Among individuals coming from abroad to live in France, some pursue long-term settlement, while others return to their country of origin or migrate onward to a different country. Data that might foster better knowledge of their characteristics are extremely rare. Whether this is a matter of a selective process is a puzzling but important question. Do people remigrate because they are less integrated? Does that mean that those who stay are the most integrated? Using data from the Permanent Demographic Sample, the author responds to these questions and shows how to take remigration into account to measure the extent to which immigrants remaining in France are integrated into the labour market.
1Since the beginning of the century, a growing quantitative literature has explored the trajectories of immigrants and their descendants into French society. Most of these studies agree in rejecting classical assimilation theory, which postulates that immigrants’ behaviours and situations become increasingly similar to those of the native population over time and across generations, especially in terms of economic integration (income, access to employment). Yet the assimilation paradigm remains central in exploring immigrants’ situation at destination. In that context, the extent to which socioeconomic and cultural differences between immigrants and natives change over time is a crucial question.
2The validity of empirical analyses on these integration processes could be affected, however, by an important demographic phenomenon: the – potentially selective – remigration of a significant proportion of immigrants. Contrary to the implicit assumption of the assimilation paradigm, not all immigrants remain permanently in the country of destination. Some eventually leave the host country, either returning to their country of origin or migrating onward to a third country. These movements, called “remigration”, seem far from negligible. In OECD countries, it is estimated that between 20% and 50% of immigrants remigrate from the host country within five years after their arrival (Dumont and Spielvogel, 2008). [1] The aim of this article is not to provide a systematic and detailed overview of the vast literature that has investigated the diversity of forms of international mobility. Suffice it to say here that the traditional framework of permanent immigration to a single destination country is being gradually replaced in migration studies by concepts such as “migratory circulation” (circulation migratoire), which better capture the diversity of trajectories (Mung et al., 1998).
3While considering immigrants’ remigration is essential for a better understanding of migration trajectories, this phenomenon must also be considered in studying immigrants’ assimilation at destination. If immigrants who leave France have specific characteristics (such as being more likely to be unemployed), then focusing only on individuals who remain in the country will not provide a comprehensive overview on the integration of the entire migrant population who came to France. Although temporary migration has often been overlooked by the literature, a growing body of empirical research has investigated immigrants’ remigration and its potential consequences for measures of their assimilation in the host society in recent years (Dustmann and Görlach, 2016). However, this question remains neglected by almost all quantitative research on immigration in France (Caron, 2016). While some studies have examined circular migration and transnational practices, emigration remains overall one of the “mysteries of the French population” (mystères de la population française) (Le Bras, 2007), and its consequences for our understanding of integration processes are seldom considered. To the best of our knowledge, no empirical study has in particular investigated the extent to which measures of immigrants’ integration in the French labour market are affected by remigration.
4This article contributes to filling these gaps by investigating two important research questions for the analysis of integration. Do the immigrants who leave France have specific characteristics? If they do, does this selective remigration affect “classical” – i.e. cross-sectional – measures of immigrants’ labour market integration in France? To explore these questions, this article uses data from the Permanent Demographic Sample (EDP) for the period 1975–1999, an ongoing panel conducted by the French National Institute of Statistics and Economic Studies (INSEE) based on population censuses and civil registers. By doing so, this study takes advantage of French data that, although relatively old, provide an interesting opportunity to document immigrants’ remigration from France and to assess the extent to which it may bias measures of their integration. Beyond this question, the aim is to explore the empirical consequences of the conceptual framework traditionally used in France, which usually focuses on the idea of an “immigrant” coming to settle permanently in France, rather than on the “migrant”, who may migrate again (Richard, 2004). While this integrationist perspective underlying French research (Spire, 1999) does not foster analysis of integration mechanisms together with mobility patterns, this study intends to examine how the demographic dimension of population movements matters for the more classical analysis of immigrants’ integration in French quantitative studies.
5The rest of the article proceeds as follows. The next section reviews theoretical and empirical research on immigrants’ selection remigration and its implications for quantitative analyses of their integration at destination. Following a brief overview of the economic and policy contexts from the 1970s onward that have likely influenced the underlying logics of migration over the period, Section III describes the data and delineates the measurement and methodological strategies to analyse out-migration from France and to test the hypothesis of bias due to selective remigration. Findings are presented thereafter in two stages. First, remigrant selectivity is examined in comparison to those who stay in France. Secondly, we investigate the extent to which this remigration affects cross-sectional measures of immigrants’ trajectories in the labour market.
I – Measures of immigrants’ integration and temporary migration
1 – Remigration, a selective process?
6Many researchers, especially economists, have pointed out that migration is often selective within the population in the country of origin (Lee, 1966; Borjas, 1985, 1987). In other words, migrants and non-migrants differ in characteristics such as age, sex, or level of qualifications. Similarly, those who remigrate from the host country rarely represent a random sample of the immigrant population in the destination society. According to the literature, immigrants with strong social, family, and cultural ties to the host society (married to a native, owning their own home, becoming naturalized citizens) are more likely to stay there in the long term (Constant and Massey, 2003; Gibson and McKenzie, 2011; Van Hook and Zhang, 2011).
7Results are less clear-cut on the selection of remigrants in terms of level of education and socioeconomic position. Until the 1980s, the neoclassical theoretical framework regarded remigration primarily as a failure to integrate into the labour market at destination (Harris and Todaro, 1970). Immigrants who remigrate should therefore be the most economically disadvantaged. Shifting the analysis of migration choices to the level of the entire household, the New Economics of Labour Migration (Stark, 1991) offers a different perspective. The migration of one member of the family can be a part of a financial strategy for the entire household. In this perspective, return is one sequence within an optimal location plan over the life course and can be part of the initial plan. Immigrants who are socioeconomically well integrated may return to their country of birth if and when they have accumulated enough savings in the host society. Empirical results on this issue are also heterogeneous. Several studies have emphasized that the likelihood of leaving the country of destination is greater for immigrants who are unemployed or who earn low wages (Edin et al., 2000; Constant and Massey, 2003; Bijwaard et al., 2014). Others found that highly educated immigrants are more likely to remigrate (Lam, 1994; Beenstock, 1996; Gundel and Peters, 2008). Selection also differs by country of origin (Edin et al., 2000; Jensen and Pedersen, 2007) and depending on whether individuals are returning to their country of origin or migrating to another destination country (Nekby, 2006; Larramona, 2013).
2 – Cross-sectional analyses, population movements, and selection bias
8A non-negligible consequence of selective remigration is that it may lead to a bias in analyses of immigrants’ integration, especially in measures of economic integration that are drawn from cross-sectional data. An important quantitative tradition, which originated from the United States, has analysed processes of assimilation by investigating wage convergence between immigrants and natives. As longitudinal data on large samples to follow the trajectories of the same individuals over time remain scarce, these first studies mainly used repeated cross-sectional data. In Chiswick’s seminal work (1978), census data were used to study the evolution of earnings of immigrants in the United States. After controlling for the characteristics of arrival cohorts, a decrease in the immigrant– native earnings gap between two dates would confirm the hypothesis of an economic assimilation of immigrants over time. However, this explanation is valid only if the composition of the population under study does not change. The implicit hypothesis is that the analytical sample contains the same immigrants at the beginning and end of the period. But if, for example, the least qualified and/or lowest-paid immigrants left during the study period, the decrease in the wage difference between immigrants and natives may reflect a mechanical increase in the average income for the immigrant population resulting from these exits. The apparent positive effect of length of stay in the host society on immigrants’ individual level of economic integration would thus be overestimated. [2] The emigration of natives could lead to a similar bias when measuring the convergence between the socioeconomic positions of natives and immigrants, as the characteristics of the reference group would also evolve.
9This potential scenario of selective remigration was largely ignored in early quantitative analyses of immigration, and empirical tests remain rare (Dustmann and Görlach, 2016). This lack of empirical analyses is partly explained by the paucity of adequate data. While providing a comprehensive and accurate count of entries in a country is already a complex task, identifying immigrants who leave is even more difficult. By simultaneously collecting data in departure and destination countries, multi-sited surveys, such as the Mexican Migration Project and the INED-coordinated MAFE (Migrations between Africa and Europe) project, allow complex migratory trajectories to be better tracked (Beauchemin, 2014). However, such projects remain rare and are by definition limited to migration trajectories among a restricted set of origin and destination countries. While retrospective data often provide information on the date of arrival, they are also limited, since they are collected only from individuals still living in the host country at the time of the survey.
10In contrast to repeated cross-sectional data, panel data follow the same individuals over time. Therefore, they can be used to distinguish individual trajectories of individual immigrants who stay in the country of destination over the full period from those of immigrants who have remigrated. By controlling for arrival cohort, differences between estimates based on cross-sectional methods (which include temporary migrants) and longitudinal methods (which exclude remigrants) can be attributed to selective remigration. This method has been used in several recent quantitative studies, with mixed empirical results. Hu (2000) and Lubotsky (2007) show that measures using census data overestimate growth in immigrants’ earnings, compared to longitudinal estimates drawn from the Health and Retirement Survey and from Social Security data. However, as Picot and Piraino (2013) point out, these cross-sectional and panel estimates are based on different data sources, which limits their comparability. As discussed below, the structure of the EDP allows this limitation to be overcome: all models can be estimated using the same census data, distinguishing remigrants from individuals who remained in the country. The study by Abramitzky et al. (2014) offers a new perspective by matching individual United States census forms from 1900, 1910, and 1920 based on criteria such as age, family name, and country of origin. Estimations from cross-sectional censuses suggested that, on arrival, immigrants held lower-paid occupations than natives but that this difference decreased over time – immigrants “catching up” with the natives. However, estimates from this reconstituted panel show instead the persistence of the initial earnings gap between immigrants and natives over time. The authors suggest that the initial cross-sectional analyses were biased by the combined effect of lower levels of qualifications among more recent waves of immigrants and a selective remigration of the least qualified immigrants. Edin et al. (2000) also find that overlooking immigrants’ remigration leads to overestimating the economic integration of immigrants who arrived in Sweden between 1970 and 1990. Constant and Massey (2003), using results from the German Socio-Economic Panel, drew an opposite conclusion: excluding remigrants did not affect measurements of immigrants’ integration.
II – The French context
1 – The invisibility of exits from France
11Remigration and its consequences for documenting integration mechanisms continue to be largely ignored by quantitative research in France. This could be explained by two factors. First, from a practical perspective, there are no precise data on departures from France (Thierry, 2008). Emigration, of both immigrants and natives, thus remains “strangely absent” (l’étrange absente) from French public statistics (Legoux and Orain, 2014). The lack of studies on remigration can also be linked to the importance of the integration paradigm in French research on immigration, particularly in sociology. The tradition of the French republican model of integration, the end of labour immigration policies, and the establishment of family reunification policy in the 1970s have contributed to deeply linking the analyses of immigration together with the issue of immigrants’ integration in French social science research. While ‘foreigners’ would only be provisional visitors, immigrants are implicitly assumed to settle permanently in the host country, and thus they “designated in the minds of demographers and statisticians a group that was ‘stable’ over time” (“désigne[rait] dans l’esprit du démographe et du statisticien un groupe ‘stable’ dans le temps”, Spire, 1999, p. 56, my translation). From that angle, most quantitative studies on immigration have focused on documenting their integration into French society. Following Sayad (1977), some studies have deconstructed this paradigm. They have pointed out that immigrants are also “emigrants” and have emphasized that pre-migration experience and characteristics matter in understanding the dynamics of integration (Charbit et al., 1997; Ichou, 2014). Intentions to remigrate as such, however, remain neglected.
12While some rare quantitative studies have investigated out-migration from France, they usually focus on specific groups. Richard (1998, 2004), for example, studied the departures of the children of immigrants, while other studies took advantage of the biographical nature of the data from the MAFE survey to examine return migrations of Senegalese and Congolese migrants (Flahaux et al., 2014; Flahaux, 2015) as well as Senegalese immigrants’ multiple moves between France, Italy, and Spain (Toma and Castagnone, 2015). The literature has also emphasized that retirement is an important moment for back-andforth movements and return migration to the country of origin (Schaeffer, 2001; Attias-Donfut and Wolff, 2005; Dos Santos and Wolff, 2010). But the consequences of considering international movement for analyses of immigrants’ integration remain little studied in France. The only notable exception is Gobillon and Solignac (2015), who investigated the effects of selection in in- and out-migration flows on immigrants’ access to homeownership in France. Using EDP data, they showed significant differences in homeownership rates between “permanent” immigrants and those who remigrate. They also observed that the departures of remigrants, who are less likely to be homeowners, have a positive effect on overall trends in immigrants’ homeownership. A working paper by Solignac (2016), based on the EDP, estimates that between a quarter and a third of immigrants observed in a census between 1968 and 1999 had left the country within seven to nine years, underlying the need to rethink the role of emigration from France in analyses of immigrants’ geographical mobility. However, this promising approach remains marginal in French research. In particular, the implications of remigration for the empirical study of immigrants’ occupational trajectories in France remain largely unexplored. This article examines this question by investigating the consequences of out-migration for the quantitative study of immigrants’ trajectories in the French labour market between 1975 and 1999.
2 – Immigrants’ employment status in France (1975–1999)
13Two connected phenomena with the potential to influence the logic of migration mark a period in France beginning in the 1970s: first, the deterioration of the economic situation after the oil crises of 1973 and 1979, and second, the end of French policies promoting immigration. As economic indicators deteriorated, the establishment of restrictive immigration policies limiting immigration to family reunification seems to have decreased (but not stopped) the flow of arrivals and contributed to increasing the proportion of women among immigrants (Safi, 2007). Moreover, immigrants were offered financial incentives to return to their countries of origin. For example, between 1977 and 1978, foreigners registered as unemployed who agreed to return to their country of origin were offered 10,000 francs. Yet, as shown by Lebon (1979), the main outcome of this measure, intended to target workers from the Maghreb, was instead to hasten the departure of immigrants who had already planned to return to their countries of origin, mainly Spain and Portugal. Zamora and Lebon (1985) estimated that 530,000 foreigners left France between 1975 and 1982. However, their method based on the aggregate difference in the “stocks” of immigrants in the two censuses fails to provide precise characterization of these remigrants.
14Figure 1 illustrates the evolution of unemployment and labour force participation rates between 1975 and 1999 for immigrants and natives and by sex. The unemployment rate increased substantially over the period for each group, reflecting the deterioration in the French economic situation after several decades of full employment. This increase is higher for immigrants, and the gap between immigrants and natives expanded, for both men and women. As for labour market participation, the proportion of economically inactive men, both immigrants and natives, remains small. Meanwhile, the period is characterized by the growing proportion of women entering the French labour market. Though the proportion of inactive women decreased by around 20%, it remained large, representing 31% and 18% of immigrant and native women, respectively, in France in 1999.
Rates of unemployment and labour force participation, 1975–1999, cross-sectional approach
Rates of unemployment and labour force participation, 1975–1999, cross-sectional approach
Interpretation: In 1975, 3.6% of economically active immigrant men were unemployed, and 43.7% of immigrant women of working age were economically active.Coverage: Individuals aged 18 to 60 years at the time of the census, not including students and retired individuals. The unemployment rate represents the percentage of individuals in the economically active population who reported being unemployed on the census. The activity rate represents the proportion of economically active individuals in the overall population.
15The greater increase in the unemployment rate of immigrants might suggest that the individual employment situations of immigrants deteriorated in relationship to natives. However, this aggregate trend could reflect two types of population movements rather than the trajectories of individual immigrants on the labour market: the arrival of new waves of migrants with different characteristics (such as being less qualified and thus more likely to be unemployed) and the selective remigration of a significant proportion of immigrants. Controlling for arrival cohort, this article explores whether the observed differences between immigrants and natives are robust when taking into account immigrants’ – potentially selective – remigration. If remigrants were positively selected – that is, if they were well integrated in the French labour market in comparison to immigrants who remained in the country between 1975 and 1999 – this raises the possibility that the deterioration of immigrants’ employment situation was overestimated or, at least, does not reflect change over time in the full immigrant population living in France in 1975. A negative selection effect in departures, with unemployed or economically precarious immigrants more likely to remigrate, would lead to the opposite conclusion.
III – Data and method
1 – Data: An original use of the Permanent Demographic Sample (EDP)
16Created in 1967 by INSEE, the EDP is a panel that contains information from censuses and civil registries of births, marriages, recognition of children, and deaths since 1968. This article uses EDP data for metropolitan France between 1968 and 1999, drawn from five census years: 1968, 1975, 1982, 1990, and 1999. The sampling method is based on a random procedure. The EDP includes individuals born on one of the first four days of October (around 1% of the population), for whom at least one census form or one of the abovementioned vital records is available. These individuals are then followed over their lifetime based on the information collected in vital statistics and censuses. “EDP individuals” born in metropolitan France are added to the panel at birth, while those born outside France are added with the census, ensuring that the sample remains representative.
17An individual can leave the panel either through death or emigration from France. Because the EDP includes death certificates for persons who died in France, attrition not explained by death can be assumed to be due to emigration. [3] The possibility of following individuals who have remained in France makes the EDP unique in the French statistical landscape, enabling indirect analyses of emigration across the entire population. In addition, the size of the EDP (around 900,000 individuals followed in 1999) enables statistical analyses of populations that are often insufficiently represented in French surveys, such as immigrants by country of origin. These analyses can cover a long period (nearly 30 years). The EDP also allows for analyses differentiated by sex, contrary to the great majority of existing research on remigration, which concentrates on men. Just as the reasons for the initial migration and occupational trajectories often differ by sex (Houseaux and Tavan, 2005; Tavan, 2006), so may the reasons for departure.
18However, using EDP data is subject to some limitations. First, individuals who declared no date of birth, the sampling criterion for the EDP, are not included in the database. [4] This proportion is larger for immigrants than for natives. Studies carried out by INSEE highlight a slight under-representation of immigrants in the EDP, particularly those from Morocco and sub-Saharan Africa (Rouault and Thave, 1997). However, these limitations should not affect the robustness of our analyses, insofar as we are not seeking to estimate the representativeness of a particular origin group. Second, the information collected in the EDP is mainly sociodemographic, and some important characteristics in analysing immigrants’ integration (such as linguistic practices and links maintained with the country of origin) are absent. The relatively long interval between censuses also leads to an underestimation of individuals’ real mobility, as movements in these intervals are unknown. Third, the analysis unfortunately cannot be extended beyond 2000 because since 2004, the census has no longer been exhaustive, but survey-based.
19Another series of limitations is more specifically linked to our definition of emigration. This indirect measure of emigration is limited, first of all, because attrition may also be due to data collection failures – a potential problem for any longitudinal data. Difficult to measure, this risk would entail an overestimation of departures. But INSEE’s centralization of population registers, along with the exhaustive coverage of metropolitan France in each census, ensure that the EDP is particularly effective in following individuals. As emphasized by Solignac (2016), the rate of omission from the census estimated by INSEE remains low (2%), although it is slightly higher for non-citizens. To minimize this problem, our definition of “departures”, which is presented in the next section, is limited to individuals absent from multiple consecutive censuses. Further, the choice to equate departures from the country with attrition net of deaths relies on the quality of the recording of death certificates. Resource restrictions led to only a partial recovery of death certificates for individuals born on 2 or 3 October. [5] Consequently, the analyses in this article were performed only on individuals born on 1 or 4 October. Our definition of emigration assumes that the same individual was identified and matched between the different censuses. INSEE’s techniques to identify individuals differ depending on place of birth. Individuals born abroad are not immediately recorded in the National Directory for the Identification of Natural Persons (RNIPP), which complicates the application of INSEE matching procedures to immigrants (Jugnot, 2014). One may nonetheless hypothesize that individuals lost to follow-up are randomly distributed within the immigrant population. In this case, they should not bias our results, as we are mainly seeking to determine whether remigrants have specific characteristics, not to estimate the size of emigration flows. Where identification is uncertain, INSEE’s systematic manual checks yield a low failure rate (around 1% according to Couet, 2006), compared to panels that are manually reconstituted by researchers using administrative databases. [6]
2 – Methodology
Identifying permanent emigrants
20An “emigrant” is here considered an individual present at time t, absent from all subsequent censuses, and for whom no death certificate between t and t + 1 was collected. Individuals absent from one or two censuses but who reappeared in the panel in 1999 are thus not considered emigrants. [7] This “conservative” approach to identifying departures limits the risk of classifying data collection failures as cases of emigration. The choice is also justified by this paper’s second research question: Do cross-sectional measures of immigrants’ labour market integration between 1975 and 1999 characterize the pattern for the entire immigrant population that was present at the beginning of the period?
21While it is possible to check that individuals who left between 1975 and 1982 did not return either in 1990 or in 1999, this is impossible for departures between 1990 and 1999 due to the switch in 2004 to a non-exhaustive, surveybased census. To ensure the most uniform possible definition of emigrants, the present study uses the 2014 version of the EDP, which has the advantage of being matched with nearly exhaustive social tax data. Available for 2011 onwards, these data offer a good overview of the individuals present in the country, as they include everyone who filed an income tax return (déclaration d’impôt sur le revenu) or paid the housing tax (taxe d’habitation) on a principal residence. This additional date is used in constructing the variable for departures (but not as an observation date) to check whether absences between 1990 and 1999 persist in 2011, again limiting the risk of falsely identifying someone as having permanently left the country. This definition of “permanent” departures over the period applies to both immigrants and natives. Failing to take into account the emigration of natives could also affect measurements of convergence between the two groups.
A comparison of models based on cross-sectional and panel data
22Section IV presents empirical analyses in two steps. To examine whether remigration is selective, a random effects logistic panel model is used to analyse the determinants of immigrants’ departures. We then test for the existence of a bias linked to selective remigration in measurements of immigrants’ labour market integration. To do this, we adopt a methodological approach whose central idea, shared with several earlier studies, consists in comparing measurements of immigrants’ labour market integration obtained from cross-sectional and longitudinal estimates. Formally, the equation to be estimated in three ways is:
24where Y is the dependent binary variable, defined only for economically active individuals, which takes a value of 1 if individual i is employed in census t and 0 if the individual is unemployed. The variable immig is the indicator of membership in the immigrant population (defined as individuals born non-French abroad); year the date of the census; and X a set of sociodemographic control variables (age, age squared, marital status, level of education, and size of the urban unit of residence). The interaction β characterizes the relationship between immigrants’ probability of being employed with respect to that of natives by census and arrival cohort. The regressions are estimated separately for men and women and for different arrival cohorts. It thus captures, in a way, the effect of time since immigration on individuals’ labour market integration. While research on the effects of remigration has generally focused on characterizing immigrants’ integration based on earnings disparities, the EDP does not include income information. [8] Empirical studies have also shown that in France, labour market inequalities between immigrants and natives result more from disparities in access to employment than from wage differences (Aeberhardt and Pouget, 2006). [9] For women, we also study the probability of being economically active rather than inactive, as women’s labour market participation was still low in the 1970s and 1980s, particularly among immigrants (Figure 1).
25As the EDP does not include the date of arrival in France, the individual’s first appearance on the panel is used as an indicator of the arrival cohort. For example, we consider an immigrant who appeared in the database in 1975 as having arrived in the intercensal period 1968–1975. Because the 1968 census is used to construct this indicator, it is not included as an observation date, and our analysis is limited to the period 1975–1999. Different age restrictions are applied to each arrival cohort to ensure that individuals were old enough to work at the beginning of the period but still young enough in 1999 to participate in the labour market. [10] A cohort of natives is associated with each wave of migration thus defined, with the same age limits to guarantee comparability. The production of separate estimates for each cohort allows for analysis of different levels of selection in different waves of migration; it also avoids imposing the same changes in employment rates over time on all cohorts.
26The objective is to compare the changes over time in immigrants’ labour market position relative to natives according to three estimates of equation (1):
- The first model uses a cross-sectional logistic regression. This statistical estimate therefore does not make use of the longitudinal character of the data (individual*year). It treats the same individual in two censuses as two different individuals. The idea here is to produce an approximate equivalent of repeated cross-sectional analyses that use successive censuses to analyse change in immigrants’ situations.
- The second model estimates equation (1) using a logit panel regression with random effects. The sample is the same as in Specification 1: individuals contribute to the estimates if they remain in the EDP. The difference between Models 1 and 2 lies in the type of statistical specification used. While cross-sectional models treat a given individual at two separate dates as two different people, this is not the case in panel regressions, which provide more robust statistical estimates. Comparing these two estimates enables us to determine whether measurements of immigrants’ integration change depending on the nature of the statistical models used.
- In the third model, we again use a logit panel regression with random effects but on a different sample, consisting only of individuals (both immigrants and natives) who did not permanently leave France during the study period. [11] Model 3 thus analyses the occupational trajectories of immigrants who remained in France until 1999, in the general socioeconomic conditions as well as the particular conditions that immigrants experienced over the study period.
27Theoretically, Models 2 and 3 differ only due to emigration, and comparing them should allow us to assess the extent to which immigrants’ remigration affects indicators of integration processes in France. If the two specifications yield significantly different results on trends over time in immigrants’ labour market position relative to natives, then there is a bias linked to selective remigration. For example, if Model 2 suggests that immigrants face a greater disadvantage relative to natives than does Model 3, this would suggest a positive selection of permanent emigrants; i.e. these immigrants were relatively well integrated in the labour market. Model 2 is an intermediate step that allows us to distinguish differences due to an improvement in the statistical specifications of the panel from those explained by the inclusion or exclusion of remigrants in the sample.
28While fixed effects models are preferable from a theoretical perspective because they make it possible to control for unobserved individual heterogeneity, we use random effects specifications. The principal limitation of fixed effects is that the variable of interest (being an immigrant rather than a native) does not vary over time and thus is not estimated. The interaction of this variable with the year nevertheless “reveals” the coefficient assigned to it (Allison, 2009). This type of model is nonetheless limited because identification is based on individuals who are subject to intra-individual variance, which imposes major restrictions on the sample. The proportion of remigrants among immigrants is divided in two when fixed effects are used for Models 2 and 3. As the choice of the specification of the error term only marginally changes the results, random effects are used instead to maintain the most representative possible sample.
Samples
29The children of immigrants are more likely to be in an unfavourable position on the labour market (Silberman and Fournier, 1999, 2006) compared to the children of natives, and they are likely to migrate for distinct reasons (Richard, 2004). They are excluded from all analyses to avoid biasing estimated differences between immigrants and natives. Students are also excluded because they may show specific forms of international mobility. Furthermore, the combination of definitions of migration waves and emigration exclude immigrants who arrived before 1968 or after 1990. Members of origin groups who mainly arrived before 1970, such as Italian and Spanish immigrants, are thus under-represented.
30Next, the samples in the two subsections of Section IV differ. In the first, which examines the factors influencing remigration, the models are estimated on immigrants who, at the time of the census, were aged 18 years or older and were not students. The second subsection, which tests for the existence of a remigration bias, includes natives but covers a more limited sample of the immigrant population. Focusing on occupational trajectories, the three specifications of equation (1) are based only on economically active individuals, with different age restrictions for each arrival cohort. Appendix Table A.1. provides the numbers of individuals used for the three specifications modelling the access to employment of immigrants who arrived between 1968 and 1975 compared to natives. For women, we include models on the probability of labour market participation, estimated on both economically active and inactive individuals.
IV – Results
1 – Immigrants’ remigration from France is selective
31Table 1 presents the proportion of departures without return up to 1999 by intercensal period and population category. Immigrants’ rate of emigration is higher, which corroborates previous results as well as the idea that a first experience of migration tends to favour a later “migratory career” (Martiniello and Rea, 2011). The greater number of departures of immigrants between 1975 and 1982 might reflect the end of labour immigration policies, which may have led migrant workers to return to their countries of immigration. Remigration rates then decreased over the period but remained significant.
32Table 2 presents estimates derived from a logistic regression with random effects which models the probability that an immigrant would leave France between t and t + 1 without returning by 1999 as a function of their characteristics in census t. Regressions control for immigrants’ first appearance in the panel and are estimated separately for men and women. As it is difficult to compare the coefficients between specifications in the context of logistic regressions, we report the marginal effects of each explanatory variable on the probability of departure. [12]
Table 1. Distribution of permanent emigrants by intercensal period
Table 1. Distribution of permanent emigrants by intercensal period
Interpretation: 38.5% of immigrants who were present in 1975 left metropolitan France during the intercensal period without returning by 1999.Coverage: Individuals born on 1 or 4 October, aged at least 18 years at the time of the census, not including students.
33The results confirm the hypothesis of selective remigration, as temporary migrants have specific characteristics. Married immigrants and divorced women are more likely to stay in France. For male and female immigrants, the probability of leaving the country increased with age. The results also highlight significant differences across countries of origin. Male immigrants from the Maghreb are more likely to remigrate, contrasting especially with those from Southeast Asia (defined here as Cambodia, Laos, and Vietnam). The role of the country of origin differs by sex; in particular, the probability of remigration is lower among immigrant women from North Africa, while European immigrant men are more likely to leave. Such differences may be linked to the influence of the geographical proximity of the countries of origin in the case of return migration.
34Determinants associated with the labour market seem to be more significant for men. This seems logical given that male immigrants primarily migrated to France for economic reasons over the period, whereas female immigrants usually arrived through family reunification. Male immigrants who are unemployed, retired, or inactive are more likely to leave the country, all things being equal, which confirms both the hypotheses of remigration associated with failures of labour market integration and of a return to the country of origin at the end of working life. The impact of the economic situation does not seem linear, however, as both unskilled manual workers and immigrants in higher-level occupations are more likely to emigrate than skilled manual workers (+5.8 and 17.3 percentage points, respectively). The results suggest that remigration is also selective in terms of education, as immigrants with no educational qualifications are significantly more likely to leave. This effect is more pronounced for men. In addition, immigrants who arrived more recently in France are more likely to remigrate.
35The causal effect of these results must be regarded cautiously. Indeed, it is difficult to distinguish whether high levels of socioeconomic and cultural integration encourage immigrants to remain in the country, or whether length of stay at destination increases the probability of marrying or finding a job in the host society. Although the direction of causality is uncertain, these results are still of interest because they show that the immigrants who left over the study period are characterized by specific employment status and levels of education. Consequently, the research question explored in what follows is whether this selective remigration affects measures of immigrants’ economic integration.
Factors influencing the probability of remigration without return between 1975 and 1999
Factors influencing the probability of remigration without return between 1975 and 1999
Significance levels: † p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001Interpretation: A male immigrant from Algeria was 11.4 percentage points more likely to emigrate than one from Portugal, all else being equal.
Coverage: Immigrants born on 1 or 4 October, aged at least 18 years on the observation date, excluding students.
2 – Immigrants’ selective remigration does not affect general trends in integration
36To answer this question, we compare estimates using cross-sectional and panel regressions, as explained in Section III. Figure 2 illustrates the marginal effect of being an immigrant on the probability of being employed at each census, as well as the 95% confidence intervals, for men and for women. The regressions from which these estimates are drawn can be found in Appendix Tables A.3 and A.4. For the sake of concision, only the graphs for the 1968–1975 arrival cohort are presented, as conclusions on the selective remigration bias remain unchanged for all arrival cohorts.
37After controlling for sociodemographic characteristics, the likelihood of being employed in 1975 is similar for active immigrants and natives. The greater increase in immigrants’ unemployment rate between 1975 and 1999 shown in Figure 1 thus cannot be explained by specific characteristics of immigrants from this cohort. The differences subsequently increased to the detriment of immigrants, particularly men, whereas the relative disadvantage of immigrant women compared to native women beginning in 1990 seems to have slightly decreased in 1999. Rather than providing information on the assimilation process, with immigrants “falling behind” natives over their years in France, this observation seems to illustrate the deterioration of the country’s economic situation between 1975 and 1999, which disproportionately affected immigrants, all things being equal. Further analyses on the other two arrival cohorts (1975–1982 and 1982–1990) confirm this idea, as the probability of being unemployed is higher for those immigrants as soon as they arrived in France than for natives with similar characteristics. Because these new waves of migration were equally (or even more) educated, this relative disadvantage on the labour market seems in large part linked to the economic crisis. The present article, however, aims not so much to test the assimilation hypothesis over the period as to investigate whether analyses of immigrants’ economic assimilation using different specifications produce different results. Immigrants’ relative disadvantage is smaller in the panel models (Specifications 2 and 3), suggesting that the cross-sectional analysis overestimates the difference with respect to natives. However, these differences are never significant, as confidence intervals are overlapping. The general conclusions remain unchanged when including those who left the country between 1975 and 1999. The results on women’s probability of participating in the labour market are also not affected by a selective remigration bias (Appendix Figure A.1).
Marginal effect of being an immigrant on the probability of being employed, active immigrants, arrival in 1968–1975
Marginal effect of being an immigrant on the probability of being employed, active immigrants, arrival in 1968–1975
Interpretation: According to Specification 1, immigrant men who arrived between 1968 and 1975 were 2.4 percentage points less likely than natives to be employed in 1982, all else being equal.Coverage: Individuals born on 1 or 4 October, aged 18 to 36 years in 1975, who appeared in the EDP in 1975 in the case of immigrants, excluding students and economically inactive people.
38Do these results differ depending on immigrants’ country of origin? Figure 3 represents the effect of origin on the probability that an economically active individual would be employed, once again for the 1968–1975 arrival cohort and with natives as the reference category. [13] The negative trend in access to employment for immigrants seen in Figure 2 seems largely due to the trajectories of those from Algeria, Tunisia, and Turkey. After controlling for sociodemographic variables, the economic positions of immigrants from Europe (in particular Spain, Italy, and Portugal) and Southeast Asia did not differ from those of native-born French citizens, corroborating previous findings (Safi, 2006). Here again, the differences between the specifications are overall not significant. However, the positions of immigrants from sub-Saharan Africa no longer appear significantly different from those of natives (their confidence interval includes the reference) in Specification 3, unlike in the two other models. Not taking remigration into account thus leads to a slight overestimation of the difference between this group and natives. The results show few differences between the three specifications in the regressions for economically inactive women by origin; these are omitted for the sake of concision.
39The remigration of some immigrants thus does not seem to significantly bias cross-sectional estimates of immigrants’ labour market integration, although the results suggest that these remigrations are selective. Several factors may explain this result. First, remigrants must make up a large enough proportion of all immigrants to distort measurements of integration. While this proportion is relatively large between 1975 and 1982, it subsequently decreased. In any case, most immigrants remained in France (Table 1), especially among those who were economically active over the whole period.
Effect of origin on access to employment among economically active persons in the 1968–1975 cohort, men and women combined
Effect of origin on access to employment among economically active persons in the 1968–1975 cohort, men and women combined
* Without Italy, Portugal, Spain.Interpretation: Immigrants from Algeria were 16.6 percentage points less likely to be employed in 1999 than natives, according to Specification 1, all else being equal.
Coverage: Individuals born on 1 or 4 October, aged 18 to 36 years in 1975, who appeared in the EDP in 1975 in the case of immigrants, excluding students and economically inactive people.
40Next, the results in Table 2 suggest that, all things being equal, those who remigrate tend to be older, retired, or inactive. Given their economic position, these individuals’ remigration behaviours are not likely to significantly affect measures of immigrants’ trajectories in the labour market. However, running the same logistic regression only on active immigrants from the 1968–1975 cohort still shows a positive effect of unemployment on the probability of remigration, and the differences in the samples used in the two parts of the analysis fail to entirely explain the absence of a remigration bias. This study focuses on the position of immigrants relative to natives in terms of access to employment. In Canada, Picot and Piraino (2013) reported different conclusions on trends in immigrants’ absolute incomes and on the relationship between immigrants’ incomes and those of natives. They found a remigration bias in the first case, but not in the second. To explain these results, they underline the role of emigration of a significant proportion of natives with the same characteristics as temporary migrants, thus compensating the effect of the emigration bias. Appendix Table A.5 provides estimations drawn from the same logistic regression as in Table 2 on the native population. While results show mechanisms that are less selective overall, male natives who are unemployed at time t are more likely to move abroad in the following period, similarly to the findings for male immigrants. This may partly neutralize the potential effect of remigration bias for immigrants. However, this hypothesis must be treated with caution. Emigration remains a small-scale phenomenon among natives, representing only around 1.5% of the population over the period (Table 1). The effect of unemployment on native men’s probability of emigrating remains small (+1.0 percentage point). Another final factor might explain the absence of a selective remigration bias. Table 2 shows that immigrants who are unemployed or in a higher-level occupation are more likely to leave France. Remigrants may therefore be both negatively and positively selected based on their situation in the French labour market. The remigration bias on measurements of immigrants’ economic integration may thus be partly masked by compensating selective effects going in opposite directions.
Conclusion
41This paper aims to describe the under-investigated phenomenon of remigration in France. It explores, for the first time, the potential consequences of remigration for quantitative analysis of immigrants’ integration in France, drawing on data from the EDP to analyse out-migration behaviours for both men and women. By comparing estimates drawn from cross-sectional and panel regressions, the data allow distinguishing the effects of a change in statistical models from a selective emigration bias in measures of immigrants’ labour market integration. Results highlight differences between temporary and permanently settled migrants in family situation, level of education, and labour market situation. However, these selective departures do not seem to significantly affect measures of the evolution of differences in access to employment among immigrants and natives between 1975 and 1999. While the two groups hold similar positions in the labour market in 1975 in a context of full employment in France, cross-sectional and panel estimates show similar increases in the unemployment rates of immigrants and natives over the study period. To some extent, these changes reflect the effects of the economic crisis beginning in the late 1970s, which had a greater impact on the economic position of new waves of immigrants and which negatively affected the labour market trajectories of immigrants who remained in France (particularly those from the Maghreb and Turkey). The remigration over the period of immigrants who were unemployed or who had no educational qualifications (and presumably less likely to remain employed) seems to have been partly offset by the remigration of immigrants in higher-level occupations and has consequently little effect on the conclusions of cross-sectional analyses of immigrants’ occupational trajectories.
42While these results shed some empirical light on an issue that has received little attention in quantitative research in France, the scope of these results is subject to some limitations. First, from a methodological perspective, the EDP does not allow distinguishing between immigrants who left within the first years after their arrival and those who stayed for seven or eight years. Yet the literature shows that the nature of selection varies with the duration of residency. Temporary migrants who leave after a short period are often unemployed or hold low-paid jobs (Duleep, 1994). Second, our analyses bear on a period characterized by a specific economic context and migration policies. While rising unemployment may have encouraged some immigrants to leave, it seems that the “closing” of borders and the development of family reunification tended instead to encourage foreign workers who initially came on a temporary basis to stay in the long term (Safi, 2007). Our results are also strongly linked to the characteristics of particular migration waves, which may have since changed. The feminization of immigration which began in the late 1970s has continued, and women now make up a majority of immigrants to France (Beauchemin et al., 2013). The initial reasons for migration (political asylum, education, family or economic reasons) affect the logic of remigration (Edin et al., 2000; Dustmann and Weiss, 2007), and the diversification of these reasons could modify trajectories and selection processes. For example, the free movement of individuals within the Schengen Area has contributed to redefining the logic of population movements, and “intra-EU” mobility is apparently more similar to internal mobility (Recchi and Favell, 2009). According to some researchers, developing transportation and communication would promote return migration (Castles, 2006). The nature of remigration may thus have changed after the 2000s: economically active immigrants may make up a larger proportion of remigrants, and analyses of integration mechanisms could be differently affected. In that perspective, Table 2 shows that the most recent migration waves have been more mobile. Similarly, recent research from INSEE offering the first indirect estimates of emigration suggests that exits from France increased between 2006 and 2015, among both immigrants and natives (Brutel, 2015).
43Finally, the absence of a selective remigration bias applies only to immigrants’ integration in terms of access to employment. Yet integration processes are not always uniform (Portes and Zhou, 1993), and the literature has shown that the theory of a “segmented” assimilation provides a useful framework to understand the dynamics of integration mechanisms in France. For example, immigrants from the Maghreb are found to be socioeconomically disadvantaged but strongly integrated culturally (illustrated, among other things, by low rates of endogamy), while the reverse is observed for Portuguese and Asian immigrants (Safi, 2006). A bias linked to the remigration of certain immigrants might therefore affect other dimensions of assimilation. Analyses of homogamy, for instance, might be affected, since the results in Table 2 suggest that remigrants are less likely to be married than immigrants who remain in France. The study by Gobillon and Solignac (2015) on the importance of selectivity in population movements for the analysis of access to homeownership goes in the same direction.
44In the end, this study contributes to enriching the conceptual and theoretical framework for the analysis of immigration in France, and calls for rethinking the role of mobility trajectories in the study of integration processes. This research programme is in line with a recent literature that highlights the upstream role played by migration projects in integration strategies (Dustmann and Görlach, 2016). This study suggests moving beyond the definition of immigrants as an a priori homogeneous statistical group and exploring the heterogeneity of this group not only by origin, but also in terms of migration trajectories. More generally, pointing to specific patterns of migration among natives calls for jointly exploring international mobility for different population groups.
Acknowledgements
My thanks to Mirna Safi for her advice and invaluable remarks, as well as to Cris Beauchemin, Marine Haddad, Morgan Kitzmann, and Félix Paquier for their feedback on the text and their encouragement. Thanks to the participants in the “Refugees and Global Justice” doctoral seminar (UCLA–Sciences Po-Humboldt) and the MUTADEMO conference (Paris) for their comments on a previous version of this article, in particular Ettore Recchi and Matthieu Solignac. I would also like to thank the editors of the journal, as well as the three anonymous peer reviewers, whose critiques and suggestions contributed substantially to the article. Thanks in addition to INSEE and the Centre d’accès sécurisé aux données (CASD) for making the data available. Finally, my thanks to Mathieu Ichou for useful discussions and feedback, which considerably improved this work.This work was supported by a public grant overseen by the French National Research Agency (ANR) as part of the “Investissements d’avenir” programme, within the LIEPP LabEx (ANR-11-LABX-0091, ANR-11-IDEX-000502), as well as through further funding under reference ANR-10-EQPX-17 (CASD).
Appendix
Sizes of samples used in regressions on access to employment in the economically active population (used for Figure 2)
Sizes of samples used in regressions on access to employment in the economically active population (used for Figure 2)
Coverage: Individuals born on 1 or 4 October, aged 18 to 36 years in 1975, who appeared in the EDP in 1975 in the case of immigrants, excluding students and economically inactive people.Descriptive statistics on remigrating immigrants and immigrants who remained in France until 1999, by census year
Descriptive statistics on remigrating immigrants and immigrants who remained in France until 1999, by census year
Interpretation: 17.15% of immigrants who left France between 1975 and 1982 without returning by 1999 were of Algerian origin. Immigrants from Algeria made up 15.94% of the population of immigrants present in 1975 who were still in France at the end of the study period.Coverage: Immigrants born on 1 or 4 October, aged at least 18 years on the observation date, excluding students.
Factors influencing the probability of being employed, economically active men, 1975–1999 (used to construct Figure 2)
Factors influencing the probability of being employed, economically active men, 1975–1999 (used to construct Figure 2)
Significance levels: † p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001Coverage: Men born on 1 or 4 October, aged 18 to 36 years in 1975, who appeared in the EDP in 1975 in the case of immigrants, excluding students and economically inactive people.
Factors influencing the probability of being employed, economically active women, 1975–1999 (used to construct Figure 2)
Factors influencing the probability of being employed, economically active women, 1975–1999 (used to construct Figure 2)
Significance levels: † p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001Coverage: Women born on 1 or 4 October, aged 18 to 36 years in 1975, who appeared in the EDP in 1975 in the case of immigrants, excluding students and economically inactive people.
Factors influencing the probability of emigration without return, natives, 1975 to 1999
Significance levels: † p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.Interpretation: A native man was 0.1 percentage point less likely to emigrate than a never-married man, all else being equal.
Coverage: Native-born individuals born on 1 or 4 October, aged at least 18 years at the observation date, excluding students.
Notes
-
[1]
For a detailed literature review, see Mezger Kveder (2013) and Dustmann and Görlach (2014, 2016).
-
[2]
It is important to differentiate between, on the one hand, studies that document the average position of the immigrant population in the labour market of the host society and its evolution over a period, and, on the other, those studying the individual trajectories of immigrants within French society since their arrival. Potential selection bias due to remigration only applies to the latter, as cross-sectional data provide a snapshot of differences between immigrants and natives who are present in the country at the time of observation.
-
[3]
This method is in line with several studies analysing emigration based on attrition in longitudinal data (Constant and Massey, 2003; Bellemare, 2007).
-
[4]
This applies only to individuals who were not integrated into the EDP by way of their vital records and who did not declare their age in the census questionnaire.
-
[5]
Solignac (2016) finds attrition between 1990 and 1999 that was 10 percentage points higher for individuals born on 2 and 3 October, although their characteristics were similar.
-
[6]
In comparison, the procedure developed by Abramitzky et al. (2014) matches 16% of natives and 12% of persons born abroad between censuses. Moreover, this technique is based on the individual’s last name and is applicable only to men.
-
[7]
Robustness tests using a less strict definition of emigration (absence from two consecutive censuses) led to similar conclusions regarding remigration bias.
-
[8]
Recent versions of the EDP provide data on paid employment through a match with annual registers of private-sector employees (DADS). However, data on wages between 1975 and 1999 are largely incomplete, particularly for immigrants who disappeared from the panel. This is explained by the changing coverage of the DADS, which only began to cover all employees beginning in 2009, as well as by problems with matching to the panel of specific individual files in the EDP.
-
[9]
Further analyses – not reported here – comparing immigrants’ and natives’ chances of entering certain socio-occupational categories once they had obtained employment provide similar results.
-
[10]
For example, we restricted the cohort of individuals who arrived between 1968 and 1975 to those between the ages of 18 and 36 years in 1975.
-
[11]
The control group thus also changed, as immigrants who were still present in 1999 were compared to natives who also had not permanently left France.
-
[12]
The characteristics of remigrants compared to immigrants who remained in France for each intercensal period are presented in Appendix Table A.2.
-
[13]
Breaking down by cohort, sex, and country of origin leads to samples that are too small to allow the estimation of certain points. To provide figures that would be easier to read and since the regressions by sex led to similar conclusions, Figure 3 is estimated on the entire sample with no distinction for sex. These analyses were also performed on all individuals in the EDP to check whether the lack of a difference was not due to the reduced size of the samples restricted to individuals born on 1 or 4 October. The resulting graphs are similar.
