The Effects of the Tramway on Mobility and Unemployment in Bordeaux: A Difference-in-Differences Analysis

Transcription

1 The Effects of the Tramway on Mobility and Unemployment in Bordeaux: A Difference-in-Differences Analysis Florent Sari Université Paris-Est, ERUDITE and Tepp-CNRS florent.sari@univ-mlv.fr 5 boulevard Descartes, Marne-la-Vallée CEDEX 2 Preliminary version - October 2012 Abstract In this work, we use a natural experiment to test the link between accessibility to jobs and labor market outcomes for inhabitants of the agglomeration of Bordeaux. We analyze the effects of the implementation of the tramway on unemployment in neighborhoods that are distant from the historical job center. We focus on the A line opened in December This line makes the junction between the two banks of Bordeaux, separated by the Garonne. It was developed in order to reduce economic and social disparities between them. Indeed, a number of neighborhoods and municipalities from the right bank were characterized by a physical disconnection to the agglomeration center, a concentration of social housing and fragile populations on the labor market. On the contrary, the true Bordeaux is located on the left bank and is more developed than the right bank, who also suffers from isolation due to the Garonne and the limited number of bridges or accesses. We use French population census of 1999 and 2006, which correspond to periods before and after the introduction of the tramway in the agglomeration. We use difference-in-differences methods to compare labor market outcomes of inhabitants of some neighborhoods located on the right bank. Some of them benefit from the tramway and others do not. In this case, we can see if increasing job accessibility of some neighborhoods has consequences on their observed unemployment rates. We also compare labor market outcomes of inhabitants of the left bank to those of the right bank. In this case, we can see if the tramway helped to reduce unemployment inequalities between the historical center of the agglomeration and the right bank. Results show that unemployment rate has globally decreased in the agglomeration on the period. However, this decrease appears to be more important for neighborhoods located close to tramway stations. More generally, it seems that the tramway project helped to reduce unemployment inequalities between the two banks of the Garonne. KEYWORDS: Spatial Mismatch Hypothesis; Unemployment; Difference-in-Differences Analysis. JEL CODE: J68; R40. Census data for 1999 and 2006 at the Iris level have been provided by the Centre Maurice-Halbwachs. 1

2 Introduction Since the late 1970 s, unemployment in France reached record levels. Such a situation was favorable to the development of many works in labor economics. On the whole, these works are distinguished by the way they try to explain unemployment. Some emphasize on the role of individual characteristics during the job search process. Some others are mainly focused on the effect of public policies or the role of institutions. A third way links unemployment to spatial organization of a given area. Several mechanisms or hypotheses are often mobilized in order to explain unemployment rate in a region, a town or a neighborhood. A large part of literature in urban economics retains the role of physical disconnection between place of residence and job centers. This phenomenon named as Spatial Mismatch is firstly introduced by Kain, in 1968 [28]. The author argues that being disconnected from jobs (or living far away from them) can have some important consequences on the unemployment process (Kain, 1968 [28], 1992 [29]). On the whole, this literature identifies two broad channels linking the Spatial Mismatch Hypothesis to the bad labor market situations of a part of the inhabitants (Arnott, 1998 [2]). The first channel is given by commuting costs. A physical disconnection between working place and residence place can lead to substantial commuting costs as most suburban locations do not have an appropriate public transportation system. In this case, workers face costs that are often too important in comparison with the salary they are offered (Coulson et al [10]). The second channel is given by different features of the job search process. First, a worker residing far away from job opportunities may encounter some difficulties in obtaining information on jobs (Rogers, 1997 [40]). Simpson (1992 [42]) argues that metropolitan areas consist in a series of "islands" with information about job opportunities, which is free within islands but has a cost among islands. In these conditions, searching a job far away from the residence area can be too costly. Jobseekers search efficiently only in a restricted area, near their residence, even if there are only poor-quality jobs (Davis and Huff, 1972 [12]; Ihlanfeldt and Sjoquist, 1990 [24]; 1991 [25]). In this work, we want to test the link between accessibility to jobs and labor market outcomes, for inhabitants of the agglomeration of Bordeaux, by using a natural experiment. Our approach is somewhat similar to those of Holzer et al. (2003 [23]), as we want to analyze the effect of implementation of the tramway on unemployment in neighborhoods that are distant from the historical job center. The tramway of Bordeaux represents a public transport system made up of three lines (A, B and C) with an overall length of 43.9 kilometers. We focus on the A line opened in December This line makes the junction between the two banks of Bordeaux, separated by the Garonne. It stretches from the center of Bordeaux (Meriadeck station), on the left bank, to the municipalities of Lormont and Cenon, on the right bank, passing through the center of Bordeaux. This line was introduced and developed in order to reduce economic and social disparities between the two banks of the Garonne. Indeed, a number of neighborhoods and municipalities from the right bank were characterized by a physical disconnection to the agglomeration center, a concentration of social housing and a concentration of fragile populations on the labor market. The true Bordeaux is located on the left bank. Historically, this area is more developed than the right bank who also suffers from isolation due to the Garonne and its limited number of bridges or accesses. We use French population census of 1999 and 2006, which correspond to periods before and after 2

3 the introduction of the A line of the tramway in the agglomeration of Bordeaux. In this way, it is possible to see if this project, which may be considered as an urban policy, helped to improve situation on the labor market of a part of the inhabitants and to reduce inequalities between the two banks. We propose to use difference-in-differences methods in order to analyze the effects of the tramway on unemployment. We first compare labor market outcomes of inhabitants of some neighborhoods located on the right bank. Some of them benefit from the introduction of this line, because they have tramway stations, while others do not. By doing this, we can see if increasing job accessibility of some neighborhoods or municipalities has consequences on their observed unemployment rates. We also compare labor market outcomes of inhabitants on the left bank to outcomes of inhabitants on the right bank. In this case, we can see if the tramway helped to reduce unemployment inequalities between the historical center of the agglomeration and the right bank. We distinguish different age categories as we consider that young people are more likely to resort to such means of transportation. Results of our analyzes show that unemployment rate has globally decreased in the agglomeration between 1999 and This decrease appears to be more important for neighborhoods located close to tramway stations. Effects are more important for young people than for others categories. In addition, it appears that the tramway project helped to reduce unemployment inequalities between the two banks of the Garonne. However, results of our analyzes are slightly subject to the parallelism assumption (i.e. in the absence of tramway the close neighborhoods would have known the same decrease of unemployment rate as those that are most distant). The paper is structured as follows: the first section presents a literature review of various empirical works that analyze the links between accessibility to jobs and labor market outcomes. The second section describes the natural experiment and its context. The third section describes the data mobilized, the difference-in-differences method retained and the econometric strategy. Finally, the fourth section presents results of our different estimations. 1 Literature review The Spatial Mismatch hypothesis and, more generally, the effects of the city structure on the local labor market outcomes are tested in many North-American and European studies. Since Kain s pioneering work on the effects of disconnection from jobs in 1968 [28], much empirical research has been conducted 1. Many works analyze to what extent neighborhood s job accessibility can explain employment probability or labor market outcomes of individuals. Some others are focused on the links between individuals mobility and labor market outcomes as they analyze to what extent car ownership or access to public transport can improve job search process. In addition, it appears that some of them confirm the likely adverse effect of living far away from jobs while others do not. The great diversity of data mobilized, population studied, methodologies retained or indicators used to measure job accessibility are all factors that may explain these divergent results. 1 Various papers provide literature review of the Spatial Mismatch Hypothesis. We can cite: Holzer, 1991 [22]; Kain, 1992 [29]; Ihlanfeldt and Sjoquist, 1998 [26]; Gobillon et al., 2007 [20]. 3

4 1.1 Testing for the Spatial Mismatch Hypothesis on survey data In his seminal paper, Kain (1968 [28]) tries to explain the high unemployment rates observed for black people in Chicago and Detroit. Results of his analyze reveal that the bad situation on the labor market of this category may be explained by an unfavorable location of employment. Residential segregation is seen as the principal cause as it tends to move away black people in neighborhoods with poor job accessibility. Ihlanfeldt and Sjoquist (1990 [24]) analyze the links between proximity to jobs center and employment probability for young people, whether they are black or white. Results from data on Philadelphia, Chicago and Los Angeles show that not living far away from jobs has positive effects on young people s employment. On the metropolitan area of Pittsburgh, Rogers (1997 [40]) finds that individuals enjoying a good accessibility to job centers are also those experiencing the shortest unemployment durations. If the author does not distinguish between black and white, he observes that less skilled workers are particularly concerned by this problem. Raphael (1998 [38]) also tries to explain unemployment inequalities in neighborhoods between black and white. His analysis on San Francisco, Oakland and San Jose in 1990 reveals that differences in job accessibility is responsible for 30% to 50% of the observed differential. From data on Chicago, Immergluck (1998 [27]) shows that vicinity to jobs centers is an important determinant of employment rates but is not the only one. The likely mismatch between skills of individuals and skills required by these jobs should be considered too. More recently, Weinberg (2000 [43], 2004 [44]) finds that black residential centralization is an important determinant of black employment status in metropolitan areas. He also finds that job centralization increases black employment rates relative to whites. Bouabdallah et al. (2002 [5]) show, for the French case, that the enlargement of the prospected area for jobs tends to decrease unemployment duration. The increase of job opportunities, allowed by such an enlargement, seems then to overcome the costs of searching further. Martin (2004 [32]) checks if Spatial Mismatch between black residential location and job location impacts on blacks labor market outcomes in U.S metropolitan areas during the period In accordance with previous works, the author shows that unemployment rate has increased in metropolitan areas where jobs are located far away from black neighborhoods. Conversely, it has decreased in areas where black people can move more easily to job centers. Covington (2009 [11]) proposes a different analysis. Using U.S data for the 1990s, his results defend the idea that decentralization of jobs hurts the job access of most workers and especially the poor workers. Indeed, it appears that the poor are much more physically isolated from employment than are the non-poor. Finally, Matas et al. (2010 [33]) have analyzed the impact of residential job accessibility on female employment probability in the metropolitan areas of Madrid and Barcelona. Their findings support the hypothesis that job accessibility by public transport is a significant determinant of female employment probability in both areas. Another part of the literature is interested in the links between individuals mobility and their labor market outcomes. In line with the contribution of Matas et al. (2010 [33]), the idea is to check if a better mobility induces the chances of finding a job. First, Ong (1996 [34], 2002 [35]) analyzes the role of car ownership on probability of finding a job for welfare recipients in California. His results reveal an important and significant positive effect. The underlying idea is that car must enable individuals to search for jobs in distant places or encourage to accept job offers located far away from home. Similarly, using a large nationally representative sample of individuals for 1992 and 1993, Raphael 4

5 and Rice (2002 [39]) find important differences in employment rates, hourly earnings or weekly hours worked between those with and without car. Kawabata (2003 [30]) examines the inverse relationship, i.e the links between access to jobs and labor market performances for individuals without car. Results for metropolitan areas of Boston, San Francisco and Los Angeles show that physical connection to job centers increases significantly employment probability and weekly hours worked. Effects are less strong for Boston because the public transport system is more developed than in the two others areas. On French data, Cavaco and Lesueur (2004 [8]) demonstrate that car ownership has positive effects on decision of mobility and that being mobile decreases unemployment duration. In 2005, Gurley and Bruce (2005 [21]) also analyze the fact of having access to car for welfare recipients but with a distinction between urban and rural. They still find that access to car increases the chances of finding a job and leaving welfare. Finally, Patacchini and Zenou (2005 [36]) show, with a theoretical model and an empirical analysis on English data, that individuals living in areas with important commuting times search less intensively a job than others. In addition, they find that access to car can increase job search intensity and so strengthen or weaken the previous effect. 1.2 Natural experiments on Spatial Mismatch Hypothesis In this work, we are in line with previous empirical works as we want to analyze the effect of having access to a new public transport system on labor market outcomes, for inhabitants of neighborhoods initially located far away from job center. The original contribution of our paper is to use a natural experiment in order to identify the causal effect. To our knowledge, very few papers are using natural experiments when testing the Spatial Mismatch hypothesis. We can cite the work of Zax and Kain (1996 [46]) who analyze the effects of a relocation of a large company in the service industry from the central business district of Detroit to the suburb on individuals mobility and employment rate. Results show that white employees whose commutes have been extended were more likely to move. On the contrary, black employees in the same situation were more likely to both move and quit employment. The authors explain this phenomenon by the restrictions on black residential choice imposed by segregation on housing market. Boustan and Margo (2009 [6]) focus on one large employer that has remained in the central city during decades: the U.S Postal Service. As this employer is historically remained in the center while others have tended to relocate, then we should observe an increase of the percentage of black employees in the total number of employees 2. They effectively find that black people choose postal work as other employment opportunities leave the central city from the 1960s to today. Nevertheless, they believe that Spatial Mismatch has become less important over time as they find a weak relationship between black postal employment and segregation for the recent period. Aslund et al. (2010 [3]) use a Swedish refugee dispersal policy in order to have an exogenous variation in individual locations 3. From their sample, they find that the fact of having been placed in a location with a poor job access plays negatively on labor market outcomes of individuals (it decreases employment probability and in a lesser extent annual earnings). The natural experiment proposed in this paper can be compared to the one of Holzer et al. (2003 [23]). The authors use a new expansion of the heavy rail system in the San Francisco Bay Area as an 2 This phenomenon is observable only if it exists a problem of physical connection to jobs for black people in the center of cities. It may be particularly true for black neighborhoods experiencing strong segregation. 3 Dealing with endogeneity of residential location is a recurrent problem for empirical works in urban economics. 5

6 exogenous change in the accessibility of inner-city minority communities to the job center. They propose to compare within-firm changes in the propensity to hire minority workers for firms located near the station to those located farther away. The comparison is made before and after the opening of the service. However, our contribution is very different from their own, especially for one point: while the authors propose to evaluate effects of the expansion of the rail system on changes in recruitment of firms, we propose to evaluate the effects of the implementation of the tramway on unemployment rates in neighborhoods. Results of their analysis show significant and positive effects for Latinos residents but little effects for black residents. These different empirical works, in an American or European context, have demonstrated the necessity to consider the problem of physical disconnection to jobs for minority communities. As they are often subject to residential segregation, they are generally located in isolated and distant areas. This situation may be considered as a major explanation of their adverse labor market outcomes. Available data in France do not permit to test easily the Spatial Mismatch hypothesis for population of different origins. For this reason, we propose to analyze the links between accessibility to jobs and unemployment for the whole population (without distinction by origin) 4. 2 The Natural experiment In this section, we briefly present the agglomeration of Bordeaux and the tramway project which is central in our analysis. Then, we present how we use this project to propose a natural experiment in order to evaluate the links between accessibility to jobs and unemployment. 2.1 The context The Urban Community of Bordeaux The natural experiment retained in this work takes place in the Urban Community of Bordeaux (CUB) located in the department of Gironde. The CUB is a inter-municipal structure that gathers 27 towns of the agglomeration of Bordeaux and is the fourth biggest in France with inhabitants (in 2007). The city of Bordeaux, considered alone, represents inhabitants. Two municipalities, Merignac and Pessac, have more than inhabitants and eight have more inhabitants. The FIGURE 1 provides an illustration of the extent of the CUB. In this work, we do not retain the whole Urban Community because the tramway project concerns a limited number of municipalities. In blue are represented the municipalities that form our study sample. Some of them are directly concerned by the implementation of the tramway while others are just neighbors (see APPENDIX 1 for the different lines of the tramway in Bordeaux). The FIGURE 1 shows clearly that the Garonne pass trough Bordeaux. It separates the historical center of Bordeaux, on the left bank, to others neighborhoods 5 and some municipalities (Floirac, Cenon, Lormont and Bassens) located on the right bank. The historical center of the city represents a true economic center with more than employment, which represents 20% of the city s employment. More generally, the left bank hosts not only administrative services, but also a large shopping 4 In addition, Wenglenski (2006 [45]) and Baccaïni (1996 [4]) have shown, for the French case, that accessibility to job centers is rather a problem of occupational status than a problem of ethnicity. 5 This part of the city is known as La Bastide. 6

7 Figure 1: The agglomeration of Bordeaux complex or sports and cultural equipments. Moreover, this part of the city concentrates a large number of favorable neighborhoods with important parts of upper and middle classes. In a recent paper, Decamps (2011 [13]) observes, for the 1990s, a gentrification process (characterized by the concentration of wealthy people in a limited number of central neighborhoods) and a location choice based on the attractiveness of the main employment centers in Bordeaux. These phenomena are partly responsible of the imbalance observed between the two banks of the Garonne. Indeed, the right bank (a part of Bordeaux and the neighboring municipalities) historically hosts lower classes and more fragile populations. Contrary to the left bank, the presence of social housing and many industrial wasteland tends to affect the image of this part of the agglomeration The tramway project In 1995, the mayor of Bordeaux -Alain Juppé- launches a new tramway project in the agglomeration of Bordeaux 6. The project is adopted by the Urban Community of Bordeaux in It has different objectives. First, to dynamize the heart of the agglomeration by facilitating access of the city to the neighboring municipalities. Second, to modernize and increase the living standards in neighborhoods concerned by the project. Third, to support the economic integration of a large number of municipalities on the right bank. Concerning the last point, the tramway project may be considered as an urban policy tool. In practice, the restoration of the connection between social housing neighborhoods and the city (via the tramway) is jointed to an urban renewal project which aims to mitigate the importance of residential buildings and to introduce social mix. The project is developed in two phases. The first phase begins in 2000 and finishes in September The second phase begins in 2006 and finishes in October The first phase involves the construction of three lines, that meet each in the city center, with a total length of 24.5 kilometers 6 The tramway already existed in Bordeaux in the first part of the twentieth century. In 1957, the mayor Jacques Chaban- Delmas decided to stop it because it was considered as obsolete and source of traffic jam for inhabitants. 7

8 and 54 stations. The first line put in service is the A line in December The two others lines are opened in spring At the end of the first phase, the A line runs from east to west and crosses the Garonne on the old bridge. It expands from the center of Bordeaux (Meriadeck station), on the left bank, to the right bank where it splits in two. It goes to Cenon (La Morlette station) in the south and to Lormont (Lauriers station) in the north. B line expands from Quinconques station, in the center of Bordeaux, to Pessac (Bougnard station) in the south-west. Finally, the C line expands from Quinconques station to Saint-Jean station (high speed train). The total cost of this first phase is 690 million, with an important part for urban planning. The second phase involves the construction of extensions to the three lines (19.6 kilometers) for a total cost of 560 million. The A line is extended to Floirac and Carbon-Blanc on the right bank. The B line is extended from Pessac center to Claveau station in the north of Bordeaux. The C line is extended from Le Bouscat to Bègles. Implementation of these lines appears to be a success in terms of daily commutes: each day there are travelers on the A line, on the B line and on the C line. Our analyze is focused on the A line because it is the only one that permits a junction between the two banks of the Garonne. It was a clear objective of authorities to stimulate development of neighborhoods on the right bank, that have always been disadvantaged by the lack of connections with the city center 7. Even today, the spatial organization of Bordeaux reveals that only two bridges (the Stone Bridge and the St Jean Bridge) enable an easy access, for inhabitants of the right bank, to the center of Bordeaux (see APPENDIX 1). This new public transport system makes travels easier between the two banks in two ways: it relieves the traffic congestion by proposing an alternative to travels by private transportation or bus; It permits to individuals, without private means of transport, to travel more cheaply. Concretely, it seems that the connection, facilitated by the tramway, changes the relationship between the two banks of the Garonne. It makes the right bank a possible place of attraction for inhabitants of Bordeaux thanks to green spaces it offers or cultural facilities that have been installed. The opposite seems equally effective as the tramway provides an easy access to the city to residents of municipalities located on the right bank. It is at least a total of inhabitants (inhabitants of Floirac, Cenon, Lormont and Bassens) who no longer feel isolated. In addition, some recent studies have shown that implementation of the tramway had effects on economic activity (see Putz and Dumas, 2009 [37]) and housing prices of the neighborhoods concerned. Between 1999 and 2009, approximatively establishments have been created in the Urban Community of Bordeaux. 436 have been created in the corridor of the tramway (for the authors of the study, the corridor represents all establishments located on the route of the tramway or in a radius of 150 meters around this route). The right bank may be considered as a direct beneficiary of this project. Indeed, if we consider the creation of service establishments, this area is the most dynamic. It has increased by 49% against 40,5% for the whole Urban Community. Concerning the housing prices, the implementation of the tramway on the right bank led to an increase between 40% and 50%. However, prices for municipalities on the right bank are still lower than in the city of Bordeaux 8. In this work, we are interested by the potential links between implementation of the tramway and 7 There is an only one bridge built in the early nineteenth century in Bordeaux. 8 See S. Pineau (2009): "Le tramway, un atout majeur pour le développement économique", venirabordeaux.com/le-tramway-un-atout-majeur-pour-le. 8

9 unemployment on the right bank. We want to answer to the following question: Is the tramway a way to reduce unemployment rate in neighborhoods close to a station? In the next subsection, we present the strategy retained in order to answer to this question. 2.2 The strategy In order to assess the effects of the tramway on unemployment for inhabitants concerned, we have to consider their situation before and after its implementation. As we need a comparison point, we replicate the same approach for inhabitants not directly concerned by the implementation of the tramway (the econometric strategy is presented in details in further section). Our analysis is made at two different scales. First, we focus on municipalities and neighborhoods located on the right bank. This area is historically and traditionally prone to isolation and the A line of the tramway is the principal measure of urban policy developed to settle this matter. For this reason, we retain a selection of municipalities of the right bank. In these municipalities, some neighborhoods have a station of the tramway or are very close to one of these stations. We will consider all these neighborhoods as treated. All others neighborhoods are more distant from the stations and so they are not considered as directly affected by the implementation of the tramway. This group of neighborhoods is name as control group. The FIGURE 2 provides an illustration of our first study sample. The exhaustive list of neighborhoods in each of these groups is given and represented in the APPENDIX 2.1. By doing this, we can compare neighborhoods that are all characterized by problems of isolation and poor accessibility to the center of the agglomeration, but some of them have benefited from the tramway while others do not have. We then compare 24 neighborhoods with the tramway (treated) to 20 neighborhoods without the tramway (controls) before and after its introduction 9. Figure 2: Study sample 1 - Neighborhoods on the right bank Second, we focus on neighborhoods and municipalities located on both banks of the Garonne. In this case, we decide to compare neighborhoods on the right bank (whether they are close or not to a tramway station) to others on the left bank (see FIGURE 3). Our objective is to check if the tramway has helped to reduce inequalities between the two banks. We consider that all neighborhoods of the right 9 Our data only permits to assess the effects of the first phase of the project. Indeed, the second phase, that begins in 2006 and finishes in October 2008, took place after the last census available to us (2006 population census). 9

10 bank are impacted by the tramway. This treated group contains 35 neighborhoods. It is opposed to the control group, that contains 96 neighborhoods from Bordeaux and some adjacent municipalities, before and after the introduction of the tramway. The list of neighborhoods for this second sample is given in the APPENDIX 2.2. A first limit of this strategy may be due to the fact that the left bank has also benefited from the implementation of the tramway on the period considered. It is the case for the A, B and C lines. However, neighborhoods located on this part of the city did not suffer from physical disconnection to the center, caused by the Garonne and the limited number of bridges or accesses. For these reasons, it is likely that the effects of the tramway on unemployment are weakened for these locations. Figure 3: Study sample 2 - Left and right banks Finally, our objective is double. On one hand, we compare labor market outcomes of inhabitants of some neighborhoods located on the right bank. Some of them benefit from the tramway while others do not. In this case, we can see if increasing job accessibility of some neighborhoods with the tramway has consequences on their observed unemployment rates. On the other hand, we compare labor market outcomes of inhabitants of the left bank to those of the right bank. In this case, we can see if the tramway helped to reduce unemployment inequalities between the historical center of the agglomeration and the right bank (neighborhood of La Bastide, municipalities of Bassens, Cenon, Lormont and Floirac). 3 Data and methods After a brief presentation of data mobilized and the difference-in-differences method, we describe the economic strategy retained in order to analyze the links between implementation of the tramway and unemployment. 3.1 Data In this work, we use the 1999 and the 2006 population census at the Iris code. The Iris code is used as a proxy of the neighborhood. It may represent a municipality or a subdivision of a municipality if the latter has more than inhabitants. The Iris code represents homogeneous spatial subdivisions in terms of housing and population with an average of inhabitants. The population census was 10

11 conducted by the French national institute of statistical and economic information (called INSEE). The population census is an exhaustive dataset giving information on all the Iris codes. It provides statistics on the types of housing, on the housing conditions, types of households, residents as well as a number of their socio-economic characteristics (gender, age, qualification, employment status, occupational status etc.). We retain two different periods for the population census because it corresponds to periods before and after the implementation of the tramway in Bordeaux (see previous section). As the A line was opened in , the 1999 population census permits to have an overview of the situation in the neighborhoods four years before. It permits to check the labor market situation of the neighborhoods considered but also to check if they are similar on their observed characteristics (one of the condition required for difference-in-differences methods). The 2006 population census is used in order to have an overview of the labor market situation three years later. It corresponds to a period long enough to observe effects of the tramway on inhabitants. Finally, we have information for the 147 neighborhoods that compose our study samples and for both periods. 3.2 Difference-in-differences method Description In order to analyze the relationship between the implementation of the tramway and labor market outcomes, we retain a difference-in-differences method. This popular method in non-experimental evaluations estimates the difference in the outcomes during a post-intervention period between a treatment group and a control group relative to the outcomes observed during the pre-intervention period. Concretely, the impact of implementation of the tramway on unemployment rate can be estimated by computing a double differences, one over time (before-after the implementation) and one across subjects (neighborhoods with the tramway and neighborhoods without the tramway). If we note t = 0 the period before the project implementation, t = 1 the period after the project implementation, Yt T and Yt C the respective outcomes for neighborhood with (treated) and without the tramway (controls) in time t, then the difference-in-differences method will estimate the average effect as follows: E (Y T 1 Y T 0 T 1 = 1) E (Y C 1 Y C 0 T 1 = 0) (1) where T 1 = 1 denotes treatment or the presence of the tramway in the neighborhood and T 1 = 0 denotes controls neighborhoods at t = 1. Subtracting the pre-intervention difference in outcomes from the post-intervention difference allows to eliminate one particular selection bias. In fact, it may exist a problem of unobserved heterogeneity that makes beneficiaries of an intervention different from non beneficiaries, even in its absence. Some unobserved factors may explain the differences in mean outcomes between the two groups, such as motivation, personality, innate ability etc. Difference-in-differences can control for this selection bias under the condition that this unobserved heterogeneity is time invariant. In this 10 We only consider the first phase of the project of tramway. As described previously, the second phase begins in 2006 and finishes in October So we do not have data recent enough to assess effects of this phase. 11

12 particular case, subtracting the pre-intervention differences cancels the bias and produces a plausible estimate of the impact of the project of tramway. The following figure (FIGURE 4) provides an illustration of the difference-in-differences method applied in this work. The horizontal and vertical axes represent time and unemployment rate, respectively. For each line, there are two points, one before (in 1999) and one after (in 2006) the tramway project is implemented. These points represent averages: two are taken in 1999 and two in 2006, respectively among the neighborhoods concerned by the tramway, and among neighborhoods not concerned by this project. The difference observed in 2006 is not the effect of the project because the difference may be caused by a selection process. Indeed, it was precisely the objective of the project to fight against adverse situation of some neighborhoods. We observe that in 2006 the difference in the unemployment rate between the two groups of neighborhoods is smaller than in It is then the reduction in the unemployment rate gap that can be considered as the effect of the project. However, the validity of the difference-in-differences method is based on an important assumption. Unobserved characteristics that create the gap between observed labor market outcomes of the two groups of neighborhoods are assumed to be time invariant. The gap between the two trends should be the same over the period. It is represented on the FIGURE 4 by the dotted line parallel to the trend observed for neighborhoods not concerned by the tramway. This line represents the counterfactual. That is to say the estimate of what would have been the unemployment rate in neighborhoods with the tramway if this project had not been developed. Figure 4: Difference-in-differences method Source: Population census 1999 and 2006, INSEE Implementation Using regression to estimate difference-in-differences Instead of taking the difference of the outcomes, we calculate difference-in-differences estimate through a regression model. Even if this methodology is less intuitive, it provides the advantage to 12

13 permit the relaxing of the assumption of "parallelism" (described in previous sub-section) by adding other covariates that may affect the outcome of interest. Such an estimation strategy permits to control for changes in variables across the years and to obtain the net effect of the project implementation on the outcome. First, the following regression equation is considered and estimated by ordinary least squares: Y i,t = α + β T i + γp t + δt i P t + ε i,t (2) where: Y i,t = unemployment rate in period t for neighborhood i. It is calculated for the years old but also for different age groups. More precisely, we consider unemployment rate for the years old, for the years old and for the years old. We believe that the use of public transport may differ according the age of individuals 11. T i = 1 if neighborhood i is concerned by the tramway; 0 otherwise. P t = 1 for the year 2006; 0 for the year T i P t = 1 only in 2006 if neighborhood i is concerned by the tramway. This interaction term is the product of the two binary variables and represents the actual treatment variable. ε i,t is the error term of the regression with variance σ 2. α, β, γ and δ are the regression parameters to be estimated. δ gives the average effect of the tramway on unemployment. An important condition must hold in order to interpret the difference-in-differences estimator, i.e the error term must be uncorrelated with the other variables in the equation: Cov (ε i,t, T i ) = 0 Cov (ε i,t, P t ) = 0 Cov (ε i,t, T i P t ) = 0 The last of these assumptions is the parallelism assumption. It means that unobserved characteristics between the two groups of neighborhoods (treated and controls) do not vary over time. In our case, it is likely that some of these assumptions are not verified. For this reason, we resort to another estimation method in order to obtain more robust results. Fixed-effects regression In a second step, we use a fixed-effects regression instead of ordinary least squares. Fixed-effects regression permits to control for neighborhoods unobserved and time-invariant characteristics that may influence the outcome of interest. We then estimate the following equation: 11 A study of Amrane (2000 [1]), on the agglomeration of Bordeaux, tends to confirm this hypothesis. It shows that the years old give priority to car for daily commutes while public transports are more often privileged by young or old people. 13

14 Y i,t = α + β T i P t + µ i + ν t + ε i,t (3) where: µ i = neighborhood fixed-effect. ν t = year fixed-effect. The others variables of the regression (Y i,t, T i P t and ε i,t ) are the same as defined above. As the previous section has shown, the project of tramway studied is not a random intervention. Indeed, an important part of the A line concerns neighborhoods with high unemployment rate and the period considered is likely to have an influence on this labor market outcomes. These phenomena are captured by the inclusion of neighborhood and year fixed-effects. Then, it allows covariance between µ i and T i P t and between ν i and T i P t. In addition, neighborhood fixed-effects (µ i ) capture differences across neighborhoods that are constant over time while year fixed-effects (ν i ) capture differences over time that are common to all neighborhoods. In a final step, the fixed-effects model is extended by adding other covariates in the regression. It permits to control for some observed characteristics that vary over time and that may have impact on the observed outcomes. We consider the following model: Y i,t = α + β 1 T i P t + β 2 X i,t + µ i + ν t + ε i,t (4) where X i,t is a vector of time-varying covariates including characteristics of the neighborhood or of the municipality in which it is located. Concerning the neighborhood, we retain information relating to demographic characteristics (the percentages of women, years old and single parents families), to socio-economic characteristics (the percentage of individuals without diploma and the average income) and to inhabitants mobility (the percentage of households with car). Concerning the municipality, we retain information relating to employment characteristics (the total employment in the municipality) 12. APPENDIX 3 shows an exhaustive list of the variables used in the estimations and provides some descriptive statistics. 4 Results We first present some descriptive statistics on the Urban community of Bordeaux and on neighborhoods concerned by the A line of the tramway. Then, we present results of the different estimations. 4.1 Descriptive statistics Unemployment disparities Interest for the agglomeration of Bordeaux is justified by the fact that it shows contrasts in spatial organization. For example, some neighborhoods of the agglomeration show concentrations of fragile 12 Data on employment at the Iris level are not available for years 1999 and This is the reason why this information is at the municipality level. 14

15 populations, whereas others are characterized by compactness of more healthy populations. To show this phenomenon, the percentage of unemployed workers in each neighborhood is mapped for 1999 (see FIGURE 5). Not surprisingly, it appears that the highest unemployment rates are concentrated in neighborhoods on the right bank. A large majority of neighborhoods of Floirac, Cenon, Lormont and Bassens have unemployment rates above 19.8%. On the contrary, many neighborhoods in the west of Bordeaux and municipalities of Bruges and Le Bouscat show relatively low unemployment rates (below 12.1%). This first map confirms that there are important inequalities between the two banks in 1999 and demonstrates the need of urban policies as the tramway project. Figure 5: Unemployment rate in 1999 Source: Population census 1999, INSEE. Figure 6: Unemployment rate in 2006 Source: Population census 2006, INSEE. 15

16 This observed opposition between labor market outcomes of the right and left banks seems still valid in 2006, although somewhat less marked. Indeed, FIGURE 6 shows that many neighborhoods of Bordeaux, located on the left bank, have now unemployment rates above 13.8%, such as neighborhoods of the municipalities on the right bank previously cited. Disparities between these neighborhoods appear less strong than before. In addition, even though unemployment has globally decreases over the period , the map gives the impression that situation of some neighborhoods concerned by the tramway has improved relative to others. However, such an assumption needs to be verified by others statistics or by a regression framework Neighborhoods on the right bank One of the objectives of this work is to check if the tramway project has an effect on labor market outcomes of inhabitants concerned. A simple comparison of unemployment rates in 1999 and 2006 for neighborhoods with and without the tramway may be a first indication. TABLE 1 shows differences in labor market outcomes and in some socio-economic characteristics in 1999 for these two groups of neighborhoods. We observe that neighborhoods in which the project must be developed have globally more unfavorable characteristics than others. Unemployment rates are higher (23.5% against 15.6% on average), whatever the age category. The observed rates are significant at the 5% level. Findings differ when we compare the situation in 2006 (see TABLE 2). We observe a downward trend of unemployment. It is now 15.5% in neighborhoods with tramway and 12.2% in others. These results show that if unemployment rate has decreased for all neighborhoods on the period considered, the decrease is more important for neighborhoods concerned by the project of tramway. It is likely that it has permitted an adjustment between the two groups of neighborhoods. Table 1: Neighborhoods on the right bank in 1999 without Tram with Tram Variables Mean Std Err Mean Std Err Unemployment rate*** 15,6% 0,057 23,5% 0,071 Unemployment rate (15-24 year olds)** 32,4% 0,089 38,3% 0,058 Unemployment rate (25-54 year olds)*** 13,8% 0,063 21,8% 0,072 Unemployment rate (55-64 year olds)** 15,8% 0,061 20,6% 0,087 % of inactive 29,4% 0,020 29,2% 0,029 % of active 71,5% 0,020 72,1% 0,029 % of women* 50,7% 0,014 51,8% 0,021 % of year olds 19,9% 0,030 20,9% 0,028 % of single parents families*** 12,3% 0,066 20,0% 0,074 % of individuals without diploma*** 17,8% 0,064 26,8% 0,067 % of individuals with a 2nd year univ. level*** 7,6% 0,027 5,7% 0,023 % of executives** 8,2% 0,051 4,2% 0,030 % of employees 14,8% 0,049 15,5% 0,058 % of blue-collar workers 23,9% 0,071 25,7% 0,077 % of social housing* 23,8% 0,252 44,2% 0,375 % of tenants 11,9% 0,069 16,9% 0,147 % of home-owners** 59,9% 0,223 36,0% 0,279 % of indiv. using public transp. to go to workplace*** 8,3% 0,051 17,7% 0,044 % of households with a car*** 88,6% 0,099 74,6% 0,078 % of workers in the same town** 19,4% 0,066 30,9% 0,176 Neighborhoods Notes: ***, ** and * denote respectively significance at the 1%, 5% and 10% level for the test for the equality of the means. Source: Population census 1999, INSEE. 16

17 However, the better situation observed may be due to a change in the socio-economic composition of the neighborhoods on the right bank. If neighborhoods have characteristics that differ before and after the implementation of the project and if they have an effect on the labor market outcomes observed, it may be difficult to believe to a causal relationship between the project and the outcome. Concretely, significant differences are observed in 1999 for the diploma level (see TABLE 1). The percentage of individuals without diploma is more important in neighborhoods concerned by the project (26.8% against 17.8%) while percentage of individuals with a second year university level are lower. Concerning occupational status, percentage of executives is also significantly lower in this type of neighborhoods. It tends to show that neighborhoods without the tramway gather population relatively more wealthy. The percentage of home-owners, more important in these neighborhoods (60% against 36%), also seems to confirm this idea. Finally, percentage of individuals using public transport to go to work place and percentage of households with a car suggest that inhabitants of the neighborhoods concerned by the tramway are less mobile and are dependent of the existing public transport structures (respectively 17.7% and 74.6% against 8.3% and 88.6%). Table 2: Neighborhoods on the right bank in 2006 without Tram with Tram Variables Mean Std Err Mean Std Err Unemployment rate* 12,2% 0,058 15,5% 0,061 Unemployment rate (15-24 year olds) 27,4% 0,102 25,9% 0,103 Unemployment rate (25-54 year olds)* 10,7% 0,054 14,0% 0,054 Unemployment rate (55-64 year olds)** 6,9% 0,045 12,4% 0,081 % of inactive** 27,7% 0,023 32,3% 0,054 % of active** 72,3% 0,023 65,4% 0,055 % of women 51,2% 0,019 52,0% 0,025 % of year olds** 18,1% 0,032 20,8% 0,037 % of single parents families** 14,6% 0,072 20,8% 0,071 % of individuals without diploma** 19,5% 0,080 26,8% 0,096 % of individuals with a 2nd year univ. Level** 9,9% 0,033 7,6% 0,037 % of executives* 8,8% 0,050 6,2% 0,041 % of employees 14,1% 0,043 15,2% 0,050 % of blue-collar workers 19,4% 0,053 20,9% 0,048 % of social housing* 23,8% 0,236 42,3% 0,365 % of tenants 12,7% 0,068 17,1% 0,144 % of home-owners** 60,8% 0,214 38,9% 0,284 % of indiv. using public transp. to go to workplace*** 12,2% 0,055 22,4% 0,063 % of households with a car*** 87,0% 0,092 77,2% 0,086 % of workers in the same town** 18,9% 0,070 31,1% 0,155 Neighborhoods Notes: ***, ** and * denote respectively significance at the 1%, 5% and 10% level for the test for the equality of the means. Source: Population census 2006, INSEE. Some of the differences observed in 1999 disappear in On the period, the percentages of executives have not evolved a lot in neighborhoods without tramway, while they have increased in others neighborhoods. Differences in terms of educational level seem less strong than before: the percentage of individual with a second year university level is more important in 2006 than in 1999 for neighborhoods with the tramway (7.6% against 5.7%). We can also suppose that the tramway has modified the behavior of inhabitants for their daily commutes, as the percentage of those using public transport to go to work place is even higher (22.4% in 2006 against 17.7% in 1999). The fact that socio-economic characteristics have changed over time in the two groups of neighborhoods could explain the differences observed for labor market outcomes, before and after the 17

18 implementation of the project. For example, if unemployment has decreased in some neighborhoods it may be due to the fact their composition has become more favorable. If the standard procedure of difference-in-differences permits to control for time-invariant characteristics, it is not the case for time-variant characteristics. These some descriptive statistics on the right bank confirm the necessity to use regressions with control variables in order to estimate difference-in-differences The opposition between right and left banks In this subsection we consider the differences in terms of socio-economic characteristics between the two banks of the Garonne. Data for 1999 and 2006 clearly show that population on the left bank is very different from the right bank. In particular, we observe that the left bank (that includes Bordeaux but also Le Buscat and Bruges) is characterized by more important percentage of individuals with a second year university level. For both years differences is significant at the 1% level. Similarly, we observe that inhabitants of the left bank are more often belonging to occupational status as executives while blue-collar workers are more present on the right bank. Even though inequalities decrease between 1999 and 2006, they remain significant. A recent study of Béoutis et al. (2012 [7]) confirms the disadvantages of the right bank. In particular, it shows that reported incomes are weak or very weak on the right bank while they are strong or very strong on the other side. Concerning residential status, we do not observe very significant difference for the percentage of home-owners. Percentages are equivalent for both banks and both periods. However, while the left bank is characterized by an important part of tenants, the right bank is characterized by an important part of social housing. Once again, these findings are valid for both periods. Concerning the mobility of inhabitants, we note some differences between 1999 and 2006 that may be due to the implementation of the tramway in Bordeaux. Indeed, it appears that the percentage of individuals using public transport to go to work place is stronger in 2006 than in 1999 for both banks. The percentage is 13.2% in 1999 and 17.8% in 2006 for the left bank. The increase is similar for the right bank since it varies from 15.7% to 19.9%. The tramway, developed in a very large part of the Urban Community of Bordeaux, is potentially a major explanation of these changes in daily commutes. The tramway is seen as a modern means of transport that permits to avoid commutes by car and that is more practical than bus, for example. The evolution of the percentage of workers in the same town (of neighborhood where they live) is more difficult to interpret. If this percentage has decreased on the left bank (from 58.8% to 55.7%), it remained constant on the right bank (approximatively 28.4%). A decrease could reflect the fact that tramway permits to have an employment far away from home more easily. Inhabitants of Bordeaux can now find an employment in neighboring municipalities. One would expect that this percentage also decreases on the right bank because, with the tramway, inhabitants can leave their municipality of residence in order to go to the job center of Bordeaux. Nevertheless, the unobserved decrease on the right bank may be explained by the fact the implementation of the tramway has been accompanied by a development of the number of firms locally. A report of the CERTU 13 (2005 [9]) shows that perspective to benefit from the tramway motivates firms to locate near the future stations. This is the case for Bordeaux but also for others French agglomerations that have experienced the implementation of the tramway. 13 The CERTU is the Center for the Study of Urban Planning, Transport and Public Facilities. It is responsible for carrying out studies on behalf of the French State. 18

19 Table 3: Neighborhoods on the left and right banks in 1999 Bordeaux s neighborhoods Left bank Right bank Variables Mean Std Err Mean Std Err Unemployment rate* 18,3% 0,090 22,2% 0,068 Unemployment rate (15-24 year olds)*** 29,9% 0,110 37,7% 0,070 Unemployment rate (25-54 year olds) 17,5% 0,086 20,3% 0,070 Unemployment rate (55-64 year olds)*** 14,2% 0,079 20,1% 0,076 % of inactive*** 33,5% 0,101 29,1% 0,027 % of active** 66,5% 0,101 70,9% 0,027 % of women** 52,6% 0,024 51,4% 0,019 % of year olds 24,6% 0,097 20,9% 0,029 % of single parents families 17,2% 0,028 18,5% 0,036 % of individuals without diploma*** 14,7% 0,111 25,2% 0,066 % of individuals with a 2nd year univ. level*** 12,8% 0,037 5,8% 0,024 % of executives*** 13,2% 0,069 4,5% 0,031 % of employees* 13,5% 0,042 15,8% 0,050 % of blue-collar workers 10,8% 0,064 26,5% 0,070 % of social housing*** 15,1% 0,224 41,3% 0,337 % of tenants*** 46,6% 0,231 15,3% 0,132 % of home-owners* 34,4% 0,181 39,7% 0,265 % of indiv. using public transp. to go to workplace* 13,2% 0,056 15,7% 0,054 % of households with a car*** 69,0% 0,140 77,7% 0,102 % of workers in the same town*** 58,8% 0,173 28,3% 0,152 Neighborhoods Notes: ***, ** and * denote respectively significance at the 1%, 5% and 10% level for the test for the equality of the means. Source: Population census 1999, INSEE. Table 4: Neighborhoods on the left and right banks in 2006 Neighborhoods on the without Tram with Tram Variables Mean Std Err Mean Std Err Unemployment rate 13,5% 0,062 15,3% 0,062 Unemployment rate (15-24 year olds)** 20,8% 0,089 27,8% 0,105 Unemployment rate (25-54 year olds) 13,5% 0,063 13,9% 0,056 Unemployment rate (55-64 year olds)** 8,7% 0,065 10,8% 0,076 % of inactive 30,8% 0,081 31,1% 0,051 % of active 69,2% 0,081 68,9% 0,053 % of women 52,4% 0,030 51,7% 0,024 % of year olds*** 25,9% 0,111 20,4% 0,036 % of single parents families 18,2% 0,065 19,8% 0,069 % of individuals without diploma*** 12,7% 0,077 26,1% 0,090 % of individuals with a 2nd year univ. level*** 14,6% 0,040 7,8% 0,034 % of executives*** 15,1% 0,065 6,3% 0,040 % of employees 14,5% 0,045 15,2% 0,047 % of blue-collar workers*** 11,4% 0,100 21,3% 0,044 % of social housing*** 14,4% 0,220 40,3% 0,329 % of tenants*** 42,7% 0,216 15,7% 0,130 % of home-owners 39,9% 0,210 42,4% 0,265 % of indiv. using public transp. to go to workplace*** 17,8% 0,080 19,9% 0,069 % of households with a car** 70,6% 0,156 78,8% 0,089 % of workers in the same town*** 55,7% 0,143 28,4% 0,142 Neighborhoods Notes: ***, ** and * denote respectively significance at the 1%, 5% and 10% level for the test for the equality of the means. Source: Population census 2006, INSEE. Finally, we observe a phenomenon already highlighted for neighborhoods of the right bank: a global decrease of the unemployment rate between 1999 and If unemployment rate is still lower 19

20 on the left bank than on the right bank, the fact remains that the decline is greater for the latter. It is 18.3% in 1999 and 13.5% in 2006 for the left bank while it is 22.2% in 1999 and 15.3% in 2006 for the right bank. In these conditions, this decline could be explained by the implementation of the A line on the right bank that ameliorate physical connection to job centers; But it could also be explained by the urban development that has accompanied the project of tramway; Or it could be explained by a more favorable socio-economic composition of the neighborhoods on the right bank consecutive to the rise of housing prices (see Pineau, 2009 op. cit.). The next sub-section proposes regressions in order to explain the evolution of unemployment rate observed on the whole period and in order to to disentangle these different effects. 4.2 Regression results Effects of the tramway on unemployment rate for neighborhoods on the right bank The following table present the effects of the tramway on unemployment rate (TABLE 5) and on unemployment rate of different age categories (TABLES 6, 7 and 8) for neighborhoods all located on the right bank of Bordeaux. Model 1 shows results for a simple OLS regression without any control variables. First of all, we observe that neighborhoods concerned by the tramway project are those with the highest unemployment rate. It confirms the fact that one of the objectives of this project was to improve the socioeconomic situation in some specific areas. We also observe a negative and significant effect for the period fixed-effect. It confirms that unemployment rate has globally decreased, on the period considered, for all neighborhoods. This observation can be explained by a favorable economic context in the agglomeration or in the region of Bordeaux. Finally, the variable Tram*Period reveals that the implementation of the tramway project has helped to reduce unemployment rate for neighborhoods close to its stations. Vicinity to tramway stations decreases unemployment rate by 4.7 per cent. In the model 2, we resort to a fixed-effects regression in order to control for neighborhoods unobserved time-invariant characteristics that may differ between neighborhoods and influence the labor market outcome. Coefficients for the period fixed-effect or for the variable Tram*Period are very similar to the previous model. The magnitude of effects is comparable but coefficients are more significant than before (they are now significant at the 1% level). Taking into account these unobserved characteristics do not seem to change observed effects. 20

21 Table 5: The effects of the tramway on unemployment rate - Right bank Variables Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Constant 0,190*** 0,028 0,258*** 0,009 0,213 0,178-0,529 0,609 Neighborhood with Tram 0,126*** 0,042 dropped 0,034* 0,021 dropped Period fixed-effect -0,034** 0,018-0,034*** 0,009-0,041*** 0,009-0,065** 0,026 Tram*Period -0,047* 0,026-0,047*** 0,012-0,033** 0,013-0,037*** 0,013 Demographic characteristics % of women 0,198 0,24 0,147 0,383 % of year olds 0,346*** 0,119 0,423** 0,187 % of single parents families 0,151* 0,096-0,003 0,176 Socio-economic characteristics % of individuals without diploma 0,226** 0,094 0,018 0,136 Average income (log) 0,001 0,001 0,002 0,002 Individuals mobility % of households with car -0,272*** 0,086-0,164* 0,101 Neighborhood fixed-effect No Yes No Yes Employment characteristics Employment in the municipality (log) -0,001 0,004 0,089 0,069 Observations R-squared 0,3256 0,857 Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively. Source: Population census 1999 and 2006, INSEE. 21

22 Models 3 and 4 represent respectively results when considering control variables that vary over time with OLS and fixed-effects regressions. We now include information on demographic characteristics, socio-economic characteristics, individuals mobility and employment. Results for the OLS regression show that the percentage of years old in the population or the percentage of single parents families tends to increase unemployment rate in the neighborhood. Explanations for these two phenomena are very different. In on case, we can argue that young people find a job less easily than others because it is likely that they have less professional experience. Employers may be reluctant to hire young people for this particular reason. In the other case, we can argue that is more difficult for single parent families to find or keep a job because of constraints imposed by childcare. Without surprise, we observe that the percentage of individuals without diploma tends to increase unemployment rate. Such a situation represents surely an obstacle to job search. Most striking, it seems that the average income in the neighborhoods does not have any significant effect on labor market outcome. Concerning the percentage of households with car in a neighborhood, it appears that it has a strong effect on unemployment rate. Having a car increases individuals mobility and represents an important advantage in job search process. Dujardin et al. (2004 [16]) find the same kind of effects for inhabitants of the Brussels metropolitan area. Finally, the total employment (expressed in logarithm) in the municipality of the neighborhood do not have any significant effect on unemployment rate. When we consider the last model (with neighborhood fixed-effects and control variables), results are slightly different. Some variables that were significant in the previous estimation are not significant anymore. We only observe effects for the percentage of years old and percentage and the percentage of households with car in the neighborhood. Introducing neighborhood fixed-effects in the regression on the small sample of neighborhoods located on the right bank tends to affect the significance of the estimated coefficients. Finally, we still observe (in models 3 and 4) an effect of the period fixed-effect. The coefficient for this variable shows that unemployment rate has decreased between 1999 and One can imagine that a lower unemployment rate in 2006 than in 1999 is due to an improvement of the socioeconomic composition in the neighborhoods. However, this hypothesis can be put into perspective because in these regressions we already control for different socio-economic characteristics. It confirms a hypothesis previously put forward that inhabitants of Bordeaux have benefited of a favorable economic context. The effect observed for the variable Tram*Period is still significant but is lower than before. When we control for other observed characteristics of the neighborhood, we find that the effect of tramway on unemployment rate is lower: respectively 3.3 per cent and 3.7 per cent for OLS and fixed-effects regression. It means that without controlling for the effects of the other variables, we overestimate the effect of this project. A part of the observed effect was due to changes in socio-economic composition of the neighborhood. The different regressions estimated on a sample of neighborhoods located on the right bank have shown that implementation of tramway has permitted to reduce unemployment rate in neighborhoods concerned. We believe that this method is a way to prove the existence of Spatial Mismatch in Bordeaux. Our results are in accordance with the work of Gaschet and Gaussier (2004 [19]) on the metropolitan area of Bordeaux. Indeed, the authors show that an enlarged area of job opportunities might have a negative impact on neighborhood unemployment or on neighborhood long term 22

23 Table 6: The effects of the tramway on unemployment rate (15-24 year olds) - Right bank Variables Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Constant 0,373*** 0,045 0,447*** 0,018 0,818** 0,303 0,047 1,065 Neighbor. with Tram 0,136** 0,057 dropped 0,019 0,045 dropped Period fixed-effect -0,049* 0,03-0,049** 0,017-0,054** 0,021-0,099** 0,045 Tram*Period -0,076* 0,04-0,076*** 0,023-0,054** 0,027-0,054** 0,023 Neighbor. fixed-effect No Yes No Yes Control variables No No Yes Yes Observations R-squared 0,253 0,672 Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively. Source: Population census 1999 and 2006, INSEE. Table 7: The effects of the tramway on unemployment rate (25-54 year olds) - Right bank Variables Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Constant 0,168*** 0,028 0,273*** 0,009 0,111 0,186-0,408 0,631 Neighbor. with Tram 0,126*** 0,041 dropped 0,039* 0,024 dropped Period fixed-effect -0,031* 0,018-0,031*** 0,008-0,041*** 0,011-0,053* 0,027 Tram*Period -0,047* 0,025-0,046*** 0,011-0,033** 0,014-0,038** 0,013 Neighbor. fixed-effect No Yes No Yes Control variables No No Yes Yes Observations R-squared 0,335 0,834 Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively. Source: Population census 1999 and 2006, INSEE. Table 8: The effects of the tramway on unemployment rate (55-64 year olds) - Right bank Variables Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Constant 0,247*** 0,028 0,269*** 0,537** 0,111 0,273-1,069 1,513 Neighbor. with Tram 0,041 0,048 dropped -0,004 0,043 dropped Period fixed-effect -0,089*** 0,016-0,089*** 0,02-0,085*** 0,017-0,151** 0,064 Tram*Period 0,007 0,029 0,007 0,028 0,011 0,026 0,008 0,032 Neighbor. fixed-effect No Yes No Yes Control variables No No Yes Yes Observations R-squared 0,335 0,595 Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively. Source: Population census 1999 and 2006, INSEE. unemployment rates. They identified two scales for the observed effects: one operating within small areas (below 2 kilometers) and corresponding to informal job-search behavior and one other relating to commuting distances resulting from a regional unbalanced structure between the east and west 23

24 parts of the urban area. Given that we work on a very small sample of the agglomeration of Bordeaux, it is mainly the first effect that operates. In a second step, we want to analyze the effect of the tramway on unemployment rate for different age categories. We remind that we make a distinction between years old, years old and years old because we believe that commutes behavior may differ according the age of individuals. TABLES 6, 7 and 8 present a synthesis of estimations made for these categories. First of all, we observe that the tramway helps to reduce unemployment rate of years old and years old. We do not observe any effect on unemployment rate of years old. This last result is note surprising because some recent studies have shown that old people use public transport less often than the average 14. In this context, it is quite normal that the tramway does not have any impact on behavior and labor market outcomes of this age category. In addition, it may be a proof that it does not hide some other mechanisms that could explain the decrease of unemployment rate. Indeed, one can believe that the project of tramway produces effects on unemployment because it has been realized together with an urban renewal project. Therefore, the project of tramway should have been favorable to all inhabitants, not only to inhabitants that use public transport the most. Secondly, when we look at results for the years old and the years old we observe important differences. When we include control variables, we find that the project of tramway is responsible for a decrease of 5.4 per cent of the unemployment rate of years old. The decrease is less strong for unemployment rate of years old: it is between 3.3 and 3.5 per cent (according to the method retained for estimation). This contrast between the two categories can also be explained by differences in terms of commutes behavior. Young people are more dependent of public transport because they do not always have resources to have their own car. The survey on households travels of the Urban Community of Bordeaux (op. cit.) reveals that 18% of years old use urban public transport (like tramway or bus) against approximatively 16% for the years old. Finally, it seems that the dependence to public transport explains the strong effect observed of tramway on unemployment rate. Thirdly, whatever the age category retained for unemployment rate, results show that the effect of the tramway is lower when we add some control variables in the regressions. In others terms, without these variables we over-estimate the effect of the tramway on local unemployment rate. It is particularly the case for the unemployment rate of years old. We find in models 1 and 2 a decrease of 7.6 per cent against only 5.4 per cent in models 3 and 4. The decline is less strong for the years old. In this case, we find a decrease of 4.7 per cent to 3.3 per cent. It means that differences in time-variant characteristics have a less important role in the explanation of the unemployment rate decrease for this category Effects of the tramway on unemployment rate for neighborhoods on the right and left banks In a second step, we want to check if the implementation of the tramway has helped to reduce unemployment inequalities between the right and the left banks of Bordeaux. We remind that we now consider that Floirac, Cenon, Lormont and Bassens and a part of Bordeaux are concerned by the tramway project. We compare labor market outcomes of the neighborhoods belonging to these municipalities 14 See, for example, the survey on households travels in the agglomeration of Bordeaux realized in 2009 by the Urban Community of Bordeaux (CUB), 24

25 to labor market outcomes of neighborhoods belonging to Bordeaux, Le Bouscat and Bruges on the left bank. The TABLE 9 proposes the same set of estimations than for neighborhoods on the right bank. In accordance with results of the previous sub-section we find that neighborhoods concerned by the tramway have higher unemployment rates than others. The coefficient for the period fixed-effect indicates that the period is favorable to a decrease of unemployment rate in the sample considered. When we introduce time-variant characteristics in models 3 and 4, we observe the expected effects. We still find a positive effect of demographic characteristics as percentage of young people or single parents families on unemployment rate. Concerning the socio-economic characteristics, the average income of the neighborhood does not have any significant effect while the percentage of individuals without diploma presents a very strong and significant effect. On the contrary, being mobile represents an important advantage as the percentage of households with car impact negatively unemployment rate. The effect of the tramway on unemployment rate appears to be more important when we are reasoning all others things being equal. The observed effect of the variable Tram*Period is low and not significant when we do not introduce other control variables. The coefficient becomes significant in model 3 when we add some characteristics of the neighborhoods. We now find a decrease of unemployment rate of 3.1 per cent. In the model 4, the introduction of neighborhood fixed-effect amplifies the impact. The observed decrease is close to 5.2 per cent. It is likely that some characteristics, not taken into account in models 1 and 2, hide the negative effect of the tramway on unemployment rate measured on the right bank. If we compare the effects found in TABLE 9 with those of the TABLE 5 for the variable Tram*Period, it appears that unemployment rate decrease is more important when we consider simultaneously neighborhoods of the right and left banks than when we only consider neighborhoods of the right bank. We can argue that the project of tramway has helped to improve job accessibility for some neighborhoods and so helped to reduce unemployment, but has mainly permits to reduce unemployment inequalities between the two banks of the Garonne. These two observations are not contradictory as the improvement of labor market outcomes of a part of inhabitants on the right bank is a necessary condition to reduce inequalities with the left bank. 25

26 Table 9: The effects of the tramway on unemployment rate - Right and left banks Variables Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Coeff. Std Err. Constant 0,231*** 0,019 0,247*** 0,009 0,259*** 0,098-0,942 0,661 Neighborhood with Tram 0,061* 0,03 dropped 0,041** 0,018 dropped Period fixed-effect -0,048*** 0,011-0,048*** 0,007-0,042*** 0,005-0,051*** 0,009 Tram*Period -0,021 0,019-0,021 0,014-0,031*** 0,009-0,052*** 0,014 Demographic characteristics % of women -0,201* 0,113 0,225 0,167 % of year olds 0,134** 0,059 0,176* 0,093 % of single parents families 0,155** 0,076 0,137* 0,085 Socio-economic characteristics % of individuals without diploma 0,477*** 0,117 0,521*** 0,063 Average income 0,002 0,002 0,004 0,004 Individuals mobility % of households with car -0,159*** 0,051-0,120* 0,074 Neighborhood fixed-effect No Yes No Yes Employment characteristics Employment in the municipality (ln) 0,003 0,002 0,091 0,06 Observations R-squared 0,144 0,813 Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively. Source: Population census 1999 and 2006, INSEE. 26

27 Conclusion In this work, we have used a natural experiment to test the Spatial Mismatch hypothesis, i.e the link between accessibility to jobs and labor market outcomes, for inhabitants of the agglomeration of Bordeaux. We have analyzed the effects of the implementation of the tramway on unemployment in neighborhoods on the right bank that are distant from the historical job center. We focused on the line A, opened in December 2003, that makes the junction between the two banks of Bordeaux. Historically, the right bank were characterized by a physical disconnection to the agglomeration center and a concentration of fragile populations on the labor market. This part of the agglomeration was isolated from the job center, on the left bank, by a limited number of bridges or accesses. Concretely, we use French population census of 1999 and 2006, which correspond to periods before and after the implementation of the tramway in the agglomeration. Firstly, we use differencein-differences methods to compare labor market outcomes of inhabitants of some neighborhoods located on the right bank. Some of them benefit from the tramway and others do not. In this case, we want to check if increasing job accessibility of some neighborhoods has consequences on their observed unemployment rates. Secondly, we use the same methods to compare labor market outcomes of inhabitants of the left bank to those of the right bank. In this case, we want to check if the tramway helped to reduce unemployment inequalities between the historical center of the agglomeration and the right bank. Our results show that implementation of the tramway has permitted to ameliorate labor market outcomes of a part of inhabitants located on the right bank. Neighborhoods of municipalities concerned by the project of tramway (Lormont, Cenon and the right bank of Bordeaux) have known a more important decrease of unemployment rate than other neighborhoods. The effects are stronger for the young people than other age categories, because they are more dependent of public transport for their daily commutes. On the contrary, we do not find any effect for the old people. Results obtained on a sample of neighborhoods located on the left and right banks reveal that a better connection between the two banks of Bordeaux has permitted to reduce inequalities in terms of unemployment rate. We can argue that the improvement of the labor market situation for some neighborhoods on the right bank has participated to a convergence of observed outcomes between the two banks of the Garonne. However, results of our analyzes are slightly subject to the parallelism assumption (i.e. in the absence of tramway the close neighborhoods would have known the same decrease of unemployment rate as those that are most distant). Indeed, the negative effect of the tramway on unemployment rate is partially explained by the changes in socio-economic composition of the neighborhoods during the period Without taking into account this phenomenon, we overestimate its effect. In terms of public policies, results of this natural experiment plead for the development of such projects in French agglomeration. The favorable socio-economic impacts are potentially multiple. Firstly, it is an easy way to improve physical connection of some neighborhoods located far away from job centers. Positive effects that arise are well known and documented by the literature on Spatial Mismatch hypothesis ( see Kain, 1992 [29]; Gobillon et al., 2007 [20] etc.). Secondly, the amelioration of the living conditions in neighborhoods concerned attracts new inhabitants. It may contribute to improve socio-economic composition of some deprived neighborhood and it develops 27

28 social mix. Various works have shown that the local composition of a neighborhood can be a hindrance to unemployment-to-work transitions (see for example Dietz, 2002 [14]; Galster, 2008 [17], Galster et al., 2008a [18], etc.). In fact, some studies on French agglomerations confirm that this phenomenon should be considered (see for example Dujardin and Goffette-Nagot, 2007 [15]; Sari, 2012 [41]). Thirdly, it also has an impact on economic activity of the neighborhoods concerned by the project. It attracts firms that locate close to the line to benefit from some positive effects of the urban renovation. Then, it increases job accessibility in two ways: it permits a better connection to job centers and it increases local employment. Additional researches could be developed in other agglomerations to check if these effects are also true in other contexts and to see if the advantages of such a project are more important than the initial costs. 28

29 References [1] E. Amrane. Les déplacements dans l agglomération bordelaise. Technical report, Le Quatre Pages Insee Aquitaine n 84, [2] R. Arnott. Economic theory and the spatial mismatch hypothesis:. Urban Studies, 35(7): , [3] O. Aslund, J. Osth, and Y. Zenou. How important is access to jobs? old question - improved answer. Journal of Economic Geography, 10(3): , [4] B. Baccaïni. Les trajets domicile-travail en ile-de-france. contrastes entre catégories socioprofessionnelles. Economie et Statistique, 294(1): , [5] K. Bouabdallah, S. Cavaco, and J-Y. Lesueur. Recherche d emploi, contraintes spatiales et durée de chômage : une analyse microéconométrique. Revue d Economie Politique, 112: , [6] L. Boustan and R. Margo. Race, segregation, and postal employment : New evidence on spatial mismatch. Journal of Urban Economics Letters, 65:1 10, [7] A. Béoutis, O. Diel, and O. Wotan. Projet bordeaux-euratlantique : diagnostic "point 0". Technical report, Le Quatre Pages Insee Aquitaine n 199, [8] S. Cavaco and Y. Lesueur. Contraintes spatiales et durée de chômage. Revue Française d Economie, 18(3): , [9] Certu. Déplacements et commerces : impacts du tramway sur le commerce dans différentes agglomérations françaises. Technical report, Amarcande, [10] E. Coulson, D. Laing, and P. Wang. Spatial mismatch in search equilibrium. Journal of Labour Economics, 19: , [11] K. Covington. Spatial mismatch of the poor : an explanation of recent declines in job isolation. Journal of Urban Affairs, 31(5): , [12] S. Davis and D. Huff. Impact of ghettoization on black employment. Economic Geography, 48: , [13] A. Decamps. La dynamique de la ségrégation urbaine à travers l évolution des profils de quartiers : Etude sur l agglomération bordelaise. Revue d Economie Régionale et Urbaine, 1: , [14] R. Dietz. The estimation of neighborhood effects in the social sciences: An interdisciplanry approach. Social Science Research, 31: , [15] C. Dujardin and F. Goffette-Nagot. Neighborhood effects, public housing and unemployment in france. Working paper, GATE, 05-05, [16] C. Dujardin, H. Selod, and I. Thomas. Le chômage dans l agglomération bruxelloise : une explication par la structure urbaine. Revue d Economie Régionale et Urbaine, 1:3 28,

30 [17] G. Galster. Quantifying the effect of neighbourhood on individuals: Challenges, alternative approaches and promising directions. Schmollers Jahrbuch, 128:1 42, [18] G. Galster, R. Andersson, S. Musterd, and T. Kauppinen. Does neighborhood income mix affect earnings of adults? new evidence from sweden. Journal of Urban Economics, 63: , [19] F. Gaschet and N. Gaussier. Urban segregation and labour markets within the bordeaux metropolitan area: an investigation of the spatial friction. Working Papers of gres, Cahiers du gres, 19, [20] L. Gobillon, H. Selod, and Y. Zenou. The mechanisms of spatial mismatch. Urban Studies, 44(12): , [21] T. Gurley and D. Bruce. The effects of car access on employment outcomes for welfare recipients. Journal of Urban Economics, 58: , [22] H. Holzer. The spatial mismatch hypothesis: What has the evidence shown? Urban Studies, 28 (1): , [23] H. Holzer, J. Quigley, and S. Raphael. Public transit and the spatial distribution of minority employment: Evidence from a natural experiment. Journal of Policy Analysis and Management, 22: , [24] K. Ihlanfeldt and D. Sjoquist. Job accessibility and racial differences in youth employment rates. The American Economic Review, 80: , [25] K. Ihlanfeldt and D. Sjoquist. The effect of job access on black and white youth employment: A cross-sectional analysis. Urban Studies, 28(2): , [26] K. Ihlanfeldt and D. Sjoquist. The spatial mismatch hypothesis: a review of recent studies and their implications for welfare reform. Housing Policy Debate, 9: , [27] D. Immergluck. Job proximity and the urban employment problem : do suitable nearby jobs improve neighbourhood employment rates? Urban Studies, 35:7 23, [28] J. F. Kain. Housing segregation, negro employment, and metropolitan decentralization. Quarterly Journal of Economics, 82:32 59, [29] J. F. Kain. The spatial mismatch hypothesis : three decades later. Housing Policy Debate, 3: , [30] M. Kawabata. Job accessibility by travel mode in u.s. metropolitan areas. Theory and application of GIS, 11(2): , [31] S. Khandker, G. Koolwal, and H. Samad. Handbook on impact evaluation: Quantitative Methods and Practices. The World Bank, [32] R. Martin. Can black workers escape spatial mismatch? employment shifts, population shifts, and black unemployment in american cities. Journal of Urban Economics, 55: ,

31 [33] A. Matas, J-L. Raymond, and J-L. Roig. Job accessibility and female employment probability : The cases of barcelona and madrid. Urban Studies, 47(4): , [34] P. Ong. Work and car ownership among welfare recipients. Social Work Research, 20(4): , [35] P. Ong. Car ownership and welfare-to-work. Journal of Policy Analysis and Management, 21(2): , [36] E. Patacchini and Y. Zenou. Spatial mismatch, transport mode and search decisions in england. Journal of Urban Economics, 58:62 90, [37] L. Putz and L. Dumas. Observatoire des effets du tramway sur l activité économique. Technical report, Chambre de Commerce et d Industrie de Bordeaux, [38] S. Raphael. The spatial mismatch hypothesis of black youth joblessness : Evidence from the san francisco bay area. Journal of Urban Economics Letters, 43(1):79 111, [39] S. Raphael and L. Rice. Car ownership, employment and earnings. Journal of Urban Economics, 52: , [40] C. L. Rogers. Job search and unemployment duration : Implications for the spatial mismatch hypothesis. Journal of Urban Economics, 42: , [41] F. Sari. Analysis of neighbourhood effects and work behaviour: Evidence from paris. Housing Studies, 27(1):45 76, [42] W. Simpson. Urban structure and the labor market. Oxford, Clarendon Press, [43] B. Weinberg. Black residential centralization and the spatial mismatch hypothesis. Journal of Urban Economics, 48: , [44] B. Weinberg. Testing the spatial mismatch hypothesis using inter-city variations in industrial composition. Regional Science and Urban Economics, 34: , [45] S. Wenglenski. Regard sur la mobilité au travail des classes populaires, une exploration du cas parisien. Les Cahiers Scientifiques du Transport, 49: , [46] J. Zax and J. Kain. Moving to the suburbs : Do relocating companies leave their black employees behind? Journal of Labor Economics, 14(3): ,

32 Appendices APPENDIX 1. THE TRAMWAY IN BORDEAUX 32

33 APPENDIX 2. LIST OF NEIGHBORHOODS APPENDIX 2.1. STUDY SAMPLE 1 Table 10: Neighborhoods with the tramway (24 obs.) Iris code Municipality Iris Bordeaux La Bastide Bordeaux La Bastide Bordeaux La Bastide Bordeaux La Bastide Bordeaux La Bastide Cenon Palmer Cenon Palmer Cenon Cavailles-Camparian Cenon Gravieres-Gravette-Beaulieu-Lagrue-Plai Cenon Saraillere-Ronceval-Seigliere Cenon Grand Pavois-Cimetiere St-Paul Cenon Maregue Cenon Bas Cenon-Cours Victor Hugo-Gambetta Cenon Bas Cenon-Cours Verdun-Testaud Lormont Cite Carriet Lormont Cite Carriet 2-La Croix Rouge Lormont Le Bourg-Le Vieux Bourg Lormont Genicart Lormont Genicart Lormont Residences des Hauts de Lormont Lormont Genicart Lormont Lormont Lissandre Zup Lormont Z.i. Quatre Pavillons-La Gardette Table 11: Neighborhoods without the tramway (20 obs.) Iris code Municipality Iris Artigues-près-Bordeaux Nord Artigues-près-Bordeaux Sud Bassens Nord Bassens Centre Bassens Sud Bassens Ouest-Zones d Activites-Quais Bordeaux La Bastide Bouliac Bouliac Carbon-Blanc Nord Carbon-Blanc Sud Carbon-Blanc Centre Floirac La Souys Floirac Gambetta Floirac Gravette Floirac Coteaux Floirac Rebedech Floirac La Maregue Etendu Floirac Plateaux Sainte-Eulalie Sainte-Eulalie Tresses Tresses Yvrac Yvrac 33

34 Figure 7: Neighborhoods on the right bank APPENDIX 2.2. STUDY SAMPLE 2 Table 12: Neighborhoods on the right bank (35 obs.) Iris code Municipality Iris Bassens Nord Bassens Centre Bassens Sud Bassens Ouest-Zones d Activites-Quais Bordeaux La Bastide Bordeaux La Bastide Bordeaux La Bastide Bordeaux La Bastide Bordeaux La Bastide Cenon Palmer Cenon Palmer Cenon Cavailles-Camparian Cenon Gravieres-Gravette-Beaulieu-Lagrue-Plai Cenon Saraillere-Ronceval-Seigliere Cenon Grand Pavois-Cimetiere St-Paul Cenon Maregue Cenon Bas Cenon-Cours Victor Hugo-Gambetta Cenon Bas Cenon-Cours Verdun-Testaud Floirac La Souys Floirac Gambetta Floirac Gravette Floirac Coteaux Floirac Rebedech Floirac La Maregue Etendu Floirac Plateaux Lormont Cite Carriet Lormont Cite Carriet 2-La Croix Rouge Lormont Le Bourg-Le Vieux Bourg Lormont Genicart Lormont Genicart Lormont Residences des Hauts de Lormont Lormont Genicart Lormont Lissandre Lormont Zup Lormont Z.i. Quatre Pavillons-La Gardette 34

35 Table 13: Neighborhoods on the left bank (96 obs.) GQ and IRIS codes Municipality Iris Bordeaux Le Lac (2) Bordeaux Bacalan (4) Bordeaux Chartrons-Grand-Parc (11) Bordeaux Hotel de Ville-Quinconces (8) Bordeaux Saint-Seurin-Fondaudege (9) Bordeaux Villa Primerose Parc Bor.-Cauderan (6) Bordeaux Lestonat-Monsejour (9) Bordeaux Saint-Augustin (5) Bordeaux Saint-Bruno-Saint-Victor (6) Bordeaux Capucins-Victoire (8) Bordeaux Nansouty (9) Bordeaux Gare Saint-Jean (5) Le Bouscat Jean-Moulin Le Bouscat Centre-Ermitage Le Bouscat La Garenne-Baudin Le Bouscat Ausone-Les Ecus Le Bouscat Jean-Jaures-Ravezie Le Bouscat Marceau-Tivoli Le Bouscat Barriere du Medoc Le Bouscat Cheneraie-Lavigne Le Bouscat Hippodrome-Lafon-Feline Bruges Tour de Gassies-Treulon Bruges Centre Ville Bruges La Marianne-Sainte Germaine Bruges Cimetiere Nord-Petit Bruges Bruges Villabois-La Hutte Note: The GQ code is the code for the large neighborhoods (a large neighborhood contains several IRIS). We do not give all IRIS codes because the list would be too long. Lecture: Number in brackets corresponds to the number of IRIS. Figure 8: Neighborhoods on the left and right banks 35