1Alternative dispute resolution (ADR) programs are well-established and largely promoted in the legal systems of most developed countries. In many of those countries, all employer-employee disputes brought to a court hearing must be preceded by a conciliation stage. The promotion of ADR for the resolution of labor disputes in the early 1980s has been largely celebrated, as it allows to reduce court workload, legal expenses and delay. However, controversies rapidly appeared when some authors suggested that ADR programs might prejudice females and minority participants (see, e.g., Delgado et al. ; Bachar and Hensler ; Craver ). To this date, the question of whether encouraging conciliation in labor disputes fosters gender inequalities has received only little attention in the law and economics literature. Yet, substantial developments in the behavioral and experimental economics literature suggest that addressing the issue of gender differences in labor disputes and pretrial settlements may be particularly relevant.
2Using experiments in the lab, economists have shown that women and men tend to differ in their bargaining behavior in various contexts. García-Gallego, Georgantzís and Jaramillo-Gutiérrez  show that females tend to adopt more aggressive strategies in the ultimatum game framework. As a sender, they send in average less to the recipient than males, whereas they reject more in average in the role of the recipient. Those findings cannot be explained by differences in risk attitude only. Dittrich, Knabe and Leipold  implement alternating offers wage-bargaining in the lab. They find that women obtain worse bargaining outcomes than men when they take on the role of employees. Solnick  implements an ultimatum game and shows that not only one’s gender, but also the gender of the counterpart affects one’s behavior, through social norms or/and stereotypes. Recently, Lambert, Peterle and Tisserand  implemented a lab experiment where participants bargain in a continuous-time environment over the division of a monetary amount. They observe that women ask for less in negotiations, thereby having a higher agreement rate than men at the price of worse monetary outcomes.
3The use of lab experiments is certainly useful when it comes to understand and identify properly the mechanisms underlying individual behavior. Few studies however extend the knowledge acquired in the lab to concrete empirical applications. It may be relevant, for instance, to explore the direct consequences of those gender differences in bargaining behavior on real-life negotiations. In this paper, we investigate whether we also observe gender differences in the in-court settlement process accompanying labor disputes in French courts.
4To answer this question, we use an original dataset compiled from legal documents and first exploited in Tisserand . The data relate to closed cases registered with Besançon Labor Court in France between January 2010 and October 2015. Our database is composed of 539 observations, one-third of which refer to pretrial settlements. 
5We find that women are significantly more likely to reach an agreement than men when they sue their employer for an irregular dismissal case, provided that the amount at stake is high enough. This is particularly striking as we observe the opposite pattern in non-dismissal cases, for which women tend to conciliate less often than men. We also show that women reach better outcomes when the case is brought to the court, controlling for some case characteristics. We believe that this finding reflects more realistic claims from women, which would facilitate pretrial agreements over lower quality cases, and a better recovery rate in the court when negotiation fails.
6The remainder of the paper is structured as follows. In the second section, we formulate our hypotheses in the light of the experimental literature. The third section presents our original dataset and reports first results through descriptive statistics, whereas the fourth section reports the econometric analysis that supports our main findings. Finally, the fifth section concludes.
Gender differences and hypotheses
7Gender differences have been extensively documented in the laboratory. Three types of behavioral differences are especially relevant to our study. First, differences in the degree of risk-taking. Second, differences in bargaining contexts, both as a proposer and as a receiver of a proposal. Third, differences in the willingness to participate in competitive tasks.
8Regarding the first of the aforementioned differences, there is a plethora of laboratory studies reporting a higher degree of risk aversion for female subjects. Specifically, as Charness and Gneezy  point out, based on a meta-analysis of 15 investment experimental games which were not designed to study gender differences, women take significantly less financial risks than men. In fact, this is also confirmed by García-Gallego, Georgantzís and Jaramillo-Gutiérrez  in a study designed to isolate how much of the observed gender differences in Ultimatum Bargaining stems from gender differences in risk attitude. In the case of pretrial bargaining, Fraisse, Kramarz and Prost  find that women are more likely to reach an agreement than men, thus avoiding the risk of losing in court.
9When risk attitudes are accounted for, gender differences persist and they are such that women are more likely to reject a given offer. In a paper designed to study gender differences in Ultimatum Bargaining, Georgantzís, Parasyri and Tsagarakis  report significant gender differences. Specifically, it is found that, contrary to Fraisse, Kramarz and Prost , females are more likely to accept a given offer, and to be willing to increase their offers (but not to decrease their demands) when a persuasive message has followed a previous transaction. Therefore, women are found to be less aggressive and more sensitive to verbal persuasion than men. Also, similar to both earlier studies by Solnick  and Sutter et al. , the study of Georgantzís, Parasyri and Tsagarakis  points to the importance of the pairing. Whether women and men hold beliefs regarding a higher probability to be matched with men in a given bargaining context is relevant to explain bargaining behaviors.
10This also relates to a more general gender difference related to the willingness to participate in a competitive task. With competitive task we mean a task in which a person’s reward is decided on his or her relative rather than absolute performance. Here the evidence shows that women are significantly less attracted by competitive tasks and show a clear preference for contexts in which their reward does not depend on their performance relative to others but, rather, on their absolute performance. A well-known result has been reported by Barber and Odean  and mostly confirmed by Niederle and Vesterlund , which associate the aversion of females to competitive environments with males’ overconfidence and females’ tendency to shy away from competition.
11Few studies have been carried out based on data from real bargaining contexts. We undertake this task based on litigation data from Labor Courts in France.  Following the previous discussion, we consider the following hypotheses to be relevant for our study:
12Hypothesis 1: Females are less likely to expose themselves to financial risks. Therefore, they will prefer to pursue a judicial decision if they feel more secure about the outcome. Once this is taken into account, we would expect cases brought to court by women to be more likely to receive a positive outcome in the court.
13Hypothesis 2: Females are less likely to engage in bargaining and competitive processes.
14Finally, the difference in risk attitudes between females and males may lead the former to behave differently across types of cases, with dismissals involving a higher risk compared to other types of case brought to court.
15Hypothesis 3: Dismissal cases imply more risks for plaintiffs than other cases.  Based on the existing risk aversion literature, we expect women to be more inclined to reach an agreement in dismissal cases.
Data and descriptive statistics
16The data are obtained through the registry of the Labor Court of Besançon in France, where we were based during the time of the database construction. Our database consists of 186 “procès verbal de conciliation” (hereafter PVC) related to all conciliation attempts registered between 2010 and 2015, and 353 “compte rendu de jugement” (hereafter CRJ) related to all first instance judgements registered between January 2014 and October 2015. 
17Each court report is manually copied in our database and treated as a single independent observation. Each PVC refers to a situation where the parties have reached a mutual agreement, whereas each CRJ refers to a situation where parties have first failed to reach an agreement and where a court verdict has been enforced. For reasons of consistency, we choose to retain only litigations in which plaintiffs are employees, which represent 98% of the overall cases. It is also worth mentioning that we do not retain the adjudications which have not gone through the conciliation phase.  Furthermore, we omit cases that were abandoned between the conciliation and the judgment phases. Since we focus on the determinants of the probability of settlement in the pretrial conciliation phase, we make no hypothesis on these particular cases.  We are nevertheless aware that these cases may have different characteristics from cases that follow the traditional procedure provided for by law.
18The conciliation rate of the sample is equal to 35.15% with no statistical difference between men (35.3%) and women (35.1%) (Proportion test, p = 0.97).  Distinguishing between dismissal and non-dismissal cases, the statistics show that the distribution of successful conciliation differs between men and women.  Table 1 gives an overview of the conciliation rates per gender and type of cases.
Conciliation rate according to gender and type of case in the dataset
|Dismissal cases only||21.31%||78.69%|
|Dismissal cases only||31.78%||68.22%|
Conciliation rate according to gender and type of case in the dataset
19While women’s conciliation rate on non-dismissal cases amounts to 41.27%, that of men rises to 52.46% (Proportion test, p < 0.05). Conciliation on dismissal cases only shows a different pattern with a significantly more important conciliation rate for women (31.78%) compared to men (21.31%) (Proportion test, p < 0.05).
20Documents also specify the economic damages estimated by the plaintiff as well as the damages that are recognized by the court. This allows us to calculate the plaintiffs’ recovery rate. Data show that the overall average recovery rate of the estimated damages is equal to 37% for both conciliation and trials. This rate significantly differs across genders: while women recover an average of 42.4% of their initial claim, men only recover an average of 32.8% (Wilcoxon rank-sum test, p = 0.01). Table 2 reports descriptive statistics. The overall agreement rate as well as the nature of the disputes are very similar for men and women. Wages and the average amount requested by plaintiffs, which are naturally correlated, are significantly higher for men.  The recovery rate, which is calculated by dividing the amount actually obtained during the judgment phase by the amount initially claimed, is significantly higher for women although the earnings of men and women in court are not significantly different.
Descriptive statistics per gender
|Mean and proportion tests|
|Dispute related variables|
|Lawyer for plaintiff in conciliation phase||66.35%||70.89%||0.30|
|Amount at stake (in €)||36,285||46,908||0.07|
|Gain at trial (in €)||9,810.14||10,856.02||0.53|
|Wage (in €)||1,890.32||2,527.1||0.01|
|Seniority (in years)||8.07||9.49||0.16|
Descriptive statistics per genderNote: a As the distribution of recovery rates is not normal, we use a Wilcoxon rank-sum test to test for gender differences in this variable.
21In this section, we analyze gender differences in the Labor Court using two different models. First, we estimate the likelihood to reach a settlement in the conciliation phase, using a binary outcome (Logit) model that distinguishes between settled and adjudicated litigation cases.  We refer to this estimation as model 1. Second, we explore the determinants of the recovery rate in trial. Recovery rates range from 0% to 100% with a probability mass at 0%. Our situation is a typical illustration of a corner solution response model as exposed in chapter 17 of Wooldridge . We find similar results when estimating a beta regression, that takes into account that the settlement rate is a proportion. To control for potential selection bias in the cases that went to trial (i.e., cases for which the conciliation phase failed), we include in the Tobit regression the Inverse Mill’s Ratio calculated from model 1.
22To answer our research question, we particularly focus on the role of gender and its interaction with some characteristics of the case. In both models, we then include a variable Female, which takes 1 when the plaintiff is a woman; a variable Dismissal, which takes 1 when the case at hand is a dismissal case; a variable Amount at stake which represents, in euros, the amount claimed by the plaintiff; the interaction variables Female × Dismissal and Female × Amount at stake.
23Control variables are identical in both models and include information regarding a) the nature of the employer (public company, association, private company); b) the fact that the employee and/or employers are accompanied by a lawyer or not; c) the reason of the conflict (dismissal, back-pay, or other reasons); d) the section of the litigation (trade, agriculture, executive, various, or industry); e) year dummies (i.e., the year the conciliation phase happened for model 1, the year the trial happened for model 2); f) the number of employees in the company.
24Complete estimation results of the determinants of the probability of settlement (Logit, model 1) and the determinants of the recovery rate (Tobit, model 2) are available in the Online Appendix II. In this article, we only display marginal effects for our variable of interest: gender. Marginal effects reported in Table 3 represent the outcome difference between a female plaintiff and a male plaintiff computed for particular values of the type of the case (dismissal or not) and the amount at stake. We choose three particular values for the amount at stake, i.e., first, second and third quartiles of the observed distribution in our dataset. All the other control variables are set to their average value. With respect to the Tobit model, the marginal effects we compute are the partial derivatives of the expectancy of the response variable conditional to the explanatory variables, as set in Wooldridge (, 674, Eq. 17.16). Inference is based on the delta method.
Marginal effects estimates on dispute outcomes
|Marginal effect of being a female in the probability of settlement (Model 1)|
|Non-dismissal case||Dismissal case|
|Low amount at stake (€5,690)||– 0.200**|
|Medium amount at stake (€21,550)||– 0.167**|
|High amount at stake (€48,516)||– 0.111|
|Marginal effect of being a female in the recovery rate (Model 2)|
|Non-dismissal case||Dismissal case|
|Low amount at stake (€5,690)||0.195***|
|Medium amount at stake (€21,550)||0.170**|
|High amount at stake (€48,516)||0.126*|
Marginal effects estimates on dispute outcomesNote: Marginal effects for gender are computed by setting particular values for the type of the case (dismissal or not) and the amount at stake (first, second and third quartile of the observed distribution in our dataset). Standard errors of marginal effects are obtained with the delta method. z-statistics are displayed in parentheses. Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
25We first investigate the settlement likelihood in the pretrial conciliation phase. The top panel of Table 3 reports the marginal effect of being a female plaintiff.
26Considering all types of cases but dismissal cases, women are less likely to reach an agreement during the conciliation phase. For low and medium amounts at stake (respectively €5,690 and €21,550), women are respectively 20 points and 16.7 points less likely than men to reach an agreement. This gender difference however tends to disappear when the amount at stake increases. An intuitive interpretation is that when the amount at stake is very large, the probability to reach an agreement is rather low, regardless of the gender of the plaintiff. A large part of non-dismissal cases are related to unpaid wages, unpaid overtime hours, or unpaid bonuses. These types of cases are straightforward: the employee only needs to prove that he/she did not perceive the money that is owed to him/her. Dismissal cases can be more tricky in that the line between a regular and an irregular dismissal is often thin. This first result is in line with our first hypothesis which states that women may be more likely to pursue a judicial decision if they feel secure about the outcome.
27In the specific situation of dismissal cases, men and women’s probability to find an agreement do not significantly differ, as long as the amount at stake is not too high. When the amount at stake is high (€48,516), women are more than 13 points more likely than men to reach an agreement. Women’s bargaining behavior significantly differs between dismissal and non-dismissal cases. Dismissal cases differ from the majority of other cases in that they are less predictable. Also, dismissal cases may imply higher stakes than the immediate monetary prejudice. Employment loss is generally associated with uncertainty, especially for women who may be disadvantaged on the labor market. This result, which is in line with our third hypothesis, may therefore reflect other gender differences in behavior, such as risk attitude or ambiguity aversion (see, e.g., Croson and Gneezy  for a documented discussion on gender differences in preferences).
28The bottom panel of Table 3 displays marginal effects computed from a Tobit estimation of the recovery rate. Our results show in particular that women tend to obtain a significantly higher recovery rate compared to men. For low amounts at stake (€5,690), the recovery rate is about 19.5 percentage points larger for women than men in non-dismissal cases, and about 16.1 percentage points larger in dismissal cases. We observe however that those gender differences tend to vanish with the amount at stake. When the amount at stake grows, one may expect that individual characteristics of the parties loose relevance in the judgment of the case.
29One could argue that these results could also reflect a difference in the nature of the demands of women and men, as well as differences in their professional activity. Nevertheless, we believe that our control for the groups of professions, as well as the very similar statistics between women and men on the nature of their request (reported in Table 2), allow us to not favor this alternative interpretation.
30Unlike laboratory experiments, our dataset does not allow us to control for individual preferences and identify the mechanisms underlying the gender effects we observe. Yet, it is possible that women are more likely to succeed at the trial phase because of case quality. As suggested in the literature, women engage significantly less often in conflicts from a general standpoint (see, e.g., Bowles, Babcock and Lai ). We thus believe that women are more likely to avoid bringing up cases they are not confident with, and to sue only when they believe the quality of their case is high enough to win. In other words, women choose to settle “lower quality” cases more often than men do during the conciliation phase, which would improve the average quality of the cases that reach the trial phase. Behavioral factors, such as pride or overconfidence, could also be responsible for men failing negotiations more often. It is worth noting that the rate with which conciliation fails for males facing a dismissal case is particularly high (78.69% of the cases in our database).
Summary and discussion
31Over these last decades, the number of empirical studies on legal data has considerably increased. The same is also true for gender studies which are now broadly documented. Gender differences in risk taking and bargaining have also been extensively studied in the lab. Still, there is little knowledge of how such differences affect behavior in real contexts involving risk and bargaining. We undertake this task based on a unique dataset compiled from Labor Court cases in France (Besançon), including information on pretrial conciliation.
32Our results show that women are less likely than men to reach a pretrial agreement on non-dismissal cases with low and medium amounts at stake. We also show that women are significantly more likely to reach an agreement when the case involves a dismissal with a large amount of money at stake. We believe the first result might relate to women’s aversion to conflict: women may (in average) want to avoid conciliation and argumentation and the possible conflict implied when they are confident enough in their chances to win at the trial phase. The second result might possibly relate to women’s risk aversion: dismissal case outcomes are hardly predictable as they often imply numerous parameters and are often followed by a period of job search with lower incomes. In this case, risk aversion might drive them to settle, securing a compensation, even if it is smaller than that they might get from the trial phase.
33Following the pretrial phase, our results also show that women are more likely to receive a larger part of their claim in the trial phase. We believe that this third result is in accordance with the two previous ones: assuming that women tend to put “safe cases” forward and to settle uncertain (dismissal mainly) cases, we can reasonably expect cases brought to court by women to be of higher quality, leading on average to a higher recovery rate. Also, this third result may relate to women’s higher aversion to conflict, leading them to make more realistic demands (more reasonable or less optimistic), which lead to higher recovery rates. Finally, this pattern is also compatible with mens’ overconfidence, also broadly documented in the experimental economics literature (Niederle and Vesterlund ).
34The heterogeneity in conciliation behavior we observe between dismissal and other cases might lead to important implications. When enforcing negotiation on parties, as it is the case in the French Labour Court, no distinction is made regarding the nature of the case. It appears that in some cases only (such as dismissal), women are more inclined to reach an agreement. One could wonder what compensation those female plaintiffs who opted out of trial would have received in a judicial environment where conciliation was optional. Making the conciliation stage mandatory for all cases might well disadvantage individuals who are less aggressive in negotiations, such as women as opposed to men, which would have direct consequences on gender inequalities.
35One of the main limitations of our paper is that our statistical analysis does not include cases that were abandoned between the conciliation and judgment phases. Some of these cases considered abandoned by the Labour Court actually give rise to out-of-court conciliations. The inclusion of these conciliations that take place outside the standard procedure provided by law could affect the results of our analysis depending on the proportion of men and women who use such practices, and their behavior during these out-of-court settlements. However, it is important to note that significant differences between men and women outside the courtroom would have to be observed for our results to be invalidated.
36While real world data provide the natural domain for an external validity test to behavioural patterns detected in the lab, they should not be seen as a substitute, but rather as a complement to laboratory studies. The latter may have often been considered suspect for unrealistic and not immediately applicable results but they provide an environment in which the researcher can control for many individual characteristics like personality traits and risk attitudes. Like in our study, the behavioral patterns detected in the lab under a controlled decision-making environment are of great help when analyzing data from analogous naturally occurring decisions in real life.
All information related to the functioning of the Labour Court in France is contained in the Online Appendix I, DOI : 10.3917/reco.706.1201.
The French Labour Court has been the object of previous empirical studies (see, e.g., Desrieux and Espinosa  or ).
Dismissal cases show a very high variability in terms of compensation awarded to the plaintiff according to his seniority and salary (Bourreau-Dubois et al. , 74, figure 2.AB). This is not the case in disputes concerning unpaid wages or bonuses where the amount that is likely to be obtained in the judgment phase is much more predictable.
We recorded all the decisions available over the periods mentioned. CRJs were filed in the Labor Court for a period of two years while PVCs, which are far fewer, were filed for five years. After these periods, documents were still archived but in an external location to which access was not granted. The time period of the data we recorded was thus shaped by this constraint. Since French labor law did not experience any major changes between 2010 and 2015 and the likelihood of settlement remained stable throughout this period both at the local and national level, the difference in time period for the two types of documents should not be an issue.
This relates to two particular types of claims: bankrupt businesses and fixed term contracts reclassification.
Detailed statistics regarding the national abandon rate in French labour courts are discussed in Desrieux and Espinosa .
This is the raw sample’s conciliation rate, which is the total number of conciliated cases divided by the total number of cases in the database. This rate is not adjusted to the time period the collection was made (60 months for PVC and 21 months for CRJ). In consequence, this conciliation rate does not match the national conciliation rate due to the data collection constraints detailed in footnote 3.
We only refer to cases of dismissals without real and serious cause and not simply to traditional claims for severance pay. A case is considered as a case of dismissal without real and serious cause if one of the employee’s requests concerns the recognition of a dismissal without real and serious cause.
Most of the monetary compensations in the Labour Code are expressed in terms of monthly wage.
The male/female distribution remains stable over the period with a shock in 2014 when the number of women is twice as low as the number of men (12 versus 24). We have again estimated the probability of agreement by excluding 2014 and our results remain unchanged. The significance of the gender variable lowers from 5% to 10%, certainly due to the loss of statistical power associated with the reduction in the number of data.