Portal de Periódicos da CAPES

Regional Difference‐in‐Differences in France Using the German Annexation of Alsace‐Moselle in 1870–1918

2009; University of Chicago Press; Volume: 5; Issue: 1 Linguagem: Inglês

10.1086/596013

2150-8372

Matthieu Chemin, Étienne Wasmer,

Labor Movements and Unions

Previous articleNext article FreeRegional Difference‐in‐Differences in France Using the German Annexation of Alsace‐Moselle in 1870–1918Matthieu Chemin and Etienne WasmerMatthieu CheminUniversity of Quebec at Montreal and CIRPEE Search for more articles by this author and Etienne WasmerSciences‐Po Paris and OFCE Search for more articles by this author PDFPDF PLUSFull Text Add to favoritesDownload CitationTrack CitationsPermissionsReprints Share onFacebookTwitterLinked InRedditEmailQR Code SectionsMoreI. IntroductionThe evaluation of labor market policies has been an expanding field in the last decades. This is partly due to the increasing availability of survey data and computing power and partly to the recognition that even complex phenomena such as the impact of laws and regulations on labor markets can be rigorously tested thanks to new empirical methodologies, such as the so‐called difference‐in‐difference approaches (also called double differences).In this approach, the outcome (education, access to employment, unemployment rate) of a treated group, that is, a group subject to a policy change, is compared to that of a control group, that is, a group made up of individuals (or any other unit of observation) as close as possible to the treated group but unaffected by the treatment. The strategy of the researchers consists in finding a sudden change in policy and building a relevant comparison group. A recent paper by Imbens and Wooldridge (2008) discusses the issues associated with the choice of the groups and surveys the literature.In countries with a federal structure such as the United States or Canada, it is straightforward to use this technique to evaluate policy changes. Indeed, many laws are specific to a state (for the United States) or a province (for Canada). When states experiencing a reform are compared to states with no change, it is straightforward to obtain inferences about the causal effects of the reform.1 In contrast, researchers studying French labor laws typically face cases in which there is no geographical variance in policy changes since the main law is supposed to apply equally to all the French territory. Therefore, policy evaluation has to rely on other control groups. A consequence of this “universal”character of the law is that the evaluation of the reduction in working time in France—from 40 to 39 hours in 1982 and then to 35 hours in 1998–2000—has been made more difficult, in the apparent absence of regional differences.2In this paper, we exploit a relatively unknown feature of French institutions: the northeastern part of France (a region, Alsace, and a subregion called a département, Moselle) has different institutions and, in particular, has a different social security system. The reason is a purely historical one. Alsace‐Moselle as it is called was part of Germany between 1870 and 1918. During that period, German laws fully applied. After the signature of the peace treaty in 1918, as a consequence of which Alsace‐Moselle was returned to France, the German code became a local law (droit local), that is, a specific body of legislation mixing up the most favorable elements of the French law code and the German one.Interestingly, there have been very few changes over time since 1918: people in Alsace‐Moselle are strongly attached to their legal specificity. The central state, quite strong in France, has never been able to generate a unified law and even recognized officially the Alsace‐Moselle specificity in a law text in 1924. Since then, many texts in the Civil Code and the Labor Code in France are amended with mentions of specific dispositions for the three départements of Bas‐Rhin, Haut‐Rhin (both of which make up Alsace), and Moselle (northeastern part of the Lorraine region).We can therefore use this unique historical accident to build difference‐in‐differences to evaluate the few reforms implemented in the rest of France differently from Alsace‐Moselle. Alsace‐Moselle can therefore be used as a control group, whereas the rest of France (which, by an abuse of language and without any meaning, we shall sometimes refer to as France hereafter when there is no ambiguity) can be used as the treatment group.In particular, we detail here three interesting institutional specificities:• Reduction in working time: Alsace‐Moselle has two days of vacation that do not exist in France: Good Friday and Saint‐Etienne (December 26). When the famous 35‐hours reform was implemented between 1998 and 2002, employers in Alsace‐Moselle first argued that these two days should be counted as part of the working time reduction and thus incorporated in the number of additional vacation days provided by the 35‐hours laws, approximately 18 annually.3 Therefore, Alsace‐Moselle experienced a 10% less stringent reduction in working time.• Absenteeism and sick leaves: Alsace‐Moselle has a very generous regime for sick employees. Employers have to pay the full salary for the total duration of an absence. Being sick is not even compulsory in order to benefit from the compensation: any fair cause is acceptable. In contrast, the rest of France has a less favorable regime: since 1945, social security covers 50% of the previous wage only after the fourth day of absence. In 1978, a law obliged all employers to give an allowance to sick and absent employees covering up to 90% of the previous wage, but only after the eleventh day of absence, during 30 days and when the employees have more than 3 years of seniority. Therefore, Alsace‐Moselle can again be considered as a control group, and the reform in France can be evaluated using difference‐in‐differences or even triple differences with respect to employees with less than 3 years of seniority.• Welfare policies: Since 1908, all municipalities in Alsace‐Moselle must provide assistance to impoverished citizens. This system was generalized to all of France in 1989 under the name RMI (Revenu Minimum d’Insertion), which is basically a subsistence income. Observers argued that the introduction of the RMI allowed municipalities in Alsace‐Moselle to reduce their subsidies by one French franc for each franc of the national subsidy. Therefore, welfare recipients in Alsace‐Moselle were left unaffected by the RMI reform and hence can be used as a control group.Of course, one may argue that Alsace‐Moselle is different because of the existence of other regional specificities. As a matter of fact, there is at least a difference reflected in the map in figure 1. The three départements of Alsace‐Moselle, labeled 57, 68, and 67 in the northeast corner of France, happen to be the only ones with a border with Germany. And this is of course not a random allocation in space!Fig. 1. View Large ImageDownload PowerPointThis has at least one undesirable consequence for researchers: since the pattern of trade between Germany and France is not homogeneous on the French territory but instead very dependent on the distance to the border as any gravity model predicts, it is quite likely that Alsace‐Moselle is disproportionately affected by the German economic cycle when it differs from the French economic cycle. In such a case, any comparison of “before and after” in Alsace‐Moselle and the rest of France will be contaminated by such spillover effects.To solve for this difficult issue, we will need to do several additional comparisons with unaffected groups in both Alsace‐Moselle and the rest of France, that is, falsification exercises or equivalently triple differences, by combining the difference‐in‐differences results of the affected and unaffected groups.In this paper, we will emphasize the richness and the diversity of the applications one can develop to better evaluate French labor policies. We present in the next three sections the three reforms in reverse order: welfare and subsistence income (Sec. II), absenteeism laws (Sec. III), and hours worked (Sec. IV) and the specificities in the Alsace‐Moselle regime. For each experience, we show aggregate results on relevant and affected variables such as long‐term unemployment, employment and unemployment, and the rate of absenteeism for both Alsace‐Moselle and France, and we point out the validity or invalidity of common wisdom of the effect of each of the three labor policies. In a series of companion papers, we go deeper into the analysis and are more specific about the various possible effects of each policy. However, the purpose of the current paper is simply to expose our idea of the Alsace‐Moselle identification strategy and how fruitful it can be. Section V describes other legal specificities in Alsace‐Moselle that could be used and presents conclusions.II. A Welfare Policy: The Minimum IncomeIn 1989, the French Parliament voted a very important law: any citizen above 25 years and below some income level became eligible for an allowance amounting to a large fraction of the minimum legal wage. This was called the RMI, and it was organized by law number 88‐1088, December 1, 1988. A natural question for labor economists is whether this welfare program had disincentive effects and generated more long‐term unemployment. This is a tricky question for empiricists given the huge amount of self‐selection and heterogeneity.By chance for econometricians, a very similar system (aide sociale) at the city level existed in Alsace‐Moselle since 1908 (lois locales des 30 mai 1908 et du 8 novembre 1909).4 For instance, in the main city in Alsace (Strasbourg), the allowance for an eligible person (single) amounts to 65% of his or her gross minimum wage (Kintz 1989).5After the introduction of the French RMI in 1989, municipalities in Alsace‐Moselle may still give an allowance to poor individuals, but this allowance reduces by the same token the RMI given by the state (Woehrling 2002).6 Consequently, after 1988, cities in Alsace‐Moselle have a direct incentive not to give this aide sociale, as emphasized by Woehrling. Poor individuals qualify for welfare payments in Alsace‐Moselle and the rest of France after 1988, but only in Alsace‐Moselle before 1988. This provides an opportunity for a difference‐in‐difference analysis before and after 1988, between Alsace‐Moselle and the rest of France, in order to evaluate the impact of the RMI.A. Methodology and ResultsIn this paper, we will present simple averages of unemployment rates in Alsace‐Moselle and the rest of France, before and after 1988. Table 1 presents long‐term (more than 2 years) average unemployment rates in France and Alsace‐Moselle, before and after 1989 (for individuals more than 25 years old). We used the Enquête Emploi, the major French labor survey, from 1982 to 2002 to calculate the ratio of the total number of people unemployed for more than 2 years and more than 25 years old divided by the total number of active individuals more than 25 years old in the area concerned and the time frame concerned.7 Active individuals are defined as the sum of employed individuals and unemployed individuals (as well as soldiers of the military contingent) more than 25 years old. For example, the long‐term unemployment rate in the rest of France before 1989 (averaged over the period 1982–89) is equal to 3.46%, as seen in row 1, column 1. The numbers in row 3 and column 3 are first differences. For example, the average long‐term unemployment rate between 1982 and 1989 is 1.12 percentage points higher in the rest of France than in Alsace‐Moselle as visible in row 3, column 1.Table 1. Long‐Term (More than 2 Years) Average Unemployment Rates in France and Alsace‐Moselle before and after 1989 (More than 25 Years Old) 1982–89(1)1990–2002(2)Difference(3)Rest of France3.464.491.03Alsace‐Moselle2.342.54.20Difference1.121.95.83***Note: The significance of the difference‐in‐differences coefficient is estimated through the bootstrap method. One hundred samples were randomly drawn with replacement from the original sample. The difference‐in‐differences coefficient was thus estimated 100 times. The difference‐in‐differences coefficient was never less than zero. Figures are long‐duration (more than 2 years) average unemployment rates for individuals more than 25 years old. They represent the ratio of the total number of people unemployed for more than 2 years and more than 25 years old divided by the total number of active individuals more than 25 years old in the area concerned and the time frame concerned. Active individuals are defined as the sum of employed individuals and unemployed individuals (and soldiers of the military contingent) more than 25 years old. The numbers in the Difference row and column are first differences. The number in the bottom right corner is a difference‐in‐differences estimate, comparing the rest of France to Alsace‐Moselle, before and after 1989.***Significant at 1%. View Table Image First differences with Alsace‐Moselle provide a misleading impression of the impact of the RMI on long‐term unemployment rates. Comparing long‐term unemployment rates in the rest of France to the one in Alsace‐Moselle after 1989 (an increase of 1.95 percentage points) provides a misleading estimate of the impact of the RMI: the rest of France might have been systematically different from Alsace‐Moselle because of time‐constant unobserved heterogeneity. Similarly, comparing unemployment rates in the rest of France before and after 1989 (an increase of 1.03 percentage points) provides a misleading estimate of the impact of the RMI: it might be that the rest of France would have witnessed an increase in its long‐term unemployment rate even without the RMI as a result of, for example, its poor economic conjuncture at the time.Only the difference‐in‐difference coefficient of row 2, column 2 (comparing the long‐term unemployment rate in the rest of France to the one in Alsace‐Moselle, before and after 1989), captures the causal impact of the RMI. The underlying assumption is that “geographic time‐constant unobserved heterogeneity” is controlled for by comparing a geographical area with itself, before and after 1989, whereas common macroeconomic trends are captured by comparing the rest of France to Alsace‐Moselle.The number in the bottom right corner is a difference‐in‐difference estimate, comparing the rest of France to Alsace‐Moselle, before and after 1989. It shows that the long‐term unemployment rate increased by 0.83 percentage point as a result of the implementation of the RMI. This economically significant result has to be tested statistically. The significance of the difference‐in‐difference coefficient was estimated through the bootstrap method, and we found it to be significant at the 1% level.8 This seems to support the “inactivity trap” argument: individuals may have less incentive to seek a low‐paying job when it does not increase disposable income by a dramatic amount compared to welfare payments.B. Falsification ExercisesThe main assumption, upon which a difference‐in‐difference estimation implicitly relies, is commonly called the “common time effects” assumption. It means that to interpret causally the difference‐in‐difference coefficient, one needs to assume that the treatment and the control groups are on the same time trend. In other words, the rest of France would have evolved the same way Alsace‐Moselle did had the RMI not been implemented. This is a strong assumption, considering some factors inherent to Alsace‐Moselle.We address this concern by performing two falsification exercises. The intuition of these falsification exercises is to look at categories of individuals knowingly not affected by the RMI. These individuals should not show differences in the rest of France compared to Alsace‐Moselle, before and after 1989, since they are not affected by the RMI. Any significant difference‐in‐difference would be evidence of different trends between the rest of France and Alsace‐Moselle, violating the common time effects assumption.First, we look at the short‐term unemployment rate. This is defined by the unemployment rate of individuals less than 2 years out of a job. Unemployment insurance in France pays a sizable fraction of the last work income during 2 years. After 2 years, the amount of unemployment insurance significantly drops. Individuals less than 2 years out of a job are thus not likely to qualify for the RMI. Indeed, income has to be less than a mere €464.05 in 2008 to qualify for RMI payments.9Table 2 is exactly the same as Table 1, except for short‐term unemployment rates. One may see that the difference‐in‐difference coefficient is a mere −0.13 percentage point. According to the same bootstrap method, this coefficient is not significant. From this table, one may conclude that nothing significant happened for short‐term unemployed individuals, earning unemployment insurance and thus not eligible for RMI payments, as expected.Table 2. Short‐Term (Less than 2 Years) Average Unemployment Rates in France and Alsace‐Moselle before and after 1989 (More than 25 Years Old) 1982–89(1)1990–2002(2)Difference(3)Rest of France6.989.522.54Alsace‐Moselle5.678.352.68Difference1.311.17−.13Note: The significance of the difference‐in‐differences coefficient is estimated through the bootstrap method. One hundred samples were randomly drawn with replacement from the original sample. The difference‐in‐differences coefficient was thus estimated 100 times. The difference‐in‐differences coefficient was less than zero 77% of the time. According to the bootstrap method, this difference‐in‐differences coefficient is thus insignificantly different from zero. Figures are short‐duration (less than 2 years) average unemployment rates for individuals more than 25 years old. They represent the ratio of the total number of people unemployed for less than 2 years and more than 25 years old divided by the total number of active individuals more than 25 years old in the area concerned and the time frame concerned. Active individuals are defined as the sum of employed individuals and unemployed individuals (and soldiers of the military contingent) more than 25 years old. The numbers in the Difference row and column are first differences. The number in the bottom right corner is a difference‐in‐differences estimate, comparing the rest of France to Alsace‐Moselle, before and after 1989. View Table Image Second, we look at individuals between 16 and 25 years old. The RMI applies only to individuals above 25 years old, whereas aide sociale in Alsace‐Moselle applies to individuals of more than 16 years old. This means that individuals between 16 and 25 years old were never affected by the RMI in France and were always affected by the aide sociale in Alsace‐Moselle. There should be no significant difference‐in‐difference coefficient of unemployment rates for these individuals between the rest of France and Alsace‐Moselle, before and after 1989. One may see from Table 3 that the difference‐in‐difference coefficient is a mere −0.09 percentage point. According to the same bootstrap method, this coefficient is not significant. From this table, one may conclude that nothing significant happened for short‐term unemployed individuals between 16 and 25 years old not affected by the RMI in the rest of France, as expected.Table 3. Long‐Duration (More than 2 Years) Average Unemployment Rates in France and Alsace‐Moselle before and after 1989 (between 16 and 25 Years Old) 1982–89(1)1990–2002(2)Difference(3)Rest of France5.554.52−1.03Alsace‐Moselle3.812.87−.94Difference1.741.65−.09Note: The significance of the difference‐in‐differences coefficient is estimated through the bootstrap method. One hundred samples were randomly drawn with replacement from the original sample. The difference‐in‐differences coefficient was thus estimated 100 times. The difference‐in‐differences coefficient was less than zero 72% of the time. Figures are long‐duration (more than 2 years) average unemployment rates for individuals more than 16 and less than 25 years old. They represent the ratio of the total number of people unemployed for more than 2 years and between 16 and 25 years old divided by the total number of active individuals between 16 and 25 years old in the area concerned and the time frame concerned. Active individuals are defined as the sum of employed individuals and unemployed individuals (and soldiers of the military contingent) between 16 and 25 years old. The numbers in the Difference row and column are first differences. The number in the bottom right corner is a difference‐in‐differences estimate, comparing the rest of France to Alsace‐Moselle, before and after 1989. View Table Image These two falsification exercises confirm that the common time effects assumption might hold. They reinforce the confidence one may have in the main result that the RMI caused unemployment.III. AbsenteeismA. Legal TextsThe general regime in France of paid sick leave was organized by a governmental text from 1945, the Ordonnance n° 45/2454 du 19 octobre 1945.10 In the original text, after a 3‐day initial period during which no indemnity is paid, half of the previous wage of the employee was paid by the social security administration with minor seniority conditions (essentially everyone having worked at least 60 hours in the last 3 months was eligible for the indemnity during 6 months of absence). In 1978, a major law obliged employers to introduce a more generous scheme for people with at least 3 years of seniority within the firm. The supplementary compensation (called indemnisation complémentaire) must be paid after the eleventh day of the sick leave (the social security compensation still starts after the fourth day). And the supplementary compensation makes the total payment at the level of 90% of the previous wage during 30 days. After the 30‐day period, the total payment reverts to 66% of the previous wage.11In contrast, the local law of Alsace‐Moselle guarantees (1) the full salary payment by the employer in case of any absence independent of one’s will (which is a fairly broad definition) and (2) no initial period during which no compensation is given: all days must be paid.12B. QuestionsA natural question is whether the generosity of the sick leave compensation could have generated a surge in absenteeism. The question can be addressed at two levels: Is it the case that in Alsace‐Moselle absenteeism has always been higher? And is it the case that the transition, in France, to a more generous system has encouraged absenteeism relative to Alsace‐Moselle, where no change took place in 1978 since the existing regime was uniformly more generous? Finally, the group of employees with less than 3 years of tenure can be used as an additional control group since they are unaffected by the 1978 law. A last exercise we can provide is to determine whether the existence of an additional seniority right has reduced the incentives to quit the firm.C. Methodology and ResultsClose to the spirit of the previous section, we will present simple averages of absence rates in Alsace‐Moselle and the rest of France, before and after 1978, to illustrate the basic intuition of a difference‐in‐difference analysis. Table 4 presents percentages of workers absent last week because of sickness for individuals with more than 3 years’ tenure (the number of individuals working less than the usual number of hours worked because of sickness divided by the number of employed individuals). For example, the absence rate due to sickness in the rest of France before 1978 (averaged over the period 1976–78) is equal to 4.26%, as seen in row 1, column 1. The numbers in row 3 and column 3 are first differences. For example, the absence rate due to sickness is 0.85 percentage point lower in the rest of France than in Alsace‐Moselle, before 1978.Table 4. Proportion of Employees Ab