Is the residential location important in access to jobs? an empirical analysis on the Paris agglomeration 1 Florent Sari2

Preliminary version (June 2009)

Abstract In this paper, we want to highlight the effects of being located in a deprived neighborhood on unemployment. Our interest is focused on problems of residential segregation. We use the 1999 Population Census for Paris and the three surrounding sub-regional administrative districts in order to estimate different models that take into account the potential endogeneity bias of location choice. We first run a simultaneous equation system that includes the residential location as an endogenous variable. We also run a probit model on the sub-sample of households living in a public housing with the idea that the location choice is exogenous. Whatever the method used, we show that living within the most deprived neighborhoods, in terms of local composition, decreases employment probability. Keywords: Unemployment, Residential segregation, Endogeneity bias. JEL Code: J64, R14, R2

1 I would like to thank Samia Benallah, Oana Calavrezo, Yannick L’Horty, Mathieu Narcy, Sanja Pekovic,

Patrick Sillard, participants of the internal seminar of CEE, participants of the internal seminar of Evry (EPEE), participants of the Spring Meeting of Young Economists 2009, participants of the 2nd Doctoral Meeting of Montpellier and participants of the Journées de Microéconomie Appliquée (JMA) 2009 for comments and suggestions that have greatly improved this paper . All errors are mine. Census data as the Iris level have been provided by the Centre Maurice-Halbwachs 2 Paris-Est Marne-la-Vallée (OEP), Centre d’Etudes de l’Emploi et TEPP (FR CNRS n°3126); [email protected], Le Descartes I, 29 Promenade Michel Simon, 93166 Noisy-le-Grand Cedex, tél.: +33 (0)1 45 92 69 74.

1

1

Introduction

The economic link between spatial structure of cities and territorial disparities has been widely developed in the U.S contrary to France. If the urban structure of cities is very different in each of these two countries, it appears that they suffer from the same difficulties: the concentration of low-skilled and minority workers and a “spatial mismatch” between the geographic distributions of labor supply and labor demand especially for these categories. However, in the spatial economics literature, the link between urban organization and socio-economics performances is not consensual. For some authors, individuals spontaneously sort themselves into a city according to their different socioeconomic characteristics or to their ability to pay a rent (Tiebout, 1956). In this context, it is not surprising to see important concentration of unemployed in some cities or neighbors. Nevertheless the link might be reversed, that is to say, unemployment in some part of cities may be due to some residential segregation problems and/or disconnection between place of jobs and place of living (Kain, 1968; Davis and Huff, 1972). In this paper, we are interested in reverse causality since our objective is to test whether city structure can be a source of unemployment in the Paris agglomeration. Paris represents an interesting case because its structure is relatively heterogeneous. The favorable composition of a sub-regional administrative district as the Hauts-de-Seine contrasts with the unfavorable composition of the Seine-Saint-Denis. The question raised is to know if the local composition of a neighbor can be a hindrance to unemployment-to-work transitions. In others terms: Are the neighborhood effects relevant to explain unemployment disparities? Empirical researches encounter difficulties in analyzing these spatial constraints’ effects on labor-market outcomes. Indeed, there is a simultaneity problem as labor-market outcomes may influence residential location choice and residential location choice may have some consequences on labor-market outcomes. There is typically an endogeneity bias that must be taken into account while estimating the effects of being located in a deprived neighborhood. If we do not control for this phenomenon we may attribute to residential location effects of some unobserved household characteristics. In the literature, previous empirical studies have developed methodologies more or less sophisticated in order to deal with this endogeneity bias. First, a solution is to work on a sub-sample of young people living with their parents. The idea is that the parents’ location choice can be seen as exogenous to the labor-market outcomes of young

2

people (Ihlanfeldt and Sjoquist, 1990). However, this approach has some shortcomings as the fact that the exogeneity of residential location holds only if the parents’ residential location choice is exogenous, which is hard to prove. A second approach is to design a simultaneous equation system or to use instrumental variables techniques in order to include the residential location as an endogenous variable (see for example Evans et al., 1992; Cutler and Glaeser, 1997; Liebman et al., 2004; Dujardin and Goffette-Nagot, 2007; Dujardin and Goffette-Nagot, 2009) In this case, the first stage models the process by which individuals sort into different neighborhoods types and the second stage analyzes the relationship between neighborhoods and a given outcome. Another approach is to focus on quasi-experimental situations which randomly assign individuals to locations with different characteristics such as Gautreaux Program or the Moving To Opportunity program (Katz et al., 2001; Oreopoulos, 2003; Kling et al., 2005). In this paper, we use two different methodologies. Firstly, we run a bivariate probit model where we explain the probability to live in a deprived neighborhood and the probability to be employed. Secondly, we develop a strategy that we believe similar to a quasi-experimental situation. We concentrate on individuals living in a public housing with the idea that location choice is rather exogenous. In this way, we hope to deal with neighborhood endogeneity when estimating its effect on employment probability. The paper is structured as follows. Section 2 briefly synthesizes economics mechanisms that could explained formation of unemployment in cities. Section 3 describes the database and our methodological approach. Section 4 gives some stylized facts about the spatial structure of the Paris agglomeration. Section 5 presents the main results and Section 6 concludes.

2

Literature Review: city structure effects on urban unemployment

Analyzing the determinants of unemployment-to-work transitions is a recurrent aim in labor economics. The job search theory developed last decades analyzes the effects of individual characteristics and public policies on the job search process and on the unemployment duration (Mortensen, 1986; Lancaster, 1990). However, traditionnals job search models do not take into account the effects of individual’s environment. For example, Holzer (1991) emphasizes the existence of a negative correlation between residence place and job search process, especially for the less-skilled workers or ethnics minorities. If individuals’ characteristics may be relevant to explain labor market outcomes, it is not sufficient as local context has a strong influence.

3

Two major explanations have been put forward to explain the existence of spatial constraints effects. The first explanation focuses on the effects of residential segregation and the social composition of neighborhood. The second underlines the role played by disconnection to jobs opportunities. In this paper, we are principally interested in the first one. Sociology and urban economics retain different conceptions of the segregation. In this paper, we use the concept of segregation by opposition to the concept of social mix. In this context, segregation results from a process of territories’ social homogenization. We can say that an area is segregated when different categories of the population do not live together on it (Tovar, 2008). In this paper, we focus on neighborhoods that concentrate low-skilled workers, ethnics minorities etc. We define such a neighborhood as a deprived neighborhood. Several mechanisms have been identified that can account for an adverse effect of residential segregation. In particular, we will see that residential segregation have effects on economics performances of individuals living in deprived neighborhoods. One of these mechanisms is based on the fact that residential segregation can be a hindrance to human capital acquisition. For example, this is because the success of a given student depends on the socio-economic characteristics of all others students in the class that the concentration of low-skill learners exerts a negative pressure on the learning process (Bénabou, 1993). Thus, in neighborhoods which concentrate low-ability students, human capital externalities can deteriorate school achievements and employability. A second consequence is that segregated or deprived neighborhoods are often exposed to the emergence of social problems that can deteriorate the job seekers’ employability. In 1991, Crane develops the epidemic theory of ghettos. His theory shows that the propensity of young people to adopt a given behavior is strongly correlated with the proportion of individuals already showed this behavior. For unemployed individuals this phenomenon is also verified: when most adults of a neighborhood are unemployed, it does not determine young people to search for a job. These fragile populations do not provide role models of social success and so they do not motivate the others to find a job. Another mechanism is based on the fact that an important proportion of jobs are usually found through personal network and to the fact that low-skilled workers, young adults and ethnic minorities, generally resort to these search methods (Holzer, 1988). If job seekers live in neighborhoods were local unemployment rates are higher than average, the probability to have contacts in unemployment will be very high and so they will not rely on their “social 4

networks”. An individual that resides in deprived neighborhoods will be confronted with a social network of poor quality. Finally, residential segregation is also likely to reduce the probability for a worker residing in a segregated area to receive a job offer since employers may discriminate against residentially segregated workers. Such an attitude may be justified by the fact that they consider that, on average, workers from stigmatized areas have bad work habits or are more likely to be criminals. Moreover, in industries and jobs in which workers are in contact with clients, employers may discriminate against residentially-segregated workers in order to satisfy the perceived prejudices of their clients (Holzer and Ihlanfeldt, 1998). A recent french study uses the findings of a correspondence testing in order to assess the potential discrimination at job access level (Duguet et al., 2009). If the authors find a negative effect of being of others nationality, they also observe an effect of residential location. All things being equal, living in an underprivilegied city decreases the probability of discrimination. A way to take into account these various effects on unemployment probability is to introduce indicators of neighborhood composition as education level, racial composition etc. However, a problem that can be raised is that these indicators are often inter-correlated. For this reason, we summarize these characteristics through two definition of a deprived neighborhoods (see section 3.2). Besides neighborhood effects, there is the spatial mismatch problem. This intuition is developed by Kain (1968) that stated that being disconnected from jobs (or living far away from them) can have some important consequences on the unemployment process. The literature identified several channels linking the spatial mismatch hypothesis to the bad labor market situations of a part of the inhabitants: Living far away from the job centre is a source of important commuting costs for job-seekers. If the wage offered by a firm does not compensate this cost, the unemployed may be weakly induced to accept the job (see Coulson, Laing and Wang, 2001; Brueckner and Zenou, 2003). Searching for a job far away from the residence area can be too costly. Job seekers search efficiently only in a restricted area, near their residence, even if there are only poor-quality jobs (Davis and Huff, 1972). Others empirical studies show that the physical distance to jobs reduces information availability regarding to job vacancies (Ihlanfeldt and Sjoquist, 1990, 1991). By mobilizing data on travelling times between municipalities, we construct certain indicators that are relevant to measure distance to jobs. It permits us to verify if some neighborhoods are suffering from this spatial mismatch problem. 5

Although the effects of the city structure on the local labour market outcomes are tested in many North-American empirical studies, in France there are very few papers on this topic. Dujardin and Goffette-Nagot (2007) estimated the effects of living in a deprived neighborhood on the unemployment level in the Lyon area. The results that they found indicate that living in the 35% more deprived neighborhoods of the Lyon area increases significantly the probability of being unemployed. Dujardin and al. (2007) / Gobillon and al. (2007) emphasize respectively the determinants of unemployment in the Brussels metropolitan area / in the Paris region. The two papers find out that residential segregation plays an important role on the unemployment rate. The results concerning spatial mismatch are more contrasted. The spatial mismatch hypothesis seems to be more valid in the Paris region than in the Brussels metropolitan area. Always on the Paris Region, Korsu and Wenglenski (2008) argue that the employment status determines the residential location more than the opposite. Nevertheless, they even highlight the effects of the social environment on employment status. This new result pleads one more time for taking into account the endogeneity bias resulting from the residential location choice. Finally, in a recent paper, Dujardin and Goffette-Nagot (2009) estimate the effect of living in public housing on unemployment probability. Mechanisms that link public housing and unemployment are potentially the same that those link the fact of being located in a deprived neighborhood and unemployment: social network effects, peer effects etc. In accordance with some works on american cities, they do not find any negative effect on unemployment probability for the French case. Most of these studies are dealing with the endogeneity bias of residential location choice by reasonning on sub-sample of young people or by running simultaneous equations system. In this work, we choose a strategy similar to those of Dujardin and Goffette-Nagot (2008, 2009) as one of our strategy is to explain simultaneously the probability of being located in a deprived neighborhood and being employed by using a bivariate probit.

3 3.1

Data and methodology The data and sample

In this paper we focus on the city of Paris and its three sub-regional administrative districts: Seine-Saint-Denis, Val-de-Marne and Hauts-de-Seine. Paris represents the most important city in France with a 762 km² and more than 6 260 633 inhabitants.

6

We mobilize data from the French Population Census (1999). It contains informations at the neighborhood level and also informations at the individual level. If the first dataset is an exhaustive source, the second represents a sample of 1/20th of the total population. Our proxy of a neighbor is the Iris. An Iris may be a municipality or a subdivision of municipality if the latter have more than 10 000 inhabitants. It represents homogeneous spatial subdivisions in terms of housing and population with on average 2 344 inhabitants. For each neighborhood we have informations on housing characteristics or socio-economics characteristics of the population. This dataset is used in order to characterize the neighbor where households live and to build different definitions of what we call a "deprived" neighborhood. Our analysis focuses on 2 596 neighborhoods. We also use this dataset in order to built some indicators that are relevant in order to control for the Spatial Mismatch hypothesis. Indeed, the French census provides some informations on the location of jobs. It will permit us to include in our regressions the average euclidian distance between place of living and place of work. We also include the total of jobs that are available in a radius of ten kilometers around the place of living. The dataset at the individual level contains informations on socio-economics characteristics as gender, age, nationality (French versus other), marital status, number of children, occupational categories, qualification level, whether individuals are houseowner, whether individuals own a car and characteristics of the others members of the household. We restrict our analysis to couple households and precisely with household head. As in Dujardin and Goffette-Nagot (2009), we do not retain women as it would imply to explain not only employment, but also labor-market participation. Finally, we retain individuals of working age (16 to 64) and residing in the Paris agglomeration. The final sample contains 47 265 household heads.

3.2

Methodological approach

This research has for objective to highlight the effects of living in a deprived neighborhood on employment probability. We use traditionnal variables as individual characteristics relative to experience, educational level, occupational status, nationality etc. We also include a dummy variable in order to indicate if the neighborhood is defined as deprived or not. It is a way for us to test the potentialy negative effects of being located in a neighborhood with bad characteristics on employment probability. The introduction of this variable raises the problem of the endogeneity of location choice (Gao and Johnston, 2004). Everything happens as if individuals 7

with similar socio-economics characteristics sort themselves in certain areas of the urban space. For example, we can imagine that individuals with well-paid jobs will choose neighborhoods with a better social environment than unemployed individuals (Fujita, 1989). While our objective is to test whether or not residential location influences unemployment probability, we have to control that individual outcomes influences the choice of a residential location. This means that individual characteristics that may influence labour-market outcomes may also influence location choices. We present our different methodologies used in order to define a deprived neighborhood and then we present strategies that have been developed to correct for the endogeneity of neighbourhood choice. 3.2.1

Definition of a deprived neighborhood

Our objective is to estimate the effects of living in a neighborhood defined as deprived on employment probability, so we have to find a relevant measure of neighborhood characteristics. Literature in urban economics generally shows that a wide variety of neighbors’ characteristics may affect individual employment probability as the unemployment rate, qualification level, percentage of blue-collars in the area etc. However, these characteristics may be very close to each others. In this case, there is clearly a colinearity problem that may potentially biases estimations. Consequently, we try to measure influence of neighborhood through a dummy variable indicating whether or not each neighborhood of the agglomeration may be considered, on the basis of the social characteristics of its inhabitants, as deprived. We get round this difficulty by using different definitions of a deprived neighborhood. Moreover, including neighborhood type in variables that affect employment allows us to test the presence of neighborhood effects. First of all, we retain a subjective definition of a deprived neighborhood. The idea is to developp a ranking of neighborhoods according different socio-economics characteristics as: percentage of individuals without diploma, percentage of individuals with at most a lower secondary diploma, percentage of individuals that are blue-collars, percentage of individuals of foreign nationality and percentage of large families. We evaluate each neighborhood for each of these characteristics. We assign scores from 1 to 10 for each neighborhood. As the size of spatials units may differ, we include the total population as a weighting. Finally, the higher the percentage is, the more the score is important. In this way, we obtain a score for each characteristics and for each spatial units. 8

In order to evaluate how a neighborhood is "disadvantaged", we aggregate the differents scores obtained. The more deprived neighborhood are those with the highest total score. As we need a dummy variable to test the relationship between residential location and labor-market outcomes, the spatial unit will be defined as deprived if it belongs to 20% of spatial units with the highest scores. This first definition is not without shortcoming. Indeed, we aggregate different characteristics but we do not rank them. Living in a district with an important part of individuals without diploma could be more detrimental that living in a district with an important part of large families. Indeed, with our final indicator we are not able to say if the spatial unit is disadvantaged because of the high percentage of individuals without diploma, because of the important part of blue-collars, because of the important part of large families... Nevertheless, this indicator presents the advantage of being in accordance with our definition of a deprived neighborhood. That it is to say a neighborhood characterized by a social homogenization. Secondly, we retain a more objective definition of disadvantaged neighborhood. We argue that the neighborhood where are located one or more Sensitive Urban Zones (called Zone Urbaine Sensible in France) are defined as disadvantaged or deprived. Sensitive Urban Zones are sub-urban areas defined by the government to be the priority target of the city’s policy, according to local considerations related to the difficulties faced by the inhabitants of these territories. In the Paris region, more than one inhabitants one eight are living in such an area 3 . There are 84 Sensitive Urban Zones in Paris and its three surrounding administratives districts. The figure 2 maps disparities between the administratives districts. With this second indicator we also take into account the various effects exposed in the previous section as the "peer effects", problems of bad quality of "social network" and especially the potentially stigmatization of some neighborhoods. Indeed, the Sensitive Urban Zone are often returning a negative signal. So, employers may be reluctant to hire individuals that are living in these districts. For this reason, we argue that an individual living in a deprived neighborhood is not only an individual living in the Sensitive Urban Zones but rather living in the neighborhood where the area is located 4 . Whatever the definition retained, it appears that population of these disdvantaged neighborhoods have characteristics that differ from those of the others neighborhoods (see Annex A for more details). 3 For the whole Paris region in 1999, the Census Population count 1 332 000 individuals living in the 157 areas of the region. 4 In the case where the Sensitive Urban Zone is smaller than the Iris - our definition of a neighborhood.

9

Figure 1: Subjective definition of deprived neighborhood

Source: INSEE, Population Census, 1999.

Figure 2: Objective definition of deprived neighborhood

Source: INSEE, Population Census, 1999.

10

Even though we use two differents definitions, figure 1 and figure 2 show that geography of deprived neighborhood is relatively stable. Both the subjective and objective definitions are showing important part of disadvantaged neighborhoods in the sub-regional administrative district of Saine-Saint-Denis, in the north-east of Hauts-de-Seine, in the south of Val-de-Marne and in some Parisians districts. Nevertheless, while these neighborhoods are more concentrated with our subjective definition, they appear more scattered with the objective definition. Spatial disparities in the second definition is due to the geography of Sensitive Urban Zone. There are 36 Sensitive Urban Zones in the Seine-Saint-Denis, while there are respectively 16 and 23 of these areas in the Val-de-Marne and the Hauts-de-Seine. On the other hand, Paris has only 9 Sensitive Urban Zones. We test our empirical model on these two definitions of a deprived neighborhood in order to see if results are stable and robusts. By doing this, we can be more confident that the effects found are not due to a bad specification of a disadvantaged neighborhood. 3.2.2

Econometric method and model

Concerning the econometric method, we retain two distinct approaches. Firstly, we design a simultaneous equation system that includes the residential location as an endogenous variable. Secondly, we focus on individuals living in a public housing with the idea that, in this case, location choice is rather exogenous. This particularity permits to establish a link between the French system and quasi-experimentals situations that randomly assigns individuals to locations with different characteristics5 . The bivariate probit model In any analysis of the relationship between employment and residential location, ignorance of the endogeneity bias will result in biased coefficients. In this context, the first step is to design a simultaneous equation system that can include both employment and residential location as endogenous variables. Our observed variables Y1 and Y2 , that correspond respectively to location in a deprived neighborhood and employment, are defined by:   ∗    1 if Y1 > 0 Y1 =     0 Otherwise 5 such as Gautreaux Program or the Moving To Opportunity program in the United States (see Katz et al., Kling

et al., 2005 for more details).

11

  ∗    1 if Y2 > 0 Y2 =     0 Otherwise Y1∗ and Y2∗ are latent variables influencing the probability of living in a neighborhood defined as deprived (neighborhood with the highest scores or with at least one or more urban renweal area) and the probability of employment. We consider the following bivariate probit model:   ∗    Y1 = X1 β1 + 1     Y∗ = X2 β2 + α Y1 + 0 2 2 Residuals of these two equations are supposed to follow a normal bivariate law with zero means and a covariance matrix that writes, after normalizations to 1 of the diagonal elements: !

1 → N (0; 2

X

), where

X

  1 ρ 12  =   ρ 1 12

     

The system is estimated by the maximum likelihood method on a sample of 47 265 households of the Paris agglomeration. It allows us to estimate effects of explaining variables and to identify if residuals of the two equations are correlated or not. In the system, X1 is a vector of exogenous variables including a constant, some individual’s characteristics as the age, nationality, education level, previous occupational status, to have a driver’s license or not, to be . We also include some characteristics of the spouse (nationality and education level), variables measuring accessibility to jobs (the logarithm of the average euclidian distance between place of living and place of work and the logarithm of the total of jobs available in a radius of ten kilometers) and dummies in order to take into account unobserved heterogeneity between the different sub-regional administratives districts due to some local policies or disparities of local labor-markets. Moreover, in order to identify the bivariate probit, we may need an additional variable that will explain probability of living in a deprived neighborhood but not relevant to explain employment probability. This kind of variable represents an instrumental variable which guarantee the identification of the model and help to estimate correlation coefficients (Maddala, 1983). However, a recent paper (Wilde, 2000) shows that exclusion restriction on the exogenous explicatives variables is not necessary to identify multiple equation probit models with endogenous dummy variables. For this author the existence of one varying exogenous variable 12

in each equation is sufficient to avoid small variation identification problems in this multiple equation probit models. Even if introducing an instrumental variable is not necessary for the identification, we make the choice to retain an instrumental variable in order to obtain more robust results. For the employment probability, the instrumental variable is the number of children in the household. The idea is that it is not likely to influence the individual’s employment probability but rather the residential location. So it is used as an exclusion restriction 6 . If we can easily believe that the number of children play an important role in the choice of the residential location, the impact on employment probability is less obvious. The fact to have children to carry out may be a reason to stay at home but rather for women. In this work, we are exclusively working on a sample of men. Moreover, to have children may have an effect on the probability of living in a deprived neighborhood because traditionnaly these type of neighborhood concentrate large families. These might be explained by the fact that housing prices are less important so it is easier to find a largest house for the family. Finally, X2 includes the same variables as X1 In this equation, number of children is not introduced as it must explain the probability of living in a deprived neighborhood but not the employment probability. Annex B shows an exhaustive list of the variables used in our estimations and provides some descriptive statistics. The probit model on the sub-sample of individuals living in a public housing In order to test the robustness of our results, we also run a probit model on the sub-sample of individuals residing in public housing. In this case, we argue that location is quasi-exogeneous. The ideas is very similar to the strategy developped by Goux and Maurin (2007) when they evaluate the effects of living in a disadvantaged neighborhood on the school achievements. In France, families with an income below a given threshold can ask for housing with a moderate rent. As the threshold is rather high, there is an important number of families asking for this type of housing. Consequently, number of families that are waiting for an answer exceeds greatly the number of availables housing and queues may be long. In addition, families have a limited control on the moment when they obtain the housing and also on its location. Indeed, the allocation from those waiting lists is seen as quotas complex system taking into account household type, household size and income floor. In this context, we can argue that households location is relatively random. However, This strategy is not without critics. For example, 6 Having children is often used as an instrumental variable in studies that try to assess the effect of living in a

public housing on labor-market outcomes (see for example Currie and Yelowitz, 2000; Dujardin and Goffette-Nagot, 2009).

13

some individuals can wait for proposals that are consistent with their wishes. An endogeneity problem still exists if individuals with unobservable favourable characteristics to employment refuse public housing offers in neighbors that are unfavourable to find a job. If we note Yi the latent variable influencing the employment probability, we can write:   ∗    1 if Yi > 0 Yi =     0 Otherwise

Then we consider the following probit model: Yi∗ = Xi βi + i Where Xi is the same vector of variables used in our previous model (individual’s caracteristics, household caracteristics, variables measuring accessibility to jobs. In this case, residuals follow a standard normal law. Parameters βi of the model are estimated by the maximum likelihood method on a subsample of 12 620 households living in a public housing.

4

Paris agglomeration: stylized facts

In this section, we provide descriptive statistics for the Paris agglomeration that are calculated from the (1999) French Population. In accordance with the theories developed in the previous section, these statistics may be relevant to explain urban unemployment.

4.1

Unemployment disparities

Our interest for this agglomeration is justified by the facts that it shows a contrasted spatial organization. Some areas of the Paris agglomeration are showing concentration of fragiles populations while others are characterized by compactness of more healthy populations. In order to show this phenomenon, we map the percentage of unemployed workers in each neighborhood (see Figure 3). If spatial organization of most American cities show high unemployment rate in the center and low in the periphery, this is not the case for Paris that shows a more contrasted geography. Indeed, it appears that the highest unemployment rates are concentrated in Saine-Saint-Denis’ neighborhoods (on average 17%) while the lowest unemployment rates are

14

located in Hauts-de-Seine’s neighborhoods (10.3%). Paris itself is also contrasted: the northwest concentrates high unemployment rates neighborhoods whereas the others neighborhoods are more heterogeneous. Finally, even if we see large parts of the agglomeration favourable or unfavourable to unemployed workers, we also see strong local disparities. This is the case in Paris or in some parts of Hauts-de-Seine and Val-de-Marne. The unemployment geography in Paris and its periphery might be explained by residential segregation, neighborhood effects or the disconnection between places of work and places of residence. However, the spatial mismatch hypothesis may be weak because Paris and the three surrounding sub-regional administrative districts are showing an important density for total employment (Guillain and Le Gallo, 2008). In this context, we can assume that spatial mismatch hypothesis should not be determinative for explaining unemployment disparities. We test these two hypothesis in further section.

Figure 3: Unemployment rate in the Paris agglomeration

Source: INSEE, Population Census, 1999.

4.2

Measures of the residential segregation

We use different index traditionally used in segregation studies: the dissimilarity index (Duncan and Duncan, 1955) and the isolation index (Bell, 1954; White, 1986)7 . The Duncan index measures the spatial distribution of a group and is bounded by 0 and 1. The value of this index 7 See Apparicio (2000) for a French review of the residential segregation index.

15

presents part of the group that would have to be displaced in order to achieve a perfect spatial distribution. It is defined as: n

DI =

1 X xi ti − xi | − | 2 X T−X i=1

where xi is the population of the group X in neighborhood i, X is the population of the group X in the whole agglomeration, ti − xi is the population of other group in neighbourhood i and T − X is the population of other group in the whole agglomeration. The isolation index is also bounded by 0 and 1 and measures the probability that a member of a group share the same neighborhood with another member of his group. A value close to 1 reflects the fact that the group is totally isolated in neighborhoods of the agglomeration. It is defined as:

PIn =

n X

[xi /X][xi /ti ]

i=1

where xi is the population of the group X in neighborhood i, X is the population of the group X in the whole agglomeration, ti is the total population in neighborhood i and n is the number of neighborhoods. We calculate these indexes for different dimension as nationality, qualification level, occupational categories and employment status. Table 1 presents these indexes for the Paris agglomeration, Paris itself and its periphery. Globally, it appears that the segregation is stronger in the Paris periphery than in Paris itself, whatever the index used. Firstly, we see that segregation is relatively high for individuals of foreign nationality. Indeed, about a quarter of the foreign people should have to be displaced in order to have an equal spatial distribution with French people in the agglomeration. They seem to suffer from concentration with an important value of the isolation index for neighborhoods of the periphery. Contrasts between qualification level and occupational status are indicative of segregation in many neighborhoods. The dissimilarity index shows that nearly 50% of blue-collars should be moved in order to obtain a perfect distribution. The index remains strong when we compare population with at most high school diploma to others. However, the high isolation measured for such a qualification level may be explained by the large definition that we have retained. Finally, except for a few neighborhoods in Seine-Saint-Denis and in Val-de-Marne, 16

Table 1: Indexes of segregation Paris agglo. Dissimilarity Index French/foreign Pop. with high school final diploma or more/ Pop. with at most high school final diploma Executives/blue-collars Unemployed workers/employed Isolation Index Foreign Pop. With at most high school final diploma Blue-collars Unemployed workers

Paris

Periphery

0,23 0,3

0,18 0,21

0,27 0,27

0,5 0,17

0,35 0,13

0,49 0,18

0,18 0,6 0,19 0,15

0,16 0,46 0,11 0,14

0,19 0,65 0,21 0,16

Source: INSEE, Population Census, 1999 Notes: the Dissimilarity Index shows that 23% of foreign people should have to be displaced in order to be mixed uniformly with French people. The Isolation Index shows that the probability of a foreign to share the same neighborhood with another foreign is equal to 0.18.

unemployed workers are not so concentrated. The dissimilarity index is small, among 0.17, for the Paris agglomeration.

4.3

The public housing system in the Paris agglomeration

In France, a large part of the public housing stock was built forty years ago, during the sixties and the following decade. Traditionnaly, Most of the public housing were built in the suburbs of cities in order to benefit from accommodation equipment and facilities. However, development of pubic housing is often seen as a cause of urban decay or criminality. Concretly, public housing policy may be a potentially cause of segregation and urban poverty traps. This is partly because of concentration of individuals with low-incomes and because of an insufficient transport network in some areas. In the agglomeration, the public housing sector represents 24.8% of the primary residences. This result hides wide disparities between sub-regional administrative districts (see figure 4). Indeed, while this sector represents 16.6% of the primary residences in Paris, it represents nearly double in Seine-Saint-Denis or in Val-de-Marne (35.8% and 29.1% respectively). Moreover, we observe a belt around Paris and some neighbors in the Seine-Saint-Denis with a part of public housing higher than 50%. Paris Center shows the lowest values with parts inferior to 3%. Contrary to unemployment geography, public housing geography in the Paris agglomeration is more scattered. Neighbors with the most important part of public housing are not necessarily those with the highest unemployment rates. We do not see any correlation (positive 17

or negative) between the fact of living in a deprived neighborhood and the fact of living in a public housing. If administrators of public housing do not allocate randomly the housing, it is clear that their selection criteria differ from those of families if they have the possibility to choose their residential location. Figure 4: Public housing in the Paris agglomeration

Source: INSEE, Population Census, 1999.

On the whole, individuals residing in public housing have different socio-economics characteristics from others (Annex B). Individuals of foreign nationality or born in overseas departments are more represented in this type of housing than overall housing (41.1% against 33.5%). They are also less educated than average (60.3% against 41.7%) and are mainly from occupational status as “office workers” or “blue collars”. However, further section shows that despite these some statistics descriptive, public housing are not exclusively located in deprived neighborhoods.

5

Results

Previous sections have shown that Paris agglomeration may exhibit a high level of residential segregation associated with high unemployment rates. This is particularly the case in SeineSaint-Denis, in north-east of Paris or in a part of Val-de-Marne. However, statistical analysis is required before one can conclude on the role of neighborhood effects on unemployment in the agglomeration. We first run a bivariate probit that allows us to take into account the endogeneity bias of residential location choice in our estimation of spatial constraints effects 18

on unemployment (Tables 2 and 3). Secondly, we run a probit on the sub-sample of individuals living in a public housing (Tables 4 and 5). Table 2 and table 3 present estimated coefficients for the bivariate probit model explaining the individual employment probability in taking into account the location in a deprived neighborhood. The first estimation is run with our subjective definition. In this case, the neighborhood is defined as deprived according to its scores on different socio-economics characteristics (see previous section for more details). In the second estimation, the neighborhood is defined as deprived as soon as there is one (or more) Urban Sensitive Zone on its territory. In both case, results are quite similar. We find that individuals’ characteristics are determinants for the probability of living in a deprived neighborhood. Individuals of foreign nationality or French born in overseas departments are more likely to live in these disadvantaged neighborhoods. Education and professional status play also an important role in the probability of living in a deprived neighborhood. Indeed, educated workers are less likely to be disadvantaged than workers with a lower education. Occupational status as “independent workers” and “executive” are more favorable than “office-worker” or “blue-collar” (these last categories have a strong influence on the probability of living in a deprived area). The spouse’s characteristics are also determinant in the location choice. Variables are playing the same role as for the head of household. All things being equals, nationality or education level of the spouse may impact negatively on the probability of living in a deprived neighborhood. The findings show that individuals tend to gather according to their socio-economics characteristics. In our case, it appears that individuals with unfavorable characteristics in terms of education, occupational status etc. are more likely to live in neighborhoods defined as deprived. Coefficients for spatial variables are more difficult to interpret. Some of these variables seem to go in the way that literature describes whereas others seem to go against. For example, the immediate vicinity seems to be favorable: neighborhoods with the best job accessibility are not defined as deprived. Indeed, we observe a negative coefficent for the variable "part of jobs that are accesible in a 10 kilometers radius". This result might be explained by some neighborhoods located in the Hauts-de-Seine or in some Parisians districts. However, jobs accessibility measured by the average ditance between place of living and place of jobs (for the whole Iris) seems to be strongly and negatively correlated with living in a deprived neighborhood. In this case, it appears that neighborhoods that are located far away from jobs are less likely to be deprived. This second result is opposed to the first. In these conditions, it is difficult to say 19

which effect prevails. The second column gives effects of individuals’ characteristics, household characteristics and spatial constraints on employment probability. Our results are globally conventional regarding individual determinants of employment status. Individuals with high diploma or with a long vocational training are more likely to have a job. Similarly, we find that individuals who were previously independent workers or executives are less unemployed than others. To be a house owner appears to be favourable. This effects might be quite surprizing because for some authors the fact to be owner is unfavourable to mobility. Home-ownership may have a positive effect on unemployment duration through the fact that it represents a constraint for the job search process and success (Oswald, 1996). Furthermore, the spouse’s characteristics seems to have an influence on employment probability. To have a spouse with an university degree increases the probability to find a job. This positive effect may be justified by the fact that the individual will benefit from the network effect, skills of his spouse. Conversely, the nationality of the spouse seems to have a negative effect on employment probability. We find a negative effect of living in a deprived neighborhood. This effect seems more stronger for our objective definition. Living in a neighborhood where is located a Urban Sensitive Zone appears to be more unfavourable in terms of unemployment-to-work transitions that living in the neighborhood that we defined as deprived. Coefficients for these two variables are relatively high and show that individuals characteristics are not the only determinants of these transitions. However, if we are evaluating there the effect of living in a segregated area, we are not able to say if this negative effect on the job search process results from "peer effects" or results from a potentially discrimination against the inhabitants of these neighborhoods. Finally, to have a drivers’ license is favorable to employment probability. This result is justified by the fact that drivers’license implies a better mobility or a better accessibility to jobs centre. A result not comforted by coefficient sign for the part of jobs that are accessibles in a 10 kilometers radius. This variable presents coefficient that is negative. In addition, the average distance between place of living and place of birth is not significant. For these reasons, it’s difficult to settle on a possible problem of spatial mismatch. The coefficient ρ12 is the correlation coefficient between the residuals of each of the two probits. As it is statistically significantly different from zero, at the 5% level, then we have to estimate the two probits simultaneously. In this case, running simples probits may give 20

biased results. It confirms also the fact that residential location is endogenous to employment probability. Moreover, we see that this coefficient is higher in the case of a presence of an Urban Sensitive Zone. It suggests that endogeneity problem is more important in this second definition of a deprived neighborhood than in the first.

21

Table 2: Results of the bivariate probit model/Subjective definition Deprived Standard Employment Standard neighborhood error error Intercept -1,047*** 0,183 0,391** 0,172 Deprived neighborhood -0,278*** 0,093 Individuals characteristics Age -0,001 0,008 0,038*** 0,007 Age square -0,000 0,000 -0,000*** 0,000 House owner -0,486*** 0,019 0,161*** 0,023 Nationality French born in France Ref. Ref. French born in overseas depts 0,299*** 0,057 0,067 0,067 Foreign nationality 0,228*** 0,022 -0,321*** 0,023 Education level University degree -0,054** 0,027 0,075*** 0,028 High school final diploma Ref. Ref. At most a lower secondary dipl. 0,085*** 0,023 0,007 0,025 Occupational status Independent workers -0,135*** 0,028 0,382*** 0,037 Executives -0,292*** 0,028 0,389*** 0,028 Intermediate professions Ref. Ref. Office-worker 0,105*** 0,026 0,144*** 0,029 Blue-collar 0,215*** 0,023 -0,070** 0,026 Spouse’s characteristics Nationality French born in France Ref. Ref. French born in overseas depts 0,247*** 0,055 0,057 0,068 Foreign nationality 0,128*** 0,021 -0,031* 0,022 Education level University degree -0,138*** 0,025 0,087*** 0,026 High school final diploma Ref. Ref. At most a lower secondary dipl. 0,131*** 0,021 0,054** 0,024 Spatial characteristics Log of Home-to-work dist. -0,281*** 0,029 0,004 0,030 % of jobs in a 10km radius -0,627*** 0,079 -0,223** 0,093 Driver’s license -0,170*** 0,021 0,429*** 0,021 Subregional admin. districts Paris Ref. Ref. Hauts-de-Seine 0,795*** 0,030 0,040 0,028 Seine-Saint-Denis 1.925*** 0,033 -0,002 0,055 Val de Marne 1,031*** 0,034 0,017 0,037 No children -0,093*** 0,023 One children 0,019 0,022 Two children Ref. Ref. Three children 0,177*** 0,028 Four children or more 0,380*** 0,036 Rho 0,108** log likehood -29 423,147 LR test 3,690 Number of observations 47 265 Variables

Source: INSEE, Population Census (1999). Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively.

22

Table 3: Results of the bivariate probit model/Objective definition Urban Standard Employment Standard Sensitve Zone error error Intercept 0,119 0,160 0,568** 0,179 Urban Sensitive Zone -0,462*** 0,132 Individuals characteristics Age 0,001 0,007 0,037*** 0,007 Age square -0,000 0,000 -0,000*** 0,000 House owner -0,212*** 0,016 0,162*** 0,023 Nationality French born in France Ref. Ref. French born in overseas depts 0,268*** 0,050 0,073 0,068 Foreign nationality 0,153*** 0,019 -0,314*** 0,023 Education level University degree -0,009** 0,024 0,075*** 0,028 High school final diploma Ref. Ref. At most a lower secondary dipl. 0,052*** 0,021 0,008 0,025 Occupational status Independent workers -0,097*** 0,028 0,376*** 0,037 Executives -0,189*** 0,023 0,379*** 0,028 Intermediate professions Ref. Ref. Office-worker 0,101*** 0,023 0,148*** 0,029 Blue-collar 0,133*** 0,021 -0,068** 0,026 Spouse’s characteristics Nationality French born in France Ref. Ref. French born in overseas depts 0,149*** 0,052 0,061 0,068 Foreign nationality 0,091*** 0,019 -0,027 0,022 Education level University degree -0,111*** 0,025 0,087*** 0,026 High school final diploma Ref. Ref. At most a lower secondary dipl. 0,053*** 0,019 0,052** 0,024 Spatial characteristics Log of Home-to-work dist. -0,210*** 0,025 0,013 0,030 % of jobs in a 10km radius -1,925*** 0,076 -0,455** 0,116 Driver’s license -0,174*** 0,020 0,415*** 0,022 Subregional admin. districts Paris Ref. Ref. Hauts-de-Seine 0,209*** 0,023 0,023 0,026 Seine-Saint-Denis 0,401*** 0,028 -0,091*** 0,033 Val de Marne 0,093*** 0,034 0,017 0,037 No children -0,025 0,021 One children 0,010 0,019 Two children Ref. Ref. Three children 0,099*** 0,025 Four children or more 0,276*** 0,033 Rho 0,201** log likehood -34 301,528 LR test 5,151 Number of observations 47 265 Variables

Source: INSEE, Population Census (1999). Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively.

23

Tables 4 and 5 present estimated coefficients of variables explaining individual employment for a sub-sample of individuals living in a public housing. As we explain in previous section, we control the endogeneity bias of location choice by restricting our analysis on these individuals living in a public housing, presuming that they do not have the possibility to choose their place of living. Results are similar to the previous model. We find a negative effect of some socio-economics characteristics as the foreign nationality or low level of education. These effects are comparable to those found in the previous model. If individuals that were previously executives or independent workers are more employed than blue-collars or office-workers, we do not find a negative effect for the category of office-workers, as in the bivariate probit model. In addition, to have a driver’s license is favorable to employment probability. Concerning spatial indicators, the vicinity to jobs seems to be without effect on probability to be in employment. The result could be explained by the fact that important part of public housing in center of Paris beneficiate of a good accessibility to jobs. Finally, our principal variable, the location in a “deprived neighborhood” or the location in a neighborhood where there is an Urban Sensitive Zone reveals a negative effect on unemployment-to-work transitions. The effect estimated in this case appears less strong than in the previous case. With this second strategy we also find a more important effect of our objective definition of a disdvantaged neighborhood than the subjective on employment probability. Finally, our results show that the unemployment probability increases with location in a deprived neighborhood. Even though we control for socio-economics characteristics of individuals, we observe a negative effect of the residential location. In others terms, all things been equal, living in a deprived neighborhood decreases the employment probability for household heads. This evidence confirms the hypothesis that residential location may affect job search behavior through phenomenon of “peer effects”, role models, social networks or territorial discrimination, even if it is hard to say which prevails.

24

Table 4: Results of the probit model for individuals living in a public housing/Subjective definition Variables

Employment

Intercept Deprived neighborhood Individuals characteristics Age Squared Age Nationality French born in France French born in overseas departments Foreign nationality Education level University degree High school final diploma At most a lower secondary diploma Occupational status Independent workers Executives Intermediate professions Office-worker Blue-collar Spouse’s characteristics Nationality French born in France French born in overseas departments Foreign nationality Education level University degree High school final diploma At most a lower secondary diploma Spatial characteristics Log of Home-to-work dist. % of jobs in a 10km radius Driver’s license Subregional admin. districts Paris Hauts-de-Seine Seine-Saint-Denis Val de Marne Pseudo R² Likelihood ratio Number of observations

0,219 -0,076***

Standard error 0,317 0,036

0,037*** -0,000***

0,013 0,000

Ref. 0,152* -0,387***

0,092 0,039

0,006 Ref. -0,074*

0,054

0,178** 0,233*** Ref. 0,117** -0,127***

0,077 0,064 0,049 0,043

Ref. -0,047 -0,075**

0,092 0,039

0,139*** Ref. 0,053

0,049

0,084 -0,223 0,485***

0,055 0,158 0,034

Ref. -0,047 -0,107* -0,099* 0,075 715,30 12 620

Source: INSEE, Population Census (1999). Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively.

25

0,044

0,039

0,049 0,058 0,057

Table 5: Results of the probit model for individuals living in a public housing/Objective definition Variables

Employment

Intercept Urban Sensitive Zone Individuals characteristics Age Squared Age Nationality French born in France French born in overseas departments Foreign nationality Education level University degree High school final diploma At most a lower secondary diploma Occupational status Independent workers Executives Intermediate professions Office-worker Blue-collar Spouse’s characteristics Nationality French born in France French born in overseas departments Foreign nationality Education level University degree High school final diploma At most a lower secondary diploma Spatial characteristics Log of Home-to-work dist. % of jobs in a 10km radius Driver’s license Subregional admin. districts Paris Hauts-de-Seine Seine-Saint-Denis Val de Marne Pseudo R² Likelihood ratio Number of observations

0,209 -0,108***

Standard error 0,317 0,033

0,037*** -0,000***

0,013 0,000

Ref. 0,156* -0,386***

0,092 0,039

0,006 Ref. -0,075*

0,054

0,180** 0,231*** Ref. 0,119** -0,127***

0,077 0,064 0,049 0,043

Ref. -0,047 -0,075*

0,092 0,039

0,137*** Ref. 0,052

0,049

0,084 -0,291 0,482***

0,055 0,016 0,034

Ref. -0,051 -0,139** -0,116** 0,075 721,68 12 620

Source: INSEE, Population Census (1999). Notes: ***, ** and * denote significance at the 1%, 5% and 10% level respectively.

26

0,044

0,039

0,049 0,055 0,058

6

Conclusion

The aim of this paper is to examine how employment probability may be influenced by location in a neighborhood defined as deprived or by poor jobs accessibility. Even if the effects of individual characteristics, public policies on the job search process or the lack of formal education are well-known, effects of spatial structure of cities are underestimated in France. This paper focuses on Paris agglomeration and tries to highlight the potentially link between its spatial structure and employment probability of individuals. We first address the endogeneity of residential choice by estimating a bivariate probit. In this simultaneous equations system we estimate probability of living in a deprived neighborhood and probability of being employed. The identification of our system is allowed by using exclusions restrictions based on the number of children. We also run a probit model on the sub-sample of households living in a public housing presuming that the location choice is exogenous. Our results show that unemployment is exacerbated by residential segregation or neighborhood effects. All things being equal living in a deprived neighborhood, characterized by a socio-economic environment of a lower quality or characterized by the presence of one or more Urban Sensitive Zone decreases the employment probability. Furthermore, concerning the Spatial Mismatch hypothesis, it seems to be partially confirmed as the fact to be mobile (to have a drivers’ license )has positive effects on employment probability. Nevertheless, robustness of this finding is discussed by the fact that our others measures of accessibility to jobs are not significant even when we concentrate on individuals living in a public housing. In terms of public policies, it means that residential location is an important determinant of labor market outcomes. In this context, a relevant recommendation is to develop and to promote social mix. This requires measures such as the U.S. Federal Moving to Opportunity (MTO) program, the Gautreaux Program or the offer of public housing in France. In the latter case, the french law SRU (Solidarity and Urban Renewal) that "encourage" municipalities to have a threshold of 20% of public housing seems to be a step forward.

27

References [1] P. Apparicio. Les indices de ségrégation résidentielle: un outil intègre dans un système d’information géographique. Cybergeo, 134, 2000. [2] W. Bell. A probability model for the measurement of ecological segregation. American Sociological Review, 32:357–364, 1954. [3] R. Benabou. Working of a city: Location, education and production. Quaterly Journal of Economic, 108:619–652, 1993. [4] J K. Brueckner and Y. Zenou. Space and unemployment: The labour-market effects of spatial mismatch. Journal of Labour Economics, 21., 2003. [5] P. Choffel and E. Delattre. Habiter un quartier defavorisé: quels effets sur la durée du chômage ? Premières informations et premières synthèses, DARES, 43.1, 2003. [6] E. Coulson, D. Laing, and P. Wang. Spatial mismatch in search equilibrium. Journal of Labour Economics, 19:949–972, 2001. [7] J. Crane. The epidemic theory of ghettos and neighbourhood effects on dropping out and teenage childbearing. American Journal of Sociology, 96:1226–1259, 1991. [8] J. Currie and A. Yelowitz. Are public housing projects good for kids? Journal of Public Economics, 75:99–124, 2000. [9] D. M. Cutler and E. L. Glaeser. Are ghettos good or bad? Quarterly Journal of Economics, 112:827–872, 1997. [10] S. Danziger and H. J. Holzer. Are jobs available for disadvantages groups in urban areas ? Working paper, Michigan State University and University of Michigan, 1997. [11] S. Davis and D. Huff. Impact of ghettoization on black employment. Economic Geography, 48:421–447, 1972. [12] E. Duguet, A. Goujard, and Y. L’Horty. Les disparités spatiales du retour à l’emploi: une analyse cartographique à partir de sources exhaustives. Economie et Statistique, forthcoming, 2008. [13] E. Duguet, N. Léandri, Y. L’Horty, and P. Petit. Are young french jobseekers of ethnic immigrant origin discriminated against? a controlled experiment in the paris area. Annales d’Economie et de Statistique., forthcoming, 2009. [14] E. Duguet, Y. L’Horty, and F. Sari. Sortir du chômage en ile-de-france : Disparités territoriales, spatial mismatch et ségrégation résidentielle. Revue Economique, forthcoming, 2009. [15] C. Dujardin and F. Goffette-Nagot. Neighborhood effects, public housing and unemployment in france. Working paper, GATE, 05-05, 2007.

28

[16] C. Dujardin and F. Goffette-Nagot. Does public housing occupancy increase unemployment? Journal of Economic Geography, forthcoming, 2009. [17] C. Dujardin, H. Selod, and I. Thomas. Residential segregation and unemployment : the case of brussels. Document de travail, INRA-LEA, 0704, 2007. [18] O. Duncan and B. Duncan. A methodological analysis of segregation indexes. American Sociological Review, 41:201–217, 1955. [19] W. M. Evans, W. E. Gates, and R. M. Schwab. Measuring peer group effects: A study of teenage behavior. Journal of Political Economy, 100:961–991, 1992. [20] M. Fujita. Urban Economic Theory: Land use and city size. Cambridge University Press, 1989. [21] S. Gao and R. Johnston. The spatial mismatch hypothesis: Critique of research and proposed methods. Working paper, 2004. [22] L. Gobillon, T. Magnac, and H. Selod. The effect of location on finding a job in the paris region. Document de travail, CEPR., 06199, 2007. [23] L. Gobillon, H. Selod, and Y. Zenou. The mechanisms of spatial mismatch. Urban Studies, 44(12):2401–2427, 2007. [24] D. Goux and E. Maurin. Close neighbours matter: Neighbourhood effects on early performance at school. Economic Journal of Labor Economics, 117:1–24, 2007. [25] R. Guillain and J. Le Gallo. Employment density in ile-de-france: evidence from local regressions. Working Paper, 2009. [26] J. Heckman. Dummy endogenous variables in a simultaneous equation system. Econometrica, 46(4):931–959, 1978. [27] H. Holzer. Search method used by unemployed youth. Journal of Labor Economics, 6:1–20, 1988. [28] H. Holzer and K. Ihlandfelt. Customer discrimination and employment outcomes for minority workers. The Quarterly Journal of Economics, 113(3):835–867, 1998. [29] K. Ihlandfelt and D. Sjoquist. Job accessibility and racial differences in youth employment rates. The American Economic Review, 80:267–276, 1990. [30] K. Ihlandfelt and D. Sjoquist. The effect of job access on black and white youth employment: A cross-sectional analysis. Urban Studies, 28(2):255–265, 1991. [31] J. F. Kain. Housing segregation, negro employment, and metropolitan decentralization. Quarterly Journal of Economics, 82:32–59, 1968. [32] J. F. Kain. The spatial mismatch hypothesis : three decades later. Housing Policy Debate, 3:371–460, 1992. 29

[33] L. F. Katz, J. R. Kling, and J. B. Liebman. Moving to opportunity in boston: early results of a randomized mobility experiment. The Quarterly Journal of Economics, 116:607–654, 2001. [34] J. R. Kling, J. Ludwig, and L.F. Katz. Neighborhood effects on crime for female and male youth: Evidence from a randomized housing voucher experiment. Quarterly Journal of Economics, 120:87–130, 2005. [35] E. Korsu and S. Wenglenski. Distance physique, proximité sociale et inégalités devant le chômage. Working paper, 2008. [36] T. Lancaster. The Econometric Analysis of Transition Data. Cambridge University Press, 1990. [37] J. Liebman, L.F. Katz, and J.R. Kling. Beyond treatment effects: Estimating the relationship between neighborhood poverty and individual outcomes in the mto experiment. Princeton University Industrial Relations Section Working Paper, 493, 2004. [38] S. Lollivier. Endogénéité d’une variable explicative dichotomique dans le cadre d’un modele probit bivarié: Une application au lien entre fécondité et activité féminine. Annales d’Economie et de Statistique, 62, 2001. [39] G. Maddala. Limited-Dependent and Qualitative Variables in Economics. Cambridge University Press, 1983. [40] D. Mortensen. Job search and labor market analysis. Ch. 15 of Handbook of Labor Economics, 2, O. Ashenfelter and R. Layard, eds:North–Holland, 1986. [41] P. Oreopoulos. The long-run consequences of living in a poor neighborhood. Quarterly Journal of Economics, 118:1533–1575, 2003. [42] A. Oswald. A conjecture on the explanation for high unemployment in the industrialised nations : part 1. University of Warwick Economic Research Papers, 475, 1996. [43] O. Regan. The effect of social networks and concentrated poverty on black and hispanic youth unemployment. The Annals of Regional Science, 27:327–342, 1993. [44] C. L. Rogers. Job search and unemployment duration : Implications for the spatial mismatch hypothesis. Journal of Urban Economics, 42:109–132, 1997. [45] H. Selod and Y. Zenou. Social interactions, ethnic minorities and urban unemployment. Annales d’Economie et de Statistique, 63-64:183–214, 2001. [46] T. Smith and Y. Zenou. Spatial mismatch, search effort and urban spatial structure. Journal of Urban Economics, 54:185–214, 2003. [47] J. F. Thisse, E. Wasmer, and Y. Zenou. Ségrégation urbaine, logement et marchés du travail. Revue Française d’Economie, 14(4):85–129, 2003.

30

[48] C. Tiebout. A pure theory of local expenditures. Journal of Political Economy, 64:416–424, 1956. [49] E. Tovar. La ségrégation urbaine : représentation économique et évaluation éthique. Une application à l’Ile-de-France des années 1990. PhD thesis, EHESS, 2008. [50] E. Wasmer and Y. Zenou. Does city structure affect search and welfare ? Journal of Urban Economics, 51:515–541, 2002. [51] M. J. White. Segregation and diversity measures in population distribution. Population index, 52(2):198–221, 1986. [52] J. Wilde. Identification of multiple equation probit models with endogenous dummy regressors. Economic Letters, 69:309–312, 2000.

31

Annex A: Definition of deprived neighborhoods

Table 6: Deprived neighborhoods and others Variables Blue-collars Individuals without diploma Individuals without low degree Large families Individuals of foreign nationality

Deprived 27,6% 29,4% 23,6% 18,4% 24,4%

Others 10,3% 13,1% 16,0% 8,5% 11,7%

Total 14,0% 16,6% 17,6% 10,6% 14,4%

Source: INSEE, Population Census, 1999

Table 7: Deprived neighborhoods and Urban Renewal Area in the Paris Agglomeration Paris Deprived neighborhoods % of Deprived Neighborhoods Number of Urban Renewal Area Neighbor. with at least one URA % of Neighbor. with at least one URA

30 3,29 9 82 8,98

Source: INSEE, Population Census, 1999

32

Hautsde-Seine 98 15,91 23 138 22,4

SeineSaint-Denis 327 54,41 36 232 38,6

Valde-Marne 106 20,42 16 129 24,6

Total 561 21,18 84 581 21,93

Annex B: Summary statistics

Table 8: Characteristics of individuals living in deprived neighborhoods and others deprived Obs. % Individual characteristics Employment status Employed Unemployed Nationality French born in France French born in overseas dep. Foreign nationality Education level University degree High school final diploma At most a lower secondary dipl. Occupational status Independent workers Executives Intermediate professions Office-worker Blue-collar Household characteristics public housing Driver’s license Household owner Spouse characteristics French born in France French born in overseas dep. Foreign nationality University degree High school final diploma At most a lower secondary dipl. Total

Others Obs. %

Total Obs. %

7956 1391

85,12 14,88

34853 3065

91,92 8,08

42809 4456

90,57 9,43

4475 555 4317

47,88 5,94 46,19

26943 657 10318

71,06 1,73 27,21

31418 1212 14635

66,47 2,56 30,96

1652 1483 6212

17,67 15,87 66,46

18430 5971 13517

48,6 15,75 35,65

20082 7454 19729

42,49 15,77 41,74

639 765 1689 1689 4510

6,92 8,18 18,07 18,07 48,25

3687 13882 8065 4708 7457

9,72 36,61 21,27 12,42 19,67

4334 14647 9754 6397 11967

9,17 30,99 20,64 13,53 25,32

5232 7494 2395

55,98 80,18 25,62

7388 32159 15434

19,48 84,81 40,7

12620 39653 17829

26,7 83,9 37,72

4945 501 3901 1618 1837 3455 9347

52,9 5,36 41,74 17,31 19,65 36,96 19,78

27449 650 9819 17987 7451 12480 37918

72,39 1,71 25,9 47,44 19,65 32,91 80,22

32394 1151 13720 19605 9289 18372 47265

68,54 2,44 29,03 41,48 19,65 38,87 100

Source: INSEE, Population Census, 1999

33

Table 9: Characteristics of individuals living in neighborhoods with at least one Urban Renewal Area and others Urban Sensitive Zone Obs. % Individual characteristics Employment status Employed Unemployed Nationality French born in France French born in overseas dep. Foreign nationality Education level University degree High school final diploma At most a lower secondary dipl. Occupational status Independent workers Executives Intermediate professions Office-worker Blue-collar Household characteristics public housing Driver’s license Household owner Spouse characteristics French born in France French born in overseas dep. Foreign nationality University degree High school final diploma At most a lower secondary dipl. Total

Others Obs. %

Total Obs. %

8218 1312

86,23 13,77

34591 3144

91,67 8,33

42809 4456

90,57 9,43

5109 476 3,945

53,61 4,99 41,4

26309 736 10690

69,72 1,95 28,33

31418 1212 14635

66,47 2,56 30,96

2456 1538 5536

25,77 16,14 58,09

17626 5916 14193

46,71 15,68 37,61

20082 7454 19729

42,49 15,77 41,74

723 1412 1858 1665 3818

7,59 14,82 19,5 17,47 40,06

3611 13235 7896 4732 8149

9,57 35,07 20,92 12,54 21,6

4334 14647 9754 6397 11967

9,17 30,99 20,64 13,53 25,32

4357 7682 2928

45,72 80,61 30,72

8263 31971 14901

21,9 84,73 39,49

12620 39653 17829

26,7 83,9 37,72

5495 425 3610 2362 1944 5224 9530

57,66 4,46 37,88 24,78 20,4 54,82 20,16

26899 726 10110 17243 7344 13148 37735

71,28 1,92 26,79 45,69 19,46 34,84 79,84

32394 1151 13720 19605 9289 18372 47265

68,54 2,44 29,03 41,48 19,65 38,87 100

Source: INSEE, Population Census, 1999

34

Table 10: Characteristics of individuals living in public housing and others Public housing Obs. % Individual characteristics Employment status Employed Unemployed Nationality French born in France French born in overseas dep. Foreign nationality Education level At most a lower secondary dipl. High school final diploma University degree Occupational status Independent workers Executives Intermediate professions Office-worker Blue-collar Household characteristics Driver’s license Spouse characteristics French born in France French born in overseas dep. Foreign nationality At most a lower secondary dipl. High school final diploma University degree Total

Total Obs. %

11019 1604

87,29 12,71

42809 4480

90,53 9,47

7465 777 4381

59,14 6,16 34,71

31438 1216 14635

66,48 2,57 30,95

7617 2193 2813

60,34 17,37 22,28

19731 7457 20101

41,72 15,77 42,51

676 1546 2621 2604 5128

5,36 12,25 20,76 20,63 40,62

4334 14647 9754 6397 11967

9,16 30,97 20,63 13,53 25,31

10311

81,68

39665

83,88

8010 710 3903 7154 2622 2847 12623

63,46 5,62 30,92 56,67 20,77 22,55 26,69

32414 1152 13723 18374 9289 19626 47289

68,54 2,44 29,02 38,85 19,64 41,5 100

Source: INSEE, Population Census, 1999

35