Introduction
1Young people leaving education and entering the work force do not necessarily find permanent employment straight away. [1] Even if they are actively seeking work, this period of transition may involve stretches of unemployment. In the French labor market over the last 30 years, unemployment among men aged 15 to 24 has remained at least twice that of the active population as a whole (15-64 years). This has spurred the development of employment policies targeted at young people. Almost a quarter of employed under 26-year-olds are in assisted contracts (Aeberhardt, Crusson, and Pommier 2011). These subsidy schemes are aimed at those whose lack of qualifications and experience leaves them with the most difficulty finding a job. In recent years, place of residence has become an important criterion for deciding who benefits from these youth employment assistance policies. The emploi franc scheme launched in 2013 is aimed exclusively at young people living in areas with particularly high unemployment rates, known as zones urbaines sensibles (sensitive urban zones) or ZUSs. [2] Similarly, those living in quartiers prioritaires de la politique de la ville (priority neighborhoods), or QPVs, benefit from privileged access to assisted contracts, including the emplois d’avenir and CIE-Starter schemes. While using place of residence as a criterion may simply be intended to focus public resources in particular areas, it may also reflect an acknowledgment that local employment levels directly or indirectly affect access to work—indirectly, if local employment levels are just one variable closely linked to other phenomena affecting the labor market; directly, if local employment levels affect a young person’s chances of finding a job.
2Beyond national averages, the difficulties young jobseekers face vary considerably from one area to another. In 1999, youth unemployment in the 10% of urban areas with the lowest rates of unemployment was below 11.5%, compared with 42% in the 10% of neighborhoods with the highest rates. [3] By way of comparison, the gap between the top and bottom deciles is only half as large, 18%, when we take the entire labor force into account. The difficulties encountered by young people from the outskirts of French cities are well known. Life in these disadvantaged areas is associated with lower chances of employment. But we need a precise description of the mechanisms explaining the effect of residential location on individual success. Geography may just be the neutral background for individual inequalities, with the spatial effects observed resulting simply from a spatial sorting among individuals by their probability of finding work. Conversely, place of residence may indeed have a causal effect on access to employment. In this second case, we can envisage a number of channels by which the effect works. Physical distance to available jobs, or discrimination on the basis of place of residence, may affect access to employment. The local employment level is also likely to have a direct impact on a young person’s ability to find work: being surrounded by people in employment may influence access to jobs by disseminating norms and information. If this is true, the fact of an individual being employed or not generates externalities, since it affects the probability that those around them will find work. We will speak of the “social” effect of local unemployment levels to designate the aggregate effect of these different mechanisms of individual interactions between employment situations. The effect is social because it does not use markets as an intermediary (Ioannides and Topa 2010): the effect of a local employment level is not related to a connection between labor supply and demand—pressures in the labor market, indirect employment generated, and so on. The effect has to do instead with the interdependence of job-seeking behavior among individuals exposed to the same environment. This social effect is endogenous if it is genuinely others’ employment situation, rather than other contextual effects, that influences the probability of being employed.
3In the presence of such a social effect, the local employment level is likely to stabilize at different values. We can speak about multiple equilibria. Employment levels in areas with similar characteristics may be lower or higher, depending on the initial employment level and individuals’ sensitivity to it. In a framework of multiple equilibria, the social effect of local employment levels may affect public labor policies. The social effect may increase their effectiveness: a spillover effect, for instance, may mean that assisted contracts bring about a lasting drop in the local unemployment rate by making a better employment equilibrium possible. Conversely, a low employment rate at the outset may hinder positive developments if public policy is unable to cause a shift from one equilibrium to another. The local social effect would then be characterized by the small number of employed individuals, which remains predominant in spite of marginal variation in the employment rate. In this case, cutting off support to the labor market may bring about a return to the initial equilibrium. The social effect of local employment levels—whether it works as a catalyst for lasting changes or a force sustaining the status quo—influences the success of public employment policies.
4This article aims to estimate the social effect of local employment. It is based on an analysis of young men moving into the labor market in relation to the employment situation of others who originally lived in the same area. We describe an empirical strategy for distinguishing the influence of neighborhood behavior on individual behavior in urban areas (Manski 1993; Brock and Durlauf 2001, 2007). Studying young people leaving the school system allows us to avoid some potential endogeneity problems. These individuals are unlikely to affect the local labor market equilibrium, thus preventing reflection effects (Manski 1993). New entrants to pre-existing markets represent only a very small part of the workforce. Furthermore, where these young people lived prior to entering the labor market is not directly determined by their employment status. This is particularly true since most young people at the end of schooling still live in the family home, in the same area where they started school. Our subjects’ place of residence is determined by their parents’ past decisions about where to live and is not directly influenced by their employment status (Ihlanfeldt and Sjoquist 1991). Finally, geo-localized data allows further control of the potential endogeneity of place of residence by allowing for an econometric approach taking advantage of local spatial variations in employment. Using nested spatial subdivisions allows us to neutralize any effects specific to the district of the neighborhood in question (Bayer, Ross, and Topa 2008). This specificity, combined with the estimation strategy used, allows us to identify local social effects on employment.
5The Génération 1998 and Génération 2004 surveys carried out by the Center for Studies and Research on Qualifications (Céreq) provide a representative sample of 60,000 young people leaving the French education system in 1998 and 2004. Precise additional information on place of residence on leaving education, made available at the end of 2010, allows us to work at a sub-municipality level. Our analysis works at the level of îlots regroupés pour l’information statistique (aggregated units for statistical information), or IRISes, a system developed by the French National Institute of Statistics (INSEE) dividing urban metropolitan France into homogeneous spaces of about 2,000 inhabitants each. Spatial variations in employment are used to analyze the relationship between employment levels in the subjects’ place of residence at the end of their studies and whether or not they are employed three years later. Using the hypothesis that residential location is exogenous for IRISes in the same area, the results highlight the effect of the employment situation within an IRIS on a young person’s probability of being employed.
6The article proceeds as follows. We begin by discussing local variability in employment access, and the potential influence of this on young people’s access to work. After this initial literature review, the second section describes the estimation strategy used. The third section presents the data, which combines the individual results of the Génération surveys with local employment information from censuses. The fourth section describes the study sample: young men at the end of their studies living in metropolitan urban France. Finally, the fifth part presents the results of the estimation strategy. They highlight the influence of the local employment situation on the probability of finding work: a 1 percentage point increase in the employment rate raises the average likelihood of obtaining a job by about 0.2 percentage point. While individual sensitivity to this marginal effect varies according to individual characteristics, this figure would be zero in areas marked by high levels of youth unemployment.
Place of residence and employment: a review of the literature
From spatial mismatch to neighborhood social effects
7The discovery of high variability in employment rates depending on place of residence within the same city has generated a body of economic literature seeking to verify the existence of a causal effect between location and employment access, to measure it, and to determine its underlying explanatory mechanisms: this literature first focused on the United States; and aimed particularly to explain the impact of African-American urban segregation on labor market inequalities. The scope of the work was then extended and adapted to other contexts—both to other populations and other urban structures. Such work relies on the idea of a spatial division between place of residence and place of employment as an explanatory hypothesis. The mechanisms it identifies can be divided into two groups, depending on whether this distance is primarily physical (“spatial mismatch”) or ethnic and social.
8The first set of explanations originates with Kain’s work (1968, 2004); they underscore the effect of physical distance between place of residence and place of employment. Residential segregation caused by housing market discrimination is a barrier to employment access. Economists have used the same principle to study other mechanisms. Gobillon, Selod and Zenou (2007), and Gobillon and Selod (2014) describe these phenomena, in which geographic distance to place of employment affects access to it. For example, transport costs between home and work may be too high relative to the salary offered, limiting the number of job offers that can be accepted. Other mechanisms involve less effective job-seeking in areas with fewer jobs, lower search intensity due to lower incentives generated by place of residence—lower cost of living, especially housing, opportunities for gray-sector work, etc.—and increased search costs that limit the size of area where jobs are sought. Employers may be concerned that travel time will lead to fatigue and absenteeism, negatively affecting productivity.
9The second set of explanations is based on ethnic and social distance between individuals seeking work and the individuals they meet while doing so. The importance of these mechanisms is not directly related to the degree of geographic distance (Gobillon, Selod, and Zenou 2007). It is based solely on differences in area of residence based on ethnic and social grouping. This may lead to reluctance to interact with people from other groups. Discrimination by consumers towards the employees they interact with (Holzer and Ihlanfeldt 1998), or by employees towards their colleagues, may limit hiring from groups distant from those that consumers or employees belong to. Moreover, the limited information an employer has about a candidate may lead them to infer their characteristics based on the average characteristics of the candidate’s neighborhood. Such statistical discrimination can lead employers to reject applications from neighborhoods with bad reputations (the “redlining effect,” Zenou and Boccard 2000). [4]
10This division into two sets of explanations becomes less clear-cut when we descend to a more fine-grained level of geographical analysis. Some mechanisms based initially on physical distance also function when based on criteria of social distance. The effectiveness and intensity of job-seeking, which are mechanisms evoked in the context of spatial mismatch, may be more affected by the composition of the immediate neighborhood than by distance to places of work. Much economic research (Ioannides and Topa 2010; Topa and Zenou 2015) on education (Goux and Maurin 2007), housing (Ioannides 2011), and employment (Bayer et al. 2008) has highlighted the importance of neighborhood composition on individual decisions and behaviors. Crane (1991) uses data from the United States to argue that disadvantaged neighborhoods are particularly vulnerable to the spread of social problems among young people (“epidemic theory”).
11Among these neighborhood effects, neighbors’ employment situation is particularly likely to affect young people’s access to employment. Two mechanisms cause this social effect. Levels of local employment can act as a social norm, affecting the intensity of job-seeking. The higher the proportion of individuals employed, the more likely there will be significant social stigma attached to joblessness (Clark 2003; Stutzer and Lalive 2004). Young people may seek to emulate peers in their neighborhood who have jobs, in addition to those normally identified as role models (Mota, Patacchini, and Rosenthal 2014). Furthermore, the surrounding population’s participation in the labor market can promote the dissemination of information on job opportunities, affecting the effectiveness of job-seeking (Ioannides and Loury 2004). The number of individuals in work increases the likelihood of hearing about vacancies and of receiving recommendations to these employers (Andersson, Burgess, and Lane 2014; Hellerstein, McInerney, and Neumark 2011; Hellerstein, Kutzbach, and Neumark 2014).
12The analysis of job-seeking methods among young people highlights the crucial role of personal acquaintances, whether friends or family (Holzer 1988). This is especially true for school leavers, who cannot yet rely on a professional network. But the development of this social work, whether among friends of family, often goes hand in hand with geographical proximity (Patacchini, Picard, and Zenou 2015). In France, Catherine Bonvalet’s analysis of family environments has highlighted the importance of intra-municipality exchanges between members of the same non-cohabiting family (Bonvalet 2003), while analyses of day-to-day interactions (Héran 1988) and mutual support networks emphasize the local character of these relationships. [5] A neighborhood influence is all the more likely as most young people have remained in the same area since starting school. Such lack of movement encourages the local character of knowledge networks.
Measuring the effect
13In practice, it is not enough to observe a correlation between an individual’s situation and that of his or her group if we are to establish the existence of a social effect (Manski 1993). Two competing mechanisms can give the same result. Similar behavior by individuals in the same neighborhood could also be explained by their exposure to similar neighborhood characteristics (contextual effects), or by a similar profile resulting from individuals being sorted spatially into homogeneous neighborhoods. Various strategies have been used to distinguish between these different effects. It is particularly important to consider whether group membership—in this case, residence in a particular neighborhood—is endogenous. Such work lies at the intersection of the literature on spatial and peer effects. [6]
14A first possible strategy for taking account of the endogeneity of place of residence follows Dahl’s (2002) method of modeling individuals’ choice of residence. A second focuses on groups whose residential location is imposed—whether, in the case of young people, by parents (Ihlanfeldt and Sjoquist 1990) or, in the case of refugees, by the state (Dustmann and Damm 2014; Damm 2014). This second approach benefits from a natural experimental framework provided by countries like Denmark and Sweden, where refugees are assigned their place of residence randomly. A third type of approach uses various econometric techniques, sometimes alongside the first two methods. Controlling for observed individual characteristics can be linked to methods that involve instrumental variables (Dujardin and Goffette-Nagot 2010), an evaluation of the magnitude of the selection effect (ibid.), or an exploitation of fixed effects on a spatial or temporal basis (Weinberg, Reagan, and Yankow [2004]). [7] Bayer et al. (2008) have proposed an approach based on the econometric exploitation of fine-grained spatial variations that highlight the influence of neighbors’ employment situations within a single city block. Related work has used simulation-based employer-employee surveys to reveal the relationship between place of residence and employment status (Hellerstein, McInerney, and Neumark 2011; Hellerstein, Kutzbach, and Neumark 2014).
15French studies tend to emphasize the importance of location for employment access. Less than a third of spatial disparities in length of unemployment in the Paris region can be explained by differences in individual characteristics (Gobillon, Magnac, and Selod 2011). Living in a disadvantaged neighborhood is associated with a significant negative effect on employment among men (Sari 2012). The unemployment rate in these neighborhoods is on average 7.5% higher than elsewhere; living in them is associated with a 6% decrease in the likelihood of being employed. The magnitude of the overall effect of place of residence may vary, however, depending on location and methods used. Dujardin and Goffette-Nagot (2010) initially found that living in neighborhoods around Lyon with an unemployment rate 11 percentage points higher than average—the most disadvantaged 25% of neighborhoods—is associated with a 2.1 percentage points increase in probability of unemployment. However, controlling for the endogeneity of place of residence leads to the conclusion that living in these districts has no significant effect on the probability of being unemployed.
16Understanding these phenomena relating to place of residence requires us to distinguish between their various underlying mechanisms. This work remains to be done; this article is part of this broader task. Certainly, correspondence audit studies have already revealed discriminatory behavior by employers in response to spatial criteria. Comparing response rates to job applications distinguished only by candidates’ place of residence allows us to identify this effect (Petit et al. 2013). Data from job sites helps us determine the effect of place of residence on job-seeking by allowing us to simultaneously observe where jobs are available and where candidates are looking (Marinescu and Rathelot 2014). However, we do not necessarily have equivalent means for identifying other mechanisms. The existence of a social effect of local employment levels on access to jobs is in general impossible to verify except in very specific cases like certain employer-employee surveys. [8] The effect of neighbors’ employment situation can only rarely be distinguished from other types of distance from employment. On the one hand, difficulties finding work in high-unemployment areas may be as much due to physical distance from places of employment as to employer discrimination. In particular, it is still difficult to measure accessibility to jobs from a particular place (Bunel and Tovar 2014), in spite of progress made in this area (Briant, Lafourcade, and Schmutz 2015). On the other hand, the employment situation of the individuals studied itself contributes to the local employment level, generating a reflection effect that can be difficult to circumvent (Manski 1993). The originality of this article’s approach lies in exploiting a new type of data for studying social effects of local employment levels in France. The use of interlocking, intra-municipality spatial divisions allows us to study the effect of the local employment equilibrium on school-leavers’ access to employment in urban areas.
Empirical approach
17We model the employment of an individual i living in a residential block-group g(i) of a neighborhood lg(i) with a binary variable yig(i) that is equal to 1 if the individual is employed:
19where Xi is the vector of observable individual characteristics, 
is the neighborhood employment level, and 
is the vector of the characteristics of the neighborhood lg(i).
20The existence of a social interaction phenomenon between individuals in the same neighborhood is shown through parameter β2. Two other phenomena are likely to lead to a similarly indirect relationship between the individual and local employment levels, biasing our estimate for this parameter. On the one hand, location may just be a confounding variable without any causal effect on employment: the apparent link between place of residence and employment would then be the result of spatial sorting according to unobserved individual characteristics. For all i, j, if lg(i) = lg(j) then 
. On the other hand, the omission of local characteristics affecting employment access can also generate this correlation, [9] since this unobserved local component is common to all individuals in the neighborhood:
22The choice of data used in this article allows us, in part, to avoid problems of the endogeneity of location with regard to employment. We join this approach with a method of identification based on the assumption that, while individuals choose a neighborhood, their location within it is random. We are influenced here by the method for exploiting nested spatial breaks developed by Bayer et al. (2008). This approach makes it possible to estimate the effect of neighborhood characteristics when there is enough local variance.
23In practice, the hypothesis of exogeneity with regard to employment at a local level is based on the following reasoning. Individuals are likely to choose a neighborhood, but their final location within it depends on specific external factors of the search for housing like the availability of apartments. Because the analysis focuses on young people leaving the school system, several elements support this idea. Being able to consider where they will live before the end of their studies and several years before they find a job limits the influence of their work on their initial place of residence. The individuals studied here generally do not make this choice: for the majority of young people living at home, their parents choose which neighborhood to live in, a choice was made years before the question of their children’s access to employment arises. Local characteristics, only one component in this initial decision, may since have evolved considerably. In the case of young people who have already left the family home, their place of residence, while not necessarily independent of their parents, [10] is primarily tied to their studies. Finally, the hypothesis that location is exogenous depending on neighborhood is supported by the small spatial scale used. Each group of housing blocks (block-group) within a neighborhood considered here is a small space of relatively homogeneous homes. Each, however, has sufficient people living within it that an individual’s situation does not significantly affect the local average calculated from census data.
24With this approach to variations in employment between block-groups, the initial specification of the probability of employment becomes:
26where the residual εi g(i) can be decomposed to take potential spatial sorting effects into account:
28assuming that ui g(i) is independent of the covariates.
29The specificity of the neighborhood is captured through the fixed effect 
. Identifying the effect of local employment on access to work is based on intra-neighborhood variations within lg(i), between the different housing block-groups g(i) that make it up. This is estimated with a logit regression. We assume the residuals ui g(i) follow a Gumbel distribution (Type 1). [11]
30Despite the small size of the neighborhoods concerned, and the fact that finding a place of residence within them is disconnected from entry into the labor market, we may still suspect that location within a neighborhood is not totally exogenous to the process of access to work under consideration. We can add control variables to capture the differences in characteristics between different places of residence in the same neighborhood:
32This empirical strategy should make it possible to identify the social effect within each neighborhood. In a given neighborhood, residential blocks are spread over a very limited area, thus they face a similar situation both in terms of reputation and distance from jobs. Within such a limited space, living in one block rather than another does not practically change one’s physical distance from available jobs. Similarly, living a few hundred meters away in a similarly labeled area is unlikely to alter perceptions about where one lives. Nevertheless, there are significant differences in employment at a very local level. If the spatial framework for these differences is so scant that it is almost uniform, we should look for their origin in terms of the influence neighbors can exert through their own employment situation. The empirical approach we adopt allows us to analyze how local employment differences are transmitted to new entrants to the labor market and influence their careers.
Using geo-coded data from the Génération surveys
33The model is developed on the basis of the Céreq Génération surveys. These are conducted regularly with a representative sample of young first-time students leaving the French education system in a given year. The individuals are surveyed three years after they have left school. Some of them are also re-interviewed once or twice in the subsequent decade. In addition to information on these young people’s entry into the labor market, these surveys provide a great deal of other information about them: their gender, age, socio-economic status, education level, household characteristics, and the place of birth and the nationality at birth of their parents.
34This paper uses the Génération 1998 and 2004 surveys, carried out in 2001 and 2007 respectively on the cohorts who left education in 1998 and 2004. The use of two sets of surveys is needed to have enough individuals for a sufficiently fine level of spatial analysis. Céreq ensures the comparability of the two surveys’ data: in addition to the similarity of the survey procedures (Aliaga et al. 2010), specific efforts are made to ensure that the data provided is strictly identical. [12] The geo-coding work carried out by these two surveys makes it possible to locate subjects’ place of residence at the end of school at the sub-municipality level. We have this information at the level of the IRIS, the smallest division for which INSEE provides data. The type of habitat within an IRIS is homogeneous, and each has about 2,000 inhabitants; their limits are based on natural boundaries in the fabric of the city like main roads, railways, and streams. [13] They allow researchers to divide municipalities with more than 5,000 inhabitants into far smaller areas. Mainland France can be divided into 50,000 such block-groups. Individuals’ residential location, denoted g(i), refers to the IRIS they live in. Our definition of neighborhoods is based on these IRISes. Such sub-municipality divisions are restricted to sufficiently populated areas. The analysis therefore excludes the rural and smaller urban areas.
35The neighborhood lg is made up of a set of block-groups (IRISes) called “grand-quartier” (district). While it may refer to the whole municipality (“commune”), most municipalities are composed of several districts. Figure 1 gives an illustration of these divisions for part of the 20th arrondissement of Paris. The smallest units are IRISes. The dark-colored IRIS covers an area of about 6 hectares (15 acres). It is part of the district of Charonne, represented by the hatched area, which has an area of less than 2.5 km2.
Definitions of a block-group (IRIS) and a district (grand-quartier)
Definitions of a block-group (IRIS) and a district (grand-quartier)
Census data on local employment levels
36The data from the Generation survey have been supplemented with contextual variables derived from the census which help us to characterize the local employment situation. This data allows us to recover local employment rates at the block-group (IRIS) and district (“grand-quartier”) levels. We thus consider all individuals in the area, and not only those interviewed in the Génération surveys or a group of local acquaintances. While small, the geographic units studied include enough people that the individuals’ influence on local unemployment rates is negligible. Moreover, the local rates used are intended to correspond to the situation at the moment when the subjects left education (1998 and 2004) and before they were surveyed (2001 and 2007). Employment indicators at the sub-municipality level are not available on an annual basis. The data from the closest censuses are used instead: the 1999 comprehensive census for the 1998 cohort, and the 2004-2008 annual census surveys for the 2004 cohort.
Characteristics of the sample
37We study the impact of local employment levels on young people’s access to work using a subsample of the Génération surveys. Firstly, the estimation strategy is restricted to urban areas divided into IRISes: municipalities with fewer than 5,000 inhabitants are thus excluded and about 60% of the individuals in the initial sample are retained. [14] Young people in sixth grade (CM2) outside mainland France are also excluded in order to keep the school curriculum homogeneous. To limit cases of inactivity when the subjects were surveyed, we consider only men. [15] To ensure that results are not distorted by extreme situations associated with very small local populations, we truncate the distribution of young people at the first and ninety-ninth percentiles according to the employment rate in their residential block-group (IRIS). As a final restriction on our estimation sample, the spatial support of the analysis is restricted to districts where at least two young respondents with different employment situations are living in different block-groups (IRIS). In order to be able to exploit intra-neighborhood variations in the explained variable, individuals observed within a district must not all be employed or all unemployed. Our estimates are therefore based on two thirds of men residing in an IRIS-based urban area at the end of their studies. Our selection does not alter the average employment characteristics of these areas, with the three indicators measuring numbers of employed persons remaining at similar levels (see Appendix 1, Table A1). Similarly, characteristics of individuals and their parents remain close. On the other hand, the proportion of employed individuals in the Génération survey three years later is lower (74.8% vs. 82.4%).
38The average age in the sample is 22. Most subjects are educated to baccalauréat level or higher. [16] Nearly three-quarters of them still lived with their parents at the end of schooling. In most cases their place of residence at the end of studies is similar to where they lived when beginning secondary education: nine out of ten still live in the same municipality where they began secondary school. Consequently, the neighborhood in which they live at the end of their studies is the one to which the majority of them have had most exposure. In more than half of cases, their parents are either blue-collar or white-collar workers (“ouvriers ou employés”), with both parents working. In 19% of the cases, one parent is an immigrant; in more than half of cases, this parent comes from an African country.
39Three years after graduation, 75% of these young people have a job. But this average conceals large—particularly spatial—disparities. There is a positive correlation between youth employment levels in place of residence at the end of schooling and their employment situation three years later (see Figure 2). The average probability of being employed varies by more than 10% depending on the neighborhood employment levels when they leave school. Employment is higher among 15 to 24-year-olds who were living in a high-employment block-group (IRIS) when they finished school. This relationship holds more strongly for individuals with fewer academic or professional qualifications.
Neighborhood employment level at the end of studies and work three years later
Neighborhood employment level at the end of studies and work three years later
The variability of employment within neighborhoods and the population homogeneity within districts
40Employment levels vary significantly at a very local level. For example, in Marseille, the unemployment rate for 15 to 24-year-olds is only 15% in certain southern districts like Saint-Giniez in the 8th arrondissement, near the Vélodrome. In contrast, it frequently exceeds 60% in the north of the city. Beyond these large city-wide spatial differences, the youth unemployment rate also varies within neighborhoods. Within Saint-Giniez, the unemployment rate ranges from 9% to 17% between different housing block-groups (IRISes). Similarly, at an equivalent distance from the town hall, close to the city center, the unemployment rate varies from 40% to 70% within the Saint-Mauront district in the 3rd arrondissement. The area concerned is only 1.5 km2. Nonetheless, analyzing the six zones defined by INSEE’s IRIS scheme highlights how long-lasting variations in employment are at this very local level.
41Our empirical approach is based on the hypothesis that, within a neighborhood, variations in employment rates between block-groups (IRISes) is not correlated with the ability of young entrants into the labor market to obtain employment, contingent on their observed characteristics. One might verify the random allocation of young people across the block-groups of each neighborhood experimentally by testing whether the distributions of predetermined individual characteristics in block-groups of the same district are identical. The size of our sample, which includes only a small number of individuals for each district, does not allow for such an experiment. But the use of census data makes it possible to analyze the importance of territorial variations in the variance of certain individual characteristics and, above all, to compare the power of different geographical divisions to explain variations in individual characteristics.
42Using the 1999 population census, we compare the proportion of this variance explained respectively by variations between municipalities (“communes”), districts (“grand-quartiers”), and block-groups (IRISes) for three individual characteristics: being educated to or beyond the baccalauréat, being born in the same department where one lives at the time of the census, and being unemployed at the time of the census (see Table 1). This variance is measured by the coefficient of determination (R2), adjusted to the number of explanatory variables obtained from the regression of each of the characteristics studied over the indicators of belonging to the geographical areas considered. [17] We compare the results obtained for the entire urban metropolitan population, as well as for sub-samples of individuals under 30 and 25 who correspond better to our population of interest. The selected sample contains 1,770 municipalities, 3,182 districts, and between 14,138 and 14,269 block-groups depending on age category.
Proportion of variance explained (%) according to geographic division
Proportion of variance explained (%) according to geographic division
Results of linear regressions carried out on data from the 1999 census. Each coefficient of determination corresponds to the regression of the characteristic indicated in each row over the indicators of membership in geographical divisions in each column. The coefficients of determination are adjusted for the number of units of geographical division (R2 adjusted).43The results show relative stability in the proportion of variance explained (adjusted R2) according to the geographical level in question. For under-25s, this is between 9.6% at the municipality level and 11.2% at the block-group (IRIS) level depending on diploma level. For the place of birth, the interval is 1.5-1.8%. The variation is greatest for the unemployment situation: the proportion increases from 3.9% to 6.4% when we take into account variations at block-group level within urban municipalities. The size of these differences in variance between and within municipalities decreases when we consider a wider population of under-30s, and even more when we consider the total labor force. At the sub-municipality level, the difference in variance explained by districts and by block-groups is relatively low: 1.1% for diploma level, 0.1% for migration, and 1.4% for unemployment, which is still the characteristic that shows the most infra-urban variability. When connected to the variance explained by differences between districts, the block-group (IRIS) variable allows us to increase the explained variance of unemployment by nearly 30%, while the education variable only allows for an increase of 10%, and geographical origin allows only for 6%.
44These results show the importance of sub-municipality variations in explaining unemployment, although such local variations are less significant for the other characteristics considered. Variations in employment happened therefore at a very local level compared to other individual characteristics considered. Even within a district with a homogeneous population, the different block-groups (IRISes) that make it up differ in employment level. But these results are neither a necessary nor sufficient condition for verifying our model’s central hypothesis. Our empirical approach is based on the absence of inter-IRIS selection, conditional on residence in a district, but also on all observed individual characteristics. Diploma level and geographical origin are therefore among the control variables used in our analysis.
Neighborhood employment levels have a significant effect on the probability of being employed three years later
45Estimating the district fixed effects model shows that an IRIS’s employment rate has a significant effect on labor market integration. The marginal effect of the local proportion of employed persons among 15 to 24-year-olds in the labor force, [18] calculated for a representative individual, [19] is 0.18, significant at the 5% threshold (see table 2, col. 2). The marginal effect is sizeable: an increase (or decrease) in the local employment level by 1 percentage point compared to the average is associated with an increase (or a decrease) in the probability of being employed of around 0.2 percentage point. This result supports the idea that the neighborhood employment rate has a significant effect on labor market entry. It is possible to test the robustness of the hypothesis that intra-district location is exogenous by introducing contextual variables specific to each block-group (IRIS). The combination of the data and estimation strategy used ensure this exogeneity, which seems to be verifiable empirically (see Table 2). By using variables specific to each block-group (IRIS), we try to determine whether the result initially obtained is likely to be explained by compositional differences within block-groups (IRISes) in the same district. The apparent link between the local unemployment level and the probability of being employed could be the result of residual structural differences between block-groups (IRISes)—type of housing, social composition, shops and nearby services, and so on—which might affect both the variables we are interested in. However, taking these contextual variations into account using aggregation coefficients (see Appendix 2) does not modify the results. While it tends to accentuate them, these differences are not statistically significant (see Table 2, col. 1). The initial result remains robust when we relax like this the hypothesis that intra-district location is exogenous.
Marginal effect of employment levels on entry into the labor market
Marginal effect of employment levels on entry into the labor market
Estimated maximum likelihood models based on Céreq’s Génération 1998 and Génération 2004 surveys.Standard deviations are given in brackets. Significance: *indicates p < 0.1, **p < 0.05, and ***p < 0.01.
The models include district fixed effects and take into account the following control variables: level of education, whether the individual retook a grade before sixth grade, age on leaving school (by level of education and deviation from normal leaving age), type of household, whether the individual is a parent, past residential mobility, parents’ socioeconomic status and job situation, and whether parents immigrated to France (from Africa, from other regions).
Like regression (1), regression (2) uses the proportion in employment among active 15 to 24-year-olds in the IRIS as its indicator. The marginal effect, calculated as 0.18, signifies that an increase (or, respectively, a decrease) of 1 percentage point in the employment rate would give the average individual an increase (or, respectively, a decrease) of 0.18 percentage point in their probability of being employed.
46The results are quite similar when other indicators of local employment are used (see Table 2, Col. 3 to 6). By relating the number of employed individuals to the total population (employment-to-population ratio) for 15 to 24-year-olds, the effect is only less significant (10% threshold). The large proportion of 15 to 24-year-olds still attending school, and therefore inactive, likely explains this difference. This second group, which has a less systematic link to the labor market, has less effect on employment levels. Finally, the effect of the employment rate of 15 to 64-year-olds becomes insignificant at the previous threshold if we do not control for contextual characteristics. These findings suggest that it is active people within the same age group, local peers, who contribute to the social interaction effect in employment access. It should be noted, however, that the different effects estimated remain close, and do not differ significantly from each other at the 5% threshold.
The heterogeneity of the effects
47For the sake of simplicity, the following analyses will be carried out, with the share of 15-24-year olds in employment among the active population of the same age as our variable of interest. This variable is likely to correspond to the reference group of neighborhood peers in terms of employment. It should be noted that the use of alternative indicators leads to similar effects and does not fundamentally change our conclusions. These do not include block-group (IRIS) characteristics, which, if taken into account alongside district fixed effects, do not negate the effects of local employment levels—indeed, the reverse is true. Regression (2) will therefore serve as the reference specification for the rest of the analysis.
48Previous estimates seem robust for the chosen specification (see Table 3, part 3-A). In the reference specification, the local employment effect is taken into account using this indicator alone (Col. 1). The aim is to verify that this effect is general and not limited to a particular group of individuals. Variables that combine the employment indicator with individual characteristics (diploma, living at home, immigrant origin) are introduced into the regression (Col. 2 to 5). We can verify that the effect of the initial variable remains significant after this modification and that its size changes only marginally. None of the combined variables affect it in a significant way. The average effect of neighborhood youth employment levels seems likely to affect each new entrant to the labor market.
Marginal effect of local employment level on entry into the labor market
| (1) | (2) | (3) | (4) | (5) | |
|---|---|---|---|---|---|
| Interactions included | |||||
| Interaction of employment level with qualifications | No | Yes | No | No | Yes |
| Interaction of employment level with living at home | No | Yes | Yes | No | No |
| Interaction of employment level with immigrant background | No | Yes | No | Yes | No |
| 3-A. Marginal effect of local employment level on entry into the labor market: accounting for combined effects | |||||
| Proportion in employment among 15 to 24-year-olds able to work | 0.1802** (0.0764) | 0.1614* (0.0902) | 0.1678* (0.0924) | 0.1787** (0.0748) | 0.1570** (0.0751) |
| 3.B. Heterogeneity of marginal effect of employment level depending on qualifications, for reference subjects distinguished only by this characteristic | |||||
| No diploma, no qualifications | 0.1828** (0.0922) | 0.1841 (0.1481) | 0.1447 (0.1015) | 0.1845** (0.0924) | 0.2452** (0.1246) |
| short professional tracks (bep/cap) | 0.1790** (0.0751) | 0.1635 (0.1457) | 0.1681* (0.0917) | 0.1773** (0.0733) | 0.1828 (0.1288) |
| Baccalauréat | 0.1813** (0.0777) | 0.1536 (0.1401) | 0.1673* (0.0934) | 0.1800** (0.0762) | 0.1629 (0.1269) |
| Advanced vocational certificate (bts/dut) | 0.1482*** (0.0546) | – 0.0056 (0.1897) | 0.1596** (0.0753) | 0.1448*** (0.0527) | – 0.0037 (0.1515) |
| Some college (Baccalauréat + 2 years’ further study) | 0.1778** (0.0744) | 0.2159 (0.2093) | 0.1684* (0.0904) | 0.1760** (0.0728) | 0.1626 (0.1889) |
| Higher graduate (Baccalauréat + 3 or more years’ further study) | 0.1552*** (0.0592) | 0.2078 (0.2123) | 0.1625** (0.0776) | 0.1522*** (0.0574) | 0.1132 (0.1659) |
| 3-C. Heterogeneity of marginal effect of employment level depending on household type at end of studies, for reference subjects distinguished only by this characteristic | |||||
| Living in a couple at end of studies | 0.1546*** (0.0571) | – 0.0060 (0.2325) | 0.0135 (0.2322) | 0.1514*** (0.0551) | 0.1296** (0.0583) |
| Living with their parents at end of studies | 0.1824** (0.0789) | 0.2310** (0.0960) | 0.2328** (0.0955) | 0.1812** (0.0775) | 0.1600** (0.0775) |
| Living alone at end of studies | 0.1792** (0.0754) | – 0.1100 (0.1890) | – 0.0954 (0.1855) | 0.1776** (0.0739) | 0.1557** (0.0743) |
| 3-D. Heterogeneity of marginal effect of employment level depending on immigrant background, for reference subjects distinguished only by this characteristic | |||||
| French parents | 0.1785** (0.0748) | 0.1860* (0.0953) | 0.1682* (0.0911) | 0.1943** (0.0783) | 0.1550** (0.0737) |
| At least one immigrant parent of African origin | 0.1863** (0.0877) | 0.2004 (0.1655) | 0.1576 (0.1005) | 0.2641* (0.1581) | 0.1678* (0.0857) |
| At least one immigrant parent of other origin | 0.1771** (0.0735) | – 0.1313 (0.2013) | 0.1685* (0.0901) | – 0.0835 (0.1886) | 0.1534** (0.0727) |
| 3-E. Marginal effect of local employment level on entry into the labor market: non-linear effects | |||||
| For an employment level smaller to 75% | 0.0366 (0.1447) | 0.0878 (0.1387) | – 0.0091 (0.0872) | 0.0275 (0.1479) | – 0.0074 (0.1384) |
| For an employment level greater than or equal to 75% | 0.4091* (0.2102) | 0.1094 (0.1217) | 0.2724 (0.2455) | 0.4158** (0.2106) | 0.3527* (0.1924) |
| N | 10,948 | 10,948 | 10,948 | 10,948 | 10,948 |
Marginal effect of local employment level on entry into the labor market
Estimated maximum likelihood models based on Céreq’s Génération 1998 and Génération 2004 surveysStandard deviations are given in brackets. Significance: *indicates p < 0.1, **p < 0.05, and ***p < 0.01.
Regressions (1) to (5) include control variables combining employment level with qualifications, living at home, and immigrant background in various ways. The other control variables are identical to those used in the initial models (see Table 3).
3-A. Regression (1) is already in Table 2, Column (2).
3-B. The marginal effect is evaluated for each qualification level on the basis of reference individuals who differ only by the value of this variable (their other variables are kept at the overall average level).
3-C. The marginal effect is evaluated for each living arrangement on the basis of reference individuals who differ only by the value of this variable (their other variables are kept at the overall average level).
3-D. The marginal effect is evaluated for each type of immigrant background on the basis of reference individuals who differ only by the value of this variable (their other variables are kept at the overall average level).
3-E. Returning to the analysis presented in 3-A, we distinguish the effect of local employment levels depending on whether this level is below or above the average (75%).
For individuals living in IRISes whose employment level is higher than the median, an increase of 1 percentage point in the employment level gives the average individual a 0.41 percentage point increase in their probability of being employed, in the situation (1) where interactions between employment level and other variables are not introduced
49The general nature of the local employment effect for 15 to 24-year-olds does not mean that it is uniform. Beyond observing limited variation in the indicator’s coefficient, we must take into account its effects when combined with other variables in calculating a marginal effect based on different individual profiles. Both sensitivity to local employment levels and degree of exposure to it are likely to have different influences depending on the profiles of new entrants. In particular, the effect will be demonstrably different depending on an individual’s education level, whether they live at home, and whether one of their parents is an immigrant.
50There are several causes of variation in effect size. The first involves the sensitivity of the link between chances of employment and local youth employment levels: [20] the same level of local employment can affect two individuals differently depending on their personal characteristics. [21] The second has to do with the characteristics of the individuals these effects apply to: [22] the same sensitivity (i.e. the same coefficient) can affect two individuals differently depending on their characteristics. [23]
51The mere fact that we are no longer considering a single representative individual, but instead individuals distinguished by their education level, [24] is enough to produce different marginal effects (see Table 3, part 3-B). These remain close when we estimate for six representative individuals differing from each other only by education level (Col. 1). But sensitivity to youth employment levels may itself differ depending on level of study. This is illustrated by the differences in marginal effects of diploma levels when the two variables are combined (Col. 5). The effect of employment levels is then clearly significant (at the 5% threshold) only for individuals without diplomas. An increase of 1 percentage point in the employment level leads to an 0.25 percentage point increase in probability of employment. Conversely, the effect is zero for graduates of an advanced vocational certificate such as brevet de technicien supérieur (Advanced Technician’s Certificate, BTS) and diplôme universitaire de technologie (two-year technical college degree, DUT). Consequently, the absence of a diploma exposes individuals more to neighborhood characteristics; getting one makes it possible to overcome local conditions. Advanced vocational certificates (BTS/DUT) appear to play a distinctively protective role. Note that the size and significance of the local effect varies depending on the combined variables introduced. This is particularly the case for the reference individual without a diploma. The effect is significant after the introduction of the variable combining employment level with immigrant origin (Col. 4), but no longer significant after the introduction of the variable combining employment level with both immigrant origin and living with one’s parents (Col. 3 and 2).
52The effect of neighborhood youth employment by type of living arrangement at the end of education on access to employment depends on whether the individual still lives with their parents, alone, or in a couple (see Table 3, part 3-C). Young people still living with their parents seem to be particularly sensitive to this social interaction effect: the parameter is higher and systematically significant at the 5% threshold. If we relax the hypothesis that the effect is uniform depending on living arrangement (Col. 3), a clear difference appears for individuals living alone or in a couple. The marginal effect increases from 0.18 to 0.23 for individuals living with their parents, and becomes zero for the others. This result holds when the employment rate is combined with the variables for education level and immigrant origin (Col. 2).
53The immediate vicinity of a family home appears to be crucial. These results suggest that sensitivity to neighborhood characteristics depends on duration of exposure to them. The family home has generally remained the same since these individuals were children. The measured effect may be less a matter of the neighborhood where these individuals finished school than of the environment they have been surrounded by since childhood. The influence of local peers is stronger when the subjects do not have a diploma. Access to employment for new entrants in the labor market is tied more closely to these peers’ employment situation. Finishing school reduces their importance.
54As before, combining parents’ immigrant status with employment level (see Table 3, Part 3-D, Col. 4) leads to a differentiation of the effects of local employment. The highest value of the parameter (0.26) is obtained for the children of African immigrants, suggesting they are more sensitive to local employment levels. But the significance of the effect remains limited, especially when employment level is crossed with the variables for education level and housing (Col. 2).
Variations in sensitivity to the effect of employed neighbors by level of employment in the neighborhood
55Galster (2008), Galster, Quercia, and Cortes (2000) and Galster, Andersson, and Musterd (2015) emphasize the nonlinear dimension of the effect of local peers on individual behavior. The influence exerted by the behavior of those in the same neighborhood is likely to become significant only if the frequency of this behavior reaches a critical threshold. [25] Such behavior must be sufficiently common if the individual is to adopt it effectively. This is one of the conditions allowing it to serve as a norm, and which means that effective social pressure to follow it is exerted on the new entrants. It is possible that neighbors’ influence is effective only if the local employment level is sufficiently high.
56To test this hypothesis, we relax the linear functional form of the social effect, replacing it with a piecewise linear one. The inflection point is set at the median of the proportion of workers aged 15-24 in employment in the areas covered by our sample, i.e. 75%. This proportion may have a different marginal effect on the probability of employment depending on whether the individual is in a neighborhood with high or low employment. In view of the results allowing us to assess variation in the marginal effects of employment levels on the probability of finding a job for young entrants (see Table 3, part 3-E), the social equilibrium effect is positive and significant only for young entrants living in an area where the proportion of young people in employment exceeds 75%. The effect for other entrants is not significantly different from zero. In areas with high youth unemployment, marginal variation in the proportion of employed workers does not affect the likelihood of being employed. These results emphasize the importance of the initial employment level. They suggest that for there to be a social interaction effect on joblessness, employment levels must be sufficiently high at the outset.
Conclusion
57This article helps to highlight the effect of neighborhood employment on entry into working life. New geo-localized data on those leaving the French school system reveals local social equilibrium effects in urban areas. Local exposure to a high proportion of employed young people increases access to employment for young men. A 1 percentage point increase in local employment rates raises the average chance of obtaining a job by around 0.2 percentage point. [26] The size of this effect is not uniform, however, and depends on individual characteristics and local employment levels. Lack of a diploma, living with one’s parents, or being a child of African immigrants makes them more open to the influence of neighborhood employment levels. Conversely, having a diploma makes it possible to escape the influence of local peers. Note, for comparison, that entering the labor market with a diploma like the cap-bep (short vocational tracks) rather than no diploma at all increases the likelihood of being employed three years later by 18 percentage points (see Table 4). For young people, obtaining a diploma is likely to have a much greater effect than a change in local employment levels.
Marginal effect of level of qualification on access to employment
Marginal effect of level of qualification on access to employment
Relative to those without qualifications, those with a Baccalauréat or higher have a 15.1 percentage points higher probability of being employed 3 years after the end of schooling.Estimated maximum likelihood models based on Céreq’s Génération 1998 and Génération 2004 surveys
Standard deviations are given in brackets. Significance: *indicates p < 0.1, **p < 0.05, and ***p < 0.01.
Regressions (1) to (5) include control variables combining employment level with qualifications, living at home, and immigrant background in various ways. The other control variables are identical to those used in the initial models (see Table 2).
58On the other hand, the effect of local employment levels on the probability of having a job appears significant only in areas where the youth employment level is sufficiently high. In neighborhoods marked by high unemployment, marginal variations in the local employment levels appear to have no significant effect on access to employment.
59Finally, this study offers the possibility of going beyond the measurement of an overall local effect, and focusing instead on one of its rarely distinguishable component parts. Neutralizing neighborhood effects allows us to study the effect of local employment levels without this being obviously affected by the effects of physical distance to jobs or statistical discrimination associated with place of residence. The analysis of the situation of young school leavers with regard to employment described in the census also allows us to avoid the endogeneity problems generally associated with this type of analysis, whether these are related to the potential effects of employment on location or the individual’s contribution to the local employment situation (reflection effects). In pursuing the hypothesis of conditional exogeneity within each neighborhood, we can control carefully for the characteristics, personal and familial, of each individual. The mechanisms contributing to this process, however, remain to be explored. The effect may also reflect the influence of information on jobs being disseminated, co-optation of access to them, and norm-following. Moreover, the relationship between these social interaction effects and other mechanisms involved in spatial variation in employment access remains to be explored. Do they add to effects of physical distance to jobs, or are there compensatory phenomena?
60It is premature at this stage of the analysis to draw strong implications for public policy. The fact that each job can generate positive local externalities on access to employment might increase the effectiveness of public policy. An assisted contract would benefit not only the young person concerned but also those around him/her. But this multiplier effect would only occur in areas where the employment level is high enough. While suggesting that targeting public policies spatially might be a valuable way of increasing their effectiveness, the result indicates that the most disadvantaged areas specifically targeted by these schemes would not benefit from this multiplier effect. On the contrary, the effect would maintain or even increase inequalities between neighborhoods. This does not mean that measures to tackle spatial inequalities are doomed, but that they must be large enough to shift the balance in the most disadvantaged areas. It is only after this first step has been taken that the local effects highlighted in this analysis might help these areas to improve.
I – Descriptive statistics and complete results of two main regressions
Sample characteristics and complete results of regressions (1) and (2) of table 2
| (A) | (B) | (C) | (D) | |
|---|---|---|---|---|
| Number of subjects of Génération surveys in work three years after leaving school | 82.4 | 74.8 | ||
| Individual characteristics at end of schooling | ||||
| Level of studies | ||||
| No diploma, no qualifications | 21.7 | 23.0 | Reference | Reference |
| Short professional tracks (bep/cap) | 19.7 | 19.6 | 0.6804*** (0.0721) | 0.6786*** (0.0720) |
| Baccalauréat | 21.6 | 21.8 | 0.6097*** (0.0901) | 0.6089*** (0.0900) |
| Advanced vocational certificate (bts/dut) | 13.6 | 12.9 | 1.2503*** (0.1266) | 1.2523*** (0.1264) |
| Some college (Baccalauréat + 2 years’ further study) | 9.9 | 9.7 | 0.7096*** (0.1553) | 0.7112*** (0.1551) |
| Higher graduate (Baccalauréat + 3 or more years’ further study) | 13.5 | 12.9 | 1.1483*** (0.1712) | 1.1443*** (0.1709) |
| Retook a grade before sixth grade | 23.8 | 24.1 | 0.1746*** (0.0669) | 0.1756*** (0.0668) |
| Distance from average time taken to obtain diploma (years) | – 0.19 (1.74) | – 0.20 (1.76) | – 0.0494* (0.0256) | – 0.0496* (0.0256) |
| Residential immobility (municipality level) | ||||
| Residential immobility from sixth grade to end of studies | 88.4 (0.32) | 88.6 (0.32) | – 0.3774 (0.2910) | – 0.3466 (0.2901) |
| Duration of residential immobility (years) | 8.64 (4.18) | 8.61 (4.16) | 0.0286 (0.0243) | 0.0269 (0.0243) |
| Living with their parents | 75.2 | 76.3 | – 0.5896*** (0.1141) | – 0.5848*** (0.1140) |
| Living in a couple | 11.1 | 10.6 | Reference | Reference |
| Living alone | 13.7 | 13.1 | – 0.4808*** (0.1255) | – 0.4799*** (0.1255) |
| With children | 7.0 | 7.1 | 0.2869** (0.1197) | 0.2875** (0.1197) |
| Parents’ characteristics | ||||
| Socioeconomic status | ||||
| Blue collar, white collar workers | 53.9 | 54.8 | Reference | Reference |
| Intermediary professions | 9.9 | 9.7 | 0.1006 (0.0891) | 0.1009 (0.0890) |
| Executives | 23.7 | 22.9 | 0.0208 (0.0703) | 0.0083 (0.0701) |
| Shopkeeper, artisans, business owners | 12.5 | 12.6 | 0.1415* (0.0793) | 0.1345* (0.0792) |
| Parents’ jobs | ||||
| Both parents working | 54.1 | 52.7 | Reference | Reference |
| One parent working | 28.2 | 28.9 | – 0.2802*** (0.0574) | – 0.2823*** (0.0573) |
| One parent having worked in the past | 15.4 | 16.0 | – 0.1854*** (0.0711) | – 0.1858*** (0.0710) |
| Neither parent having worked in the past | 2.3 | 2.4 | – 0.4602*** (0.1513) | – 0.4630*** (0.1510) |
| Parents’ background | ||||
| Born in France (both) | 81.2 | 79.9 | Reference | Reference |
| French, born abroad (one parent) | 3.5 | 3.6 | – 0.2765** (0.1308) | – 0.2726** (0.1309) |
| African immigrant (one parent) | 10.1 | 11.5 | – 0.4391*** (0.0800) | – 0.4318*** (0.0795) |
| Other immigrant (one parent) | 8.7 | 8.6 | 0.0376 (0.0926) | 0.0373 (0.0925) |
| End of studies in 2004 (%) | 35.1 | 35.1 | – 0.4849*** (0.0539) | – 0.4745*** (0.0534) |
| District (grand-quartier) Fixed Effects | Yes | Yes | ||
| Area characteristics | ||||
| Type of housing (PCA axis 1) | – 0.0243 (0.0307) | |||
| Social composition (PCA axis 2) | 0.0696** (0.0327) | |||
| Residual axis (PCA axis 3) | – 0.0350 (0.0510) | |||
| Infrastructure availability (MCA axis 1) | 0.0199 (0.0416) | |||
| Block-group employment levels | ||||
| Proportion in employment among… | ||||
| …active 15 to 24-year olds | 74.4 (11.4) | 74.1 (10.8) | 1.0925*** (0.4067) | 0.7453** (0.3493) |
| …all 15 to 24-year-olds | 24.7 (9.2) | 24.7 (8.9) | ||
| …active 15 to 64-year-olds | 85.9 (7.3) | 85.6 (7.0) | ||
| N | 16,937 | 11,014 | 10,944 | 10,948 |
Sample characteristics and complete results of regressions (1) and (2) of table 2
Column (A): characteristics of the initial sample of men living in IRIS-based urban municipalities, based on the Génération surveys.Column (B): characteristics of the subsample of men residing in IRIS-based urban municipalities used for estimation.
Column (C): regression (1) of table 2.
Column (D): regression (2) of table 2.
Columns (A) and (B): percentages, except where notes. Standard deviations are given in parentheses. Columns (C) and (D): estimated maximum likelihood models based on Céreq’s Génération 1998 and Génération 2004 surveys. The construction and characteristics of these are given in Appendix 2. Standard deviations are given in brackets. Significance: *indicates p < 0.1, **p < 0.05, and ***p < 0.01.
II – Area Characteristics
61The census data provides information on the social composition of block-groups (IRISes), including proportions of workers, executives, immigrants, and single-parent families. The first base permanente des équipements (permanent facilities database), or BPE, [27] is used for details on the infrastructures and services available in each block-group (IRIS). This gives information about the range and quantity of services in the immediate vicinity: police stations, doctors, pharmacies, childcare, hairdressers, artisans, post offices, banks, shops—bakers, butchers, stationers—shopping centers, sports grounds, cinemas, etc. Contextual variables from the 2006 census (i.e. 2004-2008 annual census surveys) are also used for: housing type (share of social housing, share of single-family dwellings); whether the home is owned or rented; the stability of the population (the proportion of those resident in the block-group (IRIS) for at least five years, compared to those who arrived in the last two years); available transport (the proportion of commuters who own cars and who take public transport); and the IRIS’s social composition (the ratio of managers to workers and the proportion of individuals without qualifications, of single-parent families, and of immigrants). This information is available at the block-group (IRIS) level, and can then be calculated at the district level.
62Aggregating characteristics:
63Contextual variables other than employment are combined and aggregated using a Principal Component Analysis (PCA) and a Multiple Correspondence Analysis (MCA).
64The use of two different methods is due to the nature of the variables, which come from two different sources. The eleven census rates characterize neighborhood and housing type, while the seventeen dichotomous variables derived from the BPE describe the proximity of different services and facilities.
65These methods allow us to summarize information from many highly correlated variables and use it to control for neighborhood characteristics.
66The first three axes of the principal component analysis account for three quarters of the total variance. The first axis, centered on type of housing, contrasts areas with concentrated social housing to areas where residents mostly own their homes. The social composition of the neighborhood is projected on the second axis: areas with high proportions of executives and newcomers who have been present for less than two years, contrasted with areas marked by a high degree of immobility during the last five years and a high proportion of people without qualifications. The third axis distinguishes people who did not move over the past few years, who are more often executives, and who use public transport.
67Residential areas are given in decreasing order of local availability of infrastructures and services. This first dimension of the analysis of multiple correspondences captures 90% of the inertia.
68These four factors can be used to account for any differences between districts. They are included in one of our main regressions (see Table 2, Col. 1, and in Appendix I, Table A1, Col. C). Only the effect of the social composition of the principal component analysis would seem to have a limited influence.
Notes
-
[1]
This work was carried out with the support of the Centre d’Études et de Recherches sur les Qualifications (Céreq) and the Secrétariat Général of the Comité Interministériel des Villes (SG-CIV), who coordinated a working group to exploit geolocalized information newly introduced in Céreq’s Generation surveys. We would also like to thank Laurent Gobillon, Pierre-Philippe Combes, Yann Bramoullé, Roland Rathelot, Camille Hémet, Mathieu Bunel, and Yves Zenou for their comments on a preliminary, more general version of this work, as well as the editor and the two anonymous readers for their suggestions.
-
[2]
This device was abandoned at the end of 2014 while still in trial stages, with only 250 signed contracts.
-
[3]
These unemployment rates were calculated using the 1999 census.
-
[4]
The results of the audit conducted at the end of 2008 and start of 2009 by L’Horty et al. (2011) show, for example, that skilled young people from a neighborhood associated in the media with the 2007 riots, Villiers-le-Bel, were discriminated against more heavily by employers than those from a similarly disadvantaged neighborhood that had received less media attention, Sarcelles.
-
[5]
This article was also part of a dossier on young people entering working life.
-
[6]
For a review of the recent literature see in particular Topa and Zenou (2015).
-
[7]
Beyond controlling invariant individual characteristics, panel data allow us to consider place of residence prior to employment or even entry into working life.
-
[8]
Hellerstein, McInerney, and Neumark (2011) and Hellerstein, Kutzbach, and Neumark (2014) used the 2000 Decennial Employer-Employee Database (DEED) and LEHD data produced by the Census Bureau.
-
[9]
For example, proximity to missions locales and other job search services, or a neighborhood’s poor reputation among employers.
-
[10]
Parents may remain involved, paying all or part of the child’s rent, serving as guarantors, and so on.
-
[11]
The functional form of this distribution makes it possible to avoid the problem of incidental parameters posed by the inclusion of fixed effects in a maximum likelihood nonlinear model.
-
[12]
An indicator distinguishing between individuals according to the survey group they belong to was systematically introduced as a control variable in the regressions presented in the remainder of this work.
-
[13]
We follow INSEE’s definition. See http://www.insee.fr/en/methodes/default.asp?page=definitions/iris.htm.
-
[14]
In addition to rural municipalities, the smallest urban ones are not necessarily subdivided into IRISes.
-
[15]
In particular, 17% of women had a child within three years of graduation, compared with 9% of men.
-
[16]
Translator’s note: The “baccalauréat” is the academic qualification that second-level students in France receive after two years of study (the final and penultimate year of secondary education). This qualification is required in order to pursue university studies.
-
[17]
These regressions were performed using census data aggregated by block-groups (IRISes). This contains data for different variables, including qualifications, employment status, and geographical origin, allowing us to analyze the variance explained by the different geographical divisions while controlling for the distribution of the population across age categories.
-
[18]
The estimation of the coefficients of the set of regression control variables is presented in Appendix 1. The complete results make it possible, in particular, to verify that the probability of being employed is higher rises with higher qualifications and when the individual’s parents are themselves employed. Conversely, having an African immigrant parent decreases the probability of being employed.
-
[19]
Each of its characteristics is fixed at the mean level observed for all of the individuals studied.
-
[20]
In other words, this is the value of the regression coefficients for the control variables associated with the employment level.
-
[21]
A good qualification may protect better against the social effect associated with the level of employment at one’s place of residence (i.e. a decrease in the value of the coefficients).
-
[22]
In other words, the value of the control variables associated with employment levels.
-
[23]
A high level of qualification may allow exposure to higher levels of local employment (see figure 2).
-
[24]
The other variables are kept at the overall average level.
-
[25]
In the French case, Issehnane and Sari (2013) show that marginal improvement in neighborhood quality only has an effect in disadvantaged neighborhoods.
-
[26]
Taking the variance of the estimator into account, this effect lies between 0.06 and 0.34 with a probability of 0.95.
-
[27]
For a detailed description of the 2007 database, see http://www.insee.fr/en/methodes/default.asp?page=sources/ope-adm-bpe.htm. Although the 1998 municipality inventory corresponds better in terms of information collection date, it does not provide information at the intra-municipality level. It assumes infrastructures have changed little in the meantime, or that this evolution is continuous with existing achievements.
