Hohmeyer and Lietzmann (2020) attribute the likelihood of welfare receipt to two mechanisms. For one matter, longer duration of welfare receipt can increase the likelihood of future benefit receipt. This is because welfare receipt decreases employment chances, both by sending a negative signal to potential employers (Blank, 1989; Lockwood, 1991) and by leading to human capital deterioration during extended periods of inactivity (Gregory and Jukes, 2001). Negative consequences for mental health and motivation can also contribute to a decrease in employment chances (Jahoda, 1982).

In addition, individuals with good prospects of employment are less likely to receive benefits. Many factors contribute to employment prospects, including job skills, the availability of employment support systems such as transportation and childcare, and mental health factors (Blumenberg, 2002). Finally, one major predictor of employment prospects is human capital (Mincer, 1958, 1991; Nickell, 1979). Human capital theory anticipates better labor market outcomes among those possessing greater skills (Becker, 1964). Primarily through education, individuals increase their skills and therefore their likelihood of employment and higher wages (Nickell, 1979; Mincer, 1991; Cairó and Cajner, 2018). Other forms of human capital include work experience (Mincer and Polachek, 1974) and host-country language skills (Chiswick and Miller, 2003).

Immigrants in Germany display heterogeneity in terms of human capital. While education levels are rising across countries of origin, immigrants from the EU tend to demonstrate higher education levels than other immigrants (Kogan, 2011; Gries et al., 2021). Conversely, immigrants from non-Western countries tend to demonstrate the greatest educational disadvantages (Brücker et al., 2016a; Spörlein et al., 2020). These differences may be due in part to education systems in immigrant-sending countries. Not all immigrant-sending countries rely on professional training and educational qualifications to the same degree as Germany (Souto-Otero and Villalba-Garcia, 2015). In many third countries, formal vocational training is much less common than in Germany and much fewer occupations require formalized training (Stoewe, 2017). For this reason, certain countries may be more likely to send more skilled immigrants than others because of overall education levels in the country.

Selection mechanisms also contribute to systematic variation in human capital between immigrants in different visa categories (Aydemir, 2011). Immigrants who migrate for work purposes are positively selected based on the skills and qualifications desired by employers (Kontos, 2011; Boeri et al., 2012). As such, immigration processes are easier for those possessing certain qualifications and skills, and there are even particular residence permits to enable highly skilled workers to more easily migrate to Germany (Cerna, 2013; Ellermann, 2020). Comparatively, refugees are less positively selected on human capital as they migrate due to humanitarian reasons (Chiswick, 1999; Dustmann et al., 2017; Brell et al., 2020). They therefore face labor market disadvantages compared to economic immigrants (Aydemir, 2011; Bevelander, 2011; Salikutluk et al., 2016). This may be one explanation as to why refugees in Germany demonstrate lower average education levels compared to other immigrants (Brücker et al., 2016a).

Although substantial literature confirms that human capital is a significant predictor of welfare receipt, there is also a large body of research indicating differences in returns to human capital for immigrants from different countries of origin. A key question is therefore whether human capital predicts welfare receipt to the same extent for all immigrants. There are several reasons why immigrants' human capital assets may bring them limited employment returns, including devaluation of origin-country human capital, skill mismatches, discrimination, and return migration.

The human capital literature distinguishes between general human capital and country-specific human capital (Chiswick, 1978; Borjas, 1985; Friedberg, 2000). Significant research has established that origin-country human capital is not valued in the labor market as much as destination-country human capital (Schoeni, 1997; Friedberg, 2000; Bratsberg et al., 2002). Furthermore, immigrants' origin country matters: non-Western degrees or degrees from countries with lower levels of economic development are associated with lower wages compared to degrees from Western countries (Lancee and Bol, 2017). Friedberg (2000) finds that returns to education are higher for immigrants from Europe and the Western Hemisphere than for immigrants from Asia and Africa. A recent analysis from Basilio et al. (2017) indicates that the native-immigrant earnings gap in Germany can be mostly explained by heterogeneity in returns to human capital by country, with higher returns for immigrants from high-income countries.

Origin-country human capital is devalued both in formal and informal ways. For one matter, foreign degrees must be recognized by the government's Central Office for Foreign Education in order to enter many professions. Degrees are recognized based on evaluations of quality and comparability to the German education system. Non-Western educational and professional qualifications are less likely to be granted recognition (Bauder, 2005; Brussig et al., 2013; Damelang and Abraham, 2016). Furthermore, legal mechanisms such as the EU Recognition Directive (2005) institute recognition of qualifications from other EU member states, the European Economic Area (EEA), and Switzerland. Immigrants from non-Western and non-European countries may therefore suffer in the labor market due to lack of formal recognition of their qualifications. Evidence indicates that degree recognition results in positive labor market outcomes in terms of employment and earnings (Brücker et al., 2021).

In addition to formal recognition, educational credentials from different countries may be valued differently by potential employers. These valuations can be based on actual or perceived quality of credentials. Immigrants from developed countries, which can devote more government resources to education, may enjoy a better quality of home country education, leading to greater cognitive skills and consequently more job market success (Altinok et al., 2018). However, Lancee & Bol provide an alternate explanation, suggesting that employers perceive non-Western degrees and credentials to be of lower value, regardless of their actual quality. They refer to signaling theory to explain this valuation process. As employers do not have complete information about job seekers' skills and productivity, they use degrees and credentials as signals or cues on which to base hiring decisions (Spence, 1973; Weiss, 1983). In this manner, employers' notions of skill and qualification replicate historical inequalities between Western and non-Western countries (Nowicka, 2014). Lancee and Bol find that the lower wages of non-Western workers can be explained at least in part by the lower signaling value of non-Western degrees. Lack of formal and informal recognition of educational credentials could lead to challenges finding employment or earning a sufficient income, increasing likelihood of welfare receipt. Because of these differences in the transferability of qualifications, I hypothesize a greater negative effect of educational attainment on welfare receipt for EU immigrants than for third country immigrants and refugees (H1).

Immigrants experience not only devaluation of education from the country of origin, but also additional labor market disadvantages that limit their returns to education. These factors may contribute to a gap in welfare receipt between immigrant groups left unexplained by human capital and sociodemographic factors. For one matter, research on labor market discrimination shows that third country immigrants experience greater discrimination based on ethnicity compared to EU immigrants (Rydgren, 2004; Constant and Massey, 2005; Kaas and Manger, 2012; Thijssen et al., 2021).

Differences between origin-country and host-country education systems and labor markets can lead to skill mismatches. Some studies have indicated that labor market returns to human capital are positively related to the sending country's GDP (Sullivan, 2010; Lagakos et al., 2012; Coulombe et al., 2014). Migrants from countries with labor markets similar to that of Germany potentially fit better into the German labor market in terms of jobs and skills needed (Duleep and Regets, 1999). Migrants from OECD and EU countries may also simply be more familiar with the German labor market system due to similarities with the origin country (Van Tubergen et al., 2004; Beyer, 2017).

Finally, immigrants who are unsuccessful in the labor market may be more likely to return to the country of origin. However, evidence regarding the link between unemployment and return migration is mixed. Several studies have indicated that unemployment spells increase return probability across immigrant groups (Constant and Massey, 2002; Gundel and Peters, 2008; Bijwaard et al., 2014). Others find that social networks, perceived discrimination, and calculations of wage differentials between the origin and destination countries can all contribute to return migration decisions (Borjas and Bratsberg, 1994; Waldorf, 1995; Tezcan, 2019). Freedom of movement within the European Union may enable EU immigrants to return home more easily if they fall upon hard times (Zaiceva and Zimmermann, 2016). By contrast refugees are unlikely to be able to quickly return to their country of origin due to ongoing violence and conflict.

All of the above factors can lead to differences in labor market success even between immigrants from different sending countries but with the same educational qualifications. A large body of evidence indicates that immigrants with credentials from non-Western countries experience lower labor market returns to their education than those from Western or OECD countries. These decreased labor market returns lead to higher likelihood of welfare receipt. Other factors including skill mismatches, return migration, discrimination, and age at migration. Because of these differences, I hypothesize that EU immigrants exhibit the lowest rate of welfare use compared to third country immigrants and refugees (H2a).

Building on previous research comparing the welfare use of immigrant groups, this paper adds refugees to the analysis as a separate group from other third country immigrants. Studies conducted after the arrival of recent refugees have indicated a refugee labor market penalty that cannot be completely explained by differences in education level or sociodemographic factors (Bähr et al., 2017; Söhn, 2019; Bedaso, 2021). However, comparison between recent refugees and other immigrants has been limited. Because of their additional disadvantages, I hypothesize an additional refugee penalty predicting higher use of welfare benefits, even after accounting for human capital and sociodemographic factors (H2b).

There are several potential explanations for the “refugee penalty.” For one matter, refugees face labor market discrimination. Recent studies have indicated that even well-qualified refugees who have learned German encounter challenges finding work and are evaluated not only based on their skills, but also on their country of origin. Employers frequently assume that refugees are less capable than other equally-educated immigrants (Khan-Gökkaya and Mösko, 2021; Kloubert and Hoggan, 2021). In addition, mental health issues stemming from traumatic events may hinder the labor market integration of refugees (Ruiz and Vargas-Silva, 2018; Brell et al., 2020). Legal regulations governing residence and employment can also negatively affect refugees' employment prospects. Employers may be met with high costs if they choose to hire refugees who have only received a delay of deportation - rather than protection status - and cannot be certain that refugees without protection status will remain in Germany (Brücker et al., 2016b). In addition, employment prohibitions upon arrival in Germany may discourage asylum applicants from seeking work later once permission is eventually granted (Brell et al., 2020; Fasani et al., 2021). The refugee penalty merits further exploration to determine how these factors may affect labor market outcomes.

Two types of welfare benefits are available in Germany for unemployed adults of working age: Unemployment Benefits I (UB I) and Unemployment Benefits II (UB II). Eligibility for UB I is based on social security contributions. Recently unemployed workers who have contributed sufficiently to the welfare system receive a benefit amount calculated from previous earnings. Those who have exhausted their UB I benefit or who have not worked long enough to qualify for UB I can receive UB II. Rather than social security contributions, UB II is a means-tested benefit, meaning that eligibility is based on income and need. The benefit amount is set according to the number of people in the household, the number of children in the household, and the ages of the children. In general, all residents of Germany are entitled to UB II, excluding tourists, seasonal workers, and asylum seekers. Recipients are also required to be between the ages of 15 and 65 and be able to work at least 15 h per week.

UB II receipt is conditional on receiving sufficient income to secure a minimum standard of living, but not conditional on employment. Individuals may be registered as employed but not earn enough to meet this minimum standard, and can therefore receive UB II to fill the gap. Conversely, unemployed individuals may not necessarily receive UB II if they have not applied or if the application process is incomplete. The relationship between unemployment and UB II receipt is therefore not always direct (Hohmeyer and Lietzmann, 2020).

EU citizens are broadly entitled to UB II in Germany under the Freedom of Movement Law (FreizügG/EU), a German national law derived from EU-level directives. There are some conditions upon recipiency: EU citizens cannot claim social security benefits for their first 3 months in Germany, as they are not permitted to enter Germany with the purpose of claiming social benefits (Mantu and Minderhoud, 2017). To become eligible, they must either work and contribute to the social security system for at least 1 year or reside in Germany for at least 5 years. As of 2017, economically inactive EU citizens cannot receive any type of welfare benefit during their first 5 years in Germany (see Bundesgesetzblatt Nr. 65, Bundesregierung 12/28/2016). If they have worked for less than 1 year, they are entitled to a maximum of 6 months of UB II benefits (see Bundesgesetzblatt Nr. 65, Bundesregierung 12/28/2016).

Migrants from several countries outside of the EU are granted special access to the welfare system under the European Convention on Social and Medical Assistance of 1953. These countries include Iceland, Norway, and Turkey. Other third country immigrants living in Germany can be eligible if they have a residence permit and legal permission to work. However, UB II receipt can increase the risk rejection or delay of residency permit extensions or German citizenship applications (Schnabel, 2020).

Refugees who have officially received legal protection status fall under largely the same eligibility requirements for UB II as German citizens (see Sozialgesetzbuch II - Grundsicherung für Arbeitsuchende § 7 Leistungsberechtigte, Bundesregierung 12/24/2003). Finally, refugees waiting for an asylum decision (or those who possess a Duldung, or temporary suspension of deportation) are entitled to an asylum seekers' benefit (Asylbewerberleistungsgesetz, AsylbLG), rather than UB II. The above information on immigrants' eligibility for UB II is also available in the Supplementary Table S1.

Data for my analyses are taken from the 2013 to 2019 waves of the core sample of the German Socioeconomic Panel (SOEP), a longitudinal panel dataset which includes representative samples of the German population (Socio-Economic Panel (SOEP), 2021). Running since 1984, the SOEP re-interviews adult household members annually and has frequently added new samples for the purpose of studying various social groups. My analytical sample includes all UB II-eligible heads of household with a direct migration background who were surveyed in the years 2013–2019. Some respondents have participated in the SOEP since its commencement, such as those from “Sample B Foreigners in the Federal Republic of Germany,” which began in 1984 and includes individuals in private households with a Turkish, Greek, Yugoslavian, Spanish, or Italian household head. Others entered the survey later, including “Sample D Immigrants,” begun in 1994. In 2013 and 2015, two new Migration Samples were incorporated, one including those who immigrated to Germany after 1995 and the second including those who immigrated between 2009 and 2013 (IAB-SOEP Migration Samples, 2021). Refugee Samples were added in 2016 and 2017 and surveyed individuals who arrived in Germany between January 2013 and December 2016 with an asylum application (IAB-BAMF-SOEP Survey of Refugees, 2021). The analytical sample therefore includes individuals who have resided in Germany from less than 1 year up to more than 50 years.

Based on eligibility for UB II, I restrict the sample as follows: first, I include only first-generation immigrant household heads surveyed from the years 2013 to 2019. The sample is further restricted to those between the ages of 18 and 65¹ who are not on parental leave and not receiving old-age, disability or civil service benefits, as receipt of these benefits precludes UB II receipt. The sample is divided into three immigrant groups: those who migrated from EU countries, those who migrated from third countries, and those who arrived as refugees. Further detail regarding excluded cases is available in the Supplementary Table S2.

As legal status shapes eligibility for social benefits in Germany (Voigt, 2020), this is also a necessary sampling criterion. Because EU citizens only become eligible for UB II after either working for at least 1 year or residing in Germany for 5 years, I exclude EU citizens who have lived in Germany for less than 5 years and do not have at least 1 year of full-time work experience². Asylum seekers are excluded from the sample, as they are not eligible for UB II and receive asylum seekers' benefits instead. Asylum recipients are included as they are entitled to UB II.

Finally, the analytical sample includes only immigrants whose highest educational qualifications were acquired abroad. As noted above, when immigrants receive their education in the destination country, they typically enjoy greater labor market returns than those educated in the origin country. Educational qualifications from Germany are therefore not comparable to those from abroad, and those immigrants who received their highest qualification from Germany are excluded.

The dependent variable in the analysis is UB II receipt. UB II receipt is measured at the household level, with a binary indicator of 1 for benefit receipt and 0 for no benefit receipt. Like most of the variables in the SOEP, this variable is self-reported, based on questions asking households which benefits they are receiving and the amount that they receive. The independent variable and control variables are operationalized dependent on how survey questions are asked in the SOEP. Education level serves as the independent variable. Control variables for the various stepwise models include: unemployment experience; human capital factors including full-time employment experience and German language skills; sociodemographic factors including gender, number of children in the household, and single parenthood; and factors related to immigration including legal status, age at immigration, and length of time the respondent has resided in Germany³. Although I provide further details regarding these variables below, an additional table explaining the precise operationalization of model components is also available in the Supplementary Table S4.

The first variables to be added in the stepwise modeling process are the human capital variables. Education level is measured through the International Standard Classification of Education (ISCED), which provides a standardized cross-national qualification of educational degrees to capture education obtained prior to arrival in Germany⁴. This measure is then simplified into three categories for analysis: basic education, which includes primary and lower secondary education; medium-level education, which includes upper secondary, post-secondary non-tertiary, and short-cycle tertiary education; and higher education, which includes bachelor's, master's, and doctoral degrees.

I also include work experience as a form of human capital. The “full-time employment experience” variable indicates the total length of full-time employment in the respondent's career up to the point of the interview. As Hohmeyer and Lietzmann (2020) attribute likelihood of welfare receipt to employment prospects as well as duration of welfare receipt to date, I also include length of time unemployed. The “unemployment experience” variable indicates the total length of unemployment in the respondent's career and is included to reflect Hohmeyer and Lietzmann's observation of the role of previous welfare receipt in predicting ongoing welfare receipt. Both variables measure length of time in years.

German language ability is a final component of country-specific human capital. I compile this variable from several similar questions about oral or spoken language abilities, all of which asked respondents to rate their language abilities on a scale of 1–5, with 5 being the highest ability; for some respondents surveyed in 2018, this question was not asked, and so I imputed language proficiency from 2017 based on the assumption that their oral German abilities would remain at least at the level of the previous year.

In addition to human capital factors, sociodemographic factors are added in the stepwise modeling process. A number of studies in the German context have indicated that a higher welfare participation rate among immigrants compared to natives is in large part due to sociodemographic factors (Riphahn, 1998; Castronova et al., 2001; Barrett and Maître, 2013; Bruckmeier and Wiemers, 2017). These factors also contribute to higher welfare receipt among non-EU immigrants compared to EU immigrants (Wunder and Riphahn, 2014). More specifically, single parenthood, a female household head, and a larger number of children increase risk of welfare dependence (Riphahn, 1998; Fertig and Schmidt, 2001; Anastossova and Paligorova, 2006). Considering this body of literature, I control for gender, single parenthood, and number of children in the household. The variable indicating the number of children is a categorical variable, with three categories: no children, one child, or multiple children.

Finally, several variables related to respondents' immigration trajectories are included in the models: legal status, age at immigration, and years of residence in Germany. Immigrants who arrive in the destination country at a younger age demonstrate labor market advantages for several reasons, including greater opportunities to gain language fluency, assimilate into the culture, join the education system and invest in other country-specific human capital (Gustafsson et al., 2017; Friedberg, 1992). Several studies indicate that immigrants who arrive in the destination country at a younger age display better labor market outcomes in adulthood (Van den Berg et al., 2014; Lemmermann and Riphahn, 2018). Age at immigration is calculated using age and years of residence in Germany. Length of residence in Germany is calculated using survey year and immigration year.

Gathering legal status information from the SOEP required the use of several variables from various questionnaires that asked respondents about their residence permits to varying degrees of detail. Since respondents were not asked for detailed legal status information in every survey year, in some cases legal status was imputed from the previous year or a more general indicator of permanent or temporary residency was utilized instead⁵. Responses were consolidated into a measure indicating four residence statuses: 1) legal protection or refugee status, 2) permanent residency, which also includes naturalized citizens, 3) temporary residency, 4) unknown or none, with 1% of respondents in the last category. Refugee samples were generally asked more detailed information about their current residence permit than the migration samples or the general sample. Detailed legal status information and information about imputed values are available in the Supplementary Table S3.

The final dataset includes 7,464 person-year observations and 4,247 unique person observations. Of this sample, descriptive statistics for person-year observations are reported in Table 1.

I employ logit random effects models in the analysis, with the binary outcome indicating whether the respondent is receiving UB II in the given survey year. Random effects models were chosen over fixed effects models due to insufficient within-individual variation. As I am interested in between-persons variation, rather than within-person variation, random effects models were also a more appropriate fit. This structure allows for an examination of the differences between immigrant groups and immigrants with different legal statuses.

I first use stepwise models to examine predictors of welfare use among the three specified immigrant groups. Model 1, which only includes the immigrant group dummy variables, demonstrates the baseline difference between immigrant groups. Model 2 incorporates human capital factors including education, work experience, and German language skills. I also control for the time respondents have already spent unemployed. Survey year is also included to control for external economic conditions and any other time-specific effects. In Model 3, I add household composition, gender, and single parenthood in order to examine whether welfare use is largely a consequence of sociodemographic factors, as indicated by several previous studies (Riphahn, 1998; Castronova et al., 2001; Barrett and Maître, 2013; Bruckmeier and Wiemers, 2017). In Model 4 I add descriptors of respondents' migration experience, including legal status, age at immigration, and years of residence in Germany. To capture country effects, I also run a model with dummy variables for the largest country of origin to ensure a sufficient number of observations in each country group. Finally, I run separate models for each immigrant group with the country-of-origin dummies. Results of all models including these dummies are available in the Supplementary Table S6.

Random effects models typically do not employ both time-variant and time-invariant variables together. However, the panel data includes a combination of independent variables that generally do not change for each respondent over time (such as gender) and variables that may change over time (such as full-time employment experience). To address this challenge, within-between random effects models (Allison, 2009, 2014; Bell et al., 2019) allow for a combination of fixed effects estimates of the effects of time-variant variables with random effects-type estimates of the effects of time-invariant variables. To measure the effects of these time-varying variables, I obtain the standard deviation of the variable by subtracting the mean of observations from its observed value in each survey year (Schunck, 2013). As such, I can estimate the effect of a change in this variable by one standard deviation.

In order to determine which variables were time-varying and therefore appropriate for this within-between random effects model treatment, I first examined the percentage of each variable that changed over time for respondents. More detail about the percentage of individuals who experience change in any of the above variables is available in Table 2 below. As less than 5% of respondents indicated changes in education level, gender, number of children in the household, single parenthood, and legal status, these variables were treated as time-invariant. Age at immigration is also time-invariant as it is an event that occurred in the past. More than 10% indicated changes in their level of self-evaluated German language skills, full-time employment experience, and unemployment experience over the survey period. Years of residence in Germany also changes with each year of the survey period. These variables were thus treated as time-varying. Language skill is employed as a continuous variable in order to treat it as time-varying.

The results of the within-between random effects logistic regressions allow for an examination of the effects of group-level characteristics over the survey period. Marginal effects of stepwise models are reported in Table 3 and regression coefficients are reported in the Supplementary Table S5. At the baseline, third country immigrants are 7.4% more likely to use UB II than EU immigrants; refugees are 43.8% more likely to use UB II than EU immigrants. After accounting for all independent variables, this difference decreases: third country immigrants are 5.3% more likely to receive benefits than EU immigrants, and refugees are 30.6% more likely to receive benefits⁶. These results confirm hypothesis (H2a), that EU immigrants exhibit the lowest rate of welfare use compared to refugees and third country immigrants, as well as hypothesis (H2b), that there is an additional refugee penalty predicting higher use of welfare benefits, even after accounting for human capital and sociodemographic factors. Differences between the three immigrant groups remain significant even after controlling for country of origin, as demonstrated in the models that employ country-of-origin dummies (Supplementary Table S6).

Several human capital factors remain significant across models. Not unexpectedly, education decreases likelihood of UB II receipt. Immigrants with a medium level of education (–0.059, p < 0.001), are less likely to receive UB II than those with a low level, and immigrants with a high level of education even less likely to receive UB II (–0.122, p < 0.001). In contrast to education, employment experience loses significance after accounting for sociodemographic characteristics, possibly because individuals lacking employment experience are engaged in childcare duties⁷.

Results of control variables related to sociodemographic characteristics are largely in line with the existing literature. Women (0.065, p < 0.001) and single parents (0.181, p < 0.001) display a higher likelihood of welfare receipt across immigrant groups. Having children in the household only becomes significant if there is more than one child (0.084, p < 0.001).

Finally, factors related to immigration significantly predict likelihood of welfare receipt; those who migrated to Germany at an older age are slightly more likely to receive benefits (0.007, p < 0.001), and those who have resided in Germany for a longer period of time are less likely to receive benefits (–0.024, p < space 0.05). Although legal status is not the focus of this paper, it is observed that legal status also predicts UB II receipt, even after controlling for immigrant group. Most notably, permanent residency or citizenship is related to a lower likelihood of welfare use (–0.086, p < 0.001). This finding corroborates previous evidence that permanent legal status or naturalization can potentially support immigrants' labor market success (Von Haaren-Giebel and Sandner, 2016; Gathmann and Keller, 2018; Riphahn and Saif, 2019). There are several possible mechanisms that may drive this relationship. Formal restrictions may limit temporary immigrants' access to certain jobs; employers may be more willing to hire immigrants who they know will stay in Germany; and immigrants may be incentivized to invest in host-country human capital once they know they will stay. That said, further research is needed to determine whether this result may be due to selection effects, where the most economically successful immigrants are also able to gain permanent residence or naturalization.

I also employ an additional model with an interaction term between immigrant group and education level. All independent variables from Model 4 are included as well. This model tests (H1) concerning returns to education by immigrant group.

Results are reported in Table 4 and indicate that education decreases the likelihood of welfare receipt for EU and third country immigrants. Compared to the baseline category of low education, coefficients for medium and high education are significant and negative for these two immigrant groups. The magnitude of these coefficients is similar as well, with a medium level of education reducing likelihood of welfare receipt by 13.9% for EU immigrants and 12.4% for third country immigrants. A high level of education reduces likelihood of welfare receipt by 24.3% for EU immigrants and 20.1% for third country immigrants. As demonstrated in Figure 1, the effect of education on likelihood of welfare receipt is therefore somewhat greater for EU immigrants than for third country immigrants.

For refugees, a medium level of education does not significantly reduce the likelihood of welfare receipt compared to a low level of education, though this coefficient just barely lacks significance. A high level of education reduces likelihood of welfare receipt by 5.9%. The effect of a high level of education is therefore significantly smaller in magnitude for refugees compared to the other two immigrant groups.

These results only partially confirm hypothesis (H1a), that there is a greater negative effect of educational attainment on welfare receipt for EU immigrants than for third country immigrants and refugees. Rather, the effect of educational attainment on likelihood of welfare receipt is similar for EU and third country immigrants. Refugees diverge from the patterns displayed by other immigrant groups as the magnitude of the effect of education is much smaller for this group. Results remained significant even after the inclusion of the country-of-origin dummy variables as a robustness check.