Double Jeopardy: How Refugees Fare in One European Labor Market

Belgium’s history of immigration does not date much further than a hundred years back, which makes it often overlooked as a country of immigration (Lesthaeghe 2000). Yet, over the last decades it has become a permanent country of settlement for diverse migrant groups. Today’s migrants, defined as those born abroad whatever their nationality, account for a high and rising share of the Belgian population: 16.5% in 2017 in comparison with 13.4% in 2009. Luxemburg, Cyprus, Austria, Sweden and Ireland are the only EU Member States that have a higher relative share of migrants (Eurostat 2018).

In the 1950s and 1960s, Belgium set out to attract primarily unskilled workers from Southern European countries and later from Northern African countries and Turkey to work in the mining and heavy industry. Backed by the lenient rules for family reunification at the time, these guest workers stayed, brought their families and settled permanently into Belgian society. Despite the introduction of a formal cap to limit economic migration in 1974, the immigration landscape kept on widening and diversifying in the next decades (Corluy 2014). Labor migration continued, especially of more highly skilled migrant workers. Family migration increased and took over from labor migration as the most important entry channel to Belgium. It was not until 1990 that asylum seekers became an important phenomenon.

Asylum applications peaked particularly in 1993, when 27,000 applications were registered; in 2000 when 47,000 first applications, or 82% of total entries were recorded; in 2011, when 26,000 asylum seekers came to Belgium, corresponding to 22% of total immigration flows; and in 2015, when 39,000 first applications or the equivalent of 35% of all entries into Belgium were registered (see Figure 1). Hence, the asylum seeker inflow in 2015 was much bigger than those seen in 1993 and 2011, but it was actually lower than that of the year 2000. The share of asylum seekers who were granted a residence status was nevertheless higher than in the past.

Figure 1

The 61% positive decision rate in 2015 was up from 47% in 2014 and 29% in 2013 (Office of the commissioner general for refugees and stateless persons 2017). Note that direct resettlement accounted for a very small share of refugee inflows in Belgium.

Between 1996 and 2016, 2,223 refugees were resettled in Belgium (source: https://www.fedasil.be/en/resettlement/history accessed on 14 January 2018).

The peak in asylum applications in 1993 coincided with the fall of the communist regimes and the disappearance of the “Iron Curtain,” as well as the flow of refugees after the breakup of Yugoslavia in the beginning of the 1990s. At the same time, the number of asylum seekers from the Zaire of Mobutu was also at a high level. The peak in 2000 resulted from the period of conflicts in Kosovo, the resurgent struggle in Congo (Democratic Republic of the Congo) and the beginning of the second war in Chechnya starting in 1999 and continuing thereafter. The peak in 2011 was mainly made up by asylum seekers from Afghanistan, Guinea, Iraq and Russia, although the Balkan countries remained strongly represented in the top 10 countries of origin. In 2015, most asylum seekers came from Syria, Iraq and Afghanistan. Together these three countries of origin accounted for 63% of the asylum applications in 2015. Somalia and Russia completed the top five (Office of the commissioner general for refugees and stateless persons, yearly reports).

Applications for asylum in Belgium accounted for 3.1% of the total number in the EU in 2015. With 35 applicants per 10,000 inhabitants, Belgium was eighth in the list of EU host countries taking in the most asylum seekers (Myria 2016). In absolute figures, the leading countries were Germany and Hungary (35% and 14%, respectively, of all applicants), while Hungary, Sweden and Austria lead in terms of applicants per capita (177, 161 and 100 applicants, respectively, per 10,000 inhabitants).

It should nevertheless be noted that Hungary is regarded as a transit country, unlike the other three countries, which constitute the final destination for potential refugees (Burggraeve and Piton 2016).

Belgium is a federal state. Residence in the country and the reception of asylum seekers belongs to federal competences. FEDASIL is the federal agency responsible for managing the reception of asylum seekers during their asylum-seeking procedure. After asylum seekers submit an application to the Immigration Office, FEDASIL allocates them a reception place based on the availability of places that day, the occupancy rate in each reception facility and the individual situation of the asylum seeker (e.g., family situation, health conditions and language proficiency). There is no real dispersal mechanism, but FEDASIL must take into account an even distribution over the municipalities in Belgium when allocating reception places (European Migration Network 2013).

In the reception centers, asylum seekers receive accommodation and meals, clothing and also social, medical and psychological support, a daily allowance (pocket money) as well as access to legal assistance and services such as interpreting and training. During the entire period that the asylum application is being examined, an asylum seeker has the right to stay at the reception structure. The right to reception ends once the procedure for international protection has finished and all possible appeals have failed. Persons who get granted the refugee status can look for their own accommodation in a Belgian municipality of their choosing. After they receive refugee status, asylum seekers tend to move away from less urbanized areas (and the Walloon Region) to big cities like Antwerp, Ghent and Brussels and medium-sized cities in Flanders (Agentschap Binnenlands Bestuur 2017).

Legislation on employment conditions also falls under the competency of the federal government. The implementation of this law, to a large extent, is part of the competency of the regional authorities, which includes among others the granting of work permits to third-country nationals. As a general rule, EU/EEA and Swiss citizens – and their family members – do not require a work permit in order to be employed in Belgium. However, third-country nationals (non-EU/EEA/Swiss citizens) and their family members who still do not have a permanent right of residence typically require a Belgian work permit.

Regarding the more specific issue of asylum seekers, the process of acquiring a work permit has undergone numerous changes since the 1970s, at times involving long waiting periods and preventions from working (see Table 1A in Appendix for an overview). Before April 2003, asylum seekers could work with a work permit B. From April 2003 to May 2007, asylum seekers declared that admissible could work with a work permit C. Such a work permit was valid for 1 year and renewable. Between 2007 and 2010, asylum seekers could no longer work as a result of the 2007 legislative changes, which removed the admissibility phase in the asylum procedure. Therefore, asylum seekers were no longer eligible for a work permit. Since 2010, asylum seekers who fulfilled certain criteria were allowed to work with a work permit C. It concerned asylum seekers who have not yet received a first instance decision on their asylum case within 6 months following the registration of their asylum application. After February 2011 up to December 2014, the same work regime was valid, but during this period asylum seekers needed to contribute to the costs of their accommodation and other material support if employed. This applied to those asylum seekers who resided in a reception center while working: they continued to be eligible both for material support and housing, but were obliged to contribute. In addition, from 2012 onward, in case asylum seekers have secured substantial income from employment, they have to leave the reception center.

Since September 2015, asylum seekers have been able to get into the labor market 4 months after they have registered with the Belgian Immigration Office (Burggraeve and Piton 2016). Following this reform, Belgium is now among the European countries with the shortest period for obtaining a work permit. Only Greece and Sweden have shorter waiting periods, as they allow immediate entry, as well as Austria, Romania and Germany, where workers have to wait 3 months (Fric and Aumayr-Pintar 2016). In Belgium, work permits for asylum seekers are not conditional on a test to make sure that no national or European resident is interested in the vacancy (unlike in Austria and Greece). There are also no limits on the sectors of activity where asylum seekers are allowed to work (unlike in Austria, the UK and Sweden), and asylum seekers are eligible for self-employment (unlike in Germany and the UK), under the condition that they apply for a professional card (Fric and Aumayr-Pintar 2016).

Given the uncertain residence for the asylum applicant, it is not allowed that the self-employed activity requires large investments.

Asylum seekers are entitled to unemployment benefits under the same terms as Belgian nationals. In practice, however, asylum seekers are unable to avail themselves of these benefits since a work history (or at least a minimum qualifying period) is required, which, of course, they cannot meet (Mussche et al. 2013). In contrast to unemployment benefits, social assistance – noncontributory guaranteed minimum resources – does not have any requirements such as waiting periods and compulsory payment of contributions. People can apply for the subsidiary minimum income if they cannot claim other social insurance benefits. Eligibility is regulated by a means test for the household, according to the de facto living arrangement. The amount of the benefit varies according to the household type. The social assistance benefit amounts are rather low by international standards, especially for single persons and couples with children. As a result, social assistance beneficiaries in all household types live with an income that is lower than the relative poverty line (Van Mechelen and Marchal 2013). Furthermore, non-take up rates are substantial (Bouckaert and Schokkaert 2011), presumably because of information costs (i.e., the cost for acquiring information about eligibility), process costs (i.e., time and effort related to the claiming process) and psychological and social costs (e.g., stigma) (see Janssens and Van Mechelen 2017 for a discussion).

Compared with other European countries, Belgium does show higher “nominal” flexibility when it comes to labor market access for refugees and asylum seekers. However, accessing the labor market is still quite far from actually getting a job (see, e.g., Baert et al. 2015; Baert and Vujić 2016 for evidence on ethnic discrimination in the Belgian labor market). Moreover, while Belgian social security is in principle an inclusive one, it is a strongly work-based system meaning that a steady link to the labor market is required in order to build up social security entitlements.

There is only a limited body of research looking at how refugees are faring in the European labor markets. An important reason for the limited evidence is the data limitations, since detailed statistical information on migration motive is not always easily accessible (Bevelander 2016). To overcome this gap in the current literature, this paper uses a longitudinal dataset linking respondents’ information from the BLFS to their administrative records Belgian Labour Market and Social Protection (LM&SP) data warehouse in order to analyze the refugees’ transition through the Belgian labor market.

The data matching between the BLFS and the LM&SP was an exact one, in the sense that the national register numbers have been used to link the individuals’ information in both datasets.

The BLFS is a representative sample from the National Register and provides, in addition to demographic characteristics, both general and more detailed data on the employment situation, such as the quality of employment and characteristics of the workplace. An added value of the BLFS versus Belgian administrative data is that the highest educational level is available, which is absolutely indispensable for research on labor market integration. The sample is also large enough to distinguish between broad categories of migrants (Corluy and Verbist 2014). However, the BLFS is a cross-sectional database, without a longitudinal dimension. Ideally, longitudinal data are needed to evaluate a potential labor market integration process. Next to its limitations in terms of longitudinal analysis, the BLFS in principle does not allow for an identification of refugees.

Exceptions are the special ad hoc modules of BLFS 2008 and 2014, containing information on migrants’ self-declared reason for coming to Belgium.

First, the LM&SP includes information on the initially issued residence permit of migrants, making it possible to categorize migrants by their legal means of access for stay or residence in Belgium. Here, we define as “refugees” migrants who obtained their first residence permit for reasons of asylum and international protection. An advantage of our data is that we can trace the socioeconomic positions of refugees and other migrants based on their self-reported year of arrival in Belgium (calculated from the years since migration variable in the BLFS).

This is in contrast to other studies that use the date on which the residence permit was obtained as a starting point (see, e.g., Bakker et al. 2016; Bratsberg et al. 2017; Ruiz and Vargas-Silva 2017; Schultz-Nielsen 2017).

Second, on the basis of the data from the participating social security institution, detailed quarterly information on the socioeconomic position is available. Depending on the position in or out of the labor market, the population is divided into: (1) employment, (2) unemployment, (3) social assistance, (4) other social insurance and (5) other unspecified. Note that the situation taken into account is invariably the situation on the last day of the quarter, giving priority to work over benefits. Employment covers self-employment, work in the regular job market and subsidized work and all possible combinations of these. Unemployment covers job seekers with a waiting allowance, an unemployment benefit or a guidance allowance. Social assistance beneficiaries are all those receiving a living allowance or other forms of financial support. Other social insurance encompasses full-time career break, incapacitation from employment and entitlement to allocation for disabled persons. The state “other unspecified” includes all persons who did not fit in any of the previous categories and thus are unknown to the participating social security institutions.

The following heterogeneous and non-exhaustive list of situations serves as an example: education and training, persons in full-time family care, cross-border workers, persons working for supra-national institutions, and persons who are known to the participating social security institutions in the course of the quarter but not on the last day of the quarter.

With these linked survey and administrative data we have, compared with other related literature, a uniquely rich longitudinal database in order to study the labor market trajectories of refugees, with different comparison groups: family migrants and labor migrants.

In order to have a large enough sample size, we pool together all quarterly BLFS samples (from Q1 2008 to Q4 2012) that were previously linked with administrative data. Data availability restricts the information on the socioeconomic position from the LM&SP to the period between 1999 and 2012.

There is a small break in the data because information on social assistance, child allowances, incapacity-to-work and invalidity benefits, occupational diseases and some pensions have only been integrated since 2003 in the LM&SP. So for the quarters before 2003 Q1, these categories are registered as “other unspecified”.

From our resulting sample, we select refugees (N = 401), family migrants (N = 4,383) and labor migrants (N = 3,387). Originally, we had 4,222 labor migrants in our data set. However, a large share of labor migrants was unknown to Belgian social security because they work for supranational institutions (e.g., EU officials, personnel of the North Atlantic Treaty Organization, personnel of the Supreme Headquarters Allied Powers Europe). Including these labor migrants in the analysis would lead to an underestimation of their “genuine” employment levels. In our dataset, we were able to single out these labor migrants (N = 835) and exclude them from the analysis.

In order to identify the labor migrants who are employed in supranational institutions, we first select the labor migrants who – according to the information in the LM&SP – are in the “other unspecified” position at the time of the BLFS. Subsequently, we look at their employment status in the BLFS based on the International Labour Organization (ILO) definition (i.e., completed at least 1 h of work in the period being measured). We find that of the 1,131 labor migrants in the “other unspecified” state at the time of the survey, 835 (74%) are employed according to the ILO definition. Presumably, these labor migrants are working at supranational institutions and therefore unknown to Belgian social security institutions.

For consistency with the existing economic assimilation literature, we also include a native-born comparison group in the initial stage of the analysis. The native sample is a 10% random sample extract of all individuals who are born in Belgium and who aged between 18 and 65 during our observation period (N = 13,267).

To answer the first research question of the paper, we present the quarterly distribution of socioeconomic positions by time since arrival for refugees, family migrants and labor migrants separately. We choose to display the distributions by time since arrival and not by calendar year because we have different arrival cohorts – in the year 2007, the socioeconomic position of those who arrived in 1999 and had therefore been in Belgium for 8 years does not mean the same as for those who arrived in 2006 and had therefore been in Belgium for only 1 year. Showing the results by time since arrival, therefore, is a more accurate way of presenting the results. We also present the quarterly distribution of socioeconomic positions for the native-born, to get a sense of the assimilation profile by migration motive. For natives, we display the distribution by calendar year.

Subsequently, with the second research question we aim to gain more insights in the labor market transitions that are behind the socioeconomic careers of refugees and other migrants. As the main focus is on transitions, the multivariate analysis is based on discrete-time event history techniques (Blossfeld and Rohwer 1995). The hazard rates of entering the first employment spell and of exiting the first employment spell are the two main dependent variables.

Although both transitions are continuous processes, we estimate discrete-time models because analysis time is measured in quarters (Jenkins 2007).

For the entry into first employment spell, duration in quarters elapsed since arrival is used as the exposure dimension in the analysis. In the analysis of first employment exit hazards, exposure is equal to the duration in quarters since employment entry. A logarithmic baseline hazard function is included for all models, meaning that hazards increase at earlier durations and decrease at higher exposures. All models use a logit link function, allowing for an interpretation of the exponentiated parameter estimates, exp(b), in terms of odds ratios.

In the models presented, we compare refugees and family migrants with labor migrants. Also included in our multivariate hazard models is a set of covariates. Sex and age are two obvious candidates. Hence, we consider age at arrival as a linear term, and we also include a quadratic term to account for possible nonlinearities in the effect. Additionally, we expect that (lack of) human capital is a very important determinant of labor market integration. The highest level of education is included, and divided here into three categories: low (ISCED 0 through 2), medium (ISCED 3 and 4) and high (ISCED 5 and 6). Since the Belgian regions differ considerably in terms of economic situation and thus in terms of employment prospects, geographic spread of migrants may provide an additional explanation for differences in labor market participation. Hence, region of residence (Flanders, Brussels, Wallonia) is added. In order to assess the influence of region of origin, migrants are sorted into six categories based on country of birth: EU Member States, Europe non-EU, Turkey and North Africa, sub-Saharan Africa, Asia and the USA. Lastly, we control for the year of arrival (10 dummies) and the year in which migrants were interviewed (4 dummies). All covariates are measured in a time-invariant way, but we do interact our migration motive dummies with elapsed duration to test whether differences between refugees and other migrants change over time spent in Belgium.

Descriptive statistics, reported in Table 1, indicate that refugees and labor migrants are pre-dominantly male, whereas family migrants are predominantly female. Of the three migrant groups, refugees arrive at the youngest age with a mean age of 29 at arrival. For family migrants, the age is 30 and for labor migrants it is 31. As expected based on the theory of self-selection (see Section 2), there exists considerable variation in terms of education: nearly 52% of refugees are low educated compared with 45% of family migrants and 29% of labor migrants. Refugees are also less frequently highly educated (only 19% is highly educated). So while labor migrants are similar to the native-born regarding their educational level, family migrants and refugees are clearly lower educated. With regard to geographic spread, we observe that migrants are clearly overrepresented in the Brussels region compared with other regions. However this is less pronounced for refugees who, in comparison with other migrant groups, more often reside in Flanders and less often in Brussels. This seems to indicate that refugees also take economic factors into account when they decide on where to settle.

There are also clear differences in terms of the dominant region of origin. Almost every refugee in our sample originates from a country outside the EU. The most common regions of origin are sub-Saharan Africa (33%), Europe non-EU (31%) and Asia (27%). In clear contrast, the large majority of labor migrants come from EU Member States (76%). An important share of family migrants comes from Turkey and North Africa (39%). The other dominant regions of origin are EU Member States (26%), Asia (12%) and sub-Saharan Africa (10%). There is some variation in the year of arrival between refugees and other migrants, but it is hard to draw any conclusions from this descriptive analysis. Finally, family and labor migrants are equally represented in the different survey years, whereas refugees are slightly overrepresented in the 2012 survey.

Before we turn to our hazard models, we want to get a first impression of the socioeconomic careers of refugees. This can be gained by showing quarterly distribution plots, where the proportion in certain socioeconomic positions is displayed by time since arrival in Belgium. We also include a quarterly distribution plot of the native-born in Figure 1A in Appendix, to get a sense of the assimilation profile by migration motive.

Figure 2 shows the quarterly distribution plots for refugees. Refugees tend to follow a particular trajectory, which is roughly as follows. In the initial years after arrival, the large majority of refugees have no connection with Belgian social security whatsoever. After entry, and hardly surprising, very few refugees are to be found in the formal labor market. Similarly, very few are dependent on social transfers. But after this initial phase follows a very quick and strong rise in dependency on social assistance, the only social protection program to which refugees can have relatively rapid access in the largely contributions-based Belgian social protection system. In the third year since arrival, the share of social assistance beneficiaries reaches a peak, just below 50%. Entry into employment remains very low during the initial years after entry. After what we could label a “transition phase”, which takes about 4 years, dependency on social assistance starts to drop and employment participation surges. After 7 years, 45% works legally (employed or self-employed). Sometime into this second phase, we also see an important fraction transitioning into unemployment benefits (19% at the end of the 10th year). Yet at the same time, an important fraction remains dependent on social assistance (19%) or stays unknown to Belgian social security institutions (12%). Figure 1A in Appendix shows that, during the same observation period, the native employment level stood at 77%. Consequently, after 10 years of residence, a substantial employment gap of more than 30% points between refugees and the native-born remains.

Figure 2

Like refugees, very few family migrants are employed in the formal labor market after entry (see Figure 3). However, family migrants are less dependent on social assistance transfers compared with their refugee counterparts, because social assistance is means-tested at a household level and family migrants usually have a partner who works and are therefore in a less precarious position. In the former group, the share of social assistance beneficiaries gradually increases to around 6% after the third year since arrival and remains quite stable thereafter. Employment levels also seem to increase more rapidly, although they converge toward a relatively low level of 44% after 10 years. Like refugees, family migrants have a very low unemployment rate in the initial years of residence, which is what we would expect considering that both groups lack a stable job history that is required to be entitled to unemployment benefits. However, for family migrants – like refugees – the unemployment rate increases steadily, peaking at a level of 10% after 10 years. What clearly sets family migrants apart is their large share of individuals in the status “other unspecified,” representing persons who are unknown to Belgian social security. About 39% of family migrants remain in this state after 10 years of residence. We can conclude that the employment level of family migrants stays well below that of the native-born over the entire observation period.

Figure 3

As expected, labor migrants clearly have the fastest labor market attachment, with an employment rate of 47% at the end of the first year, 67% at the end of the second year and 74% from the third year onward (Figure 4). After 3 years since arrival, the employment level of labor migrants is thus almost on par with the employment level of the native-born. During the initial years after arrival, labor migrants do have the highest unemployment attachment of the three comparison groups, mirroring their faster pace of employment entry. This picture changes somewhat thereafter, as their unemployment rate of 10% is equal to the unemployment level of family migrants (10%) and way below the level of refugees (19%) after 10 years of residence.

Figure 4

Lastly, social assistance take-up is only marginal among labor migrants (below 3%), compared with the dependency levels of family migrants and – in particular – refugees.

Based on these descriptive results, we can conclude that refugees indeed take significantly longer to get a foothold in the Belgian labor market compared with labor and family migrants. Due to their weak ties to the labor market in the initial years of residence, refugees are particularly vulnerable to recourse to social assistance. Over time, they do catch up to some extent and the employment gap with labor and family migrants decreases. However, once refugees get into the labor market and build up social security entitlements, they also seem to experience a greater risk of ending up in the unemployment benefit scheme. The questions following up on this are whether these differences persist and whether they are statistically significant upon controlling for important covariates. To answer these questions, in the next section, we use discrete-time hazard models to study the effect of migration motive on the duration until first employment entry. Subsequently, we analyze the risk of exiting the first employment spell to unemployment, social assistance and a heterogeneous “other” state in a competing risks framework.

Column 1 of Table 2 displays the odds ratios of the logit model for the probability of entering the first employment spell, while controlling for important covariates (an odds ratio greater than 1 indicates a positive outcome). The main result is that refugees and family migrants enter employment at a slower pace than labor migrants (the reference category), controlling for background characteristics. The disadvantage is slightly larger for refugees than for family migrants. However, as outlined earlier, differences might change over time. To test our expectations on the time dependency of relative risk of the three comparison groups, two time-varying covariates are included in column 2 of Table 2: elapsed time for family migrants (Family*time) and elapsed time for refugees (Refugee*time), where time is included in a linear fashion. Controlling for these interaction variables, the main effects observed for family or refugee migrant groups can now be interpreted as the group disadvantage (relative to the labor migrants) immediately after arrival. The odds ratio of the respective interaction variable then indicates whether the disadvantage remains constant (»1), decreases (<1) or increases (>1) over time spent in Belgium.

The odds ratio of the main refugee dummy stands at 0.14. This effect now refers to the net disadvantage of refugees relative to labor migrants concerning the likelihood of entering employment immediately after arrival. The odds ratio of the interaction effect (1.09) indicates that the initial disadvantage of refugees declines as the time spent in Belgium increases. Family migrants also experience a large net disadvantage (0.33) over labor migrants, which decreases with time spent in Belgium (as shown by the odds ratio of the interaction term: 1.02). These results confirm that, ceteris paribus, the initial disadvantage concerning labor market entry is more substantial for refugees than for family migrants, but that refugees converge toward labor migrants at a quicker pace than family migrants. This is consistent with studies looking at the labor market integration process of refugees in other European countries (Bakker et al. 2016; Bratsberg et al. 2017; Ruiz and Vargas-Silva 2018; Sarvimäki 2017; Schultz-Nielsen 2017).

Above and beyond the effects of migration motive, female migrants enter employment less smoothly than male migrants do. The impact of age at arrival is curvilinear, with the likelihood to enter the first employment spell increasing in the initial years but flattening out later on. The level of education is clearly important for labor market entry, though surprisingly, highly educated migrants do not hold any advantage over medium-educated migrants in entering the labor market. One possible explanation for this finding is that highly educated individuals try to get their home country credentials recognized before they enter the labor market. Despite some improvement on delays in degree recognition, this process remains very burdensome in Belgium (De Keyser et al. 2012). We also find that migrants residing in Wallonia have more difficulties in entering the labor market than migrants residing in Flanders and Brussels. Region of origin is clearly an important factor. In comparison with migrants from EU Member States, Asian, European non-EU, Turkish, North African, and sub-Saharan African migrants are significantly slower in entering employment, in that order. Migrants from the America do not differ significantly from the EU migrants when it comes to the speed of labor market entry, which is explained by the fact that a large number of Americans are labor migrants.

So far, our analyses show that refugees experience a net disadvantage in entering the first employment spell over family and labor migrants and that this disadvantage lessens over time since arrival. In this section, we look at the duration in the first employment spell to see when refugees and other migrants exit this first employment spell. We distinguish three types of exit, namely unemployment, social assistance and a heterogeneous “other” state (which includes both transitions to “other social insurance” and “other unspecified”).

The results of the multivariate hazard models are included in Table 3. Column 1 of the table displays the odds ratios of a logit model for the hazard of first employment exit, without considering different types of exit. The results indicate that, ceteris paribus, refugees run a higher risk of exiting their first employment spell compared with labor migrants. In columns 2–4, we discern between the different types of exit by means of a multinomial logit model (no exit is the base category). The results show that the gap in exit rates between refugees and labor migrants is primarily due to refugees’ higher exit probabilities to unemployment and to social assistance (which in the case of refugees often means reentering social assistance). The probability to exit employment to the heterogeneous other state, on the other hand, is higher for labor migrants than for refugees. Family migrants also have a higher overall exit probability than labor migrants, which is mainly caused by their higher likelihood to transition into unemployment and into the “other” state.

The result that refugees have a high likelihood of reentering social assistance corresponds to the findings of Carpentier et al. (2017). This study for Belgium shows that asylum seekers are particularly likely to return to social assistance, which is partly due to the fact that they exit social assistance mainly through a path that entails a high risk of reentry, namely paid employment. Exit to active labor market programs and exit to social insurance benefit receipt, which lead to longer-term exits, are nearly nonexistent for the group of asylum seekers. The authors refer here to institutional mechanisms (asylum seekers fall under the “Right to Social Assistance” act, which puts less emphasis on labor market integration), eligibly criteria (asylum seekers might not satisfy the eligibility criteria, i.e., an uninterrupted period of nonemployment, for certain active labor market programs) and discrimination as possible explanatory factors.

Turning to other covariate effects, results show that male migrants have a lower risk of exiting their first employment spell to unemployment and “other” compared with their female counterparts. Age at arrival is negatively correlated to exit to the “other” state, though at a decreasing rate. The direction of the education effect is as expected: the higher the level of education, the lower the risk of exiting the first employment spell. In fact, medium and highly educated individuals are less likely to experience any types of exit, compared with low educated individuals. We also find that migrants residing in Flanders are more likely to exit to the “other” state while they are less likely to exit to social assistance, compared with migrants living in Brussels. Migrants living in Wallonia, on the other hand, are more likely to exit to unemployment and social assistance. Finally, the pattern of employment exit differs largely between origin groups. Origin groups that are particularly vulnerable with regard to first employment exit are sub-Saharan Africans, Asians, Turks, North Africans and Europeans from outside the EU. For the two former groups, this is especially reflected in their higher probability to exit to the social assistance scheme.

In this section, we discuss some extensions to the main model of first employment entry in order to highlight the robustness of our results. The results of our robustness checks are reported in Table 2A in Appendix.

First, we use an alternative approach to identify refugees, family migrants and labor migrants in the dataset. For this, we limit our sample to the respondents of the ad hoc module of BLFS 2008 (Q2), which contains a question on migrants’ self-reported primary reason for migration.

This resulted in a sample of 64 refugees, 250 family migrants and 139 labor migrants. Note that in this exercise, we track migrants from 1999 to 2006 entry cohort instead of the 1999 to 2009 entry cohort.

Second, we reestimate the model of first employment entry limiting the sample to migrants who come from non-EU countries. This might be important as EU migrants have more extensive legal rights of residence and work than non-EU citizens in Belgium. They also tend to do better on the labor market and experience less difficulties in transferring human capital acquired in the home country to the labor market (Corluy and Verbist 2014). Column 2 of Table 2A in Appendix shows that it is still the case that refugees are worse off than labor and family migrants when it comes to the labor market entry of non-EU migrants.

Third, we restrict the analyses to only those migrants who obtained their highest level of education in the home country (see Table 2A in Appendix, column 3). We do this because the highest level of education is measured at the time of the BLFS and not at the time of arrival. Unfortunately, the design of the BLFS questionnaire does not allow a complete reconstruction of education histories. Combining the questions “What is the highest level of education you have successfully completed?” and “In what year did you reach that level?” does not provide sufficient information to evaluate whether an individual was continuously enrolled in education up to the date of the highest educational attainment. Here, we select only those migrants who obtained their highest level of education in their home country to see whether this changes our main results. Our exercise shows that restricting the sample to migrants who obtained their highest level of education in their home country does not change our outcomes.

Twenty-one refugees, 209 family migrants and 270 labor migrants obtained their highest level of education in Belgium and are dropped from this exercise.

As a final exercise, we drop the migrants who entered before 2003 from our sample. We do this because the LM&SP started registering information on social assistance, child allowances, incapacity-to-work and invalidity benefits, occupational diseases and some pensions from the first quarter of 2003 onward, and we want to test whether this impacts our results. Column 4 in Table 2A in Appendix shows that our results remain robust.

As mentioned earlier, several studies show that refugees have poor health outcomes and that this has important implications for their labor market success (Bakker et al. 2014; Giuntella et al. 2018). In fact, a study for Norway finds that refugees experience a substantial increase in their participation in disability programs over time, suggesting that poor and deteriorating health is an important driver behind their labor market exit (Bratsberg et al. 2014). Based on information from the social security register data, we explore whether something similar is happening in the Belgian labor market. Figure 2A in Appendix plots the share of persons who are receiving invalidity and incapacity-to-work benefits by calendar year for refugees, labor migrants, family migrants and the native-born, respectively. Results show that the share of migrants receiving such benefits is very low in the beginning of the observation window. The share gradually increases over time, but stays well below the level of the native-born at the end of our observation window (around 2% for labor and family migrants versus 4% for natives). Refugees do seem more likely to receive invalidity and incapacity-to-work benefits compared with other migrants – by the end of 2012, their share in work incapacity is slightly higher than the share for natives.

In light of these findings, we reestimate our discrete-event hazard model of first employment exit, also considering the state “work incapacity” (which was previously included in the “other” exit state). Results (data not shown, but available on request) show that, controlling for individual characteristics and duration in the first employment spell, refugees are not more likely to exit from their first employment into work incapacity compared with other migrants. One possible explanation for the difference in results with Bratsberg et al. (2014) is that our observation window is shorter.

Another possible explanation for the relatively poor labor market performance of refugees – and specifically for their higher risk of employment exit – is that they get discouraged in the types of jobs that they are able to acquire. Indeed, several studies show that refugees experience more difficulties than other migrants and natives in finding decent and well-paid jobs (Bakker et al. 2016; Bratsberg et al. 2017; Bevelander and Pendakur 2014; Lens et al. 2018; Ruiz and Vargas-Silva 2018; Sarvimäki 2017; Zwysen 2018). Unfortunately, our social security data do not include information on employment characteristics, so we cannot directly test how features of the first employment spell influence the timing until exit. However, we do have information on the characteristics of those employed at the time of the Labor Force Survey, which allows us to get some sense of the differences in job quality among employed refugees, other migrants and natives. Table 3A in Appendix shows the coefficients obtained for migrants – by migration motive – relative to natives from a series of regressions on factors predicting quality of employment. The dependent variables are, in turn, employment in a low-skilled job

This dummy variable equals 1 if a person is employed in an elementary occupation (ISCO code 9), and 0 otherwise.

The results indicate that, ceteris paribus, migrants are more likely to be employed in a low-quality job than natives. Particularly refugees and family immigrants who do work tend to do so in certain occupations and in jobs which are below their skill levels. They are also much more often to be found in involuntary part-time and temporary contracts and low-wage employment. These findings suggests that, even if these newcomers manage to find jobs quickly, the quality of the jobs they get into puts them at high risk of falling victim to cuts and redundancies (see, e.g., Bratsberg et al. 2018a for a discussion). At the same time, poor employment prospects may lead to a small overall utility difference between the states of employment and nonemployment. Consequently, this might lower refugees and family migrants’ social insurance replacement ratios encouraging some individuals who are (or could have been) capable of supporting themselves through employment instead to rely on social insurance benefits (see, e.g., Bratsberg et al. 2018b for a discussion).

Our analysis has some important limitations. First, our data lack information on knowledge of (one of the) local language(s), which is a key factor differentiating refugees from other migrants (Cortes 2004; Aydemir; 2009; Auer 2018).

Second, our data do not include residential information below the regional level (which in our case means Flanders, Wallonia and Brussels). These are too highly aggregated regional levels, containing both urban and rural areas, to provide any useful detailed information on the role of social networks. However, there is substantial evidence in the literature on the importance of networks for refugees’ labor market entry (Dagnelie et al. 2019). Neither could we take into account the effect of changing local labor market conditions, which play an important role in the employment assimilation process of immigrants (see, e.g., Barth et al. 2004; Dustmann et al. 2010). In fact, one study even showed that refugee employment is characterized by greater business cycle sensitivity than employment of other migrants (Bratsberg et al. 2018a,b). Given that our observation window entails the global economic crisis, which has hit immigrants in Belgium harder than natives (Pina et al. 2015), not accounting for unemployment effects is an important limitation of our study.

Third, we do not pick up employment in the informal market, which is an important source of employment for migrants in Belgium (Geets et al. 2007). Additionally, the information in the LM&SP also does not capture participation in training and education, which is an important drawback since the migrants in our sample are still fairly young and Belgium offers good opportunities here.

Fourth, we do not possess information on when migrants exactly received their residence permit (or in the case of asylum seekers, got recognized as a refugee). Our sample includes both migrants who have not yet completed the recognition process and migrants who have, and we cannot observe when the transition between the two states takes place.

Finally, we recognize the limitation that the first issued residence permit might not completely reflect the true motivations and aspirations of the individuals when they first migrate, as sometimes those motives are hard to fit into rigid administrative categories. Nonetheless, we believe that the first issued residence permit still provides a relevant aspect when looking at outcomes later on.