Case selection and judicial decision-making: evidence from French labor courts

Abstract

Using a database on French labor courts between 1998 and 2012, we investigate case selection and judicial decision-making. In France, judges are elected at the labor court level on lists proposed by unions, and litigants can first try to settle their case before the judicial hearing. We show that the ideological composition of the court indirectly impacts the settlement behavior of the parties but has no influence on the decision made in court. In addition, parties have self-fulfilling behavior and adapt to institutional rules. When they anticipate long judicial procedures at court, they settle more frequently and only require judicial hearings for complex cases. The duration to decide these complex cases is longer, explaining why they observe (and build their anticipation on) long case duration. Our empirical strategy uses probit, ordered probit and triprobit estimations to control for case selection.

Appendices

Appendix 1: Tables

Summary statistics for each election between 1997 and 2012 are presented in Tables 8, 9, 10, 11, and 12 .

Table 8 Summary Statistics of the share of seats obtained by each union at the 1997 elections

Table 9 Summary Statistics of the share of seats obtained by each union at the 2002 elections

Table 10 Summary Statistics of the share of seats obtained by each union at the 2008 elections

Table 11 Description of the variables

Table 12 Descriptive statistics at each decision step (corresponding to the regression samples of Table 2)

Appendix 2: Bayesian estimation of ideal points

The Bayesian estimation of ideal points is usually referred as the one dimensional item response theory. Such models originally aimed at measuring students’ performance to a test, and to locate them on a unique dimension. The objective consisted in estimating three sets of parameters: (i) an ability parameter for each student, (ii) a difficulty parameter for each question of the test, and (iii) a discrimination parameter for each question. Bayesian methods were developed to discriminate students according to their ability, by taking into account questions’ difficulty level, and by estimating their ‘relevance’ to correctly discriminate students.³⁵

These models have then be used in the political science literature, especially in the case of Supreme Court voting (Bafumi et al. (2005), Martin and Quinn (2002), Martin et al. (2005)), where researchers were willing to locate Justices on a liberal-conservative dimension. Our approach follows this literature: we aim at locating unions on a confrontation propensity axis by investigating their ability to successfully negotiate with firms owners.

More formally, our goal consists in estimating unions’ positions ($\beta _i$) on a confrontation propensity axis. To do so, as explained in footnote 18, we use a database on past inter-professional national agreements (ANI). They are country-wide agreements between worker unions and firm owners’ representatives that each union can decide to sign or not. We estimate ANI-specific parameters, i.e. their location on the confrontation axis ($\alpha _j$) and their discrimination parameter ($\gamma _j$), i.e. their capacity to discriminate unions on the confrontation dimension. The model is defined by a logistic utility model, where the latent utility depends both on unions and ANI parameters:

$$\begin{aligned} u_{i,j}=-\alpha _j + \gamma _j \beta _i + \epsilon _{i,j} \end{aligned}$$

(3)

where $u_{i,j}$ is the utility of union i to ratify ANI j, and $\epsilon _{i,j}$ is a random component.

Ideal points are assumed to be normally distributed with mean $\mu _\beta$ and variance $\sigma _\beta ^2$, and the ANI-specific parameters are assumed to be jointly distributed : $(\alpha _j,\gamma _j) \sim N_2(M,T^{-1})$. In order to avoid additive and multiplicative aliasing, as well as reflection invariance, we set parameters’ priors to default values of the MCMCpackage in R ($\mu _\beta =0$, $\sigma _\beta =1$, $M=0$, and $T=0.25$). Moreover, identification requires an additional constraint on the ideal points. Since our goal is to create a confrontational scale, we assume that the CGT, which is usually seen as the least likely to negotiate with firm owners, is more confrontational than the CFDT, which is seen as the most confrontational union. In other words, we constraint the model such that the CGT will get a positive score on the confrontational dimension, while the CFDT will get a negative score on the confrontational dimension. Of course, such a constraint does not claim that the CFDT is not confrontational at all, it only assumes that the CGT is more confrontational than the CFDT.

The first Bayesian estimate yields the results presented in Fig. 4. These results are in line with the classical distinction between confrontational unions (CGT and FO) and the others (CFDT, CGC, CFTC). It reveals however a strong heterogeneity among the two most confrontational unions, which reflects the limits of the first measure (propConfront). Indeed, the two most confrontational unions (FO and CGT) have different intensities of opposition to reforms. We use the results of this estimation to compute the second proxy for courts’ preferences for confrontation (confront).

Fig. 4

Results of the Bayesian estimation of unions’ ideal points

The second Bayesian estimate (tv_confront) allows for time-varying preferences. The results of this estimation are given in Fig. 5. They give a similar picture to the first Bayesian estimation but shows that the ideological gap between the two most confrontational unions has varied over time. We use the results of this estimation to compute the last measure of courts’ preferences for confrontation (tv_confront).

Fig. 5

Results of the Bayesian estimation of time varying unions’ ideal points