1The spatial distribution of populations is not random. For example, young people of immigrant background more often live in disadvantaged neighbourhoods than other young people. Place of residence, along with the parents’ country of origin, may influence educational outcomes and labour market integration, and it is often difficult to dissociate the respective effects of these two social factors. Are the difficulties encountered by young people of immigrant background linked mainly to origin or to residential location? This is the question addressed by Romain Aeberhardt, Roland Rathelot and Mirna Safi, using highly localized data on place of residence along with information from two CEREQ surveys on young people’s pathways to integration. They analyse the risks of dropping out of school and the probability of being in employment three years after leaving the education system by parents’ country of origin and controlling for the effect of residential location.
2The labour market situation of young people with immigrant parents in France has become a major focus of research since the 2000s thanks to the availability of new data (Aeberhardt and Rathelot, 2013; Safi, 2013). Numerous studies highlight the severe disadvantages of children of immigrants from the Maghreb and sub-Saharan Africa compared with French children born to French parents (Meurs et al., 2006). Moreover, while differences in sociodemographic characteristics, level of education or age provide a relatively good explanation for wage differentials, only one-third of the differences in employment rates can be attributed to these factors (Aeberhardt et al., 2010). These observations are confirmed by “discrimination testing” studies which link the ethnic penalties measured in statistical surveys to discrimination on the French labour market (Cédiey et al., 2008; Duguet et al., 2010a, 2010b).
Terminology
For more detailed discussion of issues relating to ethnic/racial factors, see, for example, Omi and Winant (1986), Simon (1999, 2008), Wimmer (2008), or the overview in Safi (2013).
3Despite this recent empirical evidence, the French social sciences traditionally analyse these inequalities solely in terms of differences in social environment or educational level. People of immigrant background are thought to be disadvantaged not by their origin (and the hostility that this origin may elicit), but rather by the paucity of their family, economic and social resources. The effect of residential location is central to this approach, and the concentration of these populations in the most disadvantaged neighbourhoods is seen as the main explanation for their poor outcomes. While rarely described in detail, the spatial mechanisms evoked in these studies generally refer to unequal geographical distribution of infrastructure, public transport and services, and to territorial disparities in educational resources and teaching quality. In some studies, these spatial inequalities are even seen as the driving force behind the reproduction of socioeconomic inequalities, notably via peer effects or socialization mechanisms sometimes qualified as a “culture of poverty”.
4However, the study of urban and territorial inequalities has rarely been approached via a combined measure of both ethnic and geographic factors. Recent empirical research suggests that differences in place of residence provide an incomplete explanation for the differentials in employment and unemployment observed between people of different ethnic backgrounds (Gobillon et al., 2014; Rathelot, 2014). Ethnic and geographic dimensions have also been examined jointly in “situation testing” studies (Petit et al., 2014) with the aim of identifying their respective impacts on employer response rates to job applications. While they reveal a form of territorial discrimination, these studies also show that ethnic origin plays a central role.
5In this article we use data from the Génération surveys (1998 and 2004) collected by the French Centre for Research on Education, Training and Employment (Centre d’études et de recherches sur les qualifications, CEREQ) to measure the effects of ethnic origin and place of residence on educational and occupational outcomes of young people of immigrant background. These surveys provide one of the rare sources of data for analysing the labour market integration of school leavers. They record respondents’ place of residence by infra-communal census district (IRIS), [1] making it possible to include neighbourhood effects in our models. They also enable us to compare the 1998 and 2004 cohorts to analyse changes in the young people’s situations over time. The analyses presented here aim to determine how the effect of ethnic origin on educational and labour market outcomes changes after controlling for the geographical dimension in the form of fixed effects introduced at IRIS level.
I – Explaining ethnic inequalities through geographic effects: underlying mechanisms
6The effects of place of residence have been measured for a wide range of socioeconomic indicators (Sampson et al., 2002). Cutler and Glaeser (1997), for example, estimated that a 13% reduction in spatial segregation in the United States would eliminate one third of the Black-White differences in education, employment, wages and even single parenthood.
7With respect to education, the effects of neighbourhood socialization on children’s educational achievement are the most widely studied mechanisms (Ainsworth, 2002). The concentration of educational underachievement in certain neighbourhoods is liable to trigger peer effects which modify learning preferences and discourage students from adhering to norms of academic success (Goux and Maurin, 2007). Certain studies even suggest that local socialization mechanisms directly affect children’s development and hence their performance at school (Brooks-Gunn et al., 1993). Others (Moguérou et al., 2013) draw attention to the decisive effect of family setups that are more predominant among the working-classes (lone-parent families, more siblings, etc.) and hence more frequent in these neighbourhoods.
8The impact of geographical location on educational outcomes can also be directly linked to the territorial inequality of educational provision, and even its discriminatory nature, notably towards ethnic minorities (Merle, 2012; Felouzis et al., 2005). At a high level of geographical resolution, differences between neighbourhoods may reflect differences between individual schools (teaching quality, teachers’ experience, financial and educational resources, class size, etc.). Geographical effects may also be linked to urban inequalities that are not produced directly by the school system but whose influence is considerable (public amenities such as libraries, theatres, museums, public transport, etc.). The specifically geographic mechanisms of educational inequality are still a subject of debate, however (Arum, 2000). Many empirical studies seem to indicate that they are much less decisive than the effects of individual or family characteristics (Crain and Mahard, 1983; Vallet, 2005).
9Geographical effects have also been put forward to explain ethnic inequalities in the labour market (Fernandez and Su, 2004). A first family of studies focuses on the effect of the socialization process, postulating that in neighbourhoods where a large fraction of inhabitants are unemployed, inactive and poor, people are not motivated to enter the labour market (Crane, 1991; Cutler and Glaeser, 1997; Wilson, 1996). Some research suggests that these neighbourhoods become a source of stigma in themselves, making it more difficult for inhabitants to find employment. This is known as geographic discrimination or postcode discrimination. [2]
10A second family of studies points up the existence of a spatial mismatch between the areas where minorities live and the employment areas that are accessible to them (Gobillon et al., 2007; Holzer, 1991; Kain, 2004). They look closely at the transport systems serving the residential areas inhabited by minority groups, where many are highly dependent on public transport because they cannot afford to run a car (Holzer et al., 1994). Some recent French studies have addressed these questions (Duguet et al., 2009; Gobillon, 2001; Gobillon et al., 2011), though populations of immigrant background are rarely the focus of attention.
11These various studies raise questions about the extent to which educational and labour market inequalities affecting certain ethnic minorities are linked to their concentration in the most disadvantaged geographical areas. The analyses presented hereafter seek to answer this question.
II – Data and samples
The 1998 and 2004 Génération surveys geolocated to place of residence
12The Génération survey programme was designed to shed light on the labour market integration of young people leaving the education system and the first years of their working careers.
13The survey covered all young people who had left the education system for the first time at least one year previously, who were below 35 years of age at the end of their initial education and who lived in metropolitan France at the time of the survey. The first wave was administered by telephone in the spring of the third year after respondents had completed their education, i.e. in the spring of 2001 and of 2007, respectively, for the Génération 1998 and Génération 2004 surveys that are used here. Further waves took place five, seven, and even ten years later for Génération 1998. To avoid problems linked to sample attrition, we use only the information collected after three years for our analysis.
14Around 55,000 young people took part in the first interview (after three years) of the Génération 1998 survey, and slightly below 60,000 in that of Génération 2004.
15The sampling frame was constructed ad hoc from lists obtained from schools and local education authorities. A balanced sample was then drawn on the basis of strata corresponding to the French regions and the main educational levels.
16After several questions on their identity, respondents were asked about their educational trajectory and their experience of employment, either in parallel or as part of their education. An employment calendar was then filled in to identify the respondent’s periods of employment and non-employment. This was the central core of the questionnaire. Information was also obtained about career prospects, sociodemographic characteristics (including the respondent’s and his/her parents’ nationality of origin) and living arrangements.
17The following analyses cover two distinct groups: young French people whose father and mother were both born in France (referred to as the “mainstream population”) and young French people (not necessarily French-born) with at least one parent born in North Africa, sub-Saharan Africa, Lebanon, Turkey or in the Middle East (this latter category is not listed in the dictionary of survey codes). [3] These populations grouped into a single category in our analyses will be referred to as the “minority population”. [4] The initial sample comprised 76,966 individuals: 47,734 from Génération 1998 and 29,232 from Génération 2004. The share of young people in the minority group as defined above was 10.8% in 1998 and 11.9% in 2004.
Controlling for fine-scale geographical effects
18Under a cooperation agreement with INSEE, fine-scale geolocation data were obtained (at IRIS level) for the vast majority of respondents in our sample (98% in 1998 and 92% in 2004). This made it possible to control for neighbourhood effects at a scale comparable to that used in the United States (census tract) to study segregation. Studies at this level of geographical detail are very rare in France. Moreover, the sample is large enough to control for these neighbourhood effects in the form of “fixed effects”. Individuals within a single IRIS zone can be compared, thereby eliminating the need to take account of systematic differences observed across geographical units. The analyses presented below control not only for individual characteristics (sex, level of education, parents’ origin, ethnicity, etc.), but also for unobserved characteristics of the IRIS of residence. By definition, however, these analyses can only be estimated in IRISes with at least two observations. This limits our estimation samples to 34,404 individuals in 1998 and 15,351 in 2004 (Table 1). A comparison of the full sample (Appendix Table A.1) and reduced sample (Table 3) reveals only minor differences in the inhabitants’ characteristics, suggesting that the sample reduction does not adversely affect survey representativeness. Last, ethnic composition varies slightly between the two surveys, with young people from North Africa representing 8% of the sample in 1998 (15% in 2004), and those from the rest of Africa representing 10% in 1998 (11% in 2004) (figures not shown in the Table).
Samples
Samples
Coverage: Young people with French nationality aged 15-30 who completed their education in 1998 or 2004, with parents born in France (mainstream population) or with at least one parent born in North Africa or sub-Saharan Africa (minority population).III – The situation of young people with an immigrant background: lower education and less favourable labour market outcomes
19The children of immigrants from North and sub-Saharan Africa more frequently drop out of education before completing high school (Tables 2 and 3). A detailed examination of educational disparities at all levels of qualification shows that this is the most significant disparity with respect to children of the mainstream population (8 percentage points for women and 13 points for men). While differences are also perceptible in the type of baccalauréat (high-school diploma) obtained, with more vocational baccalauréats in tertiary specialities than in industrial specialities, the disparities widen again at higher educational levels, primarily for men.
Characteristics of groups by sex and national origin, 1998 cohort (%)(a),(b)
Characteristics of groups by sex and national origin, 1998 cohort (%)(a),(b)
Note: The “Difference test” column gives the results of a t-test of the difference between the two ethnic groups.(a) Non-response. Father or mother unknown.
(b) The occupational category is not observed as the father or mother is inactive.
Significance levels: *** difference significant at 1%; ** at 5%; * at 10%.
Coverage: French people aged 15-30 who left education in 1998, IRISes with at least two observations.
Characteristics of groups by sex and national origin, 2004 cohort (%)(a),(b)
Characteristics of groups by sex and national origin, 2004 cohort (%)(a),(b)
Note: The “Difference test” column gives the results of a t-test of the difference between the two ethnic groups.(a) Non-response. Father or mother unknown.
(b) The occupational category is not observed as the father or mother is inactive.
Significance levels: *** difference significant at 1%; ** at 5%; * at 10%.
Coverage: French people aged 15-30 who left education in 2004, IRISes with at least two observations.
20Three years after completing their education, the children of immigrants from North Africa and sub-Saharan Africa are more often unemployed (around 20% versus 10% for the mainstream population) or inactive (the gap is especially wide for women in the minority population, of whom 6.3% are inactive, versus 3.1% of women in the mainstream population). Moreover, their family characteristics indicate a low level of economic and cultural capital: their mothers are more often inactive, their fathers more often unemployed or in a manual job. When they are in employment, on the other hand, their type of work contract and level of pay do not appear to be very different, on average, from those of children with French-born parents. These figures seem to indicate that the most significant differences between the two groups concern the timing of their exit from the education system and their entry into first employment – a critical stage in France.
21Last, the children of immigrants from North Africa and sub-Saharan Africa are also concentrated in the most disadvantaged areas. Among those who left the education system in 2004, 27% were living in a sensitive neighbourhood (ZUS/CUCS) [5] versus 6% of children from the mainstream group.
IV – Method
22The regression analyses presented below measure the extent to which the observed differences in educational and occupational outcomes of children of African immigrants with respect to the children of French-born parents are explained by individual, familial and geographical characteristics. Identical regressions are estimated separately for the 1998 and 2004 samples. Successive specifications aim to detect changes in the coefficient of interest associated with the ethnic origin variable (minority population “pop_min”), while also controlling for social background (parents’ occupations and employment status), educational level (only for the employment equation) and the geographic fixed effect. For each control variable, the modalities are those listed in Tables 2 and 3. The parents’ occupational category, their employment status and educational level are crossed with the sex variable in all specifications. The formal equation estimated in the absence of IRIS fixed effects is written as follows:
24where Y is the binary variable of interest (dropping out of high school with no qualifications or in employment three years after leaving the education system), Y* is the corresponding latent variable, pop_min is the dummy of belonging to the minority population, F is the dummy of being female, [6] X are the control variables, and u the error term assumed to follow a logistic distribution. Equation (1) is estimated by a logistic regression.
25When the indicator variables h corresponding to the IRIS of residence at the time of leaving education are introduced into the estimated equation, it becomes equation (2).
27Introducing fixed effects for the IRIS of residence multiplies the number of parameters in the model, with the risk of producing biased covariate estimators. These are so-called “incident parameter” biases that arise when the sample is too small with respect to the number of parameters. This problem is generally resolved by using conditional logit estimation, and we have adopted this solution here: the IRISes in which all individuals take the value 0 (or 1) for the dependent variable are not included in the model estimation, and the sample is reduced accordingly (Lancaster, 2000). To distinguish the effect of reducing the sample from that of introducing the IRIS indicator variables, the process is broken down into two stages. We first estimate the model without IRIS indicators on the reduced sample and compare the results with those obtained on the full sample. We then estimate the model with the IRIS indicators. This enables us to separate the results attributable to the change of sample size from those that can be interpreted as the effect of controlling for place of residence. The standard deviations on the marginal effects are obtained by full bootstrap (200 iterations). [7]
28It is not easy to interpret the estimated coefficients directly, or to compare them across sub-samples, since the models are non-linear. The tables therefore show the marginal effects, which provide us directly with a comparable and intuitive measure of the effects of the different variables of interest. The marginal effects correspond to the difference in the predicted probability that Y equals 1 when pop_min and F change from 0 to 1. In formal terms, the marginal effects given in the following tables correspond to the estimated versions of the three following quantities:
30The first parameter (3) measures the marginal effect on the variable of interest, for men, of having a parent of African origin (minority population) rather than a French-born parent. The second parameter (4) corresponds to the effect of being a woman rather than a man for children of French-born parents (mainstream population). The third parameter (5) measures the residual difference additional to these two effects via an interaction effect: it represents the difference between women in the minority population and men in the mainstream population, from which the two previous pop_min and F effects have already been subtracted. The sign of this interaction term indicates the relative strength of the origin effect (pop_min) for women comparative to men: a positive sign means that having an African immigrant background rather than two French-born parents has a stronger effect for women than for men. For women (compared to women in the mainstream population), the effect of belonging to the mainstream population is thus obtained by summing the first and third marginal effects.
V – The effect of origin is decisive, even after controlling for geographic effects
31Knowing where young people live when they leave education, along with their educational level and their social origin, enables us to determine whether observed differences in educational outcomes and in labour market integration for children of African immigrants reflect differences in social environment or are attributable to neighbourhood effects. In the following analyses, we first calculate the gross differences between mainstream and minority populations, then control for the young person’s social environment, educational level (in the case of labour market entry only) and place of residence at the time of leaving the education system.
32The results of the two analyses, one on dropping out before completing high school and the other on labour market integration three years after leaving education, are presented separately in Tables 4 and 5. The results for all the control variables are presented in Appendix Tables A.2 and A.3.
Marginal effects of national origin and sex on the fact of leaving education with no qualifications, controlling by social origin or residential location
Marginal effects of national origin and sex on the fact of leaving education with no qualifications, controlling by social origin or residential location
Note: The occupational category and parents’ labour market situation control variables are crossed with the sex dummy; the model estimated in columns (1) to (3) is a simple logit, for column (4) it is a conditional logit; the control variables are the same for columns (2) and (3), only the sample changes.Significance levels: *** difference significant at 1%; ** at 5%; * at 10%.
Coverage: French people aged 15-30 who left education in 1998 or 2004.
Marginal effects of national origin and sex on the fact of being in employment three years after leaving education, controlling by social origin and residential location
Marginal effects of national origin and sex on the fact of being in employment three years after leaving education, controlling by social origin and residential location
Note: The occupational category and parents’ labour market situation control variables are crossed with the sex dummy; the model estimated in columns (1) to (4) is a simple logit, for column (5) it is a conditional logit; the control variables are the same for columns (3) and (4), only the sample changes.Significance levels: *** difference significant at 1%; ** at 5%; * at 10%.
Coverage: French people aged 15-30 who left education in 1998 or 2004.
33The probability of dropping out of school is significantly higher for children of African immigrants (column 1, Table 4). This difference is more pronounced in the 1998 cohort than in that of 2004 (27 percentage points versus 20, among men). The differences attributable to ethnic origin are smaller among women (6 points lower in 2004). Overall, women drop out of school much less frequently than men (7-point difference in both sub-samples). When the two effects are summed, the probability of dropping out of school for women with immigrant parents moves very close to that of men with French-born parents.
34After controlling for the parents’ labour market situation (occupational category and labour market situation of both mother and father), the differences due to the ethnic variable decrease by 5-6 percentage points, but remain large and significant, while the differences between the sexes remain unaffected (column 2, Table 4). For an identical social background, the frequency of school dropout among men with African-born parents is 21 points higher than that of men with French-born parents in the 1998 cohort and 15 points higher in the 2004 cohort. For women, the difference is 12 points and 8 points, respectively.
35As explained in Section IV, estimating the model with dummies for the IRIS of residence implicitly restricts the sample upon which the coefficients are identified. To distinguish between the effect of restricting the sample and that of introducing dummies for residential location, the third column presents the estimate of the same specification as column (2) on the restricted sample. The fourth column of Table 4 gives the results of the estimate with dummies for the IRIS of residence. This highlights two important points. First, comparison between columns (2) and (3) shows that restricting the sample has only a small (and non-significant) impact on the coefficients of ethnic differences. By contrast, the gender differences are significantly stronger in the restricted sample. While it is beyond the scope of this article to interpret the effects of sample restriction, we can assume that it increases the weight of the most populated and most socially heterogeneous IRISes.
36The second point emerges from the comparison between columns (3) and (4): for both the 1998 and 2004 cohorts, no additional heterogeneity on top of that already taken into account with the previous variables is captured by including the neighbourhood. A young person with foreign-born parents from a given neighbourhood and social background will leave school with no qualifications more than twice as frequently as an equivalent young person with non-immigrant parents. [8]
37Taken overall, these results show that social background and residential location explain a relatively small share of the raw difference. They also suggest that the unexplained part tends to shrink between 1998 and 2004. Is this a lasting trend, a cyclical movement or a statistical artefact linked to the different sizes of our samples? While comparing the two sub-samples reveals no significant differences in their composition (Tables 2 and 3), we still cannot be certain that we are measuring a trend effect. [9]
38More generally, while our results are useful for measuring the scale of effects, they do not settle the debate on the influence of residential location on the educational outcomes of second-generation immigrants; they provide a rare opportunity to identify this influence at a detailed geographical level, but cannot identify a causal effect. Clearly, residential location is an endogenous variable: personal choices or residential constraints (structure of the property market, housing discrimination, etc.) may determine an individual’s place of residence. Numerous qualitative studies have shown that the quality of local schools is a key factor in residential choices (van Zanten, 2001, 2009). If the children of affluent, educated non-immigrant families do better at school, it is not only because their parents provide the material and intellectual resources necessary for their academic success, but also because they can exploit the system of school catchment areas to get their children into the best schools (Felouzis and Perroton, 2009; Roscigno, 1998). Specific identification strategies are needed to measure such causal effects (natural experiments or longitudinal data), and finding a credible strategy is difficult. By estimating the scale of geographic effects, this study offers a more descriptive contribution; it describes the extent to which the ethnic factor is attributable to the spatial factor. Any causal inference on the effect of geographic location would require more extensive data and more detailed empirical investigations.
39What about labour market integration? Table 5 shows the results of a similar analysis for differences in employment rates three years after leaving the education system (see Appendix Tables A.4 and A.5 for the complete results). For both sexes, there is a 13-point difference linked to ethnic origin for the 1998 cohort and a 19-point difference for the 2004 cohort. Unsurprisingly, women are less frequently in employment than men (5 points in 1998 and 3 points in 2004). Differences in social background explain only a very small share of the differences linked to ethnic origin or gender (column 2), which remain practically unchanged after controlling for the parents’ occupational category and labour market situation. The differences in educational level also explain a small share of the differences in employment rate between the two populations (column 3). Together, social background and education explain around one-third and one-half, respectively, of the difference in employment rate, but 7 points remain unexplained for the 1998 cohort and 13 points for the 2004 cohort. For an identical education and social background, we note that the gender gaps become very small, and even non-significant for the 2004 cohort.
40After controlling for the IRIS of residence, the effect of ethnic origin remains unchanged: the difference between the coefficients of columns (4) and (5) is not statistically significant. All in all, residential location appears to have a weak effect on employment rate after controlling for education and social background. Notable differences in labour market integration between the two groups – 7 points in 1998 and 14 point in 2004 – remain unexplained, even after controlling for all the numerous individual and family variables and for location. As is the case for dropping out of school with no qualifications, the imprecision of the estimate, linked to the small sample size and the detailed level of geographical resolution, prevents any further exploration of questions relating to the differences observed between 1998 and 2004.
41It should again be pointed out that for the same reasons as discussed above with respect to educational inequality (endogeneity of residential location), these findings cannot be seen from a purely causal perspective. Moreover, the educational level that we control for in our analyses of labour market integration also introduces endogeneity bias; numerous cultural and socioeconomic mechanisms are liable to result in shorter school careers and hence more limited employment prospects for children from ethnic minorities (Ichou and van Zanten, 2013). Here too, our data cannot be used for causal inference, but provide an initial empirical overview of these questions.
Conclusion
42Disentangling the effects of ethnic origin and place of residence on young people’s educational and labour market outcomes is an important theoretical concern in the literature on ethnic inequalities, but also a central preoccupation of urban policy makers. The Génération surveys are unique in providing data that can be used to separate these two dimensions. The present study makes use of the fine-scale geolocation data provided by these surveys to compare the respective amplitudes of these effects.
43The results obtained highlight the magnitude of differences between the children of African immigrants and children of French-born parents that persist even after controlling for numerous individual and familial factors. Controlling for geographic effects seems to matter, but does not challenge the conclusion that ethnic origin plays a decisive role in explaining these differences. This study corroborates the conclusions of Rathelot’s analysis (2014) based on the French Labour Force Survey (Enquête Emploi).
44While our findings provide an initial empirical overview of the scale of geographic effects, they do not reveal the exact mechanisms involved. This is a limitation shared by many studies that examine how contextual factors feed social inequalities. Indeed, while the literature leaves no doubt about the correlation between spatial segregation and the concentration of social inequalities, separating the causes and consequences of the spatial factor is fraught with methodological difficulties. More precisely, as segregation is itself the product of inegalitarian mechanisms, the distribution of individuals in space is the result of selection processes that are embedded in the inequalities linked to social background. In the United States, the Moving to Opportunity (MTO) programme was designed to measure the effects of the urban context more robustly, but the results obtained (Ludwig et al., 2008, 2012), [10] like those of our own study, call for further research on the causal effects of geographic location using more appropriate empirical tools (natural experiments linked to a new local policy, for example, panel data taking account of mobility, etc.).
45Many different interpretations are suggested in the literature to explain the persistence and stability of the ethnic origin effect. Regarding educational outcomes, while some authors focus on the role of cultural mechanisms in the reproduction of poverty (immigrant families’ relationship with the school system, educational disinvestment, oppositional culture, etc.), their findings are contested by most French empirical studies on this topic (Brinbaum and Kieffer, 2005; Brinbaum et al., 2012). There is stronger evidence for the role of teachers’ conscious and unconscious stereotypes and the effects of educational and organizational factors such as segmentation of school careers and academic tracks (in France, research has focused on optional subjects, languages, artistic and sports activities and track choices in high school) (Brinbaum and Primon, 2014; Merle, 2012; Palheta, 2012; Payet, 1995; Steichen, 2013; van Zanten, 2001).
46Regarding ethnic inequalities on the labour market, discrimination in job recruitment is a direct explanatory mechanism documented in numerous French studies. The situation of second-generation immigrants on the labour market may also be linked to the paucity of their social networks. The literature on this topic reveals the extent to which informal contacts (family, friends, acquaintances, etc.) play a decisive role in the job seeking process, for the very young in particular (Holzer, 1988). The social networks of young people of immigrant background are less fertile in terms of job opportunities because their families, friends or neighbours are often themselves in insecure employment, or employed in sectors or businesses where such opportunities are scarce (Holzer, 1987, 2001). The fact that differences in rates of access to employment are much greater than differences in wages or job quality (a fact confirmed by numerous empirical studies in France), suggests that social capital is a key factor underpinning ethnic inequalities on the job market. Specific studies are needed to test this hypothesis.
47In conclusion, there is now a broad consensus in the French social science literature that discriminatory mechanisms are key factors of ethnic inequality, be it in education, employment or socioeconomic status more generally. These mechanisms must be understood not only in interactional terms (for example, in job applications) but also in systemic terms (i.e. their production and reproduction by institutional mechanisms affecting all individuals). In addition to its direct effect as an obstacle to integration, discrimination breeds further inequality. It is sociologists and social psychologists who have explored this question of persistent inequality most fully in recent years (Lamont et al., 2014; Ridgeway, 2014).
48Our findings do not offer direct proof of discrimination mechanisms. Nonetheless, by showing that the geographic factor only marginally modifies the impact of ethnic origin on the chances of dropping out of school or finding employment, they challenge the theories which view the segregation of immigrant populations as the cause of their disadvantaged socioeconomic status. Drawing attention to the limits of strictly geographical measures, these results have important implications for urban policy makers. The existence of discriminatory mechanisms affecting ethnic minorities in school and on the labour market must be factored into public policies to combat unemployment and poor educational outcomes in disadvantaged neighbourhoods.
Acknowledgements
We would like to thank CEREQ and ONZUS for providing access to the Génération surveys geolocated to IRIS level. Our thanks also to Anthony Briant, Yaël Brinbaum, Oana Calavrezo, Thomas Couppié, Céline Gasquet, Thierry Kamionka, Matthieu Solignac, Maxime To and the other members of the analysis group for their comments. Last, we are grateful to the editors of Population and to the three anonymous reviewers whose suggestions helped us to improve our text.
Probability of leaving education without any qualifications, 1998 cohort, parameters of the logistic regression(a),(b)
Probability of leaving education without any qualifications, 1998 cohort, parameters of the logistic regression(a),(b)
Note: The occupational category and parents’ labour market situation control variables are crossed with the sex dummy; the model estimated in columns (1) to (3) is a simple logit, for column (4) it is a conditional logit; the control variables are the same for columns (2) and (3), only the sample changes.(a) Non-response. Father or mother unknown.
(b) The occupational category is not observed as the father or mother is inactive.
Coverage: French people aged 15-30 who left education in 1998 or 2004; columns (1) and (2): IRISes with at least two observations; columns (3) and (4): IRISes with at least two observations having different values for the dependent variable.
Probability of being in employment 3 years after leaving education, 1998 cohort, parameters of the logistic regression(a),(b)
Probability of being in employment 3 years after leaving education, 1998 cohort, parameters of the logistic regression(a),(b)
Note: The occupational category and parents’ labour market situation control variables are crossed with the sex dummy; the model estimated in columns (1) to (3) is a simple logit, for column (4) it is a conditional logit; the control variables are the same for columns (2) and (3), only the sample changes.(a) Non-response. Father or mother unknown.
(b) The occupational category is not observed as the father or mother is inactive.
Coverage: French people aged 15-30 who left education in 1998 or 2004; columns (1), (2) and (3): IRISes with at least two observations; columns (4) and (5): IRISes with at least two observations having different values for the dependent variable.
Probability of leaving education without any qualifications, 2004 cohort, parameters of the logistic regression(a),(b)
Probability of leaving education without any qualifications, 2004 cohort, parameters of the logistic regression(a),(b)
Note: The occupational category and parents’ labour market situation control variables are crossed with the sex dummy; the model estimated in columns (1) to (3) is a simple logit, for column (4) it is a conditional logit; the control variables are the same for columns (2) and (3), only the sample changes.(a) Non-response. Father or mother unknown.
(b) The occupational category is not observed as the father or mother is inactive.
Coverage: French people aged 15-30 who left education in 1998 or 2004; columns (1) and (2): IRISes with at least two observations; columns (3) and (4): IRISes with at least two observations having different values for the dependent variable.
Probability of being in employment 3 years after leaving education, 2004 cohort, parameters of the logistic regression(a),(b)
Probability of being in employment 3 years after leaving education, 2004 cohort, parameters of the logistic regression(a),(b)
Note: The occupational category and parents’ labour market situation control variables are crossed with the sex dummy; the model estimated in columns (1) to (3) is a simple logit, for column (4) it is a conditional logit; the control variables are the same for columns (2) and (3), only the sample changes.(a) Non-response. Father or mother unknown.
(b) The occupational category is not observed as the father or mother is inactive.
Coverage: French people aged 15-30 who left education in 1998 or 2004; columns (1), (2) and (3): IRISes with at least two observations; columns (4) and (5): IRISes with at least two observations having different values for the dependent variable.
Notes
-
[*]
CREST, Paris, France.
-
[**]
University of Warwick, United Kingdom.
Correspondence: Roland Rathelot, Department of Economics, University of Warwick, Coventry CV4 7AL, United Kingdom, email: R.Rathelot@warwick.ac.uk -
[***]
Sciences Po, Paris, France.
-
[1]
The French territory is divided into IRIS zones to form a mesh with cells of roughly equal population size. Municipalities with a population of 10,000 or more, and most municipalities with populations of 5-10,000 are broken down into IRIS zones totalling 2-5,000 inhabitants. A total of 1,900 municipalities are broken down in this way into 16,000 IRIS zones. For the others (the more than 34,000 small municipalities), the entire municipality forms one IRIS zone.
-
[2]
This “territorial discrimination” is now recognized under French law. On 14 January 2014, residential location became the twentieth criterion of discrimination in French anti-discrimination law.
-
[3]
As the minority group mainly comprises individuals whose parents are from North Africa and sub-Saharan Africa, the Middle East will not be mentioned in what follows.
-
[4]
The sub-samples are too small to define more detailed population categories. However, as numerous studies have documented the disadvantaged socioeconomic status of these populations when they are distinguished, we believe that it is useful to group them together in a study seeking to determine the extent to which this disadvantage can be explained by geographical location. The measured effects must therefore be interpreted with caution, as average effects within a potentially heterogeneous group.
-
[5]
ZUS = zone urbaine sensible (sensitive urban area) / CUCS = contrat urbain de cohésion sociale (Urban contract for social cohesion). These are priority zones defined by the government for the targeting of public support in terms of economic activity, employment and poverty alleviation. There are 2,500 neighbourhoods of this kind across urban France.
-
[6]
The question of gender-specific ethnic discrimination is regularly examined in the literature. We allow the effect of the minority population dummy to vary with sex.
-
[7]
The complete bootstrap involves drawing with replacement a large number of samples (here 200) of the same size as the initial sample and repeating the estimation procedure with each of these new samples. The distribution of the estimators obtained with this method is similar to that of the initial estimator, and the standard deviations and confidence intervals can be deduced for the coefficients of interest.
-
[8]
In the 1998 cohort, 15.3% of men and 8.8% of women left school with no qualifications (Table 2).
-
[9]
We attempted to limit the samples to IRIS zones common to both surveys. Unfortunately, this divided the sample size by 4 in 1998 and by 2 in 2004. Given the small size of the estimated effects, most are no longer statistically detectable with such small samples. For this reason, we cannot go any further in interpreting the trend effect suggested by our results.
-
[10]
The MTO team’s two studies produced paradoxical results. While the programme seems to have had substantial effects on the environment of the individuals concerned (an increase in living standards in their neighbourhoods of residence), no effect on economic performance (employment, wages) was detected. However, the programme did have major positive effects on the physical and mental health and the subjective well-being of the individuals treated.
