Back to Square One: Socioeconomic Integration of Deported Migrants 1

This paper addresses the issue of socioeconomic integration of forced return migrants, focusing on the Maghreb countries. Starting from the hypothesis that the return has to be prepared, I tested whether a disruption in the migration cycle (such as deportation) increases the individual's vulnerability and affects his integration from both a structural and sociocultural point of view, using the 2006 Migration de Retour au Maghreb (MIREM, or Return Migration to the Maghreb) survey. I found that forced returnees are more vulnerable to negative labor market outcomes compared to voluntary returnees. The absence of forced returnees from the labor market, or their underperformances, creates a net loss for the origin country and also incentives to re-migrate. The negative effect is statistically significant not only immediately after return, but also in the long run, at survey time. Forced return is also significantly and negatively correlated with sociocultural integration, reflecting a marginalization of deported migrants in their home environment, which may act as a re-emigration incentive.


INTRODUCTION
Apart from stirring civil society's outrage, the impact of forced return migration has drawn little attention from policymakers, and researchers are only beginning to study the extent of its implications. Immigrants 1 I am sincerely thankful to my advisor, Philippe De Vreyer, for his guidance and support. This paper has greatly benefited from comments by Jean-Philippe Dedieu, Fr ederic Docquier, Flore Gubert, Tom Ivlevs, Mohamed Ali Marouani, Marion Mercier, Alice Mesnard, Ewa Moraska, Christophe J. Nordman, Laure Pasquier Doumer, Anne-Sophie Robillard, Franc ßois Roubaud, and Claire Zanuso. I also thank the participants to the THEMIS conference and to the second DIAL Development Conference for useful discussions. have to face the issue of integration when they find themselves in a foreign environment, in some cases, governed by unfamiliar norms. At first glance, returning to one's home country should not entangle the difficulties faced by immigrants, but both empirics and return migrants' stories indicate otherwise. Although not of the same order, the challenges in terms of reintegration that await return migrants are comparable to the integration issues of immigrants. One of the reasons is acculturation phenomenona concept from the sociological literature which defines the change in an immigrant's behavior and culture induced by the environment at destination (Arowolo 2000). Values and expectations vary between the moment when an individual leaves the home country and the moment he returns. In addition, Preston and Brown (1993) argue that the environment in which migrants return to differs from the one they left behind.
These different aspects create the need for migrants to readjust to their home countries' contexts. This readjustment is costly. Upon returning, migrants need to reassert their socioeconomic status, a key determinant for a successful return. Finding a job or opening a business can be difficult for return migrants who lack information concerning the codes and practices in place or in the absence of social capital that could provide the needed assistance. Cassarino (2004) highlights the importance of migrants' "willingness to return" and their reintegration. Migrants wanting to return will therefore prepare their return, through resource mobilization and activating social networks.
When migrants are forced to return, whether for administrative reasons (visa or residence card non-renewed, deportation for illegal migrants, etc.) or personal reasons (health problems, family constraints, etc.), the end of the migration episode is sudden and does not leave enough time for a suitable preparation. This paper addresses the issue of forced return migrants and their reintegration from both an economic and sociocultural point of view. It mainly investigates the extent to which forced return affects socioeconomic integration, proxied by a higher vulnerability to unemployment and the degree of sociocultural integration, for which I propose a quantitative indicator based on the work by Fokkema and de Haas (2011). I found that forced return has a lasting negative impact on migrants' reintegration and discuss the implications of this result. The paper also adds a quantitative economic dimension to the existing literature on forced migration, which has included mainly sociological and legal aspects. Siegel and Shryock (1973) define a return migrant as a person moving back to the area where he formerly resided. Cassarino (2004) carries out an extended study on the existing theories on return migration, highlighting the main schools. Thus, according to neoclassical economic theory, return has been seen as the result of a failed migration experience (Todaro 1969). On the contrary, the new economics of labor migration (NELM) considers that return is a rational decision and the outcome of a predefined strategy (Stark 1991). In a structural approach theory, Cerase (1974) states that the decision to return rests mainly on contextual factors in both the home and host country. According to the transnationalism approach, return is not the end of the migration process but merely a stage (Portes, Guarnizo, and Landolt 1999). Finally, the literature relative to the social network theory (Nohria and Eccles 1992) considers that social capital considerably shapes return migration. Ghosh (2000) offers a conciliatory point of view and argues that return depends on the initial motives of migration, the length of time spent abroad, and the conditions for return. Finally, Cassarino (2004) studies preparedness and willingness to return and points out the importance of resource mobilization in the decision to return.

A BRIEF LITERATURE REVIEW
The literature on return migration is quite abundant and looks at the various outcomes of migrants upon return. De Vreyer, Gubert, and Robilliard (2009) find that return migrants in West African Economic and Monetary Union (WAEMU) countries are better-off than non-migrants. However, having migrated to an Organization for Economic Cooperation and Development (OECD) country results in a negative relative advantage in terms of labor market participation when education is controlled for. In a study on Uganda, Thomas (2008) shows that return migrants with university degrees have a higher probability of being employed compared to non-migrants having the same degree. The migration form has a significant impact on well-being, and Sabates-Wheeler, Natali, and Black's (2007) study on Ghanaian migrants suggests that traveling with formal documents increases the chances of moving out of poverty. Thus, the success of the migration experience depends on the migrant's legal status. In a recent paper, Mezger Kveder and Flahaux (2013) show that Senegalese migrants returning to Dakar get into entrepreneurial activities, but only because entrepreneurship is a "last resort" activity. Thus, entrepreneurial status indicates a failed migration experi-ence, in the sense that migrants have no financial or social capital upon return. For the Tunisian case, Mesnard (2004b) shows that, despite the fact that return migrants do not benefit from a phenomenon of human capital accumulation that would entail positive results on the labor market, their migration allows them to overcome credit constraints and invest in small businesses upon return. In a further study, she shows that the funds received by the migrants aimed at helping them overcome this financial constraint and making them return faster to their home countries might have the opposite effect and thus increase their length of stay in the destination country (Mesnard 2004a). Using the MIREM database, Mahuteau and Tani (2011) find that returning migrants who acquired university diplomas while abroad have a higher probability of becoming wage earners upon return, whereas those that chose vocational training tend to become self-employed. The authors conclude that through this channel, "they contribute to reducing poverty." Along the same lines, McCormick and Wahba (2001) show that the effects of savings from abroad on returnees' entrepreneurship vary with the education level, with a higher impact of savings for the illiterate returnees. For the case of Finland, Saarela and Finn€ as (2009) argue that return migrants have lower odds of employment than non-migrants and the trend is constant over time. They also find a negative impact of migration duration on employment rates of return migrants. Finally, they observe that return migrants experience difficulties in readapting to the Finnish society, which they explain by unobservable characteristics. Duval (2004) shows that frequent return visits generate a positive and significant probability of successful return and reintegration.
This issue of integration has been underestimated in the literature on return migration. Preston and Brown (1993) discuss this point in the context of returning refugees and argue that the environment to which the migrants return is different from the one they left when migrating and thus they need to re-adapt. Indeed, Howard (1974) and, more recently, Tannenbaum (2007) support the idea that integration upon return is more difficult than the initial integration abroad. Along the same lines, Arowolo (2000) offers a discussion on return migration, reintegration problems, and coping strategies. Finally, Heckmann (2006) suggests a framework to analyze integration and distinguishes between structural integration (which involves access to employment, housing, or education) and sociocultural integration (which cover social interactions and the feeling of belonging).
The return migration literature focuses mainly on voluntary return, and the few papers that have looked at forced return migration deal with migrants who return from camps after a forced migration. Van Hear (1995) finds that the massive involuntary return of 300,000 Palestinians has contributed to Jordan's economic recovery, but the integration of the returnees has been very difficult, with many of them experiencing a "downward" mobility. In a study on 300 Salvadoran deportees, Hagan, Eschbach, and Rodriguez (2008) highlight the cost of this forced return, both for the origin and for the destination country. Indeed, on the one hand, deported migrants are often settlers, leaving behind their families. On the other hand, origin households back in Salvador suffer as they lose their main source of income. Combining quantitative and qualitative approaches, Mezger Kveder and Flahaux (2013) take interest into "involuntary return" and conclude that migrants that were forced to return have difficulties integrating the labor market and want to re-migrate. But they do not offer any quantification of the degree of "integration."

Data and Descriptive Statistics
The data we use come from a survey of return migrants in Algeria, Morocco, andTunisia between 2006 and2007, part of the MIREM project. 2 The sampling procedure is based on a geographical stratification in the three countries, with a focus on regions with high return rates. Thus, in Algeria, the survey covers the wilayas of Algiers, Bejaia in Kabylie, and Setif eastwards of the capital and Tlemcen westwards. For Morocco, the survey was conducted in the region of Tadla-Azilal and the coastal regions of Casablanca, Chaouia-Ourdigha, and Rabat-Sal e-Zemmour-Za€ er. In Tunisia, the focus was on the northern governorates of Tunis, Ariana, La Manouba, and Nabeul as well as the governorates of Soussa and Sfax in the center of the country, and of Medenin in the south. In each country, approximately 300 individual interviews were conducted between September 2006 andJanuary 2007. 2 The project, known as the Collective Action to Support the Reintegration of Return Migrants in their Country of Origin, was launched in 2005 and was financed by the European Union and the European University Institute. The results and details for the field surveys can be found on MIREM's website www.mirem.eu.
The whole sample consists of 992 return migrants surveyed in their home countries. Of the 992, 231 (23.29%) were forced or required to return to their home countries, whereas for the rest, return was a voluntary decision. To construct the variable of forced return migration, we only keep those who answered that they were deported, those whose residence permit was not renewed, and those who answered that they were forced to return because they had administrative or fiscal problems. I expanded the sample to these last two categories as a migrant might not admit that he was deported or he could have chosen to leave the country before the authorities had him deported.
As I was interested only in the active population, I dropped those who after return are students, housewives, retired, or "other." The final sample contains 135 forced returnees.
Deported migrants come mainly from Italy (42%), France (28%), and Spain (13%). They are overrepresented in the sense that in the whole sample, 57 percent of migrants return from France, 16 percent from Italy, and only 2 percent from Spain.

Characteristics of Deported Migrants
Forced returnees are on average younger than the other returnees. This seems logical, especially according to the NELM theories, in which migrants return once they have attained their objective in terms of capital accumulation, thus after a significant lapse of time. This is also confirmed by a longer duration of stay. In terms of education acquired before migrating, the forced returnees have more often a secondary education level. Their destination is also more likely to be a southern European country, while voluntary returnees mainly come back from western Europe. I also noticed that most of the deportees had emigrated using a tourist visa (almost 65%), which is a higher proportion compared to the other returnees (30%). The characteristics of deportees during the migration episode reveal a vulnerable status, with most of them involved in seasonal work or being unemployed. Despite a similar remittance behavior between the two types of returnees, the forced returnees pay far less frequent visits to their home country. This fact may be linked to their vulnerable status in terms of legal stay in destination country and to tighter financial constraints. An intuitive consequence is a loosening of links with the origin country and a weakening of the social network back home. The migrants are therefore less prepared to return, which is accentuated by the low frequency of discussions about return among the deported migrants (more than half declared that they had never discussed about their return, while the percentage is only 13 percent among the voluntary returnees). These various elements indicate a lack of preparedness and willingness to leave the destination country which can negatively influence the success of their return. Indeed, I observed that upon return, more than half of the deported were planning to re-migrate in the near future and almost 53 percent considered that their financial situation worsened compared to the situation before having migrated. They are also less likely to be employers or business owners, and more than one-third are unemployed upon return (while only 10% of the voluntary returnees are unemployed).
To capture sociocultural integration, I constructed an index a la Fokkema and de Haas (2011), adapting it for our database and for the return context. Fokkema and de Haas (2011) compute the degree of sociocultural integration of migrants in the destination country by summing up seven indicators: (1) the extent to which migrants have social contacts with native population; (2) the ethnic nature of migrants' circle of friends; (3) the participation in either native or foreign organizations; (4) having a partner born in the receiving country; (5) the fluency in the dominant language of the destination country; (6) the degree of modernization regarding views on gender roles and parent-child relationships; and (7) the ethnic identification. By adding up these various indicators (revalued with a minimum of 0 and a maximum of 1), they obtain an index that increases with a migrant's integration in the host country (Table 1).
Unfortunately, the MIREM survey offers less information and the analysis had to be limited to fewer indicators, especially concerning integration in destination country. The first indicator taken into account when building the sociocultural integration index for the host country was frequency of contacts with native friends. The values range between 0 for no contact with native friends to 2 for frequent and very frequent contacts. The second indicator is marrying a native during migration that takes the value 0 if the answer is "no" and 1 if it is "yes." The relations with society in the host country is the third indicator. The migrants were asked to judge their relations with the host society in general using a scale going from "Very bad terms" to "Very good terms." Furthermore, migrants were asked whether they had any difficulties concerning various aspects such as access to housing, discrimination or racism, finding a job, access to healthcare, or administrative issues. By adding the various negative answers to these questions, the fourth indicator was computed relative to difficulties encountered during migration (the less difficulties the migrant experienced, the more integrated he is considered to be). Finally, each of the indicators was normalized and summed up, thus obtaining a sociocultural integration index, the value of which -ranging from 0 to 4increases with a higher degree of integration. A reintegration indicator was then computed along similar lines. As mentioned by Arowolo (2000) and Preston and Brown (1993), upon returning, migrants, who are no longer the same persons as before having migrated, find themselves in an environment that has evolved since their departure. A first indicator I included is the initial willingness to re-migrate, building on the hypothesis that an individual who wants to re-migrate will put in less effort to integrate and reciprocally a lack of integration could encourage him to re-migrate. The variable takes the value 1 for "probably or seriously considering to go back abroad," 2 for "not considering it right know or doesn't know," and 3 for "certainly not considering re-migrating." The second indicator is being a member of an association or party, taking the value 1 for "yes" and 0 for "no." Being a homeowner is the third indicator as it can be considered as a proxy for the attachment to the home country. Similar to the integration index for the host country, two indicators were added, one relative to the difficulties encountered upon return and one concerning having experienced adjustment problems and code them as to obtain an indicator that decreases with the number of difficulties and adjustment problems encountered (i.e., for those who declare having experienced difficulties in all fields and have serious adjustment problems, the indicator takes the minimum value). Finally, I included an indicator on the declared satisfaction with the return on the basis that individuals who are more satisfied with the return will be more integrated. After normalizing and adding the indicators, I computed a reintegration index whose value increases with a higher integration or what can be interpreted as "feeling of belonging" ( Table 2).
The indicators show that forced return migrants are not necessarily less integrated in the host country. However, they are far less integrated in the home country than the ones who chose to return.
The descriptive statistics might lead to think that the two populations are different and this could explain the differentiated outcomes in terms of socioeconomic integration upon return. I argue that the lower performances on the labor market and the lower sociocultural integration are not sub-sample specific, but are directly correlated with the lack of preparation linked to the unwillingness to return. I use both a linear regression and an IV approach to identify the forced return effect. However, as the validity of the instrument is difficult to assess and the available data do not allow for the construction of other alternative instruments, I complemented the simple regressions with a propensity score matching analysis and sub-sample regressions and find that the differences of outcomes remain significant when the two populations are matched. The following sections detail and present the results of these analyses.

LABOR MARKET INTEGRATION
After having observed that voluntary returnees fare better in terms of labor market outcomes than deported migrants, I tried to capture the effect of forced migration on labor performances. Unfortunately, the survey does not enable us to measure these performances through salary; thus, I focus on the probability of being unemployed, as unemployment is a crucial dimension of social exclusion and vulnerability (Bhalla and Lapeyre 1997). Given that the question of unemployment was asked both at the time of return (based on the interviewee declaration) and at the time of the survey, it is possible to capture the influence over time of the variables of interest.
I first estimated a binary choice model for unemployment just after return, controlling for the pre-migration characteristics such as age, sex, birth place, education level, employment status, migration motive, and the initial migration intention. Concerning the variables relative to the migration episode itself, I controlled for the destination, the professional status abroad, having studied, the migration duration, and the number of migration episodes. To capture the degree of preparedness to return, I controlled for the frequency of contacts with his family in the home country and for the remittance behavior, as well as for having gathered information about the return. I then ran another regression for the probability to be unemployed at the time of the survey, adding post-return variables such as the return intentions, being a member of an association or a political party and the length of time since return. Thus, the first model I estimated is the following: with U i being the probability for the individual i to be unemployed just after return, ForcedReturn i the probability of having been coerced to return, BM mi a vector of m pre-migration individual characteristics, DM ni a vector of n individual characteristics during migration, and AR ri a vector of r individual characteristics after migration. I estimated a similar model for a U i , which is the probability of being unemployed at the time of the survey: The main difference concerns the AR i vector because it includes more characteristics, namely those that had varied, and can be measured, between the time of return and the time of the survey.
As mentioned, forced return might be endogenous with respect to the probability of being unemployed, despite controlling for the pre-migration characteristics. To limit the bias, I used an instrumental approach. The main problem is that the survey does not provide enough information to find an exogenous instrument. The only variable that seems relevant in terms of determining the probability of being forced to return and not determining the socioeconomic integration after return is a variable on the type of documents used to emigrate. Return migrants were asked whether they went abroad using official documents, fake documents, or no documents. I used this variable and recoded it to take the value 1 if the individual emigrated with official documents and 0 if not. The intuition is that those who emigrated with official documents have a lower probability of being forced to return. At the same time, there is little evidence that having emigrated with official documents might impact labor market outcomes after return. One might argue that having the possibility of traveling with official documents is positively correlated with a higher socioeconomic status, thus a lower probability of unemployment, that might persist over time and lower the probability of being unemployed after return. I argue that even if this were the case, this effect is canceled, by controlling for the socioeconomic variables prior to the migration. As both the dependent variable and the endogenous variable are binary, a traditional 2SLS estimation cannot be used and, following Nichols (2011), I used different specifications with seemingly unrelated bivariate models to correct for the endogeneity. This approach implies that the probability of being unemployed is estimated at the same time as the probability of being forced to return, thus purging the coefficient of forced return of any bias. Results are presented in Table 3 and, for the sake of brevity, the non-significant controls are not presented in the table.
I observe that even if forced migration has no significant impact right after the return, it does have a positive effect over time. Therefore, if deported migrants do not find themselves more often without a job, just upon return, they are more prone to being unemployed over time. This result confirms the hypothesis that if the migrants are not willing and ready to return, they will have difficulties in activating and using their networks to find a job. At the same time, their premature return might be seen as a failure not only by themselves, but also by their social network, thus creating a feeling of rejection, as it was often mentioned in qualitative interviews in the sociological literature.
The Durbin-Watson test indicates indeed that an endogeneity correction is needed for the unemployment probability at the time of the survey. The coefficient for forced return is even stronger when I correct for the endogeneity. Along the same lines, I observe that being a member of an association or political party significantly decreases the probability of being unemployed, thus stressing the importance of being well integrated in a social institution. Also, being a recent returnee (defined as having a time since return lower than the country sample mean) increases the probability of being unemployed, indicating that a certain lapse of time is needed for the returnees to put in practice the skills acquired during their migration and become accustomed to their home country's labor market. In terms of controls, I used variables that capture migrants' characteristics before the emigration, during the migration and in the context of return. Concerning the pre-migration variables, as expected, the education level, as well as experience before having migrated (proxied by having worked before migrating), decreases the probability of being unemployed. These results are similar to those of Mezger Kveder and Flahaux (2013) who find persistence in the occupational status. Those that were employed before migrating have a higher probability of being employed after return. The financial situation before the migration (a categorical variable ranking from 1 "Very good" to 5 "Very bad") has no significant impact on the probability of being unemployed after return, but it is positively correlated with the probability of having been deported. This indicates that those who had (or at least who considered themselves as having) a worse financial situation in the origin country are also those who had a more vulnerable legal status in the destination country and thus found themselves to be more prone to deportation. This variable is probably correlated with the other control variables of economic status before the migration, which might explain the lack of significance. It might also strengthen the idea that if the characteristics before (and during) migration might control for a negative selection in the deportation and the unemployment after return, those who were more vulnerable, less endowed, and less connected are more prone to be deported and also have a lower probability to find a job after return (Table 3).
Furthermore, having studied during the migration and having worked as a wage worker before the return have no significant impact on the probability of being unemployed after return, although they are negatively correlated with the probability of being deported. Having been unemployed before returning hinders labor market insertion upon return, suggesting that the unemployment status abroad might have lowered migrants' ability to prepare their return in terms of capital accumulation.
Regarding migration duration, two effects were expected: A longer migration duration might have increased the unemployment probability through a weakening of social ties and a misperception of the origin country's labor market, or it might have decreased unemployment probability as a longer migration duration could be associated with a higher level of capital accumulation and thus a larger choice of activities upon return. The lack of significance of its coefficient indicates that these two effects might offset each other. The number of migration episodes is also insignificant with respect to the unemployment probability, suggesting that circular migrants do not suffer from a relative isolation from the local labor market, nor do they profit from their migration experiences compared to those that only have one migration episode.
The frequency of contacts with the family while abroad (a variable that goes from 1 for "Every day" to 6 for "Never") and the dummy variable indicating whether the migrant has remitted are associated with Notes: The controls include labor market outcomes prior to migration, reasons to migrate, and the intended length of stay in the destination country. ***p < 0.01, **p < 0.05, *p < 0.1. transnationalisms and should capture the attachment of migrants to their origin countries. Intuitively, among return migrants, those who had more frequent contacts with their families and who remitted might be more informed about the local labor market and might benefit more from their connections upon return. Thus, they are likely to have a lower probability of being unemployed. Although the signs of the coefficients confirm this intuition, they are not statistically significant. Also, these variables are potentially endogenous, as looking for a job in the origin country while abroad or planning to start a business upon return might increase the frequency of contacts with the origin household and the probability to remit, while also decreasing the probability of being unemployed after return.
The variable indicating whether the return migrant's main destination country is a European one or not does not appear to be correlated with the probability of being unemployed immediately after return. However, its coefficient is positive and significant in the regression of the unemployment probability at survey time, suggesting a delayed effect. This result might suggest that those who migrated in Europe were often underemployed; thus, they might have acquired fewer skills and benefited less from their migration experience upon return. Finally, the year of the first migration was added as a control to ensure the comparability of migrant cohorts as migration composition of Maghreb countries is rather various across waves, with a very low-skilled early migration, mainly for labor reasons and a more recent migration with a higher education level and a larger variety of migration reasons.

SOCIOCULTURAL INTEGRATION
The psychiatry literature highlighted the importance of sociocultural integration. Fullilove (1996) writes that "human beings have the need and the unavoidable tendency to feel uniquely and intimately related to a place they consider home." I am thus interested in knowing whether migrants who were forced to return integrate as well as those whose decision to return was voluntary. I aimed at capturing this feeling of belonging using the sociocultural integration index from Fokkema and de Haas (2011), which I adapted to a context of return.
Our dependent variable is the sociocultural "reintegration" indicator, RI i , detailed in the section "Forced Return Migrants," and the model I estimated is the following: with RI i being degree of reintegration of the individual i after return, ForcedReturn i the probability of having been coerced to return, BM mi a vector of m pre-migration individual characteristics, DM mi a vector of n individual characteristics during migration, and AR ri a vector of r individual characteristics after migration. Regarding labor market integration, I used an instrumental approach to correct for the endogeneity of return migration, confirmed by the Durbin-Watson test. As sociocultural integration is more likely to be driven by unobservable characteristics which I might not have captured appropriately, I interpret the regression coefficients as correlations and not determinants. Our variable of interest, forced migration, is highly significant in both the linear and instrumented regressions, and its coefficient has the expected negative sign. Therefore, having been deported has not only negative consequences on the structural integration (proxied by labor market outcomes), but it also seems to impede the migrants' assimilation in their home country environments.
Coming from a rural background has a positive effect on the integration degree, which might indicate stronger social ties in the rural areas compared to the urban ones. The financial situation before migration is negatively correlated with our integration variable. The straightforward interpretation would be that the wealthiest individuals are also those who are better integrated in society. However, the financial situation variable is not an objective measure of wealth, as it is based on the declaration of the interviewees' appreciation of their financial situation and its effect disappears in the instrumented regression. The number of migration episodes has the expected negative effect on the integration indicator. Indeed, a high number of migration episodes imply spending less time in the origin country and thus having fewer opportunities to bond with the home environment and to create social ties. Having migrated to Europe seems to hinder sociocultural integration upon return. Given that the main alternative is having migrated to a Gulf or MENA country, I think that the negative correlation is due to the difference in sociocultural norms between Europe and the three Maghreb countries under analysis. Migrants returning from Europe have assimilated, at least to some extent, the prevalent norms in their destination countries, and their reintegration might require slightly more effort than that of migrants to neighboring countries. As expected, having remitted is positively linked to the sociocultural integration degree as remitting allows the migrant to keep a close relationship with his social network and this will further facilitate his return. Finally, having had children after return or having invested in a project improves integration as it creates the incentive to become more involved in the home environment and, at the same time, it creates opportunities to create social bonds (Table 4).

ROBUSTNESS CHECKS Propensity Score Matching
Having been deported might be determined by unobservable characteristics such as lower risk aversion or a lower need to establish social contacts. These unobservable characteristics might also be correlated with a higher probability of being unemployed. I controlled for a maximum of observable characteristics, before and during migration, that might also be correlated with these unobservables. I also tried to instrument for the deportation and thus reduce the bias. I pushed the analysis further and used propensity score matching as an alternative to control for the observable differences in the characteristics of deported and voluntary migrants that might impact their outcomes on the labor market.
I considered the fact of being forced to return as a treatment, and thus compare the probability of being unemployed and the reintegration index of the treated to that of the untreated (voluntary returnees). By computing the average treatment on the treated, I aimed to estimate the unemployment probability (and reintegration index) difference between those who were forced to return and those who had a similar probability of being deported, but who are voluntary returnees. Following Smith and Todd (2005), I matched the two sub-samples using variables that are not affected by the treatment, but influence both the treatment assignment and the outcomes. Relying on background individual characteristics and migration experience, I first predicted the probability of being treated using a probit model. Table 5 presents the results. I thus observe that the probability to be treated (of being deported) decreases with age, the education level, and being a woman. I also used labor market characteristics to match voluntary and forced return migrants (also used in the previous regressions). I observe that having a more stable labor market situation during migration decreases the probability of being deported. Another significant determinant of forced return migration is having had children during the migration period. I supposed that for those who had children, the level of integration in the destination society and the extent of social contacts would be higher, thus lowering the probability of being deported.
Our model satisfies the balancing property, which implies that observations with the same value of the propensity score have the same distribution of observable characteristics irrespective of treatment status. I imposed a common support condition to avoid bad matching (see Figure I for the distribution of propensity scores and the region of common support). The region of common support is (0.00570506, 0.96873294), and 84.1 percent of the total sample is on common support (this percentage is 81.4 for voluntary returnees and 95.6 for forced returnees).
I estimated the effect of the treatment on the probability of being unemployed with the nearest-neighbor and kernel-based matching models, to test the robustness of the ATT coefficients. While the nearest-neighbor approach matches the treated individual to the untreated individual having the closest propensity score, the kernel estimator compares the outcome of each treated individual to the weighted average outcome of a group of nontreated individuals, with this weight being proportional to the individual's closeness to those in the comparison group. Columns (1) and (4) of Table 6 present the average treatment effect on the treated from the propensity score matching estimations. I observe that both estimators are similar. Along the lines of our results from the regressions in the previous section, I find no significant evidence of an effect of having been deported on the probability of being unemployed just after arrival. However, both estimations indicate a highly significant and positive effect of the treatment on the probability of being unemployed at the time of the survey, with the kernel estimator being slightly higher than the nearest-neighbor estimator. Thus, these estimates suggest that the causal effect of being forced to return is an increase in the probability of being unemployed by 14-15.7 percent. In the last panel of Table 6, I combined the propensity score matching model with a difference-in-difference approach, for which I used the probability to be unemployed before migration to compute the first difference. Indeed, with this approach I controlled for time-invariant unobservable characteristics, thus lowering the potential bias of the forced return. For instance, a lower employability related to cognitive capabilities might determine the individual's probability of being unemployed not only before migration, but also before return. It might also be related to an individual's ability to avoid being forced to return. When controlling for initial differences in the unemployment, I find again that forced return has a significant and positive effect on the probability of being unemployed.
The results for the reintegration upon return are also in line with the regression results and consistent between the two matching models. Indeed, both the nearest-neighbor and the kernel-based models indicate a significant and negative impact of being forced to return on the reintegration index, with very similar coefficients. However, as an "integration" index cannot be computed for the period before the migration occurred, I cannot complete the analysis with the difference-in-difference approach.

Inconstant Migrants
As the willingness to return (or lack of willingness) is a central argument in this paper, I wanted to check whether the initial migration intention plays a role in the outcomes upon return. Therefore, I used the survey's question about the initial intentions to migrate (permanently, temporary, or does not know) to identify the migrants who, despite having returned voluntarily, had the intention of staying permanently in the destination countryhereafter called "inconstant migrants." The intuition behind this choice is that these particular migrants may have experienced problems or difficulties that made them change their plans, prompting them to choose to return. In this respect, they might be comparable to the forced return migrants. I thus built a variable for migration status, which takes the value 1 if the migrant returned voluntarily and had initial plans for a temporary migration, the value 2 if the migrant returned voluntarily but his initial plans were to migrate permanently, and the value 3 if the migrant was forced to return. Among the forced returnees, I did not distinguish between those who had intended to migrate permanently or not. As I cannot know what was the intended length of stay for those who intended to migrate temporarily and at what point of this length the deportation took place, I find the distinction to be irrelevant as their plans were disrupted in any case. I thus regressed our new variable of interest on the previous outcomes, namely the probability to be unemployed just after return and at survey time and the reintegration index, and control for before, during, and after migration characteristics ( Table 7). First of all, I observe that, when compared to voluntary return migrants with temporary migration plans, the forced returnees have a significantly higher probability of being unemployed, even immediately after return, whereas this coefficient was insignificant in the previous regression.
The coefficient for what I call "inconstant migrants" is positive, but not statistically significant for the unemployment probability. This result suggests that even if their return was not in line with their initial plans, this did not affect their labor market integration upon return. The findings also imply that lower labor market integration, measured by a higher probability of Propensity Score Distribution and Common Support for Propensity Score Estimation being unemployed, is closely correlated with the deportation itself and thus probably with a lower degree of preparedness for the return migrants. However, the coefficients of forced and "inconstant migrants" are not statistically different. Furthermore, concerning the reintegration index, the results show that both "inconstant migrants" and forced migrants have significantly lower reintegration levels, suggesting that reluctance to return plays a significant role in the adjustment upon return, even for those who return voluntarily. Moreover, in this regression, the coefficient for the forced return is statistically different than for "inconstant migrants." The coefficients of the control variables do not differ significantly from the previous regressions.

CONCLUSION
Return migration is not always the end of the migration process, but it is a turning point in the life of a migrant. Its success depends on fulfilling the migrant's expectations, which is strongly linked to his "willingness" and "preparedness" to return, as highlighted by Cassarino (2004). It is also determined by the phase of the migration cycle at which the individual finds himself at the time of return. Disruptions in the migration cycle resulting in a forced return might thus have serious implications for migrants' socioeconomic well-being. Migrants who were forced to return, either by circumstances or by official institutions, are less prepared to make a living in their home countries and are also faced with a less welcoming social environment as their interrupted migration experience might be seen as a failure. Moreover, they themselves might be more reluctant to adapt to this change as it was forced upon them. Despite the increasing interest Notes: ***p < 0.01, **p < 0.05, *p < 0.1. in this topic from both a policy and a societal point of view, research on forced return remains limited, mainly due to a severe lack of data. Using a data set on return migrants in Maghreb countries, I analyzed the differentiated outcomes in terms of sociocultural and labor mar- ket integration of voluntary and forced return migrants and show that the context of migration is crucial to make return a success. I find that forcing a migrant to return will impede his integration in both structural and sociocultural terms. Forced return migrants have higher probabilities of being unemployed, suggesting lower performances on the job market and fewer opportunities. At the same time, they suffer from a lower sociocultural integration in the home environment which is synonymous with a lack of feeling of belonging. I find this result to be true even for the voluntary returnees who had initially planned to stay abroad forever, suggesting that readjustment to the home country's sociocultural environment is even more difficult when changes appear in the desired migration process. Unfortunately, the data do not allow a more in-depth analysis of other labor market outcomes or different dimensions of sociocultural integration and a more extensive data set would be needed to generalize the results.
The consequence of this difficult socioeconomic integration might be a higher incentive to re-migrate. Given their vulnerable socioeconomic status, returnees will have fewer opportunities to go abroad and may turn to illegal migration. From a public policy perspective, this is a crucial point to be taken into account, in particular in light of the high costs of deportation programs. Due to the political dimension to which these programs are often attached in the destination countries, deportation programs should be carefully evaluated in terms of their global economic efficiency, taking into account both the (short-term) reintegration in their origin country of forced return migrants and the (longer-term) impact of forced return on subsequent (potentially, illegal) re-migration. Moreover, with the economic crisis, return migration significantly increased, and at least a part of this increase can be considered as return migration forced by the circumstances. Further research is thus needed to identify barriers to return migrants' integration in the case where the decision to return was compelled, and to draw public policy recommendations toward both destination and origin countries.