WPS4483 Policy ReseaRch WoRking PaPeR 4483 Enforceability of Labor Law: Evidence from a Labor Court in Mexico David S. Kaplan Joyce Sadka The World Bank Financial Private Sector Development Department Enterprise Analysis Unit January 2008 Policy ReseaRch WoRking PaPeR 4483 Abstract The authors analyze lawsuits involving publicly- to generate predictions on how lawsuit outcomes should appointed lawyers in a labor court in Mexico to study depend on the information available to the worker and how a rigid law is enforced. They show that, even after a on the worker's cost of collecting an award after trial, judge has awarded something to a worker alleging unjust both of which are determined in part by the worker's dismissal, the award goes uncollected 56 percent of the lawyer. Differences in outcomes across lawyers are time. Workers who are dismissed after working more consistent with the hypothesis that firms take advantage than seven years, however, do not leave these awards both of workers who are poorly informed and of workers uncollected because their legally-mandated severance who find it more costly to collect an award after winning payments are larger. A simple theoretical model is used at trial. This paper--a product of the Enterprise Analysis Unit, Financial Private Sector Development Department--is part of a larger effort in the Financial Private Sector Development VPU. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted atdkaplan@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Enforceability of Labor Law: Evidence from a Labor Court in Mexico David S. Kaplan Enterprise Analysis Unit The World Bank Joyce Sadka Centro de Investigación Económica Department of Economics Instituto Tecnológico Autónomo de México December 2007 We gratefully acknowledge helpful comments from Jennifer Reinganum, Simeon Djankov, Lior Ziv, and from seminar participants at ITAM, Macalester College, and the World Bank. MailStop F4P-400, 1818 H Street, NW, Washington, DC 20433. Email: dka- plan@worldbank.org. Camino a Sta. Teresa #930. México, D.F., C.P. 10700. Mexico. Email: jsadka@itam.mx. 1 Introduction There is little dispute that Mexican labor law is extremely protective of workers. Botero, et. al. (2004), for example, perform an international comparison of labor law in which Mexico figures as one of the countries with the most onerous labor regulation from the point of view of firms. An open question, however, is to what extent this extremely protective legislation is actually enforced. In this paper, we look inside the black box of enforcement and study how labor law is applied to individual lawsuits. Specifically, we analyze alleged unjust-dismissal lawsuits from a labor tribunal in Mexico and study the process through which these suits go to trial, reach an out-of-court settlement, or are dropped. Conditional on going to trial, we analyze both court rulings and whether or not the workers manage to collect what has been awarded to them. One institutional feature we document is that it can be very costly for a worker to collect money that has been awarded at trial by a judge. Consistent with this observation, we find that it is common for trial awards to go uncol- lected, particularly for cases in which the worker had not worked for long at the firm. In this sense, it can be said that the enforcement of labor law is lax for workers with low (but not trivially low) levels of tenure. We then develop a simple theoretic framework to develop testable hypothe- ses on how outcomes should differ depending on the accuracy of the worker's information and on the worker's costs of collecting an award after the judge has made a ruling. We show that workers with better information should drop fewer small-stakes cases and more high-stakes cases. We also show that workers with high costs of collecting awards settle fewer low-stakes cases and may settle more high-stakes cases. In any court case, the information available to the plaintiff and the costs of collecting a court award are determined jointly by the worker and her lawyer. Workers may differ in terms of their knowledge, memory, or capacity to provide proof about the facts of the case, while lawyers may differ in terms of know-how and experience in similar cases. Also, as will be clear later in the paper, the collection of a payment that has been awarded by a judge certainly requires both effort from the worker and from the lawyer. Hence our model can be interpreted as predicting the effects of heterogeneity across worker-lawyer teams in terms of information and collection costs, where the heterogeneity arises from both workers and lawyers. To test the empirical implications of this model across workers, we would need data on the same worker in a number of cases. This information is not available in our data, and is generally unavailable in litigation data sets. How- ever, we can test the empirical implications of the model across lawyers. We show that informational differences across lawyers affect lawsuit outcomes and that differences in the costs of collecting awards across lawyers affect lawsuit out- comes, and therefore argue that the same differences across workers should have similar effects on lawsuit outcomes. Additionally, to the extent that we show there are systematic differences across lawyers that affect lawsuit outcomes, if workers' access to legal services is also heterogeneous, differences across lawyers 1 may tend to accentuate the differences across workers between "nominal" and "real" protections afforded by the labor law. Our empirical methodology, in addition to exploiting the fact that we have multiple observations for a given lawyer, exploits the fact that the assignment of cases to public lawyers is essentially random. Assignment of cases to lawyers is based on a short questionnaire that contains only basic characteristics of the case such as the plaintiff's gender and tenure, which we can control for in the econometric models. We therefore argue that selection of cases to lawyers based on unobservables is quite unlikely. In fact, when we focus on the 19 public lawyers whom we observe at least once in a trial and at least once not in a trial, we do not even find evidence that selection of cases to lawyers is correlated with observables. This essentially random assignment of cases to public lawyers allows us to examine differences in outcomes across lawyers and attribute these differences to the lawyers themselves, not to the unobservable characteristics of these cases. The outline of the rest of the paper is as follows. In section 2, we review the papers that are most related to what we study. In section 3, we discuss in some detail the legal framework related to alleged unjust-dismissal lawsuits in Mexico. In section 4, we discuss the data we use and present evidence that a significant fraction of tried cases result in an award going uncollected. We also present in section 4 evidence supporting our argument that the assignment of cases to public lawyers is essentially random. In section 5, we present a simple model in which a worker anticipates the possibility that it will be too costly to collect what the judge awards. This possibility affects the entire bargaining process between the worker and the firm and therefore generates several testable implications. In section 6, we present the main empirical results of the paper and relate them to the theoretical model. In section 7, we reconcile all of our results with our model by arguing that there must be heterogeneity both in terms of the accuracy of information and in terms of collection costs. In section 8, we offer our final conclusions. 2 Related literature Our paper is related to some recent papers that analyze the effects of the de facto rather than the de jure regulatory environment on economic outcomes. Lerner and Schoar (2005), for example, find that private equity investments have higher valuations and returns in countries with good enforcement mecha- nisms. Almeida and Carneiro (2007), examine the effects of differential enforce- ment across municipalities of Brazilian national labor regulations and find that increased enforcement causes formal-sector employment and unemployment to rise and causes self employment to fall. Caballero, et. al. (2006) find that the negative effects of labor-market regulation are particularly strong in countries where the regulations are likely to be enforced. Dreher and Gassebner (2007) find that corruption, and the accompanying lack of enforcement, can help the 2 process of firm creation in highly-regulated economies. Unlike our paper, these studies do not examine in depth how regulations are enforced. Rather, they use proxies for enforcement and relate these variables to other outcomes of interest. There is also increasing interest in enforcement costs in the law and eco- nomics literature. For example, Lanjouw and Schankerman (2004) argue that enforcement costs are relevant in patent litigation, and more so for relatively small and infrequent claimants. Singer (1997) reviews situations in which con- sumer debt is discharged under U.S. bankruptcy code, so that the debt is never collected by the creditor. Goodwin (2005) discusses enforcement costs and the resulting widespread problem of collecting child support payments. It is impor- tant to stress that these papers, while documenting the existence of enforcement costs, do not analyze how they affect the final outcomes of lawsuits. We believe that an analysis of the effects of these enforcement costs on individual lawsuit outcomes is an innovative aspect of our paper. A few papers attempt to measure enforcement costs and their effects on the efficiency and perceived efficiency of the legal system. Djankov, et. al. (2003) construct an index of formalism for a large group of countries. Some of the measures they consider are exactly the type of post-trial collection costs that are the focus of our paper. They consider, for example, whether the notification of a court judgment requires the participation of a court officer. They also count the minimum number of procedural actions required to enforce a court's judgment. One of their main findings is that French style civil-law countries like Mexico have legal systems that are more formalistic on average than those with other legal systems. They also find that higher formalism, including costs of collection, leads to longer duration of disputes and lower quality of legal decision-making. Elena, et. al. (2004) describe in great detail the obstacles to enforcement of court judgments faced in Peru, which like Mexico inherited a French-style civil- law system. They document the fact that in Peru all notifications in a relatively simple lawsuit require formal summons, including direct participation of a court officer. In addition, when notification does not result in immediate payment of the debt, further procedures to force payment are highly bureaucratic and complicated. They present survey evidence that excessive enforcement costs, including delays and uncertainty in the enforcement of judgments, are cited by 30% of individuals as main reasons for not using the legal system to collect a debt. Also, only 44% of respondents believed the enforcement process would result in actual collection of a debt from a small or medium-sized firm. The results from both Djankov, et. al (2003) and from Elena, et. al. (2004) indicate that enforcement costs are often excessive, and that such costs affect the quality of the legal system and levels of confidence and use of the judicial process. However, they do not document how widespread unenforced judgments are in a specific area of law, nor do they analyze the effect of this lack of enforcement on both trial outcomes and pre-trial bargaining and settlement.1 1Elena, et. al. mention evidence from a previous study claiming that on average, three years after suits have been brought 77% of judgments are still unenforced. However, they 3 Since the focus of our paper will be the enforcement of judgments, it is useful to comment on how well judgments are enforced in Mexico compared to other countries. Using the methodology described in Djankov, et. al. (2003), the 2008 Doing Business rankings place Mexico 49th out of 178 countries ranked in terms of how quickly a contract can be enforced. This time is counted from the moment the plaintiff files the lawsuit in court until payment. In terms of time to enforce a judgment, however, Mexico's rank is 121. We therefore see that the Mexican judicial system seems particularly inefficient at enforcing judgments.2 One contribution of our paper will be to show how an overly formalistic judicial system results in poor enforcement in practice. Our paper is also related to several strands of the literature on litigation. The first of these strands is the theoretical and empirical work on litigation costs, which have typically focused on two aspects of these costs. One litigation cost that has been studied is the cost of going to court, including delay in the resolution of the conflict. This work generally shows that the costs of going to court affect the probability of settlement as well as the characteristics of cases that end up in court. This means that the selection of cases that go to trial, as well as the time it takes to reach a settlement, can differ across parties with different costs of going to court. Fenn and Rickman (1999), for example, estimate a structural model and find lower litigation costs imply longer delays in reaching a settlement. Eisenberg and Farber (1997) develop a model in which the distribution from which a plaintiff's litigation cost is drawn affects plaintiff win rates and affects time to settlement. They posit that individuals are more heterogeneous in terms of their litigation costs than are corporations. They then show that, consistent with their theoretic model, individuals have higher trial rates and lower win rates at trial. Another cost that has been studied is the cost of legal services, including the rules for shifting these costs between parties to a dispute. Many studies have compared the American rule in which each party pays its own legal costs with the English rule, in which the winning party is compensated for its legal costs by the losing party. For example, Gong and McAfee (2000) show that fee-shifting increases the stakes of going to trial and therefore benefits lawyers by increasing demand for legal services. Gross and Syverud (1991) find higher settlement rates when plaintiffs pay their own litigation costs. Our paper is also related to papers that study the effects of lawyers on lawsuit outcomes. This literature has most often used a principal-agent framework to analyze moral hazard problems between clients and lawyers. Rules governing the compensation of lawyers, such as the percentage of contingency fee charged, vary across jurisdictions and countries, and this has allowed for the testing of models that predict how the incentives of the lawyer will affect litigation mention that there is very little concrete evidence on how much actual enforcement takes place. 2The data on total time to enforce a contract are available from http://www.doingbusiness.org/ExploreTopics/EnforcingContracts/. The data on time to enforce a judgment, which is a component of the total time to enforce a contract, was provided to us by the Doing Business staff and are available upon request. 4 strategy and equilibria. In this area, Helland and Tabbarrock (2003) find that contingency fees increase the quality of litigation and reduce the average time to settlement. Watanabe (2007) structurally estimates an agency model using medical malpractice data and finds that a limitation on contingency fees would reduce welfare. A few articles have considered adverse-selection problems between clients and lawyers, that is, situations in which intrinsic differences across lawyers rather than incentives dominate the effects that lawyers have on lawsuit out- comes. Along these lines, Szmer, et. al. (2007) study lawyer effects in Cana- dian Supreme Court cases and find that more experienced lawyers obtain more favorable outcomes conditional on going to trial. Nevertheless, the empirical lit- erature testing such models has been limited by the selection effect arising from the fact that clients with good cases may be more likely to select good lawyers. The literature testing moral hazard models also suffers from this selection prob- lem since they assume that lawyers' effects on lawsuit outcomes are determined solely by incentives provided through the lawyers' compensation schemes, and not by differences in the lawyers themselves or by differences in the quality of their cases. Kaplan et. al. (2008) studied the determinants of success and case outcomes in the federal labor courts in Mexico. Among other results, it was found that controlling for all observables in a lawsuit, including what the worker claims, the suit appears more successful for the worker when it concludes in settlement. This evidence is consistent with an asymmetric-information bargaining framework in which the firm is the relatively more informed party. Our theoretical model will assume that the firm has better information, which implies that workers go to court when their cases are relatively weak. 3 Legal Framework As we mentioned earlier, Mexican labor law is highly protective of workers. The law regulates hours and working conditions, health risks, fringe benefits, and firing. In this paper we analyze firing lawsuits, so a discussion of the regu- lation of firing is in order. Under Mexican law, firing can either be considered justified or unjustified. In order for firing a worker to be justified under the law, the worker must have engaged in wrongful behavior such as deliberately destroying the firm's machinery or materials, physically attacking a supervisor, showing up to work under the influence of alcohol or drugs, or being absent from work repeatedly without justification. Remarkably, firing a worker for lack of productivity or laying off a worker during downturns is not considered to be justified.3 In order to fire a worker, a firm must notify the worker in writing, stating the cause for firing the worker. Given that firms must state one of the causes 3 The discussion of Mexican labor law in this section is based on the Ley Federal del Trabajo (LFT), Title II, Chapter IV, as well as on the Reglamento Interior de la Junta Federal de Conciliación y Arbitraje (Internal Regulations of the Federal Labor Board). 5 specified in the labor code, they often fabricate causes for firing a worker who is simply unproductive, and this often results in a lawsuit in which the worker claims the dismissal was not justified. When sued by a worker, the firm is considered to carry the burden of proof in relation to the cause of firing. Certain components of firing costs do not depend on whether the firing was justified or not. In particular, any worker who is fired is entitled to unpaid overtime and wages, fringe benefits up to the date of firing, as well as severance pay equivalent to 12 days' wage per year worked at the firm. This daily wage, however, is capped at two times the minimum wage. When the dismissal is unjustified under the law, however, firing costs include several additional elements. First, a worker fired without just cause can sue for reinstatement. The firm may only refuse to reinstate for certain categories of workers such as temporary workers, those with less than one year's tenure, and at-will (typically white-collar) employees. Second, in addition to the compensation due to a worker under any type of firing, an unjustly-dismissed worker receives two additional payments. She receives back pay including benefits from the date of firing to the date of payment of the court award. She also receives three months' wage with benefits per year worked at the firm, as well as an additional 20 days' salary per year worked at the firm if she is an at-will employee. Wages for these calculations are not capped at any level. We now describe the mechanisms through which labor law is enforced. In the first place, labor code in Mexico is federal, so that private employees in any state have access to the same legally-mandated protections. The labor courts are called Juntas de Conciliación y Arbitraje. They are administrative courts that belong to the executive branch of government at both the federal and state levels. Federal labor courts resolve disputes in a number of industries listed in the federal labor code. All other labor disputes fall under local jurisdiction, so all states have at least one local junta, and large states will often have several tribunals with jurisdiction defined by the geographical location of the dispute. These tribunals are intended to serve both mediation and adjudication func- tions. The law mandates that they hold at least one conciliation hearing before proceeding to a court judgment. If the conciliation hearing concludes without a settlement, another hearing similar to a trial is held. Evidence such as expert testimony, depositions, and other documents is submitted to the judge during this hearing. After the conclusion of this hearing, the judge produces a draft ruling on matters of fact as well as matters of law and submits it to the labor board, consisting of the judge, a lay magistrate who represents firms, and a lay magistrate who represents workers. In order for the proposed draft to become a valid ruling, at least one of the magistrates must vote along with the judge in favor of the decision. Finally a hearing is scheduled in which the court's decision is read publicly in the presence of the parties to the dispute. Should the firm send a legal representative to the hearing in which the court's decision is made public, then according to the law the firm has already been duly notified of the decision. However, firms often do not send a representative to the hearing, and in this case, the firm must be notified by a court clerk. 6 In practice, in order for this notification to be carried out in a timely fashion, the plaintiff must participate in the process by making a motion to request immediate notification, as well as accompanying or having her lawyer accompany the court employee to the firm's place of business. This notification often takes some time, and firms, especially smaller ones, may do their best to avoid being notified properly. Once due notification has taken place, the firm has 72 hours to send payment to the tribunal. If the firm does not pay within 72 hours, another hearing must be scheduled in which the judge should order a court actuary to appraise the firm's assets, seize a sufficient number of assets to pay the judgment the firm owes, and proceed to a sale of these assets, after which the court pays the judgment amount to the worker directly.4 This process is akin to putting the firm through bankruptcy and therefore can be very costly, especially because the firm may block proper notification, move its place of business, or hide its assets. The court's order of an appraisal and sale of assets should be part of the same original lawsuit file from which we extract our data, however we find very few such orders. Discussions with both public and private lawyers have led us to believe that once firms have been duly notified, they generally do pay the award amount. At any point before the court's decision is announced, parties may resolve their dispute by settlement. However, unlike many other areas of law in Mexico and elsewhere, the labor courts must both approve and record settlements. Unratified settlements are not legally binding, so that parties to a dispute will generally prefer to have their settlements ratified by the court. Hence, our data from lawsuits include detailed information about settlements. Apart from the protections in the federal labor code, the federal government and the states provide workers under their jurisdictions with free legal repre- sentation through public agencies generally called Procuradurías de la Defensa del Trabajador. The public prosecutors who work for these agencies are licensed lawyers or interns in their fourth year of law school. Public lawyers are not al- lowed to receive any compensation from their clients, who are assigned to them by the agency. They are paid a salary by the agency, which does not depend, at least not explicitly, on their performance. For methodological reasons that will be explained later, these public lawyers will be the focus of our empirical work. 4 Data and Preliminary Statistics We have assembled a data set comprised of all lawsuits filed in the Junta Local de Conciliación y Arbitraje del Estado de México - Valle de Cuautitlán, during 2000 and 2001.5 This tribunal is located in an industrial area towards the northern part of the Mexico City metropolitan area. Overall 718 cases were initiated in 2000 and 1,850 cases were initiated in 2001. Cases involving public lawyers, 4This procedure is governed by Title 15 of the LFT, Articles 939-975. 5These data were obtained by the authors using a new law governing freedom of govern- mental information in Mexico. 7 which will be the focus of this paper, account for 174 cases initiated in 2000 and 491 cases initiated in 2001. There were many more lawsuits filed in 2001 because of the dramatic decline of the maquiladora sector, which represents a large fraction of cases filed in this tribunal. For all lawsuits, we observe the motive for filing, which is typically the allegation of an unjust dismissal, as well as the date of filing. From the initial filing made by the worker's lawyer, we observe a description of the job held, the dates the worker started and stopped working for the firm, the salary with and without fringe benefits, hours per week, the worker's demands, gender, and date of birth. In firing law suits, workers generally demand reinstatement, back-pay, overtime, fringe benefits, and severance pay. In terms of the lawsuits' outcomes, we observe three modes of termination: dropped suits, settlements, and trials leading to a judgment by the court. We record the date of conclusion of the procedure and the payment received by the worker under a settlement or a court judgment. For trials, we observe a trial result stated by the court. This result classifies the decision as being in favor of the firm, in favor of the worker, or mixed in the sense that the court only concedes part of the worker's claim. We also observe the votes of the judge and the magistrates representing labor and management in favor of or against the judgment, and the facts of the case as recognized by the judge, including any payments that the firm previously made to the worker. Often a court ruling will result in constitutional appeals by one or both parties, and in these cases, we record the number of constitutional appeals, who files the appeals, and we extract data on the first and last court ruling. We now present some descriptive statistics from the data set. Table 1 presents summary statistics for lawsuits in our sample separately for lawsuits involving private lawyers, lawsuits involving the 49 public lawyers observed in the data at least once, and for lawsuits involving the 19 lawyers who we observe going to trial at least once and not going to trial (dropping or settling) at least once. The main difference we see between lawsuits involving public and private lawyers is that final payoffs are substantially bigger in cases involving private lawyers. We also see that private lawyers tend to go to trial more often. Some of our empirical models will be identified by lawyers for whom we observe both lawsuits that go to trial and lawsuits that do not go to trial. Restricting the data set to these lawyers essentially removes interns (those who have not yet completed their law degrees) from the data set. We see from table 1 that this restriction does not substantially affect the descriptive statistics. The 30 lawyers eliminated by this restriction account for only 85 observations. Perhaps the most important feature we see from table 1 is that, both for cases involving private lawyers and cases involving public lawyers, it is quite common for positive awards at trial to go uncollected. In the case of private lawyers we see that, of 202 lawsuits in which a positive amount was awarded at trial, this amount was left uncollected 123 times. Similarly in the case of public lawyers we see that of the 45 lawsuits in which a positive amount was awarded at trial, the award was left uncollected 25 times. It is important to note that these are not judgments that were overturned on appeal. As far as 8 the court knows, the worker simply decided no to (or was unable to) undertake the procedures necessary for the collection of the award. The main reason for focusing on public lawyers is that we believe the assign- ment of lawsuits to these lawyers was not based on unobservable characteristics. Court personnel assured us that case assignment was based on a short question- naire that contained only basic information such as tenure and gender which we observe. In fact, we were told that tenure of the worker was the most important factor in determining the assignment of cases to workers. This essentially random assignment of cases to lawyers will allow us to at- tribute differences in lawsuit outcomes to the lawyers themselves. If we find evidence that differences across lawyers in terms of their information and in terms of their collection costs are important determinants of lawsuit outcomes, it will seem natural to conjecture that these same differences across workers have similar effects. We attempt to verify this view of the assignment process in table 2. We estimate linear models with lawyer fixed effects for two characteristics of the case: a female worker dummy and years of tenure. Table 2 presents the results of the F-tests of the null hypothesis that there is no heterogeneity across lawyers. The results for private lawyers are quite strong; both gender and years of tenure are strongly correlated (at the 0.01 level) with the lawyer fixed effects. That is, case assignment is far from random. When we use all public lawyers, we see that years of tenure is strongly correlated (at the 0.01 level) with the lawyer fixed effects, but gender does not appear to be correlated with these lawyer fixed effects. These results are consistent the assertions of court personnel that tenure was the main variable used to assign cases to lawyers. When we restrict our analysis to the 19 public lawyers for which we observe at least one case that went to trial and at least one that did not, we no longer see any evidence of non-random assignment. That is, neither gender nor years of tenure appear to be correlated with the lawyer fixed effects. We believe that the results from table 2 are encouraging for our analysis. The assignment of lawsuits to lawyers could not have been based on things like the strength of the worker's claim because there would be know way to read such information from the short questionnaire filled out by the plaintiffs. When we restrict our analysis to the 19 lawyers for whom we observe both at least one lawsuit that goes to trial and at least one that does not, we do not even observe a significant correlation between the observable characteristics and the lawyer fixed effects. These 19 lawyers can be viewed as the basic staff of the court. We now turn to the issue of whether different lawyers indeed appear to act differently. In table 3 we investigate whether there are significant differences across lawyers in their probabilities of a lawsuit ending by being dropped, by being settled, or going to trial. We estimate random-effects logit models with no independent variables in which the dependent variable is one of the three possible modes of termination. We present the chi-bar-square statistics of the test of the null hypothesis that all lawyers have equal probabilities that the case will be dropped, settled, or go to trial. Looking first at the models for private lawyers, we reject the null hypothesis 9 at the 0.01 level for all three termination modes. One may suspect, however, that these results are strongly affected both by differences in observable and unobservable characteristics of the cases across lawyers. When we use all public lawyers, we reject the null hypothesis that lawyers have the same probabilities of dropping and settling their cases at the 0.01 level. We only reject the null hypothesis that all public lawyers have the same probabilities of going to trial at the 0.10 level. Using only the 19 public lawyers with one trial and one non- trial outcome, we again reject the null hypothesis that lawyers have the same probabilities of dropping and settling their cases at the 0.01 level and find no significant differences in their probabilities of going to trial. We will exploit the fact that we find strong differences in settling and dropping probabilities in the subsequent analyses. Since cases in which lawyers do not collect a positive trial award will be a key focus of our analysis, we want to explore these cases a bit more. The cases in which a positive award is left uncollected do not appear to be of trivial stakes. In the case of private lawyers, a judge awarded a positive amount to the worker in 202 cases. In the 123 cases in which the positive award was left uncollected, average years tenure was 3.76 and the median was 2.46. The analogous figures for the 79 cases in which a positive award was collected are 3.43 and 1.59. Surprisingly the cases in which a positive award is not collected appear if anything to be higher stakes cases, although the non-random assignment of cases to private lawyers makes these comparisons suspect. When we analyze the data for public lawyers, we see that the judge awarded a positive amount in 45 cases. In the 25 cases in which the award was not collected, average tenure was 1.92 with a median of 1.51. In the 20 cases in which a trial award was collected, average tenure was 5.98 with a median of 2.59.6 Another way to see that awards in high-tenure cases get collected is to note that there were seven cases in which a worker with more than seven years of tenure was awarded a positive amount at trial. In all seven of these cases the award was collected. It is clear that, at least in the case of public lawyers for whom we believe that the assignment of cases to lawyers is close to random, cases in which a positive award is not collected tend to be lower-stakes cases. Nevertheless, these uncollected awards do to appear to be from trivially small cases. Razú (2006), for example, finds that 75% of newly-hired workers in Mexico do not stay continuously with the employer for one year. Kaplan, et. al. (2007) find that about 38% percent of formal-sector workers in Mexico were hired within the past year. We therefore see that a substantial fraction of employment at any given time has tenure below the median tenure observed for uncollected awards. 6The results from the 19 public lawyers with at least on trial and one non-trial outcome look nearly identical to the results for all public lawyers. 10 5 Simple Bargaining Model with Collection Costs In order to derive testable implications about the bargaining process, we con- sider a model in which a worker brings a lawsuit against a firm. We assume that the worker maximizes her expected payment net of legal costs. We assume that, if the case goes to trial, the judge will award V . For simplicity we assume that the firm has perfect information both about the lawsuit and about the worker. We will further assume that the worker always observes V , and observes with probability . The timing of the game is as follows: 1. The worker observes V . With probability , the worker also observes . With probability 1 - , the worker does not observe . In this case, the worker simply knows that is drawn from a uniform distribution on the unit interval. 2. The worker decides whether to drop the case or not. If the case is dropped, the payoff to the worker is zero. If the case is not dropped, the worker pays a cost of CO to proceed to the offer stage. 3. If the case has not been dropped, the worker makes a take-it-or-leave-it offer to the firm. The worker asks to receive a payoff of S. If the firm accepts the offer, payment is made and the game ends. If the firm rejects this offer, the case goes to trial and the judge awards V to the worker. 4. If the worker pays a cost of CC, she receives the award. If not, the worker receives nothing. We will assume that CC > CO. The model can be solved quite simply. First, consider the cases in which the worker observes . If V < CC, the case will be dropped. If not, the worker will make a settlement offer of V , that will always be accepted by the firm. Hence, when the worker observes the true value of the case, the lawsuit will never end up in court. Now consider the case in which the worker does not observe . If V < CC and the parties have reached the offer stage, the firm will not accept any offer since the firm knows that the judge's award will not be collected. Conditional on V CC, an offer CC or less will be accepted with probability one. Therefore the worker will never offer less than CC. The expected payoff (excluding the cost of making an offer which would have already been paid by this point) to the worker can be written as: + CC E () = -CC + µS + S (1) 2 ¶¸µS -VCC ¶ µV V- S¶. How do we arrive at this expression? With probability CC , the judge's award V would be too small to be collected, so any offer will be rejected and the payoff to the worker will be zero. With probability S-VCC , the offer will be rejected by the 11 firm even though the award will be large enough to be collected. The expected judgment conditional on being in this situation is S+CC , but the worker will be 2 forced to pay a cost of CC to collect the award. With probability V -S, the offer V will be accepted and the payoff is simply S. It is straightforward to show that the optimal offer made by the worker is7 S = ½ V - CC if V 2CC (2) CC if V < 2CC. We now consider two potential sources of heterogeneity across workers in order to derive testable implications of the model. The first form of heterogene- ity is that the workers differ in their values of , that is, workers differ in the accuracy of their information about the case. If this were true, workers with better information would be less likely to drop low-stakes (low V ) cases and more likely to drop high-stakes (low V ) cases. How can we see that this is true? Note first that a worker who never observed ( = 0) would have a cutoff level of V below which she will always drop the case and above which she will never drop the case. A worker who always observed ( = 1), on the other hand, would drop cases if and only if V < CC. This means that, even if V is very close to CC, the perfectly-informed worker will have a positive probability of not dropping the case. Furthermore, even if V is extremely large, the perfectly-informed worker will have a positive probability of dropping the case. What other predictions do we have about workers if we assume they only differ in terms of the quality of their information ()? Since all cases get settled when V CC and the worker observes , better-informed workers should always have settlement probabilities that are at least as high as those of workers with worse information. Further, better-informed workers should always have lower probabilities of a trial than workers with worse information, but since we observe relatively few trials in the data, this hypothesis will be difficult to test. As mentioned in the introduction, these testable implications could relate to comparisons of the outcomes of different cases for the same worker. Since we do not observe workers multiple times in the data, we cannot use workers to test these implications. We do, however, observe lawyers multiple times in the data. We will therefore test these hypotheses using lawyers, implicitly making the reasonable assumption that the information used by the worker- lawyer team is a combination of worker information and lawyer information. The essentially random assignment of cases to lawyers guarantees that there should be no correlation between the quality of worker information and the quality of lawyer information. Nevertheless, the effects of differences in worker information and differences in lawyer information should be the same. We will not, unfortunately, observe any proxy for the quality of the lawyer's information (). We will, however, observe a proxy for the stakes of the case 7It is very easy to add a cost of going to trial to the model. Assume, for example, that the worker's lawyer has to pay a cost of Ct if the case goes to trial. The resulting optimal offer would be S = V - CC - Ct if V - CC - CT CC and S = CC if V - CC - CT < CC. 12 (V ). In particular, we argue that the tenure at the firm of the dismissed worker is a good proxy for the stakes of the case. Assuming that lawyers only differ in terms of the quality of their information, we can rewrite the testable hypotheses in the following way: i) Lawyers with high probabilities of dropping low-stakes cases will have low probabilities of dropping high-stakes cases. ii) Lawyers with high probabilities of settling low-stakes cases will have high probabilities of settling high-stakes cases. The other potential source of heterogeneity that we consider in this paper is that workers differ in their costs of collecting awards (CC). The first (trivial) testable implication in this case is that, conditional on any value of V , workers with high collection costs will have dropping probabilities that are at least as high as those for workers with lower costs. We now turn to settlement probabilities assuming workers differ in their collection costs. As (the probability of observing the true value of the suit) approaches 1, all cases that are not dropped will settle, since both parties will know the true value of the case. Also, since the worker and her lawyer know the true value of the case, for any value of V , the case will be dropped with a higher probability when collection costs for the worker-lawyer team are higher. Since settling and dropping are the only two case outcomes, this means that for any value of V , a worker with higher costs of collection is less likely to settle. As approaches zero, however, the story is more complicated. Note first that, conditional on the suit not being dropped and conditional on not being observed, settlement will occur whenever the true value of the suit (V ) is greater than the settlement offer (S) given by equation 2. Simple inspection of equation 2 reveals that the optimal settlement offer is higher for high-cost workers when V is low and is lower for high-cost workers when V is high. We therefore see that if CO = 0, which would imply that no suits are dropped, workers with high collection costs would have lower probabilities of settling low-V suits and higher probabilities of settling high-V suits. The intuition behind the above result is straightforward. When the stakes of the case are high, a firm views offers from high- and low-cost workers similarly since, conditional on going to trial, all workers will collect with probability close to one. In the bargaining stage, however, a high-cost worker will ask for less money and therefore settle more often since she is more anxious to avoid the trial. Hence for high-stakes cases, a high-cost worker is more likely to settle. This is exactly how a standard cost of going to trial operates in the literature. When the stakes of the case are high, which implies that awards will almost never be left uncollected, a cost of going to trial and a cost of collecting an award are effectively the same thing. When the stakes of the case are low, however, the firm anticipates that a high-cost worker will not collect the award. Therefore in a low-stakes case a high-cost worker is less likely to settle because the firm views a trial as a good outcome. In our model this translates into settlement occurring whenever the 13 true value of the case exceeds the worker's collection costs. This implies a lower probability of settling for workers with high collection costs and low values of V . The possibility that a high-cost worker will settle with a lower probability, even if cases are never dropped, differentiates our model from those with costs of simply going to trial. How do we incorporate dropped cases into our analysis of the effect of col- lection costs on settlement probabilities? The fact that a high-cost worker will have a higher cutoff level of V required to not drop the case only reinforces the result that, when is small, high-cost workers will have lower settlement prob- abilities for low-V cases. To see this, one only has to note that the high-cost worker will have a higher cutoff value of V in order to proceed with the case. If the high-cost worker is below her cutoff value of V , her probability of settling will be zero. Once V is high enough, dropped cases cease to be an issue and our previous analysis that high-cost worker will settle with higher probabilities remains intact. Once again, we will use lawyers as a way of informing us about the effects of these costs on lawsuit outcomes. We do not observe any proxy for collection costs, but we will continue to use tenure as a proxy for the stakes of the case. Assuming that lawyers only differ in terms of their collection costs, we therefore summarize out testable implications in the following way: iii) Lawyers with high probabilities of dropping low-stakes cases will have high probabilities of dropping high-stakes cases. iv) Lawyers with high probabilities of settling low-stakes cases may have low probabilities of settling high-stakes cases. Although the relation between the settlement probability for low-stakes cases and the settlement probability for high-stakes cases is theoretically ambiguous if lawyers only differ in terms of collection costs, settlement probabilities will still be central to our empirical analysis. If we find evidence that lawyers who settle low-stakes cases tend not to settle high-stakes cases, we will be able to reject the hypothesis that lawyers differ only in terms of their information. Such a finding would therefore imply that differences in collection costs across lawyers affect lawsuit outcomes. Since it can be also extremely costly for the workers in terms of time to collect awards after trials, it seems likely that differences across workers in collection costs should have similar effects on outcomes. The primary goal of the empirical section will be to test the above hypotheses empirically. We will begin, however, by presenting evidence that these costs of collecting awards significantly impact the bargaining and trial outcomes we study. We will also present some results we believe are interesting, although not strictly related to the theoretical model. 6 Empirical Analysis Our first goal in this section is to demonstrate that costs associated with col- lecting awards must be taken into account in order to understand how lawsuits 14 are resolved. For the rest of the paper, we will only use data from the 19 lawyers with at least one trial and one non-trial outcome. We will do this because some of our models, like the one we present below, compare outcomes of lawsuits that go to trial to outcomes of lawsuits that do not go to trial for the same lawyer. Lawyers who do not have at least one lawsuit that goes to trial and at least one that does not go to trial contribute very little to these estimations. The inclusion of these lawyers, however, would require the estimation of many more parameters in a non-linear model. As we mentioned earlier, the inclusion of these lawyers would require the estimation of an additional 30 lawyer effects while only adding an additional 85 observations. Consider now the following model: posil = 1tenureil (1 - trialil) + 2tenureil trialil+ 3genderil (1 - trialil) + 4genderil trialil + 5trialil+ (3) l (1 - trialil) + ltrialil + il where the subscript i denotes the case and the subscript l denotes the lawyer. The dependent variable posil is a dummy variable equal to one if the worker recovers a positive award. If the case ended in a trial ruling, the dummy will be equal to one if the worker was awarded a positive amount at trial and if this award was in fact collected. If the lawsuit did not end in a trial ruling, then posil is simply a dummy variable for whether the case was settled (all settlements are for positive amounts) as opposed to being dropped. We consider two observable characteristics: gender and tenure, and allow the effects of these variables to be different for trial and non-trial outcomes. We also allow trials to have a different intercept than do lawsuits that do not end as trials. We estimate the parameter l for each lawyer, which among other things captures the differences in settlement probabilities across lawyers, controlling for gender and tenure and conditional on the case not going to trial. Note also that , through the parameter l, also affects the probability of recovering a positive amount at trial. Since the two parameters and l enter multiplicatively in the last term of the equation, we estimate this model with non-linear least squares.8 The first column of table 4 presents the results of the estimation of equation 3. The first result to point out is that the estimate of 2 is 0.04 and statistically significant at the 0.01 level. This result tells us that, conditional on going to trial, workers with high tenure tend to collective positive awards. We also see that the estimate of is -1.06 and is significant at the 0.05 level. This means that lawyers who tend to settle cases that do not go to trial (ones with high values for l) tend not to collect positive awards for cases that go to trial. We therefore see that cases that go to trial are more likely to end with the worker collecting something when the worker was employed for a long time at the firm and when the worker's lawyer drops a high fraction of cases that do not go to trial. One simple explanation for these results is that these types of 8Some cases are grouped together in the same lawsuit. We calculate the standard errors of our estimated parameters allowing for arbitrary heteroscedasticity and allowing for an arbitrary correlation of the error terms among cases grouped together in the same lawsuit. 15 cases receive more favorable rulings at trial. Another explanation is that these types of cases do not receive more favorable rulings, but that awards in these cases are more likely to be collected. The results from columns 2-5 of table 4 support the latter explanation. In column 2 we estimate a model similar to equation 3 in which the dependent variable continues to be posil for lawsuits that do not go to trial. For lawsuits that do go to trial, however, we use as the dependent variable a dummy for whether the judge declares her ruling to be favorable for the firm. Columns three and four present the analogous estimations examining the trial rulings of favorable for the worker and mixed respectively. Since the none of the estimated values of 2 or from columns 2-4 are statistically significant, we see no evidence that high-tenure cases are more likely to receive favorable rulings at trial and no evidence that lawyers who drop a lot of cases receive more favorable rulings at trial. In column 5, however, we use a dummy variable for "not collecting a positive amount awarded at trial" as the trial outcome measure. Since the estimate of 2 is -0.03 and significant at the 0.01 level, we see that awards from high-tenure cases tend to be actually collected. Furthermore, since the estimate of is 1.49 and significant at the 0.05 level, we see that lawyers who tend to settle many cases also tend to leave positive awards uncollected at trial. Combining the information from all columns of table 4, we see that workers in high-tenure cases tend to collect something at trial, not because they do better in terms of trial outcomes but rather because the awards are actually collected. Similarly, lawyers who settle many cases tend not to collect positive amounts for their clients at trial not because they do worse in terms of trial outcomes, but rather because they simply tend not to collect positive awards for their clients. The results on tenure are obviously consistent with the theoretical model if we view tenure as a proxy for the stakes of the case (V ). But what do the results on lawyers have to do with the theoretical model? Perhaps the easiest interpretation of the results on lawyers from table 4 is that the lawyers who drop a lot of cases do so because they have better information. Consistent with the model, these lawyers should tend not to go to trial when the amount awarded is likely to be too small to bother collecting. Indeed, we will present further evidence in favor of this hypothesis? Can the results on lawyers from table 4 be consistent with the idea that lawyers simply differ in their costs of collecting trial awards? If we made the (ridiculous) assumption that gender and tenure were the only variables observed by the lawyer, the results in table 4 would seem to contradict the predictions of this hypothesis. According to this hypothesis, lawyers who settle (do not drop) a lot of (non-trial) cases should have low costs of collecting trial awards, and therefore should tend to collect positive amounts with higher probabilities. Furthermore, lawyers who settle (do not drop) a lot of (non-trial) cases should have lower probabilities of not collecting positive awards at trial. Of course the lawyer should observe many things that we do not observe as econometricians. Suppose, for instance, that we observe two cases with different lawyers in which tenure has a low value. Suppose further that neither of these 16 cases is dropped. If we know that one lawyer has low costs of collecting trial awards, the fact that we observe that she did not drop the case might not be surprising. If, however, we know that the other lawyer has a high cost of collecting a trial award, it is surprising to see that the case has not been dropped. It is therefore likely that some unobservable (to the econometrician) characteristics of the case are quite favorable. We could therefore rationalize the results in table 4 in the following way. Lawyers with high costs of collection end up dropping many cases. Conditional on going to trial, these lawyers with high costs therefore have cases that are stronger for unobservable reasons. Since the tried cases for high-cost lawyers are in fact stronger, it is quite natural to see that the high cost lawyers tend to recover positive awards for their clients. If you thought that lawyers with high costs of collecting awards also had high costs of proceeding with the case in the first place (that is, not dropping the case early on), this "selection effect" would be even stronger. We will in fact present evidence in favor of the hypothesis that differences in collection costs also play an important role. In summary, we do not believe the results on lawyers from table 4 are par- ticularly helpful in testing our hypotheses. We do believe, however that table 4 demonstrates that the collection costs, which are the emphasis of our entire paper, are important factors both for explaining why low-tenure cases do not collect awards after trial and for explaining differences across lawyers. We now turn to empirical exercises that are more closely linked to our hypotheses. If we believe that workers differ in the quality of their information, the theoretical model makes a clear prediction on dropped cases. Workers with better information should drop fewer low-stakes (low V ) cases, because they will be able to separate out the few low stakes cases that are very likely to lead to a judgment that is worth collecting. Workers with better information should also drop more high-stakes (high V ) cases since they will be able to recognize the few high V cases that are not worth the effort. The model therefore predicts that workers who are more likely to drop small cases should be less likely to drop large cases. As mentioned earlier, we would need multiple observations on workers to test this hypothesis with workers directly. Since we do observe lawyers multiple times in the data, we can use lawyers to test the general hypothesis that informational differences are important determinants of lawsuit outcomes. To the extent that this hypothesis is confirmed with lawyers, it seems likely to be true for workers as well. In order to test this prediction, we estimate the following equation: dropil = l + 1femaleil + 2tenureil + l tenureil + il. (4) Equation 4 also has to be estimated by non-linear least squares. The pa- rameter l measures the lawyer's propensity to drop a lawsuit when tenure is equal to zero. A negative value for the parameter would imply that, for a large enough value of tenure, lawyers who are more likely to drop when tenure is low are less likely to drop when tenure is high. We present the results of estimating equation 4 in the first column of table 5. 17 As predicted by the theoretical model when lawyers differ in the accuracy of their information, our estimate of is negative (-0.14) and statistically signifi- cant at the 0.01 level. According to this estimation, lawyers would be predicted to have the same dropping probabilities when tenure is 7.20 years. This figure is a bit worrisome since tenure of 7.20 years corresponds to the 90th percentile of the tenure distribution in our data, that is, there are very few observations with a tenure level higher than 7.20. To address this concern, we estimated an equation similar to equation 4, but with a more flexible functional form for tenure. Specifically we estimated dropil = 1femaleil + l + 2tenureil + 1l tenureil+ 3tenure2il + 2l tenure2il + 4tenure3il+ (5) 3l tenure3il + 5tenure4il + 4l tenure4il + il. We present the results of estimating equation 5 in column 2 of table 5. Importantly, lawyers with high probabilities of dropping when tenure is low are now estimated to have lower probabilities of dropping when tenure is greater than 3.58, which is at the 75th percentile of the tenure distribution. To make the results of table 5 more transparent, we plot in figure 1 the estimated values of the derivative of the dropping probability with respect to l, for all tenure values up until 23.98 which is the 99th percentile of the tenure distribution. We do this both for the equation in which tenure is entered linearly and for the case in which tenure is entered as a quartic. Since we believe this "switching point" in the probabilities of dropping is a crucial test of the hypothesis that lawyers differ in terms of the accuracy of their information, we explore this issue further. In column 3 of table 5, we present estimates of the following equation: dropil = 1femaleil + l (tenureil < 3.58) + 2 (tenureil 3.58)+ (6) l (tenureil 3.58) + il. The cutoff value of 3.58 to separate low and high tenure was chosen because our estimation of equation 5 indicated that lawyers with high probabilities of dropping when tenure is low were estimated to have lower probabilities of drop- ping when tenure is greater than 3.58. The estimated value for is -1.42 and is statistically significant at the 0.01 level. In column 4 of table 5 we re-estimate equation 6, but only using tenure values in the bottom quartile (tenure 0.55) or tenure values in the highest quartile (tenure 3.595). The idea behind this estimation is to throw out the observations from tenure ranges in which the differences between lawyers are estimated to be small. Our estimate of is now -1.61 and is significant at the 0.05 level. Overall we believe that there is considerable empirical support for the model's prediction that lawyers who drop a high percentage of low-stakes cases will drop a low percentage of high-stakes cases. In other words, the results from table 5 support the hypothesis that lawyers differ in terms of the accuracy of their information. It is also worth noting that we could not rationalize the 18 results of table 5 if we thought that lawyers differ only in their costs of collecting awards. A high-cost lawyer would be more likely to drop all cases. The results from table 5 are therefore consistent with the hypothesis that informational differences affect lawsuit outcomes. Although we confirmed this hypothesis using heterogeneity across lawyers, there is no doubt enormous het- erogeneity across workers in terms of their information. In this sense, the results from table 5 almost certainly indicate that labor law will be enforced less strictly for workers who lack the information necessary to defend their rights. Now that we have presented evidence that informational differences are im- portant determinants of lawsuit outcomes, we turn to evidence that the costs of collecting awards are also important determinants of lawsuit outcomes. Recall that if workers differ in terms of their collection costs, it is possible for workers with high probabilities of settling low-stakes cases to have low probabilities of settling high-stakes cases. Such a result, however, would be inconsistent with the hypothesis that workers only differ in terms of the quality of their infor- mation. We therefore estimate models like in table 5 (equations 4, 5, and 6), but use settlement as the dependent variable instead of the case being dropped. Once again, we use differences across lawyers to establish that differences in collection costs are important determinants of lawsuit outcomes. We present the results of estimating settlement probabilities in table 6. In column one we present the results of estimating an equation analogous to equa- tion 4, but with settlement as the dependent variable instead of dropped cases. Once again we estimate to be negative (-0.13) and statistically significant at the 0.01 level, implying that those lawyers with high settlement probabilities when tenure is low have lower settlement probabilities when tenure is high. This "switching point" occurs when tenure is 7.97 years, which is the 91st percentile of the tenure distribution. When estimating the analogy of equation 5 for set- tlement probabilities, we estimate that the switching point occurs at a tenure of 3.46 years, which is the 74th percentile of the distribution of tenure. In figure 2, we once again plot the estimated values of the derivative of the probability of dropping with respect to l. We again present the results of some "less parametric" models like equation 6, this time using 3.46 years as the cutoff between high and low tenure. When we use all of the data, we estimate to be negative (-0.46) but not statistically significant (p-value of 0.106). When eliminating observations from the middle two quartiles of the tenure distribution, we now estimate to be negative (-0.79) and statistically significant at the 0.01 level. Overall table 6 presents evidence that lawyers may also differ in terms of their costs of collecting awards. In particular, lawyers who settle with high probabilities when the stakes of the case are low (lawyers with low collection costs in the theoretical model) settle with lower probabilities when the stakes of the case are high.9 The results in table 6, therefore, support the hypothesis that heterogeneity 9Since the results of analyzing equations like equations 4, 5, and 6 for trial outcomes do not give clear empirical results and do not relate to the theoretical model in an obvious way, we do not report the results of these models. We are happy to provide these results upon request. 19 in terms collection costs affects lawsuit outcomes. Since we have found evidence for heterogeneity in terms of collection costs across lawyers, it seems extremely likely that this same sort of heterogeneity exists across workers. In fact, the main cost of collection is that both the worker and the lawyer accompany the court clerk when she attempts to notify the firm about the judge's ruling. Certainly the value of time varies across workers much as it does across lawyers. In this sense, it seems likely that workers with high collection costs do not receive the full benefits to which they are entitled. They will drop many cases when they have a legitimate case, they may accept low settlement amounts in order to avoid trying to collect, or they may leave awards uncollected after trials. We view the results in tables 5 and 6 as the results that are most directly linked to our model. In table 7, however, we present some models that we believe are interesting although not related in a clear way to our theory. In particular, we estimate the following equation in column one of table 7: posil = 1tenureil (1 - trialil) + 2tenureil trialil+ 3genderil (1 - trialil) + 4genderil trialil + 5trialil+ (7) l (1 - trialil) + 1ltrialil + 2ltenureil trialil + il. The parameters l capture (much like in equation 3), among other things, the differences in settlement probabilities across lawyers, controlling for gender and tenure and conditional on the case not going to trial. The parameter 1 now captures how settlement probabilities conditional on not going to trial (l) affect the probability of recovering a positive amount at trial when tenure equals zero. The key feature of this model is that, through the parameter 2, the differences in recovering something at trial between lawyers who settle or drop most of their non-trial cases can vary with tenure. We see from column one that the estimate of 2 is -0.18 and statistically significant. That is, lawyers who drop a lot of cases do comparatively worse in low tenure cases, which one may argue is consistent with the theoretical model although we certainly have not resolved the selection issues that made our interpretation of the results from table 4 difficult. We think, however, that the more interesting results come from analyzing the rulings of the judge. Our theoretical model has an exceedingly simple view of a trial. In the model, the judge simply reveals the truth and does not need to communicate with the two litigants. One might conjecture, however, that a more complex model would predict that lawyers with high costs of collecting awards would tend to exaggerate their claims for low-stakes cases. After all, why would a lawyer ask for a "reasonable" amount if the lawyer would not bother collecting a "reasonable" amount? In column two of table 6 we estimate a model similar to equation 7 in which the dependent variable continues to be posil for lawsuits that do not go to trial. For lawsuits that do go to trial, however, we use as the dependent variable a dummy for whether the judge declares her ruling to be favorable for the firm. The parameter l continues to measure, among other things, the lawyer's propensity to settle cases as opposed to dropping them. Since we do not estimate 20 a significant coefficient for 2, we find no evidence that lawyers who settle a high fraction of non-tried cases have differential propensities to lose high- or low-stakes cases outright. In column three, however, we analyze the outcome of the judge's ruling being favorable to the worker. Our estimate of 2 is -0.34 and significant at the 0.01 level, which implies that lawyers who drop a lot of cases (presumably those with high costs of collecting trial awards) are comparatively less likely to win low- stakes cases outright. Finally, we analyze the probability of a "mixed" ruling in column 4. Since we estimate that 2 is 0.25 and statistically significant at the 0.01 level, we find evidence that lawyers who drop a lot of cases (presumably those with high costs of collecting trial awards) are comparatively more likely to get mixed rulings. Our interpretation for these results on trial outcomes is the following. The results on rulings that are favorable to the firm tell us that, when the stakes of the case is low, judges do not tend to rule that lawyers who drop a lot of cases (presumably those with high costs of collecting awards) bring for cases with no merit. The results on rulings favorable to the worker tell us that judges tend not to accept the entire claims of lawyers who drop a lot of cases when the stakes of the case are low. Rather, the results on mixed rulings tell us that the judges tend to say that, for low-stakes cases, lawyers who drop a lot of cases tend to be exaggerating their claims. The results from table 7 may explain why some workers make "ludicrous" demands. Kaplan et. al. (2008) document that some workers make claims that seem unreasonable. Workers who make these claims tend to go to trial more often and tend not to be rewarded for these claims. Based on the evidence from table 7, one might conjecture that these workers have high costs of collecting awards. Our model would therefore be consistent with the observation that these workers tend not to settle. Firms would be anticipating that these workers would not collect their awards after trials and therefore would not be willing to settle the cases. 7 Reconciling Theory and Evidence Tables 5 and 6 present evidence that neither of the two sources of heterogeneity across lawyers on their own can explain our empirical results. Recall that table 5 told us that the lawyers who drop low-stakes cases tend not to drop high-stakes cases. Recall further from table 6 that lawyers who settle low-stakes cases tend not to settle high-stakes cases. Since trials form a relatively small percentage of outcomes, it would appear that those lawyers who drop low-stakes cases (and tend not to drop high-stakes cases) are also those who tend not to settle low- stakes cases (and do tend to settle high-stakes cases). We confirm this fact by looking at the correlation of the estimated values for l for the 19 lawyers across tables 5 and 6. The correlations are -0.93, -0.93, -0.87, and -0.81 using the estimates from columns 1, 2, 3, and 4 respectively. We will now argue that the two sources of heterogeneity that we consider, 21 when taken into account simultaneously, can be reconciled with the empirical evidence. Let us suppose, for example, that the lawyers who disproportionately drop high-stakes cases and disproportionately do not drop low-stakes cases have better information. The fact that these lawyers disproportionately settle low- stakes cases is perfectly consistent with having better information. The question then becomes how we can reconcile the fact that these lawyers also dispropor- tionately do not settle high-stakes cases? This result could only be reconciled with our theory if the better-informed lawyers also had lower costs of collecting awards. It therefore seems that the best explanation of what we observe in the data is that the lawyers with more accurate information about the quality of their clients' cases also have lower collection costs. Perhaps having more accurate information and lower collection costs are in fact, simply reflections of being a well-informed lawyer. A well-informed lawyer should understand the law better and therefore should have a more accurate signal of the quality of the case. A well-informed lawyer should also know how to handle the evasive techniques employed by some firms, and therefore should have lower costs of collecting judgments awarded by the judge. 8 Conclusions Government regulations, combined with the mechanisms through which regula- tions are enforced, have a crucial impact on a country's business climate. In this paper, we analyzed the interaction between an extremely rigid labor law and a court system that is inefficient at enforcing the law. In particular, we used data from a labor tribunal in Mexico to show that 56% of awards "won" by workers were not collected. This never occurred in cases in which the worker had more than seven years of tenure with the firm. Although we could not analyze worker heterogeneity in lawsuit outcomes directly, we could analyze heterogeneity across the lawyers representing them. We showed empirically that those lawyers who drop a lot of cases tend not to leave trial awards uncollected. One interpretation for this result is that better- informed lawyers anticipate cases in which they would be unlikely to collect the amount awarded at trial and drop these cases at earlier stages. Another interpretation is that lawyers with high costs of collecting awards drop all low- stakes cases and only go to trial with high-stakes cases. In order to help us sort through these two interpretations, we developed a simple theoretical model to help interpret the effects of having a cost of collecting awards after a trial. The model generated distinct testable hypotheses of how workers (and the lawyers representing them) would act differently depending on differences in the accuracy of their information and on differences in their costs of collecting awards. We find evidence that lawyers are different both in terms of the accuracy of their information and in terms of their collection costs. We therefore see that the distinction between de facto and de jure labor reg- ulation is a complex one. We show that differences in the information available 22 to the worker affect the application of labor law. We also show that when the worker is more willing to defend her rights, either because the potential benefits are high or because her costs are low, labor law is applied more strictly. More generally, we show that the worker herself is a crucial determinant of the degree to which labor law is enforced. References [1] Almeida, Rita and Pedro Carneiro (2007). Inequality and Employment in a Dual Economy: Enforcement of Labor Regulation in Brazil, Mimeo, The World Bank. [2] Botero, Juan C., Simeon Djankov, Rafael La Porta, Florencio Lopez-de- Silanes, and Andrei Shleifer (2004). The Regulation of Labor, Quarterly Journal of Economics 119(4): 1339-82. [3] Caballero, Ricardo J., Kevin N. Cowan, Eduardo M.R.A. Engel, and Ale- jandro Micco (2006). Effective Labor Regulation and Microeconomic Flex- ibility, Cowles Foundation Discussion Paper No. 1480. [4] Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer (2003). Courts, Quarterly Journal of Economics 118(2): 453-517. [5] Dreher, Axel and Martin Gassebner (2007). Greasing the Wheels of Entre- preneurship? The Impact of Regulations and Corruption on Firm Entry, CESIFO Working Paper No. 2013. [6] Eisenberg, Theodore and Henry S. Farber (1997). The Litigious Plaintiff Hypothesis: Case Selection and Resolution, Rand Journal of Economics 28: S92-112. [7] Elena, Sandra, Alvaro Herrera, and Keith Henderson (2004). Barriers to the Enforcement of Court Judgments in Peru. Winning in Court is only Half the Battle: Perspectives from SMEs and Other Users, IFES Rule of Law Occasional Working Paper Series. [8] Fenn, Paul and Neil Rickman (1999). Delay and Settlement in Litigation, Economic Journal 109(457): 476-491. [9] Gong, Jiong and R. Preston McAfee (2000). Pretrial Negotiation, Litiga- tion, and Procedural Rules, Economic Inquiry 38(2):218-238. [10] Goodwin, Jennifer (2005). Domestic Relations, Georgia State University Law Review, 22:73-82. [11] Gross, Samuel R. and Kent D. Syverud (1991). Getting to No: A Study of Settlement Negotiations and the Selection of Cases for Trial, Michigan Law Review 90: 319-393. 23 [12] Helland, Eric and Alexander Tabarrock (2003). Contingency Fees, Settle- ment Delay, and Low-Quality Litigation: Empirical Evidence from Two Datasets, Journal of Law, Economics, and Organization 19(2):517-542. [13] Kaplan, David S., Gabriel Martínez González, and Raymond Robertson (2007). Mexican Employment Dynamics: Evidence from Matched Firm- Worker Data, World Bank Policy Research Working Paper No. 4433. [14] Kaplan, David S., Joyce Sadka, and Jorge Luis Silva-Mendez (2008). Litiga- tion and Settlement: New Evidence from Labor Courts in Mexico, Journal of Empirical Legal Studies, forthcoming. [15] Lanjouw, Jean O. and Mark Schankerman (2004). Protecting Intellectual Property Rights: Are Small Firms Handicapped? Journal of Law and Economics 47(1):45-74. [16] Lerner, Josh and Antoinette Schoar (2005). Does Legal Enforcement Af- fect Financial Transactions? The Contractual Channel In Private Equity, Quarterly Journal of Economics 120(1): 223-46. [17] Razú, David (2006). Competencia Desleal entre Políticas Públicas en Méx- ico: El Modelo de Seguridad Social vs. Programas Asistenciales, Mimeo, Instituto Mexicano del Seguro Social. [18] Singer, George H. (1997). Section 523 of the Bankruptcy Code: The Fun- damentals of Nondischargeability in Consumer Bankruptcy, The American Bankruptcy Law Journal 71: 325-405. [19] Szmer, John, Susan W. Johnson and Tammy A. Sarver (2007). Convincing the Court: Two Studies of Advocacy: Does the Lawyer Matter? Influenc- ing Outcomes on the Supreme Court of Canada, Law and Society Review 41:279-300. [20] Watanabe, Yasutora (2007). Estimating the Degree of Expert's Agency Problem: The Case of Medical Malpractice Lawyers, mimeo. Northwestern University. 24 2.32 4.22 6.12 8.02 02 to 2.91 4.81 6.71 respect zero 8.61 interaction with 61 equals 2.51 4.41 tenure 6.31 linear dropping tenure 8.21 of 21 tenure when 2.11 of 4.01 years interaction 6.9 probability settling 8.8 of of tenure 8 2.7 4.6 polynomial Derivative 6.5 probability 1: 8.4 the 4 2.3 Figure 4.2 6.1 8.0 0 1 0 1.5 0.5 -1 -2 -0.5 -1.5 -2.5 2.32 4.22 6.12 8.02 to 02 2.91 4.81 respect zero 6.71 8.61 with 61 interaction equals 2.51 4.41 tenure 6.31 linear settlement tenure 8.21 of 21 tenure when 2.11 of 4.01 years interaction 6.9 probability settling 8.8 tenure of of 8 2.7 4.6 polynomial 6.5 Derivative probability 8.4 2: the 4 2.3 Figure 4.2 6.1 8.0 0 1 0 1.5 0.5 -1 -2 -0.5 -1.5 -2.5 Table 1: Descriptive Statistics All suits with private lawyers N Mean Std Dev Min Max tenure 1,906 3.76 4.85 0 39.86 gender 1,906 0.32 0.47 0 1 final payment (2000 pesos) 1,906 15,967 74,518 0 1,683,751 case settles 1,906 0.50 0.50 0 1 case dropped 1,906 0.28 0.45 0 1 case goes to trial 1,906 0.22 0.41 0 1 positive award at trial uncollected 202 0.61 0.49 0 1 All suits with publicly-appointed lawyers N Mean Std Dev Min Max tenure 665 3.12 4.86 0 47.08 gender 665 0.34 0.48 0 1 final payment (2000 pesos) 665 6,779 21,914 0 385,212 case settles 665 0.63 0.48 0 1 case dropped 665 0.26 0.44 0 1 case goes to trial 665 0.11 0.31 0 1 positive award at trial uncollected 45 0.56 0.50 0 1 Only publicly-appointed lawyers with at least one trial and at least one non-trial N Mean Std Dev Min Max tenure 580 3.02 4.60 0 34.91 gender 580 0.35 0.48 0 1 final payment (2000 pesos) 580 6,751 22,972 0 385,212 case settles 580 0.63 0.48 0 1 case dropped 580 0.26 0.44 0 1 case goes to trial 580 0.11 0.31 0 1 positive award at trial uncollected 42 0.57 0.50 0 1 Table 2: Assignment of Cases to Lawyers (F-statistics on joint significance of lawyer fixed effects) Dependent Variable female tenure All suits with private lawyers: N=1906, F(989, 916) 1.682 *** 2.581 *** All suits with public lawyers: N=665, F(48, 616) 1.255 3.214 *** Public lawyers with at least one trial and at least one non-trial: N=580, F(18, 561) 1.141 1.157 Notes: The F-statistics correspond to tests of the joint significance of the lawyer fixed effects in models with no other independent variables. We use the notation of *** to denote significance at the 0.01 level. Similarly ** denotes significance at the 0.05 level and * denotes significance at the 0.10 level. See text for details. *** * trial 1.69 0.35 We the for effects) 192.88 of text Variable *** *** *** variables. denotes See random ** settle 175.34 12.79 5.74 significance level. lawyer joint Dependent *** *** *** independent Similarly 0.10 Termination of the no the level. of drop of at 118.72 11.79 6.59 with tests 0.01 to Modes the significance at models at significance and and logit joint on trial ersy correspond law significance denotes* Lawyers one 19 3: and statistics statistics lawyers: least random-effects denote level Table ersy lawyers: at N=580, in to *** private law public ersy 0.05 with effects of the law at with 990 with 49 non-trial: chi-bar-square lawyers one The random notation (Chi-bar-square suits suits the All N=1906, All N=665, Public least Notes: lawyer use significance details. Not *** * ** ** the use the 0.00 0.22 0.10 1.49 at Positive Award collected -0.03 (0.01) (0.01) (0.12) (0.05) -0.59 (0.42) (0.59) squares trial,a the We ** least notis dropped.is that significance Mixed Ruling 0.01 correlated. (0.01) 0.00 (0.01) -0.10 (0.13) 0.10 (0.05) 0.57 (0.37) -0.31 (0.53) case Outcomes non-linear outcome theif possibility be may Trial ** the denotes the ** to with zero for Worker Wins 0.00 (0.01) 0.00 (0.01) 0.10 (0.13) 0.10 (0.05) 0.44 (0.40) -0.11 (0.59) which and in Rates proceeding Similarly ** estimated settled,is are case same level. Firm Wins -0.01 (0.01) 0.00 (0.01) 0.00 (0.12) 0.10 (0.05) 0.07 (0.27) 0.30 (0.40) observations theif the 0.01 level. Settlement models heteroscedasticity into the 0.10 *** ** ** ** All For one to for at the at Trial grouped Relating Recover at 0.04 (0.01) 0.00 (0.01) -0.09 (0.10) 0.10 (0.05) 0.90 (0.37) -1.06 (0.51) lawyers. Something equal allowing 19 been significance parentheses. Models in from 4: dummyais have significance calculated that denote errors to are Table trial) trial) *** denotes* settlement observations variable cases of errors in and Standard 580 notation level tenure*(trial) tenure*(not female*(trial) female*(not trial (lawyer's fraction)*(trial) Notes: using dependent Standard outcomes the 0.05 Table 5: Models Predicting Dropped Cases case is case is case is case is dropped dropped dropped dropped female -0.09 ** -0.09 ** -0.08 * -0.06 (0.04) (0.04) (0.04) (0.06) tenure 0.03 *** 0.07 (0.01) (0.05) tenure2 0.00 (0.01) tenure3 0.000 (0.001) tenure4 0.00000 (0.00002) tenure >= 3.58 0.63 *** 0.73 *** (0.15) (0.24) (tenure)*(lawyer's dropping -0.14 *** -0.29 ** fraction when tenure=0) (0.03) (0.13) (tenure2)*(lawyer's dropping 0.00 fraction when tenure=0) (0.03) (tenure3)*(lawyer's dropping 0.001 fraction when tenure=0) (0.002) (tenure4)*(lawyer's dropping -0.00002 fraction when tenure=0) (0.00004) (tenure >= 3.58)*(dropping -1.42 *** -1.61 ** fraction when tenure < 3.58)) (0.51) (0.74) tenure level when lawyers have same probability of 7.20 3.58 dropping Only tenure in lowest (<.55) or No: No: No: Yes: highest quartiles (>=3.595) N=580 N=580 N=580 N=289 Notes: Standard errors in parentheses. All models are estimated with non-linear least squares using 19 lawyers. The dependent variable is a dummy equal to one if the case is dropped, zero if the case is not dropped. Standard errors are calculated allowing for heteroscedasticity and for the possibility that the outcomes in cases that have been grouped into the same proceeding may be correlated. We use the notation of *** to denote significance at the 0.01 level. Similarly ** denotes significance at the 0.05 level and * denotes significance at the 0.10 level. Table 6: Models Predicting Settlement case is case is case is case is settled settled settled settled female 0.08 * 0.07 0.07 0.04 (0.04) (0.04) (0.04) (0.06) tenure 0.08 *** 0.21 ** (0.02) (0.08) tenure2 -0.01 (0.02) tenure3 0.000 (0.001) tenure4 0.00001 (0.00002) tenure >= 3.46 0.91 *** 1.14 *** (0.18) (0.19) (tenure)*(lawyer's settlement -0.13 *** -0.32 ** fraction when tenure=0) (0.03) (0.13) (tenure2)*(lawyer's settlement 0.01 fraction when tenure=0) (0.03) (tenure3)*(lawyer's settlement 0.001 fraction when tenure=0) (0.002) (tenure4)*(lawyer's settlement -0.00002 fraction when tenure=0) (0.00004) (tenure >= 3.46)*(settlement -0.46 -0.79 *** fraction when tenure < 3.46)) (0.28) (0.30) tenure level when lawyers have same probability of 7.97 3.46 settling Only tenure in lowest (<.55) or No: No: No: Yes: highest quartiles (>=3.595) N=580 N=580 N=580 N=289 Notes: Standard errors in parentheses. All models are estimated with non-linear least squares using 19 lawyers. The dependent variable is a dummy equal to one if the case is settled, zero if the case is not settled. Standard errors are calculated allowing for heteroscedasticity and for the possibility that the outcomes in cases that have been grouped into the same proceeding may be correlated. We use the notation of *** to denote significance at the 0.01 level. Similarly ** denotes significance at the 0.05 level and * denotes significance at the 0.10 level. Not * ** ** the use the 0.00 0.24 0.10 0.92 0.13 Positive at Award collected -0.12 (0.07) (0.01) (0.12) (0.05) -0.20 (0.52) (0.75) (0.09) squares trial,a the We ** ** *** ** *** least notis dropped.is that significance Mixed Ruling -0.17 (0.07) 0.00 (0.01) -0.07 (0.13) 0.10 (0.05) 1.06 (0.38) -1.03 (0.52) 0.25 (0.09) case correlated. Outcomes non-linear outcome theif possibility be Trial *** ** ** *** the may denotes the with zero for ** to Worker Wins 0.24 (0.08) 0.00 (0.01) 0.07 (0.12) 0.09 (0.05) -0.54 (0.44) 1.31 (0.60) -0.34 (0.11) which and Rates in ** estimated settled,is proceeding Similarly same level. Firm Wins -0.04 (0.05) 0.00 (0.01) 0.01 (0.12) 0.10 (0.05) 0.17 (0.29) 0.16 (0.42) 0.04 are case (0.07) Settlement observations theif the 0.01 level. models heteroscedasticity into the 0.10 ** ** ** All For one to for at the Trial at Relating Recover at 0.17 (0.07) 0.00 (0.01) -0.11 (0.10) 0.10 (0.05) 0.45 (0.42) -0.41 (0.59) -0.18 (0.09) grouped Something lawyers. equal allowing 19 been significance Models parentheses. in from 7: dummyais have significance calculated that denote errors to are Table trial) trial) *** denotes* settlement settlement observations variable cases of errors in and Standard 580 notation level tenure*(trial) tenure*(not female*(trial) female*(not trial (lawyer's fraction)*(trial) (lawyer's fraction)*(trial)*tenure Notes: using dependent Standard outcomes the 0.05