Are There Gender Differences in Quantitative Student Evaluations of Instructors?

Abstract

Recent research conducted at numerous universities has found evidence of instructor-gender differences in student evaluations of teaching (SET). This paper examines whether such gender effects exist in “instructor overall” ratings within a database of SET that includes almost 600,000 observations from the past 11 years for the Faculty of Arts and Sciences (FAS) at a large research university in the northeastern United States. First, using multivariate OLS regression analysis, we tested 32 hypotheses of gender differences within discipline-rank combinations. Of the 32, only two hypothesis tests showed statistically significant gender differences in the instructor overall rating; one discipline-rank combination had higher average scores for male instructors, and one discipline-rank combination had higher average scores for female instructors. Second, using quasi-experimental data from calculus courses, we found that mean instructor overall scores of female instructors were different from those of male instructors only for Teaching Assistants (TAs) and Teaching Fellows (TFs), with higher scores for female TAs and TFs. Overall, we find no evidence of systematic gender differences in our analysis.

Appendices

Appendix 1

See Table 4

Table 4 Results of recent literature examining gender bias in SET scores

.

Appendix 2

See Table 5

Table 5 Definitions of predictor variables

.

Appendix 3

See Table 6

Table 6 Multivariate regression results, cross-divisional analysis

.

Appendix 4

Regression Models for Calculus Instructors’ SET Scores

We test for gender effects using the following ordinary least squares (OLS) multivariate models:

$$SET_{ict} = \theta \cdot {\text{female}}_{i} + \beta_{0} + \varepsilon_{ict} ,$$

(1)

$$SET_{ict} = \theta \cdot {\text{female}}_{i} + \beta_{0} + \beta_{1} \cdot {\text{rank}}_{it} + {\upbeta }_{2} \cdot {\text{age}}_{it} + \beta_{3} \cdot {\text{age}}_{it}^{2} + \varepsilon_{ict} ,$$

(2)

$$SET_{ict} = \theta \cdot {\text{female}}_{i} + \beta_{0} + \beta_{1} \cdot {\text{rank}}_{it} + \beta_{2} \cdot {\text{age}}_{it} + \beta_{3} \cdot {\text{age}}_{it}^{2} + \beta_{4} \cdot {\text{female}}_{it} \cdot {\text{rank}}_{it} + \beta_{5} \cdot {\text{year - term + }}\beta_{6} \cdot {\text{female}}_{it} \cdot {\text{year-term }} + \varepsilon_{ict} ,$$

(3)

where index i enumerates instructors, c is the course offering,

$${\text{c}} \in \left\{ {{\text{Intro to Calculus}};{\text{ Calculus}},{\text{ Series and Differential Equations}};{\text{ and Multivariable Calculus}}} \right\};$$

and time is indexed by

$$t \in \left\{ {{\text{Fall }}\;2006,{\text{ Spring }}\;2007, \ldots { },{\text{ Spring }}\;2017} \right\}.$$

Here, SET_ict is the average score of instructor i for course c received in semester t, and female is the dummy variable that takes value 1 if the instructor is a female and 0 otherwise. The control variable rank represents a set of four dummy variables for the instructor being a Lecturer/Preceptor, Professor, Senior non-ladder instructor, or Teaching Assistant/Teaching Fellow. For example, if an instructor is a preceptor in a given semester, then the rank variable for this instructor will be (1, 0, 0, 0), while a teaching post-doc will be represented by (0, 0, 0, 0). We notice that faculty rank may change over time and thus, generally, rank is time-dependent. Finally, age_it is the instructor age in years at the beginning of a given semester. One can see that model (1) is equivalent to a t-test for a difference between mean SET scores in the two gender groups; in model (2) we control for instructor rank and age; and finally, interactions between faculty rank and gender and also between year-term and gender are added in (3). Here, year-term denotes 1, 2,…, 22 which correspond to Fall 2006, Spring 2007,…, Spring 2017, respectively. Alternatively to the OLS models (1, 2 and 3), we consider their variations as follows:

$$SET_{ict} = \theta \cdot {\text{female}}_{i} + \beta_{0} + \alpha_{ct} + \varepsilon_{ict} ,$$

(4)

$$SET_{ict} = \theta \cdot {\text{female}}_{i} + \beta_{0} + \alpha_{ct} + \beta_{1} \cdot {\text{rank}}_{it} + {\upbeta }_{2} \cdot {\text{age}}_{it} + \beta_{3} \cdot {\text{age}}_{it}^{2} + \varepsilon_{ict} ,$$

(5)

$$SET_{ict} = \theta \cdot {\text{female}}_{i} + \beta_{0} + \alpha_{ct} + \beta_{1} \cdot {\text{rank}}_{it} + \beta_{2} \cdot {\text{age}}_{it} + \beta_{3} \cdot {\text{age}}_{it}^{2} + \beta_{4} \cdot {\text{female}}_{it} \cdot {\text{rank}}_{it} + { }\beta_{6} \cdot {\text{female}}_{it} \cdot {\text{year-term}} + \varepsilon_{ict}$$

(6)

where α_ct are the instructor invariant fixed effects that satisfy the following constraints:

$$\mathop \sum \limits_{c} \mathop \sum \limits_{t} \alpha_{ct} = 0$$

in each of (4, 5 and 6) cases.

Appendix 5

See Table 7

Table 7 Regression results for SET scores of calculus instructors using multivariate regression models (1, 2 and 3) and (4, 5 and 6)

.