Discrimination against students with foreign backgrounds: evidence from grading in Swedish public high schools

Abstract

We rigorously test for discrimination against students with foreign backgrounds in high school grading in Sweden. We analyse a random sample of national tests in the Swedish language graded both non-blindly by the student's own teacher and blindly without any identifying information. The increase in the test score due to non-blind grading is significantly higher for students with a Swedish background. This discrimination effect is sizeable, about 10% of the mean or 20% of the standard deviation of the blind test score.

Keywords:

• discrimination
• education
• grading
• teacher bias

Acknowledgements

We thank George Farkas and Tore Ellingsen, the editor Colin Green and two anonymous referees for their helpful comments, the Institute for Labour Market Policy Evaluation (IFAU), the Jan Wallander and Tom Hedelius Foundation, the Swedish Research Council, and the Swedish Council for Working Life and Social Research for financial support.

Notes

1. For example, Statistics Sweden (2007) reports that in about 60% of the Swedish localities, almost all children have a Swedish background. In the larger cities, areas where a majority of children have foreign backgrounds are common. In areas that are dominated by foreign children, the children are generally from Africa, Asia, or a non-EU-25 country.

2. See Ouazad and Page (2011) for some new experimental evidence.

3. As in previous studies, we measure the relative difference in outcomes between two groups. We cannot separate between one group being positively discriminated (favouritism) from that the other being negatively discriminated, as discussed by Feld, Salamanca, and Hamermesh (2013).

4. For evidence, see Ferguson (2003).

5. For a detailed description, see Hinnerich, Höglin, and Johannesson (2011).

6. Recently, the Swedish Schools Inspectorate has begun to partly evaluate schools.

7. See the regulatory curriculum in Skolverket (2011).

8. In the appendix, we also provide results from an ordered logit, not assuming cardinality, and the results when the grades have equal space using a scale of 0, 1, 2, and 3, not 0, 10, 15, and 20. The results are robust to these alternative specifications.

9. In Table 1, we use the scale 0, 10, 15, and 20 for Fail, Pass, Pass with Distinction, and Excellent.

10. The correlation between the non-blind and the blind grade is 0.41 (for both the Pearson and the Spearman correlation). The scores are identical for about 46% of the students.

11. The smallest cluster is of reasonable size (42 external graders). Thus, small sample problems should be of little importance. Interestingly, the standard errors are somewhat larger when we cluster only on the school level, However, with respect to critical levels of significance (1% and 5%), there is no difference in Tables 2 and 4 when we use either schools as level of clustering or both schools and external graders. The results from clustering only on the school level can be received upon e-mailing the corresponding author.

12. Since there was one test date in the spring and one in the fall, there were 18 topics in the data set. Moreover, there was one category in which the topic was unclear.

13. A standard test of the same proportions shows that of 19 different topics, the proportions are significantly different from zero (5% level) in four topics between the groups, indicating some sorting depending on the student's background.

14. Marginal effects from estimating Equation (1) as an ordered logit are presented in Table A1.

15. A student with a foreign background could have two different classifications, being both non-European and European, if, for example, the mother was born outside Europe but the father was born in Europe (but not Sweden). We defined the groups by first using information about whether the student with a foreign background was born in Sweden, in Europe, or outside Europe. If she was not born in Sweden, she was classified according to her country of birth. If she was born in Sweden, we based our presented results on the country of birth of the mother. However, the results are similar when we use the country of birth of the father, as this distinction only changed the classification of two students from the European group to the non-European group.