Class size, student performance and Tiebout bias

Abstract

If students tend to choose instructors with high teaching ability, a negative link between class size and student performance may remain obscured until this kind of sorting has been taken into account. Using data from student evaluation surveys, we confirm that it is crucial to control for instructor fixed effects.

Notes

¹ As the examinations take place several weeks after the evaluations and evaluations are based on an anonymous survey, students' test scores cannot be linked to their assessments. Furthermore, it is not possible to link individual or average test scores to class size as students anonymously attend class and examinations are independent of the instructors.

² Ordered probit regressions yield very similar results. They are available from the authors upon request.

³ Ignoring clustering in the regression is likely to produce downward biased standard errors (Moulton, 1990) as class characteristics and individual student performance differ in the level of aggregation.

⁴ To avoid perfect collinearity with the course dummy variables, the control group consists of three instructors. Note that the estimated influences of class size and instructor characteristics are independent of the concrete instructors included in the control group. As a further check of the robustness we defined a broader course dummy only differentiating between economics courses and quantitative courses. This reduced the collinearity problem between course and instructor dummies allowing us to exclude only one instructor dummy. The regression results are very similar to those reported in table 1. They are available from the authors upon request.

⁵ Note that the coefficient estimates imply a negative relationship between class size and student performance for our entire sample. The largest class in our sample has 225 students. The turning point of the U-shaped relationship lies with 244 students outside the sample.