Search

Several studies have reported that student evaluation of teaching (SET) presents important problems. First, depending on the area, there are significant differences in the evaluations. Second, numerous noninstructional biases exist, such as when those teachers who award better grades obtain better SETs. Correcting the rankings by considering these biases (e.g., adjusting SETs according to the class grade) has been proposed. In this paper, we analyse a third problem: it is impossible to correct the biases because they are specific to each area, level, and even class. On a sample of 15,439 SETs, we compared the biases present in two very close areas (accounting and finance) and at two levels (undergraduate and postgraduate). Then, we used a procedure based on the analysis of residuals in OLS models to eliminate area- and level-specific biases. However, there are still latent biases apparently linked to each specific group of students. [Display omitted] • Noninstructional biases generate huge distortions in Student Evaluation of Teaching. • These biases are markedly different according to area and level. • A bias correction procedure based on the analysis of residuals in OLS models is proposed. • After incorporating area- and level-specific corrections, there are remaining latent biases. • It does not seem possible to turn SETs into reliable instruments of teaching quality. [ABSTRACT FROM AUTHOR]