VoxEU Column Labour Markets Migration

Language training and refugees’ success

The labour market integration of refugees and immigrants is key to their ability to contribute to the economy of the receiving country and to enhancing the fiscal sustainability of more open immigration policies. Using the quasi-random assignment of Danish refugees to language training, this column shows that language acquisition significantly increased the lifetime earnings of refugees. Refugees with language training became more likely to work in communication-intensive jobs and obtained additional education. The positive effects are transmitted to the next generation in terms of improved schooling outcomes for male children of refugees.

Newly arrived refugees are substantially underemployed compared to natives and to people who migrate for economic opportunities. Underemployment and low income imply initial fiscal and economic costs to the receiving economies (Dustmann et al. 2016, Fasani et al. 2018, and Ruist 2016), and as the employment gap is persistent, the ability of refugees to contribute fiscally is also diminished in the long run. 

European policymakers are still debating how to best facilitate the economic integration of refugees, especially in the wake of major refugee inflows in 2015 and 2016. European governments have taken different approaches. Some have focused on teaching refugees language skills while others have focused on helping job search and stimulating job finding.

One reason for the dissimilar approaches is the lack of evidence (and consensus) on the effectiveness of each strategy. Two challenges have limited progress in understanding this issue. First, the more able immigrants typically acquire better language skills and earn more (an omitted variable issue), so isolating the causal link between language proficiency and earnings requires quasi-experimental research designs. 

Second, language acquisition is a gradual process. The effects of acquiring a language may take long to materialise compared with competing job-oriented approaches, making long-run longitudinal data necessary to assess the full gains from attending language school. 

Our recent paper (Arendt et al. 2021) makes progress on these dimensions. While other studies (Sarvimäki and Hämäläinen 2016, Heller and Mumma 2020) have used random or quasi-random assignment to language classes and found positive labour market effects, we follow refugees for a longer time period to also examine the effects on the next generation and consider several dimensions of economic integration.

Danish refugee integration reform

We analyse a major reform that significantly expanded language training for refugees in Denmark. The reform was introduced in 1999 and has several features that make it interesting to study. First, the sudden and discontinuous implementation of the reform creates as-good-as-random assignment to the policy for refugees admitted around a cut-off date of 1 January 1999. Second, the reform significantly increased the duration and quality of language training. Third, the long time horizon since implementation and rich administrative data allow us to study outcomes over an extended period, including the effects on the children of the refugees. 

The Danish reform involved other smaller changes to integration efforts, such as a temporary 13-month reduction in welfare benefits for a subgroup as well as more gradual changes to the geographical dispersion of refugees. We discuss those changes together with the results below.

Language training facilitated access to education and higher-paying jobs

Our study of the Danish reform yields several key findings, as illustrated in Figure 1, which represents long-run outcomes for refugees, by month of arrival, around the cut-off date to qualify for the enhanced language training (1 January 1999). 

We observe: (1) an average yearly earnings and employment increase by 34% and 23% of the average, respectively; (2) strong and significant improvements in occupational task complexity; and (3) no effect on adult crime rates. 

Figure 1 Long-run integration by month of admission


Notes: The graphs show sample means of average cumulative earnings, employment, task complexity conditional on employment, and criminal charges in year 18 after the reform by one-month bins of admission dates.

A detailed analysis of the timing of the effects and additional outcomes strongly suggests that these effects are driven by improved language skills. In fact, the improved labour market outcomes for treated refugees only occurred after the completion of the language classes. Additionally, treated refugees accessed higher-paying occupations with stronger communication requirements. Treated refugees, especially young adults, were more likely to enrol and complete education in Denmark, especially vocational training. Finally, refugees who spoke languages that were linguistically far from Danish and unfamiliar with the Latin alphabet gained most from the more extensive training.

Other reform consequences

The temporary welfare benefit reduction for a subgroup of the treated, those older than 25 years of age or with children, resulted in a one-year drop in disposable income. In the same period, it temporarily increased shoplifting crimes among this group. The impact on crime disappeared when the benefit reduction ended after one year. 

We also find that the mandatory nature of language training, imposed simultaneously with the reform, led to short-term reductions in mobility from rural areas during the training programme, as people attended the classes. It had no economic effects, however, and there were no significant differences in mobility in the long run (after ten years).

Finally, possibly because of the improved economic situation and language proficiency, children born to treated refugees improved in some outcomes. Boys whose parents were both treated were more likely to complete lower secondary school and less likely to engage in criminal activities as teenagers and young adults. Girls, however, were not affected, possibly because they were already at lower risk of negative outcomes. These results are in line with the importance of male role models on outcomes such as crime and dropouts from primary school, especially for boys.

Policy implications

Our work has several policy implications. First, there is a strong case for providing language training to disadvantaged immigrants, as this would contribute to closing the large employment and earnings gap for refugees. A cost-benefit analysis shows that the returns to language training were up to $15 for $1 invested when we evaluate the policy over an 18-year period. 

Second, the employment and earnings improvements from language training are permanent, in contrast to estimated impacts of economic incentives that appear to be temporary (e.g. Andersen et al. 2019). 

Third, the evidence shows that welfare benefit reductions may have unintended impacts on crime, while this is not the case for language-training provision.


Andersen, L H, C Dustmann and R Landersø (2019), “Lowering welfare benefits: Intended and unintended consequences for migrants and their families”, CReAM Discussion Paper 05/19, Centre for Research and Analysis of Migration, Department of Economics, University College London.

Arendt, J N, I Bolvig, M Foged, L Hasager and G Peri (2021), “Language training and refugees’ integration”, NBER Working Paper 26834.

Dustmann, Christian, Francesco Fasani, Tommaso Frattini, Luigi Minale and Uta Schӧnberg (2016), “On the economics and politics of refugee migration”, VoxEU.org, 18 October.

Fasani, F, T Frattini and L Minale (2018), “(The struggle for) refugee integration into the labour market: Evidence from Europe”, VoxEU.org, 9 April.

Heller, B, and K S Mumma (2020), “Immigrant integration in the US: The role of adult English language training”, EdWorkingPaper 20-288, Brown University.

Ruist, J (2016), “Fiscal cost of refugees in Europe”, VoxEU.org, 28 January.

Sarvimäki, M, and K Hämäläinen (2016), “Integrating immigrants: The impact of restructuring active labor market programs”, Journal of Labor Economics 34(2): 479–508.

735 Reads