[author]Lin Changqing

[content]

Learning or Networking? The Causal Effect of Judges Sitting by Designation

*Author Lin Changqing

Professor, Department of Economics, National Cheng Kung University, Taiwan

*Author Zhang Yongjian

Clarke Chair Professor at Cornell Law School

Abstract: When lower court judges and higher court judges jointly form a collegial panel to hear cases, the former can learn from the latter through this process, or establish a personal relationship with the latter, or both. Lemley and Miller's empirical study on US federal courts found that judges who serve as junior college judges for one week, after returning to the lower court, have a significant decrease in the second instance revocation rate of patent infringement cases they subsequently hear. This study utilized a unique dataset covering nearly one million cases, particularly 2591 appeal cases, to investigate whether the short-term promotion system for judges in Taiwan would lead to a decrease in case revocation rates, and further analyzed the reasons for this. The author applied a double difference model and found that when the second instance appointed judge or presiding judge of an appeal case was a colleague of the original judge, the revocation rate would significantly decrease (reflecting the networking effect). In addition, after secondment, the overall revocation rate of appeal cases significantly decreased, which is consistent with the learning effect; However, this result was not obtained through strict causal identification design.

1.Introduction

In judicial practice, judges often intentionally or unintentionally favor the courts they have worked for. Previous studies have shown that judges of the United States Supreme Court tend to vote in favor of the rulings of the federal appellate circuit courts they have previously served in. And this favoritism may extend to (or perhaps stem from) judges' favoritism towards old colleagues. For example, Lemley and Miller found that in the US federal court system, when district court judges are appointed to participate in the United States Court of Appeals for the Federal Circuit (which has exclusive jurisdiction over patent cases and is one of the 13 appellate circuit courts of the US federal court) to hear cases, their revocation rates in patent infringement cases decrease when they return to the district court. It can be seen that judicial actions may be influenced by bias in different ways.

The changes in the second instance revocation rate (as well as other judicial judgments) mentioned above can be attributed to the "learning effect", "network effect", or a combination of both. The 'learning effect' refers to the practical experience gained by district court judges while working in the appellate court; The 'network effect' refers to the intentional or unintentional favoritism or informal respect shown by appellate court judges towards familiar regional court judges. Learning is one of the important purposes of the judge sizing by design system and is usually encouraged. However, establishing personal relationship networks poses normative issues as it violates the principles of the rule of law.

In Taiwan, there is a similar system called the "Transitional Promotion" system, commonly known as the "Three Specialized Students" system, where judges in the district court typically work in the appellate court for three years before returning to the district court. By using the DiD model, the author analyzes whether the system reduces the case revocation rate ("reversal" in Taiwan refers to the "annulment of the original judgment" in the second instance - see Articles 450 and 451 of the Civil Procedure Law of Taiwan, because most of the readers of this official account come from the mainland, and this concept is roughly consistent with the "annulment of the original judgment" in the mainland, so here we translate "reversal" into "annulment" - translator's note), and further identify whether the change in the revocation rate can be explained by the network effect and learning effect.

2. Institutional background

2.1 The judicial system in Taiwan

There are three levels of courts in Taiwan responsible for adjudicating civil cases: 22 regional courts, 6 high courts (appellate courts), and the Supreme Court of Taiwan. The appellate court may revoke the judgment of the lower court based on legal or factual issues. The courts involved in this study include 11 regional courts and 5 high courts.

The ideal recognition environment needs to meet four conditions. Firstly, cases in regional courts are randomly assigned to judges. Secondly, cases in the appellate court are randomly assigned to judges. Thirdly, the collegiate bench of the appellate court is randomly formed or formed according to strict predetermined rules. Fourthly, appeal cases are randomly assigned to a certain appellate court or allocated according to strict predetermined rules. The judicial system in Taiwan meets these four conditions:

2.1.1 Randomly allocate all cases

In Taiwan, all cases are randomly assigned to judges. When a case is submitted to a regional court, the allocation of the case depends on the amount claimed by the plaintiff and the nature of the dispute. If the plaintiff's claim amount exceeds 6 million New Taiwan Dollars (approximately 200000 US dollars), it will be allocated to the case pool with high litigation amount, otherwise it will be allocated to the case pool with ordinary litigation amount. From 1999 to 2019, 85% of district court cases were assigned to one of these two case pools. Similarly, during the aforementioned period, 63% of the cases in the appellate court were allocated to three types of case pools based on the amount of the litigation subject matter: relatively low amount (25%), ordinary amount (26%), and high amount (12%).

Other cases (15% of cases in regional courts and 37% of cases in high courts) are assigned to specialized case libraries, covering areas such as international trade, labor law, maritime law, consumer law, election law, and medical accidents. Although some specialized cases are managed by specialized courts within the court and randomly assigned to judges within that court, it is more common for specialized cases to be randomly assigned to any judge in the court.

The author tested the random allocation situation using linear probability models (LPMs). The dependent variable includes one of the five case characteristics: whether the plaintiff is a natural person, whether the plaintiff is a government agency, whether the defendant is a natural person, whether the defendant is a government agency, and the disputed amount. The only independent variable is the fixed effect of judges (including only judges who have worked in specific years and courts). Through regression analysis of approximately 6000 cases per year from 22 regional courts and 6 appellate courts between 1999 and 2019, the results showed that although there were some cases of non random allocation, the majority of cases (over three-quarters of cases) were mostly randomly assigned to judges.

2.1.2 Composition of the collegiate bench

Each high court involved in this study contains multiple administrative divisions, typically consisting of three to four units. The most senior judge in the administrative court is usually appointed as the presiding judge by the administration, and the fourth judge in the four units is usually a junior college judge.

The appellate courts in Taiwan are always composed of a panel of three judges to hear cases, including assigned judges, presiding judges, and side judges. Among them, the appointed judge is determined by the algorithm of random case division, responsible for all preparation work of the case and writing the judgment. The most senior judge in the collegiate bench, usually the presiding judge, will serve as the presiding judge. Accompanying judges participate in panel discussions, but their role is relatively passive. In the collegiate bench of the appellate court, the role of a junior college judge is usually the appointed judge or accompanying judge, rather than the presiding judge.

In general, judges from the third vocational school usually only evaluate together with two or three judges in the same courtroom. Three specialist judges may rotate to other courts after one year, or permanent judges of the appellate court may rotate to the courts where three specialist judges are located after one year, providing regular opportunities for three specialist judges to work with other judges. In addition, illness or other unexpected events may also lead to the joint trial of several cases by the three specialized judges and judges from other courts. But in any case, these arrangements are determined according to a predetermined fixed sequence and have been announced. Because of this legal design, any pair of judges usually do not (or rarely) form a collegiate bench together (see Figure C and Figure D in Figure 1). In other words, the judges or panel of the appellate court cannot select cases based on personal preferences.

Finally, regarding the fourth condition, the appeal case is assigned to the appellate court according to strict statutory jurisdiction rules and cannot be manipulated by the third specialist judge or his/her former colleagues. The jurisdiction of the appellate court did not change during the research period.

2.2 Three year college judge system

In Taiwan, judges in regional courts can apply to participate in short-term promotion programs after working for about 15 years. This plan typically promotes district court judges to appellate court judges for three years, and then transfers these three specialist judges back to the district court. If there is a vacancy in the appellate court after returning to the district court for a few years, some of the junior college judges will be promoted to appellate court judges. The relevant regulations indicate that the secondment plan is based on performance selection. It can be inferred that only excellent judges will be selected.

3. Research Design and Data

3.1 DiD research design

The author's observation unit is appeal cases, identified through the random allocation of each appeal case (the original judgment was made by a three specialist judge working in the district court). The dataset of this study includes district court rulings made by three professional judges during the three-year period before and after their secondment.

The processing group of this study is assigned to cases of appellate court judges who have previously worked with or will work together with junior college judges in the appellate court panel. The control group includes cases assigned to appellate court judges who have not co tried cases with junior college judges during their tenure as junior college judges. The LPM model studied is as follows:

Among them, the dependent variable Dijd refers to whether the judgment initially made by the third vocational judge j in the district court d in the appeal case i has been overturned (=1) or upheld (=0) by the appeal court. Postijd is used to indicate whether the aforementioned case occurred before the judge was promoted to the appellate court (=0) or after the judge returned to the district court (=1).

The term 'Colleague' corresponds to a dummy variable 'Any Judge', or three dummy variables: 'presiding judge', 'assigned judge', and 'side judge', which refer to whether a colleague of a junior college judge is present in the collegiate bench of an appeal case (yes=1, no=0). The author conducted two sets of regression analyses. In the first group, if any of the three judges in the panel is a colleague of a junior college judge, then the value of 'Colleague' is 1. In the second group, three types of judges in the collegiate bench were considered separately.

In addition, future or former colleagues are defined as judges who have co tried at least Q cases with three professional judges, and this Q-value is referred to as the "cutoff" by the author. In the regression results table, 25, 50, 75, 100, and 125 cases were used as cutoff values, while in the graph, 150 different cutoff values were used, including 1-150 cases. The coefficient β 2 of Colleague is used to analyze the difference in revocation rates between appeal cases handled by future colleagues before secondment and those that will not become future colleagues.

Colleague x Post "involves one or three interaction items. In the first set of regression analysis, it is "Any Judge x Post". In the second set of regression analysis, they are respectively "presenting judge x Post", "assigned judge x Post", and "side judge x Post". Their coefficient β 3 is used to measure the overall impact of former colleague judges handling appeal cases on the revocation of first instance judgments in the appellate court. If the relationship with former colleagues is effective, then these coefficients should be negative and statistically significant.

X includes the following variables: whether the plaintiff is a natural person; Whether the defendant is a natural person; Is the case in the district court heard by a panel of three judges with three professional judges serving as presiding judges (Yes=1; No=0); Case category.

In addition, if the appeal case is assigned to a professional court (such as labor, election, medical accident, etc.), and the three professional judges have participated in at least five appeal hearings of the relevant professional court, then the variable "Specialized" takes a value of 1, otherwise it takes a value of 0. The interaction item 'Specialized x Post' aims to analyze the learning effect.

η j represents the fixed effect of three professional judges. η d represents the fixed effect of the regional court. η t represents the fixed effect of the year. The error term ε ijd is used to identify clustering effects in cases heard by the same junior college judge j.

3.2 Data

This study conducted an in-depth analysis of 62 judges who were promoted from district courts to appellate courts through a secondment program between January 1, 2002 and December 31, 2014. The author tracked all cases handled by these 62 junior college judges in the three years before promotion and within three years after returning to the district court. Since the dependent variable is whether the first instance judgment has been revoked, the author only studied cases that have been appealed to the appellate court and have already made a second instance judgment.

In the main regression analysis, 2591 appeal judgments related to 62 junior college judges were used. 31 junior college judges were excluded from the main analysis due to being assigned to a district court under a different jurisdiction than the appellate court they had worked for after their term as junior college judges (for example, the appellate court where junior college judges temporarily served was Court A, and they were later assigned to a district court under the jurisdiction of Court B - translator's note). In other words, since all the cases of these 31 junior college judges who have been seconded have not been handled by their former colleagues, they cannot be affected by the network effect, and therefore the DiD framework is not applicable. The author conducted separate chi square analysis and LPM regression on the changes in the revocation rates of these judges' cases to reveal the learning effect.

4. Research results

4.1 Network effect

Table 3 shows the effectiveness of former colleagues in handling appeal cases (if any judge in the three member panel meets the criteria of "colleague", then the appeal case is considered to be handled by the former colleague).

The results indicate that when an appeal case is handled by a colleague judge who was once a junior college judge, the likelihood of the first instance judgment being overturned is lower (compared to cases handled by non colleague judges).

In order to further analyze which judge in the appeal case panel has an impact on the revocation rate, the author distinguished the judges and conducted a second regression. Table 4 presents this result in detail:

The results indicate that when the presiding judge in the appeal panel is a former colleague of a junior college judge, the likelihood of a judgment made by the junior college judge being overturned is lower (these effects are significant at the 5% significance level across almost all cutoff values); When this former colleague was appointed as the appointed judge, the likelihood of the judgment of the junior college judge being overturned was also low (at most cut-off values, these effects were significant at the 10% significance level, and sometimes significant at the 5% level); But when this former colleague served as the accompanying judge, the revocation rate did not have a significant impact.

The above results cannot be explained by the learning effect, as in Taiwan, the majority of judges are professional judges (although in recent years, a few judges have been selected from practicing lawyers), who have received 1.5 to 2 years of training at the Judicial Academy. Due to the highly similar backgrounds and centralized training of these judges, it is unlikely that different legal principles will develop among the various panels of the appellate court. In fact, the judges interviewed by the author did not mention these differences. It should also be noted that the judges of the appellate court consider the Supreme Court of Taiwan as the highest authority. Therefore, any possible differences in the panel will be overwhelmed by the requirement to follow the legal principles and style of the Taiwan Supreme Court.

The author points out that the network effect is a more reasonable explanation for the above results, that is, it is related to personal relationships. Another explanation is that the three major student program has created a high-level hypothesis among former colleagues. After all, only district court judges who have demonstrated good reputation and academic performance can be selected to participate in the Three Professional Student Program. Therefore, the judges of the third vocational school are clearly outstanding among their peers. A junior college judge may leave a deep impression on colleagues in the appellate court during their term of office. When these appellate court judges subsequently review the judgments made by these former colleagues (junior college judges), they may compromise their colleagues' judgments. Although 'saving face for colleagues' may be a partial motivation for rejecting an appeal, assuming diligent former colleagues must have' thoughtful consideration 'may also prompt appellate court judges to' let go '. The existing research design cannot reveal which mechanism plays a dominant role, but larger scale data and causal inference designs yield results consistent with Lemley and Miller's (2015) study, demonstrating that the network effect in Taiwan's judicial system (and even any judicial system) should not be ignored.

The author's research results also indicate that the higher the familiarity between appellate court judges and junior college judges, the stronger the network effect. Appeals court judges who have repeatedly observed the handling of cases by junior college judges may tend to compromise their judgments in future appeal cases.

4.2 Learning effect

The regression model shows a significant learning effect. The coefficient of the Post variable is always negative, approximately -0.25, and is significant at all cutoff values (at a significance level of 10%), and significant at a significance level of 5% at many cutoff values.

However, a potential challenge in explaining the learning effect is that the decrease in revocation rates after secondment may not only be attributed to learning in the appellate court, but also to judges having more experience. Even if these three college judges are not promoted to appellate court judges and continue to stay in the district court, their revocation rate will still decrease because these judges have accumulated experience by handling more cases during this period.

To verify this hypothesis, the author calculated the annual average revocation rate of each judge in their Nth enforcement year. The author categorizes judges into three types of associate judges (A, B, and C in Figure 3) and all non associate judges (D in Figure 3), and reports a box plot of the annual average revocation rate. Figure 3 does not show a decreasing trend in the revocation rate as the judge's experience increases. This indicates that the negative value of the Post variable cannot be simply attributed to the natural decline in revocation rates during a judge's career.

The results indicate that when the presiding judge in the appeal panel is a former colleague of a junior college judge, the likelihood of a judgment made by the junior college judge being overturned is lower (these effects are significant at the 5% significance level across almost all cutoff values); When this former colleague was appointed as the appointed judge, the likelihood of the judgment of the junior college judge being overturned was also low (at most cut-off values, these effects were significant at the 10% significance level, and sometimes significant at the 5% level); But when this former colleague served as the accompanying judge, the revocation rate did not have a significant impact.

The above results cannot be explained by the learning effect, as in Taiwan, the majority of judges are professional judges (although in recent years, a few judges have been selected from practicing lawyers), who have received 1.5 to 2 years of training at the Judicial Academy. Due to the highly similar backgrounds and centralized training of these judges, it is unlikely that different legal principles will develop among the various panels of the appellate court. In fact, the judges interviewed by the author did not mention these differences. It should also be noted that the judges of the appellate court consider the Supreme Court of Taiwan as the highest authority. Therefore, any possible differences in the panel will be overwhelmed by the requirement to follow the legal principles and style of the Taiwan Supreme Court.

The author points out that the network effect is a more reasonable explanation for the above results, that is, it is related to personal relationships. Another explanation is that the three major student program has created a high-level hypothesis among former colleagues. After all, only district court judges who have demonstrated good reputation and academic performance can be selected to participate in the Three Professional Student Program. Therefore, the judges of the third vocational school are clearly outstanding among their peers. A junior college judge may leave a deep impression on colleagues in the appellate court during their term of office. When these appellate court judges subsequently review the judgments made by these former colleagues (junior college judges), they may compromise their colleagues' judgments. Although 'saving face for colleagues' may be a partial motivation for rejecting an appeal, assuming diligent former colleagues must have' thoughtful consideration 'may also prompt appellate court judges to' let go '. The existing research design cannot reveal which mechanism plays a dominant role, but larger scale data and causal inference designs yield results consistent with Lemley and Miller's (2015) study, demonstrating that the network effect in Taiwan's judicial system (and even any judicial system) should not be ignored.

The author's research results also indicate that the higher the familiarity between appellate court judges and junior college judges, the stronger the network effect. Appeals court judges who have repeatedly observed the handling of cases by junior college judges may tend to compromise their judgments in future appeal cases.

4.3 Learning effect

The regression model shows a significant learning effect. The coefficient of the Post variable is always negative, approximately -0.25, and is significant at all cutoff values (at a significance level of 10%), and significant at a significance level of 5% at many cutoff values.

However, a potential challenge in explaining the learning effect is that the decrease in revocation rates after secondment may not only be attributed to learning in the appellate court, but also to judges having more experience. Even if these three college judges are not promoted to appellate court judges and continue to stay in the district court, their revocation rate will still decrease because these judges have accumulated experience by handling more cases during this period.

To verify this hypothesis, the author calculated the annual average revocation rate of each judge in their Nth enforcement year. The author categorizes judges into three types of associate judges (A, B, and C in Figure 3) and all non associate judges (D in Figure 3), and reports a box plot of the annual average revocation rate. Figure 3 does not show a decreasing trend in the revocation rate as the judge's experience increases. This indicates that the negative value of the Post variable cannot be simply attributed to the natural decline in revocation rates during a judge's career.

To further explore the mechanism of learning effects, the author further examined cases in professional courts. Professional courts mainly handle cases in specific legal fields (which are randomly assigned among judges), as well as other types of cases. In order to enable more judges to accumulate professional knowledge in these legal fields, the various divisions of the appellate court often take turns handling professional cases year by year. Therefore, judges from vocational colleges may or may not have worked in professional courts. And even if they have worked in a professional court for a year, they may not have handled many related cases. However, some judges have indeed dealt with these cases, and the author investigated whether the revocation rate would decrease if these judges were assigned to specialized legal fields after returning to the district court.

The results indicate (Tables 3 and 4) that the key interaction term (Specialized x Post) did not reach a significant level statistically. However, the author points out that due to the small number of specialized cases and large standard errors, it is impossible to clearly confirm whether there is a learning effect.

5. Conclusion

The short-term promotion system for judges in Taiwan has produced a statistically significant and substantial network effect: the network effect has reduced the revocation rate of first instance judgments by approximately 8% to 21% (between the three years before and after the secondment). Regression analysis also indicates the presence of significant learning effects. At the same time, the author emphasizes that caution must be exercised when interpreting research results. More empirical research is needed to verify whether the network effect is widespread and whether it cannot be distinguished from potential learning effects.

The original article was published in The Journal of Legal Studies, Volume 54, Number 1, January 2025, and was reprinted from the WeChat public account "Empirical Methods Research ".