This paper draws attention to a subtle, but concerning, empirical challenge common in panel data models that seek to estimate the relationship between student transfers and district academic performance. Specifically, if such models have a dynamic element, and if the estimator controls for unobserved traits by including district-level effects, then model validity does not allow for a district's academic performance, in turn, to impact future transfers. Yet it seems reasonable that families, having access to publicly available aggregated information on standardized test results, seek to move their children to better-performing districts. In this paper, we demonstrate that, not only is such feedback quantitatively and qualitatively important, but also that allowing for such feedback substantially alters the estimated relationship between transfers and district performance.

This article uses longitudinal student-level data from the American University of Sharjah, a large comprehensive university in the Middle East, to examine the relationship between student evaluations of teachers and current and future student achievement. Our model strategies control for the observed and unobserved heterogeneity of students and use unique instruments. We find that when all disciplines are examined together there is a positive relationship between current evaluation and current grade point average (GPA) but a negative relationship between past evaluations and current GPA. Discipline-specific estimations find the same results in the math and science course subsample, but for other course types there is little relation between evaluation and GPA.