The Extent of Bias in Grading

Do biased perceptions and behaviors affect teachers’ assessment of students? To investigate this question, a number of studies use data on two different scores for the same individuals: one non-blind score based on classroom tests assessed by the student’s own teacher and one blind test score based on a national exam marked externally and anonymously. In the absence of bias in teachers’ assessments, it is argued, there should not be significant differences in the gaps in blind and non-blind scores between different groups. This article present a parsimonious econometric framework that distills out the assumptions necessary to identify group bias in teachers’ assessment from such a comparison of blind and non-blind scores. This framework lays the foundation for our empirical analysis, where data from the Norwegian school system are employed to estimate and interpret differences between nonblind and blind assessments. The results show that the relationship between the subject ability and non-blind results tends to be different from the relationship between subject ability and blind results. Evidence of this is found both when grades are recorded when teachers grade the same test and when they grade based on different assessments that are meant to test the same skill. The difference between non-blind and blind will therefore be a function of the skill tested. This leads to different estimates of the group bias when holding ability fixed.