Solanki2018 PDF

American Educational Research Journal
Month XXXX, Vol. XX, No. X, pp. 1–35

DOI: 10.3102/0002831218759034
Ó 2018 AERA. http://aerj.aera.net
Looking Beyond Academic Performance:

The Influence of Instructor Gender on
Student Motivation in STEM Fields
Sabrina M. Solanki
Di Xu
University of California, Irvine
Recruiting more female faculty has been suggested as a policy option for
addressing gender disparities in science, technology, engineering, and math-
ematics (STEM) fields given its ability to engage female students through
a role model effect. While a small but growing body of literature has exam-
ined the role of instructor gender at the higher education level, it typically
focuses only on academic outcomes. This paper utilizes a unique data set
that includes not only information about student course performance in
STEM but also a number of motivation-related measures. We find that hav-
ing a female instructor narrows the gender gap in terms of engagement and
interest; further, both female and male students tend to respond to instructor
gender. We conclude by discussing the policy implications of these findings.
KEYWORDS: female faculty in STEM, instructor gender, role model effects in

college, student-teacher gender match
Introduction
Although women have made tremendous gains in terms of college
enrollment, degree attainment, and representation in traditionally male-
dominated fields, researchers and policymakers remain concerned about
their presence in science, technology, engineering, and mathematics
SABRINA M. SOLANKI is a doctoral student in the School of Education at the University of

California, Irvine, 3200 Education, Irvine, CA 92697-5500; e-mail: [email protected].
Her research focuses on higher education policy, teacher effectiveness, STEM educa-
tion, and the evaluation of education interventions.
DI XU is an assistant professor of educational policy and social context at the
University of California, Irvine, a research fellow with the Community College
Research Center (CCRC) at Columbia University, and a visiting fellow at the
American Enterprise Institute. Her research uses quasi-experimental designs to exam-
ine the impacts of postsecondary educational programs, policies, and practices on
student academic performance, degree completion, and labor market outcomes.
Solanki, Xu
(STEM) fields. Even today, the choice of a STEM major in college remains
highly gender-dependent. Roughly 14.5% of female college students—as
compared to 32.9% of males—major in a STEM field, and 32.4% of female
students leave STEM to study unrelated fields, whereas 25.5% of males do
so (Chen, 2009). It is not surprising, then, that women comprise a small pro-
portion of fields like computer science, engineering, and physics (National
Science Board, 2016), which generally confer high earnings (Bureau of
Labor Statistics, 2013). Gender gaps in STEM education translate to dispro-
portionate gains in long-term economic mobility, earnings inequality, and
occupational segregation when students enter the labor market. While a vari-
ety of policies have been implemented to recruit more college students from
underrepresented groups into STEM, the question of persistence is left unan-
swered; how do female college students remain in and best progress
through the STEM pipeline? To address this question, this study examines
the impact of instructor gender on student motivation in STEM fields.
Recent literature suggests that female STEM success may be strongly
influenced by students’ educational experiences in college (Chang,
Sharkness, Hurtado, & Newman, 2014). Theoretical discussions have nomi-
nated a few reasons why gender may be relevant in this context. Prominent
among them is the idea that a student-instructor gender match can affect the
identity development process; the identity-based motivation (IBM) model in
particular provides context for understanding how identities are dynamically
constructed and made salient by situational cues, such as instructor gender.
In this paper, we use the IBM model as one way to explain why female
instructors have the potential to influence how female students interpret
their STEM experience, behave in their STEM classes, and respond to diffi-
culties. This theory suggests that a female instructor can help female students
shape and maintain a STEM-related identity (Gilmartin, Denson, Li, Bryant, &
Aschbacher, 2007), become more engaged during the learning process
(Schunk, 1995), and become confident that a future in STEM is attainable
(Oyserman, 2007).
A small but growing number of studies have examined the role of
instructor-student gender match in the college classroom (e.g., Bettinger &
Long, 2005; Canes & Rosen, 1995; Carrell, Page, & West, 2010; Hoffman &
Oreopoulos, 2009; Neumark & Gardecki, 1998; Robst, Keil, & Russo,
1998). With the exception of Carrell et al. (2010), who found substantial
same-gender effects, these studies generally found small positive effects
for having a same-gender instructor in regard to academic indicators of suc-
cess, such as course performance. While academic measures are important,
a rich body of literature also shows that an inviting learning context might
benefit students by improving their motivation, attitude, and engagement
without necessarily boosting their immediate course performance. Further,
improved motivation, engagement, and attitude in the college setting may
have an influential impact on students’ later academic and career choices
2
Same-Gender Instructors in College STEM Courses
(Karabenick & Urdan, 2014), therefore influencing their educational and
employment trajectories (Gutman & Schoon, 2013). The role of instructor-
student gender match therefore cannot be fully understood without explor-
ing course-specific motivation-related measures.
Two studies have looked beyond course grades to directly examine stu-
dent interest. Using a large college administrative data set from the state of
Ohio, Bettinger and Long (2005) examined whether same-gender instructors
impact student interest in a subject using academic-related measures, such as
additional credit hours in a subject and academic major. They found that
female instructors positively influenced these measures of interest in a num-
ber of disciplines. Similarly, Carrell et al. (2010) found that female instructors
positively influenced female students’ likelihood of taking future STEM
courses and likelihood of graduating with a STEM degree. While these
studies provide important initial information about how student-instructor
gender match may influence motivation-related measures, more direct meas-
ures of student motivation and course engagement are still needed to illumi-
nate how learning processes and attitudes may be influenced as a result of
instructor gender.
In this study, we build on existing literature and extend it by directly
examining the impact of student-instructor gender match on a myriad of
course-specific motivation-related measures. Drawing on a unique data set
that combines college-level student achievement and demographic data
with course survey data, this study examines the impact of instructor gender
on students’ self-reported values for behavioral engagement (measured
using class attendance, participation, and interpersonal interaction with
the instructor), interest in a subject, academic self-efficacy, and utility value
in a number of different STEM gateway courses. We choose to focus on gate-
way courses because students’ learning experience in these courses often
affects their interest and success in subsequent STEM learning, also influenc-
ing important academic decisions, such as major choice and even college
persistence (Chambliss & Takacs, 2014).
Our focus on course-specific motivation-related measures fills an impor-
tant gap in current literature that mainly uses academic performance as the
key outcome measure. Not only are nonacademic outcomes important meas-
ures of student experience and success in and of themselves, but they also
illuminate the channels through which instructor-student gender interactions
may influence the educational experience. Indeed, the larger constructs rep-
resented by our nonacademic measures have been associated with academic
persistence and positive post-college outcomes (Gutman & Schoon, 2013).
The major methodological challenge in estimating the impact of a same-
gender instructor on students’ learning outcomes is student self-selection
into courses taught by male and female instructors. To minimize selection
bias, we build on existing literature using a two-way fixed effects model
(e.g., Fairlie, Hoffman, & Oreopoulos, 2014; Hoffman & Oreopoulos,
3
Solanki, Xu
2009) that controls for both student-level fixed effects and class fixed effects
(i.e., a course-section level), therefore eliminating potential bias that is con-
stant at either the individual level (e.g., academic capacity) or classroom
level (e.g., instructional quality and grading criteria).
We begin our inquiry by examining whether introductory STEM courses
with female instructors narrow the gender achievement gap as measured by
course grades and the probability of successfully completing a course with
a grade of a C or better. As detailed in the following, our analysis yields
results similar to those of previous studies: Female students, on average,
are less likely to successfully complete a STEM gateway course than male
students, and the gender achievement gap shrinks in the presence of
a female instructor. These results motivate the main research question in
this study: Does having a same-gender instructor influence the gender gap
in regard to course-specific, motivation-related measures?
To address this question, we utilize self-reported student pre- and post-
course survey data measuring four important motivation-related items: stu-
dent course behavioral engagement, self-efficacy in a particular field, interest
in this field, and utility value of this field. Although we find that female stu-
dents tend to be less engaged and interested than male students in general,
the gender gap in course behavioral engagement and attitude narrows when
a female instructor teaches a STEM course.
In the following section, we introduce a theoretical framework as
a means of discussing relevant studies that link instructor gender to student
academic performance and motivation. Afterward, we describe our data and
research context, sample, and methodology, eventually presenting our
results and concluding with a discussion of our findings and their implica-
tions for policy and practice.
Theoretical Discussion and Empirical Evidence

for Student-Instructor Gender Match
Existing literature exploring the impact of gender on student learning
outcomes has furnished three major explanations as to how a female instruc-
tor might influence female students’ academic performance and motivation:
a role model effect, differences in teaching style, and teacher bias. We dis-
cuss these three explanations sequentially, review empirical evidence in sup-
port of each theory, and consider implications for our study.
Role Model Effects: Identity-Based Motivation Theory

The university experience is a crucial setting in which to study student
development and the influence of factors such as instructor gender.1
During this developmental time, students shape their identity (Erikson,
1968; Waterman, 1985) and envision their future selves while making
4
connections between the college classroom and the larger society it repre-
sents. The idea that instructor gender can contribute to a role model effect
for college students is rooted in identity-based motivation theory, which pos-
its that identities are not stable but dynamically constructed within contexts
that help individuals interpret experiences and react accordingly (Oyserman,
2007). When identity is made salient, often by situational cues such as
gender, it influences meaning-making—how one interprets his or her
experience—and impacts how one might behave in response to an identity
interpretation (Oyserman, 2007).
IBM theory helps us understand how a STEM course with a female
instructor has the potential to change aspects of students’ STEM identities.
The STEM classroom is a place where many aspects of identity are at play.
STEM classes are often male-dominated and taught by male professors. In
addition, male students generally earn higher grades, are more outspoken,
and are often perceived as being more knowledgeable in the STEM classroom
than their female counterparts (Grunspan et al., 2016). The classroom environ-
ment therefore provides a positive interpretation for male students: STEM is
associated with men. In addition, a male student has ample opportunity to
interact with and be influenced by professionals of the same demographic
group, thereby receiving more reassurance than a female that his imagined
future self—as a physicist or an engineer, for example—is attainable.
On the other hand, if a female student aims to become a physicist or an
engineer, she might come to question her goal in a male-dominated class-
room environment as she interprets the classroom in the same way as
a male student. A female instructor, however, can change this interpretation,
making it more likely that the female student will think STEM is for ‘‘people
like [her].’’ This new interpretation fosters identity congruence. The IBM
model posits that identity congruence has implications for action; if an action
feels identity-congruent, individuals are more likely to further engage in the
action, even in the face of difficulty. In fact, difficult tasks and challenging
circumstances are less likely to be interpreted as impossible but rather
important to one’s identity (Oyserman, 2014).
Prior studies have shown that improved identity compatibility is directly
linked to improved academic performance and increased motivation for
females (see London, Rosenthal, & Gonzalez, 2011; L. Rosenthal, London,
Levy, & Lobel, 2011; Settles, Jellison, & Pratt-Hyatt, 2009). They have also
demonstrated that identity compatibility protects against negative conse-
quences that arise from stereotype threat, defined as decreased performance
resulting from the fear of confirming a negative stereotype (see Marx &
Roman, 2002; McIntyre et al., 2005). This is an important finding given
that the stigma of females in STEM fields is often responsible for female
STEM attrition (Corbett, Hill, & St. Rose, 2008). Thus, if more STEM courses
were taught by female professors, female students could theoretically expe-
rience the identity-congruent benefits of being more engaged in STEM,
5
Solanki, Xu
exerting more effort to overcome academic challenges, and generally per-
sisting more in moving toward STEM-related goals.
Studies indirectly testing IBM theory and role model effects have been con-
ducted extensively in the K–12 context, and results vary (e.g., Antecol, Eren, &
Ozbeklik, 2015; Dee, 2005; Ehrenberg, Goldhaber, & Brewer, 1995; Lavy, 2004;
Nixon & Robinson, 1999; Zirkel, 2002). For example, Dee’s (2005) seminal work
found that middle school girls were more likely to report that they did not look
forward to science class and/or did not think science would be useful for their
future when they had a female science teacher. On the other hand, Zirkel
(2002) found that middle school students with at least one race- and gender-
matched role model had better long-term academic outcomes, reported more
achievement-oriented goals, enjoyed achievement-related activities to a greater
degree, and thought more about their future when compared to students with-
out a race- and gender-matched role model.
Given recent efforts to diversify faculty within college departments,
a small but growing body of literature has focused on instructor-student gen-
der interaction at the postsecondary level (e.g., Bettinger & Long, 2005;
Canes & Rosen, 1995; Carrell et al., 2010; Hoffman & Oreopoulos, 2009;
Neumark & Gardecki, 1998; Robst et al., 1998). In terms of academic
achievement measures, results are mixed, with some studies finding a posi-
tive impact for initial course performance (e.g., Carrell et al., 2010) and
others finding a null effect (e.g., Hoffmann & Oreopoulos, 2009).
Unfortunately, due to data limitations, the majority of gender match studies
focus only on academic performance and do not contain measures directly
related to motivation. Thus, the motivation-related mechanisms driving stu-
dent academic success are, at this point, ambiguous.
Teaching Style
Instructional technique can also explain the impact of student-instructor
interaction on achievement and engagement. Indeed, prior research has
shown that no single approach to teaching works for every student, partic-
ularly in STEM instruction (Carrell et al., 2010), and that teaching preferences
are different for each gender. For example, in one study, researchers found
that male college students in a STEM lab course preferred multimodal
instruction, whereas most female students in the class preferred single-
mode instruction utilizing a ‘‘learning by doing’’ approach (Wehrwein,
Lujan, & DiCarlo, 2007). Learning preferences have consequences for stu-
dent success as students are more likely to be engaged in courses that
employ their preferred instructional techniques and strong engagement
can lead to improved performance (Ferrara, 2012).
While we do know that student gender matters in terms of preferred
instructional approach, less is known about whether teaching style is a prod-
uct of instructor gender. In one study, female instructors were found to often
6
use interactive teaching techniques, such as class discussion, small group
discussion, and group projects. Male instructors, on the other hand, used
less interactive approaches, tending to rely more heavily on lecture
(Starbuck, 2003). Current evidence suggests that pedagogical gender differ-
ences like these do have implications for student learning; specifically,
female students prefer certain types of teaching styles that are associated
with instructor gender (Carrell et al., 2010).
Teacher Bias
Another student-instructor factor that might influence student learning
and engagement outcomes is teacher bias, whereby teachers may treat their
students differently based on gender. Literature has mainly looked at teacher
bias through the lens of expectations, often citing the Pygmalion effect, or the
theory that people perform better when more is expected of them (R.
Rosenthal & Jacobsen, 1968). For example, males are often stereotyped as
being better at math than females, and STEM teachers might therefore have
higher expectations for them, interact more with them, and overestimate their
performance. These actions could improve male achievement and engage-
ment while at the same time creating challenges for females in STEM disci-
plines (see Hyde & Jaffee, 1998; Rebhorn & Miles, 1999; Sadker & Sadker,
1985). Additionally, in a recent study, Lavy and Sand (2015) compared the
scores of middle schoolers’ math tests with the scores of the same tests graded
by external teachers who were assumed to be completely unaware of student
gender. The students’ teachers tended to give female students lower scores
and male students higher scores than the external group of teachers.
The possible existence of teacher bias explains why academic outcomes
alone are insufficient for understanding both the experience of students in
STEM courses and the impact of instructor gender. For example, if a teacher
is more likely to give higher grades to a student who shares his or her gen-
der, a positive estimate of student-instructor gender match regarding test
scores would not translate into a positive learning experience for female stu-
dents; it would instead simply reflect teachers’ response to students’ gender
in assigning a grade. Existing studies have addressed teacher grading bias by
exploring long-term outcomes, such as students’ subsequent enrollment and
performance in STEM courses and their likelihood of graduating with
a STEM degree (e.g., Bettinger & Long, 2005; Carrell et al., 2010). Our study
contributes new evidence to existing literature by directly examining course-
specific motivation-related measures.
The Current Study

Overall, there is limited evidence about the effect of student-instructor
gender match at the higher education level. Even among studies that identify
a positive impact on student academic performance, researchers fail to reach
7
Solanki, Xu
a consensus on the mechanisms driving these results. One important reason
for conflicting interpretations is a lack of ability—given limited data—to look
beyond academic-related outcomes and directly measure student behavior,
engagement, and motivation.
Our study builds on previous studies by providing a current perspective
on the educational relevance of gender in the college setting that is congru-
ent with the experiences of many U.S. college students. Most importantly, it
fills a gap in educational literature with an examination of the influence of
same-gender instructors on nonacademic outcomes using a unique data
set that includes pre– and post–self-reported motivation-related variables
used as key dependent variables. To our knowledge, a thorough examina-
tion of the connection between instructor gender and student motivation
in the college context is unstudied in prior literature, thereby substantiating
the purpose of this study.
Study Background
Data and Research Context
Our analysis is based on detailed student and administrative data from the
University of California, Irvine (UCI), a large public university in California. UCI
is part of the University of California system, which consists of 10 campuses
with a total enrollment of approximately 200,000 undergraduate students in
the 2013 school year. UCI enrolled approximately 12.5% of these students.
The UCI student population consists of approximately 45% Asians and
30% Caucasians, with roughly a quarter of the student body belonging to
an underrepresented ethnic or racial minority group (URM). Of UCI’s
25,000 undergraduate students in the 2013 school year, more than half
were female, and 30% of degrees awarded in 2013 were in a STEM field.
Beginning in the spring of 2013 and continuing for the duration of three
academic quarters, a team of researchers at UCI collected the data set used in
this paper (Denaro, Reimer, Solanki, Xu, & Warschauer, 2018), observing
full- and part-time undergraduate university students enrolled in 23 different
STEM courses. These courses include introductory courses in the field of bio-
logical sciences (e.g., DNA to organisms), chemistry (e.g., general chemis-
try), engineering (e.g., introductory engineering), information and
computer sciences (e.g., introductory programming), math and statistics
(e.g., introduction to linear algebra), and physical sciences (e.g., Classical
physics), all of which were prerequisites for other mandatory courses in
one or more STEM majors and were offered in multiple sections during
the year. Each course typically enrolled 200 or more students.
Supplementary Table S1 in the online version of the journal provides
descriptive statistics for the list of courses used in this study.
8
Table 1
Summary Statistics at the Student-Class Level
Panel A: Student Demographic Characteristics in Male Versus Female-Taught Classes
(1) (2) (3) (4)

Full Male-Taught Female-Taught t Test of
Sample Class Class Difference
Female 0.531 0.481 0.595 –16.157

White 0.127 0.135 0.118 3.506
Black 0.015 0.016 0.015 0.729
Hispanic 0.200 0.195 0.206 –1.961
Asian 0.501 0.491 0.514 –3.257
Other 0.063 0.059 0.068 –2.236
Nonresident 0.092 0.103 0.079 5.891
First-generation status 0.536 0.536 0.538 –0.357
Low-income status 0.380 0.376 0.388 –1.794
SAT-math 626.51 633.254 618.118 12.654
SAT-verbal 561.59 560.879 562.476 –1.153
High school GPA 3.91 3.908 3.915 –1.891
N 20,209 11,286 8,923
Panel B: Student Demographic Characteristics by Student Gender in All Classes, Male-Taught Classes, and Female-Taught Classes
All Classes Male-Taught Classes Female-Taught Classes
(1) (2) (3) (4) (5) (6) (7) (8) (9)

Male Female
Student Student t Test Male Female Male Female
Sample Sample of Sample Sample t Test of Sample Sample t Test of
Mean Mean Difference Mean Mean Difference Mean Mean Difference
White 0.141 0.115 5.467 0.152 0.117 5.525 0.124 0.114 1.312
Black 0.013 0.017 –2.498 0.013 0.019 –2.469 0.013 0.016 –1.096
Hispanic 0.194 0.205 –1.866 0.196 0.195 0.080 0.193 0.215 –2.583
(continued)
9
Table 1 (continued)
10
Panel B: Student Demographic Characteristics by Student Gender in All Classes, Male-Taught Classes, and Female-Taught Classes
All Classes Male-Taught Classes Female-Taught Classes
(1) (2) (3) (4) (5) (6) (7) (8) (9)

Male Female
Student Student t Test Male Female Male Female
Sample Sample of Sample Sample t Test of Sample Sample t Test of
Mean Mean Difference Mean Mean Difference Mean Mean Difference
Asian 0.476 0.522 –6.635 0.460 0.524 –6.836 0.502 0.522 –1.772
Other 0.063 0.063 0.047 0.060 0.058 0.460 0.068 0.068 –0.010
Nonresident 0.111 0.075 8.759 0.119 0.087 5.571 0.100 0.065 6.098
First-generation status 0.515 0.555 –5.624 0.515 0.558 –4.499 0.516 0.553 –3.408
Low-income status 0.349 0.409 –8.753 0.346 0.407 –6.661 0.354 0.411 –5.465
SAT-math 647.09 608.84 32.572 652.289 613.399 25.252 638.863 604.211 18.983
SAT-verbal 564.97 558.68 4.863 565.153 556.421 5.019 564.697 560.986 1.892
High school GPA 3.88 3.93 –12.572 3.886 3.931 –9.379 3.889 3.932 –8.152
N 9475 10,734 5,857 5,429 3,618 5,305
Detailed student-level information, collected from the UCI Office of
Institutional Research (OIR), includes demographic data regarding gender,
ethnicity, age, first-generation status, and low-income status; academic prep-
aration variables, such as SAT scores and high school GPA; and postsecond-
ary data, such as the type of courses students take during a particular quarter,
number of credits students take in a quarter, undergraduate GPA, and pro-
grams of study. For the purpose of this study, student administrative records
were matched to courses in the data set, which include information about
instructor name, gender, and academic rank.
Motivation-Related Measures
During the year of data collection, pre- and post-experience surveys
were administered in students’ STEM courses. These surveys were adminis-
tered during Weeks 2 and 9 of each 11-week quarter and were completed by
approximately 65% of the students in our sample.2 The survey items repre-
sent different motivational constructs, such as interest value, utility value,
and self-efficacy; they also identify students’ level of course engagement
as measured by students’ reports of attending class, listening attentively to
lectures, engaging in formal help-seeking behavior, and participating in class
discussion. These items were standardized by course and used in our anal-
ysis as outcome variables.
Specifically, the survey included 25 items designed to measure students’
experience in a course and motivation in regard to the course subject; 12
items were not relevant to our study and therefore not included.3 The items
used in our study are divided into four major categories: behavioral engage-
ment, academic self-efficacy, interest, and utility value. We include details for
each category and a Cronbach’s internal reliability alpha score in the follow-
ing. This score is an index of internal consistency reliability that assesses the
degree to which responses are consistent across a set of multiple measures
of the same construct—usually for self-reported items (Warner, 2013).
Behavioral Engagement
This category includes four post-experience survey items. Specifically,
students were asked, ‘‘During this course, about how often have you done
the following: (i) attended lectures, (ii) listened attentively to lectures, (iii)
asked the professor or TA for help in this class, and (iv) asked questions
and contributed to the discussion in lecture?’’ Students were asked to indi-
cate their responses on a 5-point Likert scale, with 1 = never, 2 = rarely,
3 = sometimes, 4 = often, and 5 = always. The Cronbach’s internal reliability
alpha is equal to 0.62.
11
Solanki, Xu
Academic Self-Efficacy
In the pre- and post-experience surveys, students were asked to respond
to the following three items using a 5-point Likert scale ranging from 1 = not
at all true to 5 = very true: (i) ‘‘I’m certain I can master the skills taught in this
class,’’ (ii) ‘‘I’m certain I can figure out how to do the most difficult course
material,’’ and (iii) ‘‘I can do almost all the work in the class if I don’t give
up.’’ The Cronbach’s internal reliability alpha is equal to 0.86.
Interest
In the pre- and post-experience surveys, students were asked to respond
to the following three items using a 5-point Likert scale ranging from 1 = not
at all true to 5 = very true: (i) ‘‘I find many topics in this course to be inter-
esting,’’ (ii) ‘‘Solving problems in this class is interesting for me,’’ and (iii) ‘‘I
find this class intellectually stimulating.’’ The Cronbach’s internal reliability
alpha is equal to 0.89.
Utility Value
In the pre- and post-experience survey, students were asked to respond
to the following two items: (i) ‘‘Having a solid background in the material
taught in this course is worthless’’ and (ii) ‘‘After I graduate, an understand-
ing of the material in this course will be useless to me.’’ The items were
reverse-coded so that on a 5-point Likert scale, 1 = very true and 5 = not
at all true. Cronbach’s internal reliability alpha is equal to 0.73.
Sample and Summary Statistics

The analytic sample consists of 20,209 course observations from 9,766
students who were enrolled in STEM courses at UCI. Table 1 presents sum-
mary descriptive statistics at the student-class level. As shown in Panel A
(Column 1), compared to the general UCI student population described pre-
viously, students in our analytical sample consist of more Asian students,
a smaller proportion of students from URM groups, and a slightly smaller
proportion of females.
Table 1 Panel A also documents differences in student baseline charac-
teristics between male- and female-taught classes (Columns 2 and 3). While
a number of these differences do not reach statistical significance, female-
taught classes include a larger proportion of female students than male-
taught classes. In addition, the average SAT math score is lower in female-
taught classes.
Panel B includes separate summary baseline characteristics for the male
and female student samples (Columns 1 and 2). Compared to the students in
our male-only sample, a larger proportion of female students are categorized
as both first-generation status and low-income status. Our data also show
12
a gender gap in prior math achievement, with a male student average SAT
math score equal to 647.1 and female student score equal to 608.8. The
38.3-point difference between these categories is statistically significant.
Lastly, Panel B includes demographic characteristics, sorted by student
gender, for female- and male-taught classes (Columns 4–9). A particularly
notable point illuminated by Panel B is that the gender gap in prior math
achievement persists regardless of instructor gender. However, it is most pro-
nounced in male-taught classes. Lastly, the gender gap regarding high school
GPA is almost the same in male-taught and female-taught classes regardless
of instructor gender.4
Table 2 Panel A documents differences, sorted by student gender, for
our two main academic outcomes and the full set of nonacademic outcomes.
The first academic outcome listed is course grade. Course grades are deter-
mined on a 4-point scale whereby an A is 4 points, an A– is 3.7 points, a
B1 is 3.3 points, and so on. An F is 0 points. The academic gender achieve-
ment gap is equal to 0.16 grade units; as shown in Table 2 Panel A, this
difference—although small—is statistically significant. This academic gender
gap persists regardless of instructor gender (Panels B and C), although it is
smaller in magnitude in female-taught classes. Also, as shown in Table 2,
male students seem to be subject to a performance decrement when in
a female-taught class (Mfemale = 2.57), as compared to a male-taught class
(Mmale = 2.67), while the average performance of female students seems to
be fairly stable (Mmale = 2.48 vs. Mfemale = 2.47). In all subsequent regression
analyses, course grades are standardized by course so that grade changes
can be understood as effect sizes.
Table 2 also documents each gender’s likelihood of receiving a C or bet-
ter in STEM classes. We chose C or higher as a threshold mainly because
dropping below a C average places students on academic probation at
most universities, with academic probation being a strong predictor of col-
lege attrition (Seymour & Hewitt, 1997). Both male and female students
are more likely to receive at least an average grade (C or higher) with
male instructors than they are with female instructors. However, female stu-
dents are less likely than male students to receive a C or better in general
(Panel A), and the gender achievement gap this outcome represents is pres-
ent irrespective of instructor gender (Panels B and C).
As mentioned previously, Table 2 documents differences, sorted by gen-
der, for our 13 motivation-related variables. All variables have been stan-
dardized to have a mean of zero and a standard deviation of one within
a course. The descriptive evidence provided in Panel A illustrates gender
gaps (favoring males) for our motivation-related items; in fact, male students
reported higher values for three of four items measuring behavioral engage-
ment—and for all items measuring interest and self-efficacy—than female
students. For the most part, these differences are statistically significant.
13
Table 2
14
Summary Statistics for Outcome Measures
Panel A: All Classes

Full Sample Male Sample Female Sample t Test of Difference
Academic outcomes
Course grade 2.547 2.631 2.474 10.505
C or better 0.855 0.865 0.847 3.766
Nonacademic outcomes
Behavioral engagement items
Attend lectures 0 –0.081 0.052 –6.899
Listen attentively to lectures 0 0.005 –0.003 0.432
Ask the professor or TA for help in this class 0 0.003 –0.002 0.267
Ask questions and contribute to class discussion 0 0.085 –0.055 7.205
Interest items
I find many topics in this course to be interesting 0 0.095 –0.062 8.071
Solving problems in this class is interesting for me 0 0.109 –0.071 9.315
I find this class intellectually stimulating 0 0.051 –0.033 4.360
Self-efficacy items
I’m certain I can master the skills taught in this class 0 0.202 –0.131 17.383
I’m certain I can figure out how to do the most difficult course material 0 0.214 –0.139 18.431
I can do almost all the work in the class if I don’t give up 0 0.077 –0.050 6.551
Utility value items
Having a solid background in the material taught in this course is worthless 0 –0.125 0.084 10.46
(reverse-coded; higher values indicate not at all true)
After I graduate, an understanding of the material in this course will be useless to me 0 –0.112 0.074 9.691
Panel B: Female-Taught Classes
Academic outcomes
Course grade 2.509 2.567 2.469 4.208
C or better 0.839 0.840 0.837 0.354
Attend lectures 0.052 –0.025 0.094 –4.444
(continued)
Table 2 (continued)
Panel B: Female-Taught Classes
Listen attentively to lectures 0.028 0.046 0.018 1.008

Ask the professor or TA for help in this class 0.005 –0.008 0.012 –0.729
Ask questions and contribute to class discussion –0.011 0.063 –0.053 4.112
Interest items
I find many topics in this course to be interesting –0.020 0.064 –0.066 4.614
Solving problems in this class is interesting for me –0.021 0.102 –0.089 6.800
I find this class intellectually stimulating –0.011 0.043 –0.042 3.015
Self-efficacy items
I’m certain I can master the skills taught in this class –0.021 0.181 –0.133 11.224
I’m certain I can figure out how to do the most difficult course material –0.022 0.196 –0.143 12.116
I can do almost all the work in the class if I don’t give up –0.027 0.047 –0.069 4.099
Utility value items
After I graduate, an understanding of the material in this course will be useless 0 –0.160 0.062 7.903
to me (reverse-coded; higher values indicate not at all true)
Panel C: Male-Taught Classes
Academic outcomes
Course grade 2.578 2.671 2.479 9.772
C or better 0.869 0.881 0.856 3.967
Attend lectures –0.049 –0.126 0.008 –4.775
Listen attentively to lectures –0.027 –0.028 –0.027 –0.040
Ask the professor or TA for help in this class –0.005 0.012 –0.017 1.112
Ask questions and contribute to class discussion 0.011 0.102 –0.057 5.929
Interest items
I find many topics in this course to be interesting 0.019 0.120 –0.057 6.564
Solving problems in this class is interesting for me 0.020 0.115 –0.052 6.210
I find this class intellectually stimulating 0.011 0.058 –0.024 3.045
(continued)
15
16
Table 2 (continued)
Panel C: Male-Taught Classes
Self-efficacy items
I’m certain I can master the skills taught in this class 0.020 0.218 –0.128 13.162
I’m certain I can figure out how to do the most difficult course material 0.021 0.228 –0.134 13.753
I can do almost all the work in the class if I don’t give up 0.026 0.100 –0.029 4.881
Utility value items
After I graduate, an understanding of the material in this course will be useless to 0 –0.075 0.088 6.097
me (reverse-coded; higher values indicate not at all true)
The difference between male and female students’ reported values is
particularly pronounced for items measuring self-efficacy. In fact, for the
item ‘‘I’m certain I can figure out how to do the most difficult course mate-
rial,’’ the raw difference is 0.35 standard deviation (SD) units. The gender dif-
ference in reported values for the items measuring interest ranges from 0.08
to 0.18 SD units. We also find a significant difference in reported behavioral
engagement values for the item ‘‘asked questions/contributed to class dis-
cussion.’’ Lastly, we note that female students reported higher values than
male students for all items measuring utility value, indicating that descrip-
tively, female students find STEM course material more useful for their future
than male students do.
When we examine gender gaps in our motivation-related items that are
sorted by instructor gender (Panels B and C), many of the patterns discussed
previously persist. However, the overall gender motivation gap is smaller in
female-taught classes in regard to a number of these items.
Table 3 presents summary descriptive statistics for our STEM courses.
Irrespective of gender, each student took an average of 2.1 courses.
Twenty-one percent of females took classes in biology, whereas 12% of
males did so. Roughly 54% of females and 41% of males took classes in
chemistry. Percentages for female and male involvement in mathematics
and statistics classes are roughly the same. Substantially fewer females
took computer science and engineering classes than males. These gender
differences in course-taking are similar to what national data indicate
(Chen, 2009). It is important to note that there is also gender disparity in
instruction; 39% of instructors across all courses in our data set are female
(34% are female, on average, per course). The data set includes 73 sections
(a section is defined as a course delivered at a specific time for which stu-
dents meet with an instructor; it is equivalent to a ‘‘class’’) within 23 courses
over three academic quarters, with three sections, on average, per course.
Estimation Strategy
Basic Ordinary Least Squares Estimation
In this study, we explore whether moving from a male instructor to
a female instructor improves female students’ performance and motivation
in STEM-related gateway courses relative to male students (i.e., the gender
gap). We begin with a basic ordinary least squares (OLS) model (Equation 1):

Yijkst 5 b0 1 b1 Female instructorj 1 b2 ðFemale studenti Þ 1
: ð1Þ
b3 Female instructorj 3 Female studenti 1 Xi 1Rankj 1 dt 1 uijkst
Our data are organized at the student by course section level. Yijkst rep-
resents outcome Y for student i in course k section s, with instructor j in term
(i.e., quarter) t. Female_instructorj is dichotomous and equal to 1 if
17
Solanki, Xu
Table 3
Summary Statistics Describing (a) the Number of Students Enrolled in Courses
by Department, Student-Class Level and (b) the Number of Sections (n = 73)
Taught by Females Within Each Department, Instructor-Class Level
Students Instructors
Full
Sample Female Male Female Male
Percent Percent Percent Percent Percent
(n = 20,209) (n = 10,734) (n = 9,475) (n = 32) (n = 42)
Biology 0.17 0.21 0.12 0.22 0.10

Chemistry 0.48 0.54 0.41 0.41 0.40
Engineering 0.06 0.03 0.11 0 0.17
Information and 0.07 0.04 0.11 0 0.10
computer sciences
Math/statistics 0.16 0.14 0.18 0.28 0.17
Physics 0.05 0.02 0.07 0.06 0.07
Average number of 2.08 2.11 2.07
courses taken
instructor j is female and 0 if instructor j is male. Female_studenti is dichoto-

mous and equal to 1 if student i is female and 0 if student i is male. Xi is a vector
of student control variables and includes student ethnicity, first-generation sta-
tus, low-income status, SAT math score, SAT verbal score, and high school GPA.
Rankj is an indicator of academic rank of instructor j. dt represents term fixed
effects and reflects the specific quarter when a student took a course.
The interpretation of Equation 1 relies on three key estimates. Using
standardized course grade as an example, b1 represents the average differ-
ence in standardized course grade for male students with a female instructor
relative to male students with a male instructor. b2 represents the average
changes in standardized course grade for female students—relative to
male students (i.e., the gender achievement gap)—with a male instructor.
b2 1 b3 represents the average changes in standardized course grade for
female students—relative to male students (i.e., the gender achievement
gap)—with a female instructor. The interactional effect (b3) therefore repre-
sents the decrease (or increase) in the gender achievement gap and is the
main parameter of interest.
Two-Way Fixed Effects Model

Using a basic OLS model to estimate an unbiased interaction effect
between female students and instructors is quite challenging given the
18
nonrandom assignment of students to instructors. For example, we might be
concerned that students who prefer taking a course with a female instructor
have certain characteristics that also correlate to their potential course per-
formance. If any of these characteristics (e.g., motivation and ability) are
absent from the data set, the estimates from a basic OLS model are biased.
In this case, a positive interaction effect might be due to student sorting
between male and female instructors and not necessarily a student-instructor
gender match.
Bias could also arise from differences between classes.5 For example,
the classroom environment and student-instructor dynamics in general
chemistry taught in the fall by female Professor X may be different from
those in general chemistry taught during the same term by male Professor
Y. Even within the same course, female instructors might be different from
male instructors in terms of their instructional approaches and grading pol-
icies, and these differences may impact student learning and engagement.
However, since our key variable of interest is the interaction between stu-
dent and instructor gender, we can employ a class fixed effects term to con-
trol for variation across classes within a particular course.
As such, to address potential biases at either the student level or the class
level, we take advantage of the panel data structure in our study—multiple
observations per student—which enables us to employ a two-way fixed
effects model (Fairlie et al., 2014).
Yic 5b3 ðFemale instructorc 3 Female studenti Þ 1 gi 1 Fc 1 Fkg 1 uic : ð2Þ
Compared to Equation 1, Equation 2 includes two additional sets of var-

iables: gi and Fc, which are fixed effects terms for student and class, respec-
tively. gi controls for both observable and unobservable student-level
characteristics that are constant for an individual (e.g., ethnicity and aca-
demic motivation).6 This term does not, however, control for how student
achievement motivation applies to some courses but not others.
Fc represents class fixed effects that control for observable or unobserv-
able class-level characteristics, such as class size, delivery format, grading
policies, term, and time of day. It is worth noting that this specification
also implicitly controls for characteristics that are fixed at the instructor level,
such as instructor quality and instructional approach. Another advantage of
including a class fixed effects term, as explained in more detail in Fairlie et
al. (2014), is that it implicitly standardizes testing procedures across student
groups we compare since all students are subjected to the same class char-
acteristics mentioned previously. It is also worth noting that the main effect
for instructor variables (i.e., gender and rank) and student gender has been
removed due to controlling for class fixed effects and student fixed effects,
respectively. Lastly, Fkg are course by gender fixed effects, which allows
gender differences to vary across courses.
19
Solanki, Xu
b3 in Equation 3 represents whether the gender achievement and/or
motivation gap narrows in a female-taught class compared to a male-taught
class. A positive b3 indicates that the gender gap narrows in the presence of
a female instructor. As noted in our earlier theoretical discussion, a positive
b3 for student course grade does not uniquely identify a role model effect,
defined as a social psychological effect in which, for example, female stu-
dents are encouraged as a result of simply observing a female instructor.
Instead, a significant interaction between student and instructor gender
could be the result of any number of mechanisms, including not only the
role model effect but also favoritism or hostility in assessing male or female
students. Additionally, other mechanisms driving an interaction effect might
include instructors providing different levels of attention, advice, and
encouragement to students based on their gender. For example, male pro-
fessors might encourage and elicit participation from male students (e.g.,
calling on male students more frequently than female students); female
instructors might provide different levels of attention and advice during
office hours to female students as a means of encouraging them.
Examining course-specific motivation-related measures can shed light on
the mechanisms that may be related to situations like these.
Concerns About Differential Sorting

While the two-way fixed effects model is able to address many threats to
the internal validity of our study, we are still concerned about differential
sorting. The two-way fixed effects estimates would be biased if, for example,
highly motivated female students are more inclined to enroll in classes
taught by female instructors and, similarly, highly motivated male students
in classes taught by male instructors. If that were the case, an interaction
effect, indicated by a positive b3 coefficient, might be due to students’
endogenous sorting rather than positive effects of student-instructor gender
match.
To examine the extent of endogenous sorting, we conducted a sorting
test similar to that used in Fairlie et al. (2014). The main goal of the sorting
test is to examine whether observable female and male student characteris-
tics tend to be different in female-taught versus male-taught classes.
Specifically, to conduct this test, we first aggregated the data at the class level
and calculated gender-specific averages of the following baseline character-
istics for female and male students separately for each class: SAT score, high
school GPA, minority status, race, low-income status, and first-generation
status. These gender-specific averages (denotedXgc ) serve as the dependent
variable in Equation 3. We then use a difference-in-difference model to
examine whether the average gender difference in any of the student char-
acteristics mentioned previously in a class varies depending on the gender of
the instructor teaching that class:
20
Table 4
Sorting Regressions
High First- Low-

SAT- SAT- School Non- Generation Income
Math Verbal GPA URM White Asian Other resident Status Status
Interaction 1.110 3.835 0.008 0.012 0.019 –0.014 –0.009 –0.008 0.001 0.003
(2.962) (4.149) (0.012) (0.019) (0.011) (0.029) (0.011) (0.020) (0.023) (0.016)
Note. Each cell is associated with a different regression; column titles indicate the outcome
variable. We present the interaction term Female Instructor 3 Female Student, which is
interpreted as the extent to which female students sort into courses taught by female
instructors. Each regression includes course fixed effects and term fixed effects.
Standard errors are in parentheses and clustered at the course level. URM = underrepre-
sented ethnic or racial minority group.
Xgc 5 d1 3 Female instructorc 1 d2 3 Ig 1 d3 3 Female instructorc 3 Ig 1 ygc ; ð3Þ
where Ig is a dummy variable equal to 1 if the average is computed for

female students and 0 if it is computed for male students. d3 is the parameter
of interest, which represents whether gender gaps in the baseline character-
istics listed previously vary across female- and male-taught classes.
Table 4 presents the results of our sorting regression.7 All standard errors
are clustered at the course level. As shown in Table 4, none of our estimates
are statistically significant; thus, we do not find evidence of differential
sorting.
Concerns With Survey Data

Given that our study is focused on student nonacademic outcomes
retrieved from survey data, similar to the previous discussion, a potential
concern is endogenous sorting in responding to the survey. We might be
concerned that instructor-student gender interaction influences a student’s
probability of completing a course survey. For example, if more engaged
and motivated students are more likely to respond to a course survey in clas-
ses taught by a same-gender instructor, a positive interaction effect for stu-
dent motivation and engagement measures might be due to student
sorting between male and female instructors in completing a course survey,
thereby not necessarily capturing students’ actual engagement and motiva-
tion in a course with a same-gender instructor.
To investigate this issue, we first examine descriptive characteristics
comparing survey respondents to nonrespondents in Supplementary Table
S2 (in the online version of the journal). While respondents are more likely
to be female, Asian, and low-income than nonrespondents, the differences
between respondents and nonrespondents won’t bias our estimates unless
there is a student-instructor gender match effect for students’ probability
21
Solanki, Xu
Table 5
Estimated Role of Female Instructor Status for Student Academic Outcomes
Standardized Course Grade Receiving a C or Better

(1) (2) (3) (4)
Female instructor –0.065** — –0.054*** —

(0.032) — (0.013) —
Female student –0.252*** — 0.020 —
(0.076) — (0.026) —
Interaction 0.007 0.034 0.023* 0.023
(0.032) (0.050) (0.012) (0.016)
Course fixed effects X X
Class fixed effects X X
Student fixed effects X X
Student controls X X
Note. Each cell reports the coefficients from a separate linear probability regression. The
standardized grade has a mean zero and standard deviation one. All models include a con-
trol for instructor academic rank and term fixed effects. Student controls are ethnicity, first-
generation status, low-income status, SAT verbal score, SAT math score, and high school
GPA. Missing values have been adjusted using a dummy variable approach. The sample
size is 19,672. The interaction term represents a female student by female instructor inter-
action. Standard errors are in parentheses and clustered at the class level.
*p \ .10. **p \ .05. ***p \ .01.
of responding to a survey. Thus, we ran a series of regression models, using

Equation 2, to test for this possibility. In these models, responding to a survey
was coded as 1. Our results are listed in Supplementary Table S3 (in the
online version of the journal); the response rate gender gap does not vary
based on instructor gender (Row 3). This null effect alleviates the concern
that our survey response rate might be biased due to the presence of
a same-gender instructor.
Results
Student Achievement Outcomes
Table 5 presents our results estimating the impact of female instructors on
the gender achievement gap in college-level STEM courses. Model 1 uses
a course fixed effects term and all available student-level control variables.8
Model 2 represents our two-way fixed effects model using a class fixed effects
and student fixed effects term. Standard errors are clustered at the class level.9
Our findings are consistent with prior work on gender match, although
our point estimates are not always significant. As shown in Column 1, the
main effects of instructor gender (Row 1) echo the descriptive information
22
presented earlier. Specifically, male students receive lower grades in courses
taught by female instructors than courses taught by male instructors. On
average, taking a course with a female instructor reduces the standardized
course grades of male students by a statistically significant amount of 0.07
SD units. Prior studies have reported point estimates ranging from 0.04 to
0.08 SD units (Carrell et al., 2010; Hoffman & Oreopoulous, 2009).
The second row of Table 5 displays the b2 coefficients, which, as noted
earlier, represent the gender achievement gap in male-taught classes. We
find a significant gender achievement gap in courses taught by male instruc-
tors. In Column 1, which controls for both course fixed effects and student-
level controls, the gender achievement gap is equal to 0.25 SD units. On a 0-
to 4-point scale, this approximately represents the change from a B1 to a B,
for example.
One limitation of using standardized course grade as a continuous vari-
able is that it does not fully capture the possible impact of instructor-student
gender match for the full distribution of letter grades. Considering that
receiving a C or better is particularly important for students’ academic prog-
ress and success (e.g., maintaining a C average in college-level courses is
necessary for avoiding academic probation; a quarterly GPA that falls below
a C average can disqualify students from some majors as well), Table 5
presents students’ probability of receiving a C or better in a course.10
Although small in magnitude, the b3 coefficient representing the interac-
tion term is positive across both model specifications and equal to 0.02 per-
centage points. This indicates that the gender achievement gap for the
probability of receiving a C in a STEM course narrows in female-taught clas-
ses. This finding mainly reflects the fact that male students are less likely to
receive a C or better in female-taught classes, although female students
experience a very small absolute gain as they move from a male- to
a female-taught class. The point estimate in Row 1, Column 3 indicates
that male students are 5 percentage points less likely to receive a C or better
in female-taught classes than male-taught classes.
Heterogeneous Treatment Effects

Prior research has found the effects of instructor gender to be more pro-
nounced for highly skilled female students (as measured by SAT math score)
(see Carrell et al., 2010). We follow this line of inquiry and explore the val-
idity of a common treatment effect across the student sample by investigat-
ing whether the effects of instructor gender differ by student achievement
level. In this analysis, we first divide our sample of students into three cate-
gories based on SAT math score: highest scoring students (with a score
greater than or equal to 660), middle scoring students (with a score greater
than or equal to 590), and lowest scoring students (with a score less than
590). Next, we examine student performance for each subgroup using
23
Solanki, Xu
Table 6
Estimated Role of Female Instructor Status for Student Academic Outcomes,
SAT High Scoring Students Only
Standardized Course Grade Receiving a C or Better

(1) (2) (3) (4)
Female instructor –0.088** –0.041***

(0.037) (0.013)
Female student –0.773*** –0.091***
(0.116) (0.032)
Interaction 0.076 0.073 0.049** 0.041*
(0.059) (0.060) (0.021) (0.023)
Course fixed effects X X
Class fixed effects X X
Student fixed effects X X
Student controls X X
Note. Each cell reports the coefficients from a separate linear probability regression. The
standardized grade has a mean zero and standard deviation one. All models include a con-
trol for instructor academic rank and a term fixed effects. Student controls are ethnicity,
first-generation status, low-income status, SAT verbal score, SAT math score, and high
school GPA. Missing values have been adjusted using a dummy variable approach. The
sample size is 6,312. The interaction term represents a female student by female instructor
interaction. Standard errors are in parentheses and clustered at the class level.
*p \ .10. **p \ .05. ***p \ .01.
models that include (1) course FE and student control variables and (2) class
FE and student FE. The results are presented in Table 6.
Among the three different performance subgroups, we find treatment
effects only for the highest scoring students; the results for this particular
group are consistent with the full sample analysis results in Table 5. For
example, we find no significant interaction terms (Row 3) for the standard-
ized course grade outcome measure. We do, however, find that the achieve-
ment gap narrows in female-taught courses in regard to our second
achievement outcome, earning a C or better in a STEM class, as indicated
by the positive b3 coefficient (Columns 3 and 4). It is important to note
that the estimated effect for this subgroup of students (b3 = 0.05) is almost
twice as large as the effect for the average student. It therefore seems that
the gender interaction effect observed in Table 5 is driven mainly by students
with the highest SAT scores. This outcome is similar to that of Carrell et al.
(2010), which found the most significant effects for female students with
the strongest preexisting math skills.
24
Student Course-Specific Motivation Outcomes
Our first step in this paper was to estimate the gender achievement gap in
classes taught by male and female instructors. While we do not find evidence
to suggest that having a same-gender instructor has any impact in terms of
average course grade, we do find that the achievement gap narrows by a small
amount in female-taught classes for the achievement-related outcome receiv-
ing a C or better. We now turn to our main results, examining whether course-
specific motivation-related measures, such as course engagement and attitude,
narrow in female-taught classes.
Table 7 includes estimates for our course-specific motivation-related
measures. Model 1 includes a course fixed effects term and all available
student-level control variables.11 Model 2 represents our preferred model
and includes a student and class fixed effects term. It therefore indicates
whether replacing a male instructor with a female instructor increases female
students’ self-reported motivation in comparison to male students’ motiva-
tion in the same class.
Panel A includes the following four survey items measuring behavioral
engagement: attended lectures regularly, listened to lectures attentively,
asked the professor for help, and asked questions or contributed to class-
room discussion. The results are generally consistent across model specifica-
tions for each item in Panel A. Interestingly, for the first item, attended
lectures, we observe a gender gap in male-taught classes favoring females,
as indicated by the positive b2 coefficient (Column 1, Row 2). The negative
b3 coefficient (Row 3) is statistically insignificant in both model specifica-
tions, indicating that the gender gap is similar in female-taught classes.
The remaining three items measuring behavioral engagement in Panel A
indicate the presence of a gender engagement gap in male-taught classes
that favors male students, mirroring the descriptive evidence in Table 2
described earlier. This gender gap is represented by the negative b2 coeffi-
cients (Columns 3–7, Row 2). Further, our results suggest that a gender
match can mitigate this gap.
For example, in our class fixed effects model, female students’ average
reported value for the variable asked the professor for help (Column 5, Row
2) is 0.23 standard deviations lower than that for male students in the same
male-taught class. However, in a female-taught class, female students’ aver-
age reported value is only 0.04 (b2 1 b3) standard deviations lower than that
of their male counterparts. The decrease in the gender gap with a female
instructor is significant and robust across both model specifications, as indi-
cated by the positive b3 coefficient. Using the most conservative value, this
decrease represents an 83% reduction in the gender gap for this particular
engagement item.
We can better understand why the gender gap narrows by looking at
absolute losses/gains as students move from a male-taught to a female-
25
Solanki, Xu
Table 7
Estimated Role of Female Instructor Status for Student Nonacademic Outcomes
Panel A: Behavioral Engagement Items
Ask the Ask Questions/

Attend Listen Attentively Professor Contribute to
Lectures to Lectures for Help Discussion
(1) (2) (3) (4) (5) (6) (7) (8)
Female instructor 0.233*** — 0.094* — –0.103** — –0.073** —

(0.058) — (0.056) — (0.047) — (0.031) —
Female student 0.362*** — –0.036 — –0.229* — -0.097 —
(0.040) — (0.129) — (0.121) — (0.128) —
Interaction –0.047 –0.006 0.021 0.045 0.190*** 0.228*** 0.063* 0.124**
(0.029) (0.072) (0.039) (0.069) (0.047) (0.060) (0.036) (0.058)
Panel B. Interest Items
I Find Many Topics Solving Problems I Find This Class

in This Course to in This Class Is Intellectually
Be Interesting Interesting for Me Stimulating
Female instructor –0.103** — –0.067** — –0.085* —

(0.040) — (0.033) — (0.045) —
Female student –0.124 — –0.317*** — –0.134 —
(0.120) — (0.036) — (0.081) —
Interaction 0.075** 0.065 0.025 0.063 0.057 0.118*
(0.035) (0.061) (0.033) (0.062) (0.040) (0.064)
Panel C: Self-Efficacy Items
I’m Certain I’m Certain I Can Do Almost

I Can Master I Can Figure Out All the Work in
the Skills Taught How to Do the Most the Class if I
in This Class Difficult Course Material Don’t Give Up
Female instructor –0.033 — –0.046 — –0.019 —

(0.034) — (0.036) — (0.036) —
Female student –0.180** — –0.229*** — 0.107** —
(0.084) — (0.059) — (0.041) —
Interaction –0.007 0.093 0.001 0.020 –0.058 –0.050
(0.038) (0.060) (0.046) (0.075) (0.039) (0.055)
(continued)
26
Table 7 (continued)
Panel D: Utility Value Items
Having a Solid After I

Background in Graduate, an
the Material Understanding of
Taught in This the Material in
Course Is This Course Will
Worthless Be Useless to Me
Female instructor 0.005 — 0.008 —

(0.038) — (0.030) —
Female student 0.308*** — 0.215** —
(0.043) — (0.093) —
Interaction –0.0231 –0.012 –0.054 0.117*
(0.043) (0.067) (0.048) (0.060)
Course fixed effects X X X X
Class fixed effects X X X X
Student fixed effects X X X X
Student controls X X X X
Note. Each cell reports the coefficients from a separate linear probability regression. Model
1 includes a control for instructor academic rank and term fixed effects. Student controls
are ethnicity, first-generation status, low-income status, SAT verbal score, SAT math score,
and high school grade point average. Missing values have been adjusted using a dummy
variable approach. The sample size is 11,284. The interaction term represents a female stu-
dent by female instructor interaction. Standard errors are in parentheses and clustered at
the class level.
*p \ .10. **p \ .05. ***p \ .01.
taught class (outlined in Supplemental Table S4 in the online version of the

journal). The gender engagement gap for the item asked the professor for
help narrows in female-taught classes due both to females reacting positively
to female instructors and male students reacting negatively to female instruc-
tors. The point estimates in Column 5 suggest that male students experience
an absolute loss in female-taught courses as they reported average values
0.10 SD (Row 1) units lower for female-taught courses than those reported
for male-taught courses. On the other hand, female students experience
an absolute gain as they move from male- to female-taught courses.
Female students in female-taught courses reported values 0.09 (b1 1 b3 in
Column 5) SD units higher than those reported for male-taught courses.
As a result, the gender gap narrows in female-taught courses by 0.19 SD
units, as indicated by the b3 coefficient (Column 5, Row 3).
It is important to note that we also find a positive association between
instructor gender and our last behavioral engagement item, asked questions
27
Solanki, Xu
or contributed to class discussion (Columns 7 and 8). As indicated by posi-
tive b3 coefficients (Row 3), the gender engagement gap decreases by 0.06
to 0.12 SD units with a female instructor; this decrease is mainly due to male
students experiencing an absolute loss as they move from a male- to female-
taught class. As the point estimates in Column 7 suggest, male students’
reported values are 0.07 SD units lower (Row 1) for female-taught courses
than those reported for male-taught courses (see Supplemental Table S4 in
the online version of the journal). This negative reaction by males to a female
instructor is a consistent phenomenon that will be discussed in depth later.
Table 7 Panel B includes estimates for our variables that measure inter-
est, including three survey items regarding whether a student: finds topics in
the course interesting, finds solving problems in the class interesting, and
finds the class intellectually stimulating. Like in our behavioral engagement
findings, male students reported being less interested in female-taught
courses than male-taught courses (point estimates range from 0.07 to 0.10
in Row 1).
We also find that the gender interest gap, favoring males, decreases
more in female-taught courses than male-taught courses for one item mea-
suring interest in the course topic (Columns 1 and 2). The b3 coefficients
indicate that replacing a male instructor with a female instructor shrinks
the gender interest gap by 0.07 to 0.08 SD units; this result becomes statisti-
cally insignificant, however, in the preferred two-way fixed effects model
reported in Column 2. The point estimates reported in Column 1 suggest
that the narrowing of the gender interest gap is driven mainly by male stu-
dents experiencing an absolute loss as they move from male- to female-
taught courses, represented by the b1 coefficient and equal to 0.10 SD units.
The gender interest gap for the second item (Columns 3 and 4) indicates
that female students in a male-taught class reported much lower values for
the item I find solving problems interesting than male students in the
same class (b2 = –0.32 in Column 3, Row 2). The interaction term, however,
indicates that this gap is not significantly different for female-taught classes.
Lastly, the interaction term for the item I find this class intellectually stimu-
lating is positive and significant in Column 6, which includes a class and stu-
dent fixed effects term.
Table 7 Panel C includes estimates for the course-specific motivation
measure, self-efficacy. This panel includes three survey items measuring
whether a student feels certain that he or she can: master the skills taught
in class, figure out how to do the most difficult course material, and do
almost all the work in the class by not giving up. The gender efficacy gap,
favoring male students, is significant in male-taught classes for all items, as
indicated by the b2 coefficients in Row 2. However, replacing a male instruc-
tor with a female instructor does not decrease this gap.
28
Discussion
Women make up 47% of all U.S. employees yet represent only a quarter
of positions held in mathematical sciences and constitute a mere 13% of
engineers (NSF, Science and Engineering Indicators, 2016). In this study,
we have discussed the educational relevance of an instructor’s gender in
regard to student outcomes in college-level STEM courses with the idea
that while statistics such as these reflect the dramatic gender gap within
STEM classrooms, examining instructor gender might yield information
that can help reduce this gap.
Identity-based motivation theory suggests that the STEM gender gap is
particularly problematic as female students are less likely than their male
counterparts to find STEM compatible with their gender identity. According
to the IBM model, identity incongruence is accompanied by negative conse-
quences. A female student is likely to interpret STEM difficulty as evidence
that STEM is not for ‘‘people like [her]’’ and thus become less likely to pursue
a future in STEM (Karabenick & Urdan, 2014; Oyserman, 2014). A competent
female instructor has not only the capacity to serve as a role model but also
the ability to communicate the idea that there is a place for females in
STEM. In male-dominated disciplines like those STEM represents, this idea
is particularly important as it conveys that females can surmount both aca-
demic difficulties and societal stigma.
IBM theory provides a compelling framework to investigate issues
related to gender and student-instructor interactions in the STEM classroom.
Most of the higher education literature about this topic and how it relates to
STEM success is focused solely on achievement outcomes; our study is the
first to look beyond academic performance in examining the relationship
between instructor gender and nonacademic student outcomes.
Based on information collected through pre- and post-experience sur-
veys from a large swath of STEM gateway courses, we found that student
intellectual engagement can be improved as a result of instructor gender,
as the gap between female and male students’ course engagement and atti-
tude toward a STEM subject is reduced when a course is taught by a female
instructor. Specifically, while female students reported being less inclined to
ask a male STEM professor for help than male students did (a difference of
0.23 SD units), having a female instructor almost completely offsets this gen-
der difference in help-seeking. In a female-taught class, female students’
reported value is only 0.04 SD units lower than that reported by male stu-
dents. In addition, female students reported higher values for the item
attended lectures in female-taught classes than they did for this item in
male-taught classes. These outcomes are examples of how students posi-
tively modify their behavior in courses taught by instructors who share their
gender, validating the premise of this study. These outcomes also provide
some evidence that the psychological processes described by the IBM model
29
Solanki, Xu
might be occurring. In addition, the IBM model is partially supported by our
results because female students do not experience the same decrement to
their academic and engagement outcomes as male students (see
Supplemental Table S4 in the online version of the journal).
While these results provide compelling evidence that student-instructor
gender congruence influences the learning process, future studies are
needed for more concrete policy implications. On one hand, relationship
effects could simply be a product of gender stereotypes influencing student
perception. If this is true, it might be beneficial for colleges to increase the
presence of female instructors and advisors, especially in male-dominated
fields. On the other hand, student-instructor relationships may not be gender
exclusive and instead could be the product of pedagogy. For example, if
female instructors exhibit more empathy toward the learning needs of stu-
dents and this is more important to female students than male students, it fol-
lows that this emphasis on empathy would be responsible for females
reacting positively to female instructors. This type of story suggests that
the instruction might be more effective if instructors are better aware of
the distinct needs of different subpopulations. Thus, future research would
benefit from micro-detailed instructor information on pedagogy used and
how this pedagogy influences student-instructor interactions.
Even though the current policy climate is focused on promoting women
in science, the results that we found for males should not be ignored. The
existing literature on student-instructor gender match has mixed findings
regarding how male students react to female instructors. For example, based
on data from the United States Air Force Academy, Carrell et al. (2010) found
that instructor gender had no effect on male students’ degree attainment in
STEM in this particular setting. In contrast, other studies (e.g., Dee, 2005;
Hoffman & Oreopoulos, 2009), including the current one, found that male
students react negatively to female instructors on a variety of measures.
These different findings suggest that the way students react to instructor gen-
der may be highly context bound. In our specific case, this negative reaction,
or mismatch effect, is greater than the mismatch effect for female students
(see Supplemental Table S4 in the online version of the journal). Further,
in a couple of instances, we even find that the positive interaction is entirely
driven by the decrement among male students in reported values as they
moved from a male- to female-taught class.
The question that still remains is: Why might male students react nega-
tively to female instructors? Some researchers suggest that gender stereotypes
within the college environment—specifically regarding student perception of
female and male instructors—play a role (Miller & Chamberlin, 2000) in prop-
agating the negative reaction this study captures. In fact, a recent study found
that male students evaluated their male instructors more favorably than their
female instructors regardless of teaching style and grading criteria (Boring,
Ottoboni, & Stark, 2016). Other less related work in the area of neighborhood
30
effects has shown that there are gender differences in adapting to new envi-
ronments; psychological stress stemming from neighborhood environments
has a more serious effect on males than females (Zaslow & Hayes, 1986),
for example. It might be the case that because males expect male instructors
to teach STEM courses, they feel less comfortable in female-guided STEM
environments.
Given that a salient outcome from our study is that male students
respond negatively to female instructors, future studies should dedicate
themselves to understanding the variables advancing this negative reaction
and also consider how female instructors can better connect with male stu-
dents. Investigating these issues is important in terms of discerning appropri-
ate policy solutions.
Conclusion
Bringing about gender equality in STEM fields is important, and a com-
mon policy prescription has been to increase the number of female faculty in
STEM fields so they can inspire and encourage young women to pursue
STEM degrees. Indeed, researchers and policymakers continue to argue
that female professors serve as role models and thus have the potential to
improve female student academic and nonacademic outcomes, informing
our decision to use identity-based motivation theory as a lens through which
to view STEM instructor impact.
Given that male students are indeed affected by female instructors, our
study’s results suggest that the policy implications of such a finding are not
clear. What is abundantly clear, however, is that student-instructor interac-
tions matter and that the role of gender in higher education should be
made an explicit point of departure for future studies. Indeed, what is being
taught—in addition to who is teaching it—sends strong messages to both
male and female students, so it is suggested that institutions of higher learn-
ing enact guiding principles and policies in keeping with shifting gender
roles and perceptions of equality.
The college years are a critical time in which students imagine their
future selves and take action to achieve their career goals. If females con-
tinue to be the minority in STEM classes and are not exposed to readily avail-
able STEM female role models, it is likely that they will indeed believe STEM
fields are largely inaccessible, causing female STEM involvement to decline
further. This study takes an important first step toward understanding the
mechanisms that make gender relevant in the higher education classroom.
Future research should continue to examine the role of nonacademic out-
comes in regard to female student success so that researchers and policy-
makers can identify potential strategies at the college level that can help
female students thrive in the STEM environment.
31
Solanki, Xu
Notes
This material is based on work supported by the National Science Foundation under
Grant No. 1256500. The opinions expressed are those of the authors and do not represent
the views of the National Science Foundation. We would like to thank Mark Warschauer,
Lynn Reimer, and Amanda Nili for providing high-quality data and expert guidance on the
course context of this research. We are also indebted to Greg Duncan, George Farkas,
Rachel Baker, Emily Penner, and Phoebe Camilletti for their valuable comments and sug-
gestions on drafts of this paper.
1
Instructor race is an equally important factor to examine in terms of student devel-
opment in the classroom. Unfortunately, we were unable to examine the implications of
race due to lack of variation in instructor race.
2
Sixty-five percent of students completed the pre-experience survey, 55% of students
completed the post-experience survey, and 50% of students completed both surveys. A
potential concern regarding these survey items is that the instructor-student gender inter-
action may also influence a student’s probability of completing the course survey. We
directly examine this issue and provide empirical evidence rejecting this possibility in
the methodology section.
3
The excluded survey items could not, theoretically, be influenced by same-gender
student-instructor interactions. ‘‘I have to give up a lot to do well in this course’’ or ‘‘In
comparison to the rest of your classmates, how do you rate your achievement?’’ are
two examples of excluded items.
4
Baseline survey items are similar for female and male students across instructor
gender.
5
A class is a course (general chemistry) taught during a specific term (winter 2014) at
a distinct time with a given instructor. The term class is interchangeable with the term
course-section.
6
Of all students in our sample, 55% took courses in multiple fields, therefore contrib-
uting to the student fixed effects estimator. However, including students who did not take
courses in multiple fields does not bias our results as long as selection bias, if any, is con-
stant for an individual. A smaller degree of within-individual variation would be problem-
atic if it yielded an imprecise estimator; with large sample sizes, however, this is typically
less of a concern. In a robustness check, we also limited our sample to students who took
courses in multiple fields, and the resulting effect sizes and significance levels were almost
identical to those reported here. For a detailed discussion of the properties of the fixed
effects estimator and key assumptions underlying fixed effects models using panel data,
see Wooldridge (2002).
7
We include course fixed effects and term fixed effects in the model. Ideally, we want
to include a course by term fixed effects, but we do not have variation in instructor gender
at this level. The main source of variation in our model comes from different gendered
instructors teaching the same course over time.
8
We include Model 1 to capture student and instructor main effects.
9
We also estimated our models by two-way clustering standard errors at the student
and course levels; standard errors increase—but do not change—our significance levels.
10
We also ran models in which receiving a grade of B or better and D or better func-
tioned as outcome variables. These results were not significant.
11
Given that female-taught classes include a larger share of female students, one
potential concern is that the student-teacher gender match effect might be driven by
peer effects. As an additional robustness check, we include in Model 1 an additional con-
trol for the interaction between class gender composition and the gender of the student for
all outcome measures. As shown in Supplemental Table S5 (in the online version of the
journal), the point estimates and standard errors remain consistent for the student-teacher
gender match interaction term with this additional control.
32
References
Antecol, H., Eren, O., & Ozbeklik, S. (2015). The effect of teacher gender on student
achievement in primary school. Journal of Labor Economics, 33, 63–89.
Bettinger, E., & Long, B. (2005). Do faculty serve as role models? The impact of
instructor gender on female students. American Economic Review, 95, 152–157.
Boring, A., Ottoboni, K., & Stark, P. (2016). Student evaluations of teaching (mostly)
do not measure teaching effectiveness. ScienceOpen Research, 1, 1–11.
Bureau of Labor Statistics. (2013). Labor force statistics from the CPS. Retrieved from
https://www.bls.gov/cps/lfcharacteristics.htm#occind.
Canes, B., & Rosen, H. (1995). Following in her footsteps? Women’s choices of col-
lege majors and faculty gender composition. Industrial and Labor Relations
Review, 48, 486–504.
Carrell, S., Page, M., & West, J. (2010). Sex and science: How professor gender per-
petuates the gender gap. The Quarterly Journal of Economics, 125, 1101–1144.
Chambliss, D., & Takacs, C. (2014). How college works. Cambridge, MA: Harvard
University Press.
Chang, M., Sharkness, J., Hurtado, S., & Newman, C. (2014). What matters in college
for retaining aspiring scientists and engineers from underrepresented racial
groups. Journal of Research in Science Teaching, 51, 555–580.
Chen, X. (2009). Students who study science, technology, engineering, and mathe-
matics (STEM) in postsecondary education. Washington, DC: National Center
for Education Statistics.
Corbett, C., Hill, C., & St. Rose, A. (2008). Where the girls are: The facts about gender
equity in education. Retrieved from https://www.aauw.org/files/2013/02/
Where-the-Girls-Are-The-Facts-About-Gender-Equity-in-Education.pdf
Dee, T. (2005). A teacher like me: Does race, ethnicity, or gender matter? American
Economic Review, 95, 158–165.
Denaro, K., Reimer, L., Solanki, S., Xu, D., & Warschauer, M. (2018). Documenting
instructional practices in STEM lecture courses [Data file and code book]. Ann
Arbor, MI: Inter-university Consortium for Political and Social Research
[distributor].
Ehrenberg, R., Goldhaber, D., & Brewer, D. (1995). Do teachers’ race, gender and
ethnicity matter? Evidence from the national educational longitudinal study of
1988. Industrial and Labor Relations Review, 48, 547–561.
Erikson, E. (1968). Identity, youth, and crisis. New York, NY: Norton.
Fairlie, R., Hoffman, F., & Oreopoulos, P. (2014). A community college instructor like
me: Race and ethnicity interactions in the classroom. American Economic
Review, 104, 2567–2591.
Ferrara, M. (2012). The relationship of teaching style, gender preference, and gender
match in single gender and coeducational settings. Advances in Gender and
Education, Montgomery Center for Research in Child & Adolescent Development
(MCRCAD), 3, 14–27.
Gilmartin, S., Denson, N., Li, E., Bryant, A., & Aschbacher, P. (2007). Gender ratios in
high school departments: The effect of percent female faculty on multiple
dimensions of students’ science identities. Journal of Research in Science
Teaching, 44, 980–1009.
Grunspan, D. Z., Eddy, S. L., Brownell, S. E., Wiggins, B. L., Crowe, A. J., &
Goodreau, S. M. (2016). Males under-estimate academic performance of their
female peers in undergraduate biology classrooms. PLoS ONE, 11(2).
doi:10.1371/journal.pone.0148405
33
Solanki, Xu
Gutman, L., & Schoon, I. (2013). The impact of non-cognitive skills on outcomes for
young people. London: Institute of Education.
Hoffman, F., & Oreopoulos, P. (2009). A professor like me: The influence of instructor
gender on university achievement. Journal of Human Resources, 44, 479–494.
Hyde, S. J., & Jaffee, S. (1998). Perspectives from social and feminist psychology.
Educational Researcher, 27, 14–16.
Karabenick, S. A., & Urdan, T. C. (2014). Motivational interventions. Bingley, UK:
Emerald Group Publishing.
Lavy, V. (2004). Do gender stereotypes reduce girls’ human capital outcomes?
Evidence from a natural experiment. Journal of Public Economics, 92, 2083–
2105.
Lavy, V., & Sand, E. (2015). On the origins of gender human capital gaps: Short and
long term consequences of teachers’ stereotypical biases (NBER Working Paper
No. 20909). Cambridge, MA: National Bureau of Economic Research.
London, B., Rosenthal, L., & Gonzalez, A. (2011). Using experience sampling meth-
odology to capture the impact of identity, support, and gender rejection on the
academic engagement of women. Journal of Social Issues, 67, 510–530.
Marx, D., & Roman, J. (2002). Female role models: Protecting women’s math test per-
formance. Society for Personality and Social Psychology, 28, 1183–1193.
McIntyre, R., Lord, C., Gresky, D., Ten Eyck, L., Frye, G., & Bond, C. (2005). A social
impact trend in the effects of role models on alleviating women’s mathematics
stereotype threat. Current Research in Social Psychology, 10, 116–136.
Miller, J., & Chamberlin, M. (2000). Women are teachers, men are professors: A study
of student perceptions. Teaching Sociology, 28, 283–298.
National Science Board. (2016). Science and engineering indicators 2016. Arlington,
VA: National Science Foundation.
Neumark, D., & Gardecki, R. (1998). Women helping women? Role model and men-
toring effects on female PhD students in economics. Journal of Human
Resources, 33, 220–246.
Nixon, L., & Robinson, M. (1999). The educational attainment of young women: Role
model effects of female high school faculty. Demography, 36, 185–194.
Oyserman, D. (2007). Social identity and self-regulation. In A. Kruglanski &
T. Higgins (Eds.), Handbook of social psychology (pp. 432–453). New York,
NY: Guilford Press.
Oyserman, D. (2014). Identity-based motivation: Core processes and intervention
examples. In S. A. Karabenick & T. Urdan (Eds.), Motivational interventions
(pp. 213–240). Bingley, UK: Emerald Group Publishing.
Rebhorn, L., & Miles, D. (1999). High-stakes testing: Barrier to gifted girls in mathe-
matics and science? Social Science and Mathematics, 99, 313–319.
Robst, J., Kell, J., & Russo, D. (1998). The effect of gender composition of faculty on
student retention. Economics of Education Review, 29, 429–439.
Rosenthal, L., London, B., Levy, S., & Lobel, M. (2011). The roles of perceived identity
compatibility and social support for women in a single-sex STEM program at
a co-educational university. Sex Roles, 65, 725–736.
Rosenthal, R., & Jacobsen, L. (1968). Pygmalion in the classroom: Teacher expecta-
tion and pupils’ intellectual development. New York, NY: Holt, Rinehart and
Winston.
Sadker, M., & Sadker, D. (1985). Sexism in the classroom. Vocational Education
Journal, 60, 30–32.
Schunk, D. H. (1995). Self-efficacy and education and instruction. In J. E. Maddux
(Ed.), Self-efficacy, adaptation, and adjustment: Theory, research, and applica-
tion (pp. 281–303). New York, NY: Plenum Press.
34
Settles, I. H., Jellison, W. A., & Pratt-Hyatt, J. S. (2009). Identification with multiple
social groups: The moderating role of identity change over time among women
scientists. Journal of Research in Personality, 43, 856–867.
Seymour, E., & Hewitt, N. (1997). Talking about leaving: Why undergraduates leave
the sciences. Boulder, CO: Westview Press.
Starbuck, G. (2003, April). College teaching styles by gender. Paper presented at the
meeting of the Western Social Science Association, Las Vegas, NV.
Warner, R. (2013). Applied statistics: From bivariate through multivariate techniques.
Thousand Oaks, CA: Sage Publications.
Waterman, A. (1985). Identity in adolescence: Processes and contents. San Francisco,
CA: Jossey-Bass.
Wehrwein, E., Lujan, H., & DiCarlo, S. (2007). Gender differences in learning style
preferences among undergraduate physiology students. Advances in Physiology
Education, 31, 153–157.
Wooldridge, J. (2002). Econometric analysis of cross section and panel data.
Cambridge, MA: The MIT Press.
Zaslow, M., & Hayes, C. (1986). Sex differences in children’s response to psychoso-
cial stress: Toward a cross-context analysis. In M. Lamb, A. Brown, & B. Rogoff
(Eds.), Advances in developmental psychology (Vol. 4, pp. 285–338). Hillsdale,
NJ: Lawrence Erlbaum.
Zirkel, S. (2002). Is there a place for me? Role models and academic identity among
White students and students of color. Teachers College Record, 104, 357–376.
Manuscript received March 8, 2017
Final revision received December 21, 2017
Accepted January 13, 2018
35

Solanki2018 PDF

Uploaded by

Copyright:

Available Formats

Solanki2018 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solanki2018 PDF

Uploaded by

Copyright:

Available Formats

American Educational Research Journal

Month XXXX, Vol. XX, No. X, pp. 1–35

Looking Beyond Academic Performance:

KEYWORDS: female faculty in STEM, instructor gender, role model effects in

SABRINA M. SOLANKI is a doctoral student in the School of Education at the University of

Theoretical Discussion and Empirical Evidence

Role Model Effects: Identity-Based Motivation Theory

The Current Study

Panel A: Student Demographic Characteristics in Male Versus Female-Taught Classes

(1) (2) (3) (4)

Female 0.531 0.481 0.595 –16.157

All Classes Male-Taught Classes Female-Taught Classes

(1) (2) (3) (4) (5) (6) (7) (8) (9)

All Classes Male-Taught Classes Female-Taught Classes

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Sample and Summary Statistics

Panel A: All Classes

Panel B: Female-Taught Classes

Panel B: Female-Taught Classes

Listen attentively to lectures 0.028 0.046 0.018 1.008

Panel C: Male-Taught Classes

Panel C: Male-Taught Classes

Biology 0.17 0.21 0.12 0.22 0.10

instructor j is female and 0 if instructor j is male. Female_studenti is dichoto-

Two-Way Fixed Effects Model

Yic 5b3 ðFemale instructorc 3 Female studenti Þ 1 gi 1 Fc 1 Fkg 1 uic : ð2Þ

Compared to Equation 1, Equation 2 includes two additional sets of var-

Concerns About Differential Sorting

High First- Low-

Xgc 5 d1 3 Female instructorc 1 d2 3 Ig 1 d3 3 Female instructorc 3 Ig 1 ygc ; ð3Þ

where Ig is a dummy variable equal to 1 if the average is computed for

Concerns With Survey Data

Standardized Course Grade Receiving a C or Better

Female instructor –0.065** — –0.054*** —

of responding to a survey. Thus, we ran a series of regression models, using

Heterogeneous Treatment Effects

Standardized Course Grade Receiving a C or Better

Female instructor –0.088** –0.041***

Panel A: Behavioral Engagement Items

Ask the Ask Questions/

Female instructor 0.233*** — 0.094* — –0.103** — –0.073** —

Panel B. Interest Items

I Find Many Topics Solving Problems I Find This Class

Female instructor –0.103** — –0.067** — –0.085* —

Panel C: Self-Efficacy Items

I’m Certain I’m Certain I Can Do Almost

Female instructor –0.033 — –0.046 — –0.019 —

Panel D: Utility Value Items

Having a Solid After I

Female instructor 0.005 — 0.008 —

taught class (outlined in Supplemental Table S4 in the online version of the

You might also like

Female instructor –0.065 — –0.054* —

Female instructor –0.088 –0.041*

Female instructor 0.233*** — 0.094* — –0.103 — –0.073 —

Female instructor –0.103 — –0.067 — –0.085* —