Gender bias in teaching evaluations: the causal role of department gender composition

Abstract

Women are underrepresented in academia's higher ranks. Promotion oftentimes requires positive student-provided course evaluations. At a U.S. university, both an archival and an experimental investigation uncovered gender discrimination that affected both men and women. A department's gender composition and the course levels being taught interacted to predict biases in evaluations. However, women were disproportionately impacted because women were more often in the gender minority. A subsequent audit of the university's promotion guidelines suggested a disproportionate impact on women's career trajectories. Our framework was guided by role congruity theory, which poses that workplace positions are gendered by the ratios of men and women who fill them. We hypothesized that students would expect educators in a department's gender majority to fill more so essential positions of teaching upper-level courses and those in the minority to fill more so supportive positions of teaching lower-level courses. Consistent with role congruity theory when an educator's gender violated expected gendered roles, we generally found discrimination in the form of lower evaluation scores. A follow-up experiment demonstrated that it was possible to change students' expectations about which gender would teach their courses. When we assigned students randomly to picture themselves as students in a male-dominated, female-dominated, or gender-parity department, we shifted their expectations of whether men or women would teach upper- and lower-level courses. Violating students' expectations created negative biases in teaching evaluations. This provided a causal link between department gender composition and discrimination. The importance of gender representation and ameliorating strategies are discussed.

Keywords: gender bias; teaching evaluations; university policy.

Fig. 1.

Illustrated are the findings from the archival study (N = 4,700) in which department gender composition and course-level significantly predicted teaching evaluation scores for men and women differently. In male-dominated departments, women tended to have higher evaluations than did men in lower-level courses; in upper-level courses men had higher evaluations did than women. In contrast, in female-dominated departments men tended to receive higher evaluations than did women in lower-level courses, and in upper-level courses women received higher evaluations than did men. There were no such differences at the mean department gender composition. Error bars indicate ±2 SEs.

Fig. 2.

In study 1, (N = 4,700), because the majority of women held gender-minority status (72.6% of departments were male dominated), and most classes taught were upper-level courses (72.3%) the potentially negative impact of gender bias was greater for women than for men. An estimated 32% of men and 52% of women at this university were potentially negatively impacted by gender bias in their teaching evaluations.

Fig. 3.

From study 2 (N = 803), illustrated are the percentages of women and men that students expected to teach upper- and lower-level courses from our model estimates. Percentages are the predicted values from the statistical model that tested students’ expectations. Students guessed which of two professors (one male and one female) would teach four courses presented in random order. Students in the male-dominated, gender parity, and female-dominated departments had different expectations about whether a man or a woman should teach upper and lower-level courses.

Fig. 4.

Illustrated are the findings from study 2 (N = 803) in which random assignment to different department gender compositions and course levels significantly predicted different evaluation scores for men and women. Error bars indicate ±2 SEs.

Fig. 5.

In study 2 (N = 803), students viewed a “Faculty” webpage, in which we manipulated the percentage of men and women to create male-dominated, parity, and female-dominated departments. The photos were not blurred in the actual stimuli.