Source country differences in test score gaps: evidence from Denmark

Abstract

We combine data from three studies for Denmark in the PISA 2000 framework to investigate differences in the native–immigrant test score gap by country of origin. In addition to the controls available from PISA data sources, we use student‐level data on home background and individual migration histories linked from administrative registers. We find that second‐generation students from Lebanon and Pakistan increase their reading scores substantially compared with the first generation, while there is no improvement for students from Turkey, the single largest immigrant group. Native–immigrant gaps in mathematics are generally smaller than in reading skills, suggesting that part of the native–immigrant gap is due to lower language proficiency of immigrant students.

Keywords:

• immigrants
• PISA
• country of origin
• native‐immigrant gap

Acknowledgements

Thanks to Helena Skyt Nielsen and Astrid Würtz Rasmussen for comments and suggestions. Financial support by the Rockwool Foundation Research Unit and the Danish Social Science Research Council is gratefully acknowledged.

Notes

1. There is also a (small) related literature on indigenous test score gaps in Latin America (McEwan 2004, 2008).

2. Throughout this study, we use the terms source country and country of origin (COO) interchangeably.

3. Nevertheless, the immigrant flow from the guest‐workers’ countries of origin continues, driven by marriage migration.

4. We would have liked to present results for the heterogeneity of the estimates of socio‐economic background variables across countries of origin, too. However, probably due to the small sample sizes, the results were hardly significant and showed no systematic patterns. We therefore do not present these results in this article.

5. Since immigrant generation and language at home is a variable defined/varying only for immigrants, we need to add the coefficient estimate of the main effect of immigrant generation to that of the interaction term to make the relevant comparison.

6. These data‐sets contain observations with valid information on the key variables, namely source country and reading scores. In both the PISA Copenhagen and the PISA Ethnic data‐sets we lose a number of observations due to the non‐reporting of civil registration numbers by school heads, which means that register information cannot be linked to the PISA data.

7. In Rangvid (2007) it is shown that students with one immigrant and one Danish parent and immigrant students from western countries perform at similar levels in the PISA test to native Danes.

8. We include separate indicators for 10 source countries (those with more than 50 student observations in our combined sample). These countries are: Turkey, former Yugoslavia, Pakistan, Lebanon, Iraq, Morocco, Somalia, Vietnam, Iran and Afghanistan. Immigrants from other non‐western countries are grouped into a residual category. All observations are included in all regressions, but we have chosen to report results for the four largest countries of origin only (those with more than 150 observations) due to the very small samples for the remaining countries.

9. As is seen in the table, the subsample of the international PISA 2000 study, which is drawn to mirror the entire population composition in Denmark, has a much lower immigrant share (5%) than both the Copenhagen sample and the PISA Ethnic data‐set (27% and 31%), which sample students at schools with high immigrant shares.

10. A potential disadvantage of an analysis based on linear regressions is that observations in the different country of origin groups may not fall within a region of common support (McEwan 2008). That is to say, immigrant children may be observationally quite dissimilar from native Danish children, and the linear specification implicitly relies on projections of outcomes outside the observed range for such students. To assess the sensitivity of the results, we estimated a probit, regressing the immigrant dummy variable on the full set of family characteristics listed in Appendix 1. We then calculated propensity scores for each student. The region of common support includes 93% of immigrant students and 90% of Danish students. We then dropped observations that lay outside the region of common support and re‐estimated the regressions. The estimates of the gaps were not substantively different from the full sample estimates.