Press, 1993). These accounts are detailed in Ramseyer & Rasmusen, Judicial, supra note 1, at 283-5. For other anecdotes, see Ramseyer & Rosenbluth, supra note 34, at chs. 8-9; J. Mark Ramseyer, The Puzzling (In)dependence of Courts: A Comparative Approach, 23 J. Legal Stud. 721 (1994). For a different interpretation, see John O. Haley, Judicial Independence in Japan Revisited, 25 Law in Japan 1 (1995). See Ramseyer & Rasmusen, Judicial, supra note 1, at 278; Ramseyer & Rasmusen, Political, supra note 1. See Ramseyer & Rasmusen, Judicial, supra note 1, at 283-85; Ramseyer & Rasmusen, Political, supra note 1. We dropped those cases that did not give the name of the deciding judge (14 convictions, 6 acquittals), those cases whose outcome was hard to code (11 cases), and those judges who did not appear in our career records (they were probably prosecutors seconded to the courts – an occasional practice). In addition, we dropped those judges who had been appointed less than a year before the reference year, and those who quit less than two years after the reference year (about 20 judges). For the 2 judges for whom we lacked the information on Exams Failed we used the sample-wide mean. We also examined the high court opinions in 1976 and 1979 and traced back the judges who wrote the district court opinions. Perhaps because of the wider variation in years written, however, the results are more haphazard and inconclusive. We thus limit ourselves to the more uniform sample of district court opinions written in 1976 or 1979. Daiichi hoki shuppan, K.K., ed., Hanrei taikei CD-ROM [A Systematic Case Law: CD-ROM] (Tokyo: Daiichi hoki, biannually updated). Nihon minshu horitsuka kyokai, ed., Zen saibankan keireki soran: kaitei shinban [Biographical Information on All Judges: New Revised Edition] (Tokyo Konin sha, 1990). The third edition has since (in 1998) been published. Shiho undo kenkyujo, ed., Osorubeki saiban [Fearsome Trials] (Tokyo: Zenbo sha, 1969). Compare this spread, e.g., with Ramseyer & Rasmusen, Taxpayer, supra note 1, at 581 (mean of 2.88 and maximum of 20.6), at 581 (mean 2.91, max 29.02); Ramseyer & Rasmusen, Judicial, supra note 1, at 264 (mean 2.02, max 10.42); William M. Landes, Lawrence Lessig & Michael E. Solimine, Judicial Influence: A Citation Analysis of Federal Courts of Appeals Judges, 27 J. Legal Stud. 271, 278-79 (1998) (for federal appellate judges, mean of 28 signed opinions per year and maximum of 82--- noteworthy, if not as extreme as in Japan). Ramseyer & Rasmusen, Judicial, supra note 1; Ramseyer & Rasmusen, Political, supra note 1; Ramseyer & Rasmusen, Taxpayer, supra note 1. Japanese courts do not consider this double jeopardy. We have used the conventional two-sided t-tests throughout this article, even though arguments could be made for using one-sided tests in many places (which would be more likely to show coefficients significantly different from zero). Saiko saibansho, supra note 2, at tab. 29. Saibansho ho [Courts Act], Law no. 59 of 1947, § 26. Saiko saibansho, supra note 2, tab. 29. The published acquittal rates are higher both (i) because acquittals are generally more newsworthy to commercial case reporters, and (ii) because the unpublished convictions mostly involve uncontested proceedings. We give the details of these calculations in Ramseyer & Rasmusen, supra note 31. Note that the Secretariat will always hold strong priors that a prosecuted suspect is guilty, quite rationally. This follows from the prosecutor's incentive to economize on his resources by only prosecuting guilty suspects. This idea is explored in Eric B. Rasmusen, Predictable and Unpredictable Error in Tort Awards: The Effect of Plaintiff Self Selection and Signalling, 15 Int'l Rev. L. Econ. 323 (1995). For an empirical study finding evidence that Japanese judges were punished for taking non-LDP positions in labor cases, see Shiro Kashimura, Rodo jiken ni okeru saibankan no keireki to hanketsu [The Opinions and Careers of Judges in Labor Cases], 41 Kobe hogaku zasshi 325 (1991). See J.M. Ramseyer & M. Nakazato, supra note 6, at 168-72. Citizens also want minimal error, of course. We have no reason to think the Japanese system has greater error than the U.S., though that is of course partly the subject of this paper. For a summary of the literature suggesting a high accuracy rate in the Japanese judiciary, see Johnson, supra note 3, at ch. 1. J.M. Ramseyer & M. Nakazato, supra note 6, at 175-78 (1993 data). It is also noteworthy that murder rates in Japan have actually declined since the 1950's. Id., at 176. For sources, see supra, footnote 27. For speculation about why the violent crime rate in Japan is so low, see J.M. Ramseyer & M. Nakazato, supra note 6, at ch. 7. Not all, obviously. After all, how likely the defendants are to be guilty is exactly the issue we try to explore indirectly in this article. The implication here is that, holding constant the level of guilt within a given pool of defendants, the conviction rate for the pool will be higher in Japan than in the U.S. Restated, part -- though not all -- of the answer to the question posed in the title to this article is simply the absence of jury trials in Japan. On why the returns to private practice are low in Japan despite the small number of lawyers (essentially, the ease with which one can circumvent the bar’s monopoly on legal practice), see J.M.Ramseyer & M. Nakazato, supra note 6, at ch. 1. Japanese Conviction Rates: Page 33