Stereotype threat,anxiety,instructor gender,and underperformance in women

Abstract

Whatever their major, students are often required to take at least one course in statistics. After graduation, statistics is a key skill in numerous workplace settings. However, for many, it is a particularly difficult course. One factor that may play a role is the lingering misconception that women are not as good as men in mathematics subjects such as statistics. Belief in this gender stereotype can lead women to avoid taking this class and ultimately could contribute to their underrepresentation in many professions. Instructor gender may also be a factor that affects student performance. This study examined whether a female role model would reduce the detrimental effects of a gender/mathematics stereotype threat in women and improve their academic performance and retention rate. Several types of anxiety were measured to determine what aspects of anxiety might be most relevant to stereotype threat. For men, anxiety and performance were not related to the gender of their instructor or endorsement of the gender/mathematics stereotype. For women, mathematics anxiety and anxiety-about-the-specific-class were related to their endorsement of the stereotype. Having a female instructor initially led to higher rates of underperformance on the first examination among women as compared to men. Continued interaction with a female role model, however, reduced this deficit for women by the end of the semester. Understanding this process may help educators better prepare women for success in both academia and the professional world.

Keywords

anxiety and mathematics performance female role models instructor gender stereotype threat women and statistics

Stereotype threat

Stereotype threat (Steele and Aronson, 1995) refers to a process in which negative stereotypes associated with individuals’ identities lead them to expect that an unfavorable evaluation will be made of them. On the basis of superficial characteristics, others may have certain expectations about their capabilities and behavior. These expectations can act as a threat to the integrity of one’s self-image, resulting in harmful consequences including poor self-esteem, self-imposed isolation, inability to focus mentally, and academic underperformance (Steele, 2010).

Stereotype threat has been shown to negatively impact women’s performance in several contexts, especially in mathematics (e.g. Danaher and Crandall, 2008; Spencer et al., 1999). Although the belief that men are better at mathematics than women has been repeatedly disproven (e.g. Hyde et al., 2008, surveying 69 countries), this gender stereotype continues to burden women. Numerous studies confirm how widespread this stereotype is (e.g. Lummis and Stevenson, 1990, in the United States, Japan, and Taiwan; Nosek et al., 2009, in 34 countries). Thus, this is a global issue.

The persistence of this gender/mathematics stereotype impacts women’s lives by undermining their self-confidence, affecting occupational choices, and limiting their earnings and influence. American women, for example, are less likely to take advanced mathematics classes in college (Amelink, 2010) and to select mathematically based majors than men (George-Jackson, 2011). As a result, they are underrepresented in mathematics-intensive professions (Ceci et al., 2009). Despite its ubiquity, however, one should not assume that the gender/mathematics stereotype is universally accepted. Nosek et al. (2009), for example, found that 30% of respondents did not subscribe to this belief.

While stereotype threat effects are well established for mathematics, little is known about these effects in women studying statistics. Although statistics overlaps mathematics, it is multidimensional, involving additional skills such as logical reasoning and interpretation (Onwuegbuzie and Wilson, 2003). Nevertheless, most students equate statistics with mathematics (Bond et al., 2012). There is also considerable correspondence between mathematics and statistics anxiety (Onwuegbuzie and Wilson, 2003; Zeidner, 1991). There is, therefore, a need to investigate how stereotype threat impairs women’s performance in statistics in actual classrooms and how its effects might be ameliorated.

Steele (2010) maintains that stereotype threat most negatively impacts the strongest students, those working at “the achievement vanguard of their group” (p. 138) and who care the most about what is being tested (p. 97). He reports, for example, on an unpublished study involving performance on a difficult verbal examination in which only African American high school students who were the most motivated performed worse than White students under stereotype threat conditions (Steele, 2010: 54–59). Similar results were found by Smith et al. (2007) with undergraduate women on a computer science task. Surprisingly, several studies (e.g. Aronson et al., 1999) simply showed that strong students fell victim to this threat but did not include a comparison group of weak students. Another goal is, therefore, to examine the strength of stereotype threat in strong (high academic achievement) students compared to others.

How stereotype threat impairs performance

Early researchers assumed that it was anxiety caused by stereotype threat that impaired performance, diminishing cognitive processing in several ways. Steele proposed that anxiety induced by stereotype threat causes a vicious cycle of rumination and distraction. The “racing mind” (Steele, 2010: 121) of the stereotype-threatened individual may also reduce working memory capacity (Schmader and Johns, 2003). Under such chronic stress, students’ thinking can become rigid. They may waste time and energy “over-efforting” and isolate themselves with solitary studying rather than seeking out assistance from others (Steele, 2010: 104).

The ability of standardized self-report anxiety measures to predict performance deficits due to stereotype threat remains unclear. Miller and Bichsel (2004) found that math, but not trait or state, anxiety was the strongest predictor of mathematics performance. Cherney and Cooney (2005) also found that mathematics anxiety scores were predictive of undergraduates’ grades in statistics classes. However, Hunsley (1987), who measured both mathematics and test anxiety, found that only test anxiety was related to performance.

External cues that moderate stereotype threat

Stereotype threat is aroused to different degrees by external cues, even very subtle ones. For example, the gender/mathematics stereotype may be activated by simply asking women to indicate their gender on a test form. Gathering identity information after rather than before taking a mathematics test reduced gender performance differences by 33% (Danaher and Crandall, 2008). Another environmental factor which has proven potent in reducing stereotype threat is the presence of powerful figures (experts or role models) similar to oneself, a phenomenon called the stereotype inoculation model (Stout et al., 2011). In a Swiss study, Latu et al. (2013) found that improvement in self-evaluation and leadership performance resulted when business women were exposed to a successful female role model while men’s attitudes and behaviors were not affected.

The influence of role models is not always positive, however. Hoyt and Simon (2011) and Parks-Stamm et al. (2008) showed that women’s self-ratings of competence and aspiration for leadership in business declined after exposure to high-level female role models. This did not occur when the role models were explicitly described as being typical women who had educational backgrounds like the participants (Asgari et al., 2012; Lockwood, 2006, in Canada).

Stereotype inoculation has been demonstrated numerous times in the academic setting. A large US study found that African American and Latino college students’ underperformance was reduced in classes taught by minority professors (Massey et al., 2002). In mathematics, female role models are limited. Women make up a relatively small percentage of mathematics teachers in primary through high schools and become even less frequent at the college level (Pillsbury, 2010). Female mathematicians and scientists are also scarce in the media (Kitzinger et al., 2008, in the United Kingdom). Thus, it is not surprising that the gender/mathematics stereotype appears early (Ambady et al., 2001) and is widespread among girls (Muzzatti and Agnoli, 2007, in Italy; Keller, 2002, in Germany).

The effect of instructor gender on student performance is neither simple nor consistent. Marx and Roman (2002) had highly mathematics-identified students take a difficult mathematics test with either a male or female proctor whose competence in mathematics was established by indicating that they were the authors of the test. Female students performed as well as male students when the proctor was female, but performed more poorly and showed lower self-esteem when the proctor was male. Similarly, Stout et al. (2011) found that female STEM (Science, Technology, Engineering, or Mathematics) majors attempted more questions on a difficult mathematics test and showed stronger self-identification with mathematics when the proctor was a woman rather than a man.

Positive mathematics outcomes in women can result from a single exposure to a female role model (Marx and Roman, 2002) but may also develop gradually as students continue to interact with the role model. Stout et al. (2011) found that female students in a college calculus class initially attempted to answer more questions when the instructor was male (11%) rather than female (7%). By the end of the semester, this percentage had risen to 46% for women with female instructors but remained unchanged for those having male instructors. Approximately 12% of female students approached instructors of both genders for help early in the semester. But at the end of the semester, the percentage had risen to 14% for those with female instructors, while it dropped to zero for those with male instructors.

Measurement of stereotype threat

Stereotype threat has been operationalized in a number of ways. Goff et al. (2008) used a social distance measure, while Steele and Aronson (1995) used the completion of word fragments and ratings of preferences for various types of activities (music and sports) that are often stereotypically associated with particular groups. Simple endorsement of stereotype statements has also been used as a measure. In Blanton et al. (2002), two statements regarding men’s superiority to women in mathematics were embedded in other gender-related statements. Overall, 19% of undergraduate women rated the statements as not at all true. While the sample as a whole showed a low level of endorsement, 41% rated the statements as having at least some truth. Activating the stereotype by giving women social comparison information about their abilities had the greatest detrimental effect on those with moderate to high levels of stereotype endorsement.

Steele et al. (2002) argue that simply being aware that a negative stereotype might be applied to one’s performance can induce stereotype threat even though one does not personally endorse the attitude. However, Steele (2010) later suggested that the detrimental effects of stereotype threat would be greatest in those who hold the belief most strongly. Schmader et al. (2004) found that 24% of female students in mathematics-related majors indicated that there might be some truth to stereotype statements about mathematics ability. Stereotype endorsement was negatively correlated with confidence, performance, self-esteem, and the desire to attend graduate school in their current majors.

Although examination scores and grades are often used as outcomes measures, underperformance is a conceptually more relevant measure because stereotype threat is specifically defined as causing underperformance (Steele, 2010: 30). Steele and Aronson (1995) operationalized underperformance as the difference between how African Americans performed relative to European Americans on a difficult test when it was presented as a measure of verbal reasoning (assumed to arouse stereotype threat) versus as a task to measure problem solving in general (assumed to not arouse stereotype threat). Spencer et al. (1999) compared the gap between the grades in mathematics and English classes earned by equally talented and prepared male and female students.

This study was designed to increase our understanding of the effects of role models and anxiety on mathematics/gender stereotype threat and performance in statistics. Specifically, its goals are to demonstrate that (1) a gender/mathematics stereotype threat exists and causes underperformance in statistics classes and is related to dropout rate, (2) a simple measure of agreement with the stereotype can identify students likely to be affected by it, (3) a stereotype threat in statistics differentially affects the highest achieving students, (4) a role model can reduce the detrimental effect of stereotype threat, and (5) confirm that some types of anxiety are more associated with stereotype threat and underperformance than others.

Five specific hypotheses were tested:

Women will show negative effects of stereotype threat on the first examination of the semester by underperforming more than men.

Women’s underperformance on the first examination will be greatest for those who agree with the stereotype statement and will be particularly evident for those who are the highest achieving as measured by cumulative academic achievement.

Among women, there will be a positive correlation between stereotype endorsement score and some anxiety measures.

Underperformance in women will be reduced from the first examination to the final grade in the class for those who had female but not male instructors.

Dropout rates among high stereotype-endorsing women compared to low stereotype-endorsing women will be lower for those who have female rather than male instructors.

Methodology

Participants

Participants came from 11 semester-long Introductory Statistics for the Social Sciences classes at a large urban California community college These classes were given over two consecutive semesters. California community college admission requirements are high school graduation or being 18 years of age or older. The specific requirement for taking a first course in statistics is passing a class in intermediate algebra. In total, 121 students (65% women) from four classes participated in the first semester, and 330 different students (62% women) from seven classes participated in the second semester. Eight outlying students, who did not otherwise differ from the others on grade point average (GPA) or age but who had earned z-scores of −2.0 or lower on the first examination, were eliminated from the sample. GPA is a cumulative average of one’s academic achievement which ranges from 0.0 = F or fail to 4.0 = A or superior. All analyses, except those examining dropout rates, included only data from students who took both the first and final examinations.

Six of the classes were taught by a single male instructor and five classes were taught by three female instructors, one of whom was also a researcher. There was no significant difference in GPA or age between the male and female students. However, students in the female instructors’ classes (M = 23.44, standard deviation (SD) = 8.07) were significantly older than the students in the male instructor’s classes (M = 21.53, SD = 5.36), F(1, 378) = 6.560, p = 0.011, partial eta squared = 0.017. This was because the only evening class, which normally attracts older students, was taught by a woman. There was no significant difference in the ratio of men to women in the classes taught by the male or female instructors. Analyzing gender of students by gender of instructor, there was a significant difference in GPA between male and female students in the male instructor’s classes only, with female students having higher GPAs (see Table 1).

Table 1.

Mean age and GPA of male and female students by instructor gender.

Instructor gender	Male students			Female students
Instructor gender	Mean	(SD)	n	Mean	(SD)	n	t	df	p	d
Male
Age	21.00	(3.26)	66	21.83	(6.20)	121	−1.008	185	ns
GPA	2.88	(0.70)	59	3.10	(0.59)	111	−2.272	168	0.024*	0.358
Female
Age	21.44	(5.63)	77	24.11	(9.33)	114	−1.543	187	ns
GPA	3.06	(0.72)	74	3.03	(0.63)	107	0.268	179	ns

GPA: Grade Point Average; SD: standard deviation.

Age is given in years, GPA is based on a scale of 0.0 = F or fail to 4.0 = A or superior.

*p < 0.05.

Instruments

Participants responded to two Likert-type items on a pre-test survey. One item was a self-report measure of anxiety-about-the-specific-class from 1 = not at all anxious to 7 = extremely anxious, which was significantly correlated with other standardized measures of anxiety (see Table 2). The second item was a measure of the gender/mathematics stereotype based on Blanton et al.’s (2002) measure. Students responded to the statement “Men are better in mathematics than women and, therefore, usually do better in mathematics classes,” from 1 = strongly disagree to 7 = strongly agree with the middle point specifically indicating no opinion.

Table 2.

Correlations of age, GPA, agreement with the math stereotype, underperformance, and anxiety by sex of student.

		1	2	3	4	5	6	7	8	9	10	11	12
Males
1. Age	Corr.	1.00
	Sig.
	n	143
2. GPA	Corr.	−0.130	1.00
	Sig.	0.137
	n	133	133
3. Stereotype statement	Corr.	−0.020	−0.122	1.00
	Sig.	0.814	0.163
	n	143	133	144
4. UPI exam 1	Corr.	0.019	0.165	0.076	1.00
	Sig.	0.834	0.075	0.412
	n	119	118	119	119
5. UPI final grade	Corr.	0.078	−0.104	0.000	0.648**	1.00
	Sig.	0.403	0.266	0.996	0.000
	n	117	116	117	116	117
6. UPI difference	Corr.	−0.019	0 281*	0.063	0.515**	−0.319**	1.00
	Sig.	0.838	0.002	0.504	0.000	0.000
	n	116	115	116	116	116	116
7. AMAS	Corr.	0.201*	−0.050	−0.156	0.095	0.097	−0.037	1.00
	Sig.	0.016	0.570	0.062	0.306	0.301	0.691
	n	142	132	143	118	116	115	143
8. Anxiety class	Corr.	0.303**	−0.109	−0.036	0.236**	0.133	0.129	0.560**	1.00
	Sig.	0.000	0.210	0.665	0.010	0.153	0.167	0.000
	n	143	133	144	119	117	116	143	144
9. E-scale	Corr.	0.259**	−0.019	−0.060	0.151	0.169	−0.056	0.618**	0.416**	1.00
	Sig.	0.002	0.828	0.475	0.102	0.069	0.553	0.000	0.000
	n	143	133	144	119	117	116	143	144	144
10. W-scale	Corr.	0.135	−0.127	0.079	0.172	0.191*	−0.036	0.508**	0.376**	0.727**	1.00
	Sig.	0.109	0.144	0.348	0.061	0.039	0.702	0.000	0.000	0.000
	n	143	133	144	119	117	116	143	144	144	144
11. TAI	Corr.	0.240**	−0.073	0.025	0.185*	0.205*	−0.047	0.606**	0.437**	0.940**	0.902**	1.00
	Sig.	0.004	0.402	0.765	0.044	0.026	0.618	0.000	0.000	0.000	0.000
	n	143	133	144	119	117	116	143	144	144	144	144
12. Trait anxiety	Corr.	0.142	0.119	−0.030	0.002	0.016	−0.053	0.341**	0.186*	0.490**	0.488**	0.525**	1.00
	Sig.	0.091	0.171	0.724	0.980	0.864	0.574	0.000	0.026	0.000	0.000	0.000
	n	143	133	144	119	117	116	143	144	144	144	144	144
Females
1. Age	Corr.	1.00
	Sig.
	n	235
2. GPA	Corr.	0.148*	1.00
	Sig.	0.029
	n	218	218
3. Stereotype statement	Corr.	0.030	0.034	1.00
	Sig.	0.649	0.616
	n	233	216	234
4. UPI exam 1	Corr.	0.016	0.098	0.075	1.00
	Sig.	0.820	0.177	0.302
	n	193	193	191	193
5. UPI final grade	Corr.	−0.044	−0.030	−0.069	0.525**	1.00
	Sig.	0.551	0.692	0.360	0.000
	n	182	182	180	182	182
6. UPI difference	Corr.	0.070	0.139	0.137	0.506**	−0.469**	1.00
	Sig.	0.349	0.061	0.067	0.000	0.000
	n	182	182	180	182	182	182
7. AMAS	Corr.	0.026	−0.026	−0.159*	0.141	0.150*	−0.019	1.00
	Sig.	0.692	0.702	0.015	0.051	0.044	0.797
	n	235	218	234	193	182	182	236
8. Anxiety class	Corr.	0.223**	−0.136*	0.113	0.195**	0.155*	0.013	0.525**	1.00
	Sig.	0.001	0.045	0.085	0.007	0.038	0.858	0.000
	n	234	217	234	192	181	181	235	235
9. E-scale	Corr.	−0.016	0.007	−0.026	0.098	0.011	0.092	0.446**	0.324**	1.00
	Sig.	0.804	0.915	0.695	0.174	0.884	0.216	0.000	0.000
	n	234	217	233	192	181	181	235	234	235
10. W-scale	Corr.	0.016	−0.160*	0.073	0.154*	0.075	0.070	0.469**	0.387**	0.748**	1.00
	Sig.	0.811	0.018	0.269	0.033	0.313	0.351	0.000	0.000	0.000
	n	234	217	233	192	181	181	235	234	235	235
11. TAI	Corr.	0.004	−0.072	0.020	0.131	0.035	0.093	0.487**	0.382**	0.940**	0.916**	1.00
	Sig.	0.955	0.288	0.759	0.069	0.642	0.213	0.000	0.000	0.000	0.000
	n	234	217	233	192	181	181	235	234	235	235	235
12. Trait anxiety	Corr.	0.097	0.144*	0.054	−0.031	−0.015	−0.018	0.280**	0.176*	0.384**	0.381**	0.410**	1.00
	Sig.	0.139	0.034	0.411	0.666	0.841	0.812	0.000	0.007	0.000	0.000	0.000
	n	233	216	232	191	180	180	234	233	233	233	233	234

GPA: grade point average; UPI: Underperformance Index; Stereotype statement: “Men are better in math than women and, therefore, usually do better in math classes;” AMAS: Abbreviated Mathematics Anxiety Scale; Anxiety class: anxiety-about-the-specific-class; E-scale: the Emotion subscale of the Test Anxiety Inventory; W-scale: the Worry subscale of the Test Anxiety Inventory; TAI: Test Anxiety Inventory; and Trait Anxiety: Spielberger’s Trait Anxiety Scale.

p < 0.05; **p < 0.01.

Three additional types of anxiety were measured: (1) Spielberger’s Test Anxiety Inventory (TAI; Spielberger et al., 1980), (2) Spielberger’s State-Trait Anxiety Inventory for Adults (STAI; Spielberger et al., 1983), and (3) the Abbreviated Mathematics Anxiety Scale (AMAS; Hopko et al., 2003). Both the TAI and STAI consist of 20 statements to which the student answers “almost never,” “sometimes,” “often,” or “almost always.” A sample item from the STAI (Cronbach’s α = 0.86) is “I worry too much over something that really doesn’t matter” and from the TAI (Cronbach’s α = 0.96) is “I feel confident and relaxed while taking tests.” The TAI also provides two subscale scores: Emotionality (E-scale, Cronbach’s α = 0.91; “While taking examinations I have an uneasy, upset feeling”) and Worry (W-scale, Cronbach’s α = 0.91; “Thinking about my grade in a course interferes with my work on tests”). The AMAS (Cronbach’s α = 0.90) consists of nine situations, such as “Taking an exam in a math course” which the student rates as causing 1 = low to 5 = high anxiety.

Procedure

During the first week of classes, participant rights were explained to the students. They then signed a consent form and took the pre-test survey, TAI, STAI, and AMAS. Scores on the first exam, given within the first month of the semester, and the overall class grade were provided by the instructors. GPA and age were obtained from the registrar with permission of the students.

Measuring underperformance

Simply finding that men outperformed women would not demonstrate underperformance because this difference could be due to a number of uncontrolled factors such as men previously having had more mathematics classes. Additionally, a student could have received high test scores compared to other students but still have underperformed relative to their own usual level of performance. The measure of underperformance used here compared participants’ performance in this particular class with their overall performance in all of their previous college classes, that is, their GPAs.

An Underperformance Index (UPI) was calculated by dividing each student’s GPA by 4.0 to change it to a percentage, transforming the student’s z-scores on the first exam and final class grade into percentages and then subtracting the percentages derived from the earned z-scores from the percentage equivalent to their GPA. A positive UPI indicates worse-than-expected performance (underperformance), while a negative UPI indicates better-than-expected performance (overperformance).

A blocking analysis was done examining the UPI scores by class. No differences were found between the classes, with p-values ranging from 0.101 to 0.965. Thus, underperformance data were grouped across classes when examining the influence of instructor gender.

To examine changes within individuals, difference scores were calculated by subtracting students UPI scores on the final grade in the class from their UPI scores on the first examination. A negative difference score indicates increased underperformance, while a positive score indicates the opposite.

Results

It was hypothesized that if stereotype threat existed in female students, women should underperform more than men on the first examination. Overall, UPI scores for the first examination ranged from −0.42 (overperformance) to 0.74 (underperformance). Women’s first examination mean UPI (M = 0.22, SD = 0.22) was significantly greater than men’s (M = 0.16, SD = 0.24), F(1, 311) = 5.130, p = 0.024, partial eta squared = 0.016. Although these two means attest to the difficulty of the class for most students, many did quite well as evidenced by the range of UPI scores.

It was predicted that women who agreed with the stereotype statement and were the highest achieving would show the greatest underperformance. Endorsement scores were not significantly correlated with underperformance on the first examination for either males or females, even for the highest performing students (GPA > 3.5), as can be seen from Table 2. The distribution of scores for the stereotype statement was highly skewed. Both women and men disagreed more often than they agreed with it (women: 169 vs 28, or 72% vs 12%), χ²(1, n = 197) = 100.92, p < 0.001; (men: 83 vs 22, or 58% vs 15%), χ²(1, n = 105) = 35.44, p < 0.001.

Underperformance was further examined by categorizing endorsers into those who agreed and disagreed with the stereotype. There were no differences among the endorsers or non-endorsers in terms of GPA or age in either men or women. The majority of students disagreed with the stereotype statement in classes taught by both the male (134 non-endorsing, 88% vs 18 endorsing, 12%), χ²(1, n = 152) = 88.53, p < 0.001, and the female instructors (118 non-endorsing, 77% vs 32 endorsing, 23%), χ²(1, n = 150) = 49.31, p < 0.001, as can be seen from Table 3.

Table 3.

Number of student responses to the math stereotype statement measured categorically.

Instructor gender	Students
	Endorsers		No opinion		Non-endorsers
	Male	Female	Male	Female	Male	Female
Male	11	7	17	18	39	95
Female	11	21	22	20	44	74

n = 379.

Instructor gender did not relate to male students’ endorsement of the stereotype statement. Among female students, however, those with the male instructor were more likely to reject the statement (95, 93% non-endorsing vs 7, 7% endorsing) than those with a female instructor (74, 78% non-endorsing vs 21, 22% endorsing), χ²(1, n = 197) = 9.37, p = 0.002.

For male students, there was no difference between the endorsers and non-endorsers in the UPI scores on the first examination. For female students, there was a significant difference in UPI between those who agreed and disagreed with the stereotype: M (endorsers) UPI = 0.32, SD = 0.23; M (non-endorsers) UPI = 0.21, SD = 0.22, t(158) = 2.254, p = 0.035, d = 0.034 with higher UPI scores for endorsers.

Agreement and rejection of the stereotype statement both indicate some level of engagement with the concept. Although it served as the mid-point of a continuous scale, the “no opinion” option indicates that students may have either not thought about the issue or felt it was not relevant to them and, thus, were not engaged with it. Since research indicates that a stereotype must be considered self-relevant to be activated, examining the “no opinion” group gives a fuller picture of the stereotype threat effect. Men were proportionately more likely to indicate “no opinion” (39 or 27%) than women (38 or 16%), χ²(1, n = 456) = 4.27, p = 0.039, suggesting that more women had some opinion on the topic. On the first test, “no opinion” women, but not men, showed a difference in underperformance from the endorsement group F(2, 189) = 3.344, p = 0.037, partial eta squared = 0.034, with “no opinion” women (M underperformance = 0.18, n = 32) showing the lowest underperformance and being significantly different from endorsers (M underperformance = 0.32, n = 24), Hochberg’s GT2 post hoc test = 0.145, p = 0.045.

For men, as predicted, there was no correlation between anxiety levels and stereotype endorsement scores, as can be seen from Table 2. For women, as predicted, there was a significant correlation between endorsement scores and anxiety for both mathematics anxiety (AMAS), r = 0.159, p = 0.007, n = 234, and anxiety-about-the-specific-class, r = 0.113, p = 0.043, n = 234, one-tailed. Examining data categorically, for women only, endorsers reported more mathematics anxiety (M = 27.21, SD = 5.81) than non-endorsers (M = 23.63, SD = 5.86), F(2, 232) = 4.419, p = 0.013, partial eta squared = 0.037, and more anxiety-about-the-specific-class (M = 5.21, SD = l.62) than non-endorsers (M = 4.28, SD = 1.45), F(2, 231) = 4.713, p = 0.010, partial eta squared = 0.039.

It was predicted that only women having female instructors would show a reduction in UPI scores from the first examination to the final course grade. Table 4 shows the UPI scores for the first examination and final course grade by student gender, instructor gender, and endorsement category. For male students, the mean UPI score either increased or did not change from the first examination to the final grade regardless of endorsement category and instructor gender. For men and women who were non-endorsers, in the male instructors’ classes, there was a significant increase in underperformance from the first examination to the final course grade. Women endorsers taught by women initially showed the highest underperformance, but as predicted, this group showed the greatest decrease in UPI although this change was not significant.

Table 4.

Underperformance (UPI) of male and female students on the first exam and final grade in course by instructor gender and endorsement category.

Sex of student	Non-endorsers					Endorsers					No opinion
	UPI 1	UPI final	t-test	df	p	UPI 1	UPI fina l	t-test	df	p	UPI 1	UPI final	t-test	df	p
	M (SD)	M (SD)	t-test	df	p	M (SD)	M (SD)	t-test	df	p	M (SD)	M (SD)	t-test	df	p
Males
Instructor
Male	0.09 (0.18)	0.20 (0.23)	−3.672	32	0.001*	0.11 (0.09)	0.17 (0.17)	−1.630	10	ns	0.22 (0.30)	0.21 (0.20)	0.067	11	ns
Female	0.21 (0.28)	0.21 (0.25)	−0.028	33	ns	0.12 (0.18)	0.22 (0.19)	−1.543	7	ns	0.19 (0.28)	0.24 (0.21)	−1.413	17	ns
Females
Instructor
Male	0.18 (0.22)	0.26 (0.24)	−2.979	75	0.004*	0.23 (0.18)	0.27 (0.20)	−0.536	6	ns	0.13 (0.19)	0.16 (0.18)	−0.620	14	ns
Female	0.25 (0.23)	0.25 (0.22)	−0.034	48	ns	0.36 (0.24)	0.27 (0.19)	2.001	16	ns	0.22 (0.19)	0.23 (0.18)	−0.373	16	ns

UPI: Underperformance Index; SD: standard deviation.

UPI 1 is underperformance on the first exam and UPI final is the underperformance on the final grade in the course.

All t-tests were paired t-tests.

p < 0.0048.

Examining individual difference scores for male and female students separately, only for women was there a main effect for instructor gender. Female students in the male instructor’s class (M = −0.07, SD = 0.23) showed increased underperformance, while those in the female instructors’ classes showed a decrease in underperformance (M = 0.01, SD = 0.19), F(1, 181) = 7.413, p = 0.007, partial eta squared = 0.040. To examine the changes more closely, a series of paired t-tests were performed. Since the use of multiple ts necessitates a higher level of significance, the Bonferroni correction was calculated. Consequently, only p-values of 0.004 or lower were reported as significant.

Regardless of their endorsement category, approximately twice as many male students’ UPI scores increased than decreased, indicating their performance had declined. For women, endorsement category was related to difference scores. While “no opinion” women were as likely to show increases as decreases, endorsing women more often earned positive than negative difference scores (16, 70% vs 7, 30%). Outcomes were reversed in non-endorsing women (48, 39% vs 75, 61%); Yates’ χ²(2, n = 149) = 6.057, Yates’ p = 0.048.

Endorsing women who improved over the course of the semester came, with only three exceptions, from the female instructors’ classes. Non-endorsing women with female instructors were about equally likely to show increases as decreases in their difference scores, while those with the male instructor increased rather than decreased difference scores twice as often. Regardless of instructor gender, “no opinion” women were equally likely to show negative or positive difference scores. To summarize, the performance of endorsing women in female instructors’ classes improved over the course of the semester. Conversely, the performance of non-endorsing women in the male instructor’s classes worsened. As expected, only women with GPAs higher than 3.5 (the upper quartile of the distribution), who were in women’s classes and were endorsers, showed a significant correlation between difference and endorsement scores, r = 0.524, p = 0.015, n = 21, indicating less underperformance by the end of the semester.

Contrary to predictions, there was no significant difference in GPA between those who dropped and completed the course. Male and female students dropped the class in similar proportions: 17% for men and 16% for women. The overall dropout rates in classes with male and female instructors were also similar (15% and 17%, respectively). Male students’ dropping of the class was unrelated to their stereotype endorsement. For female students, however, endorsers tended to drop the class more often (28%) than did non-endorsers (16%). In female instructors’ classes, 19% of endorsing women dropped the class, while in male instructors’ classes, the dropout rate among endorsing women was 46%, as can be seen from Table 5.

Table 5.

Number of dropouts by student gender, instructor gender, and endorsement category.

	Students who dropped the course
	Non-endorsers			Endorsers			No opinion
	Number of dropouts	Percent	Total students	Number of dropouts	Percent	Total students	Number of dropouts	Percent	Total students
Male students
Instructor gender
Male	5	11.4	44	1	8.3	12	2	10.5	19
Female	9	17.0	53	1	8.3	12	6	21.4	28
Female students
Instructor gender
Male	16	14.3	111	6	46.2	13	2	10.0	20
Female	15	16.9	89	5	19.2	26	2	9.1	22

n = 449.

The “no opinion” students (men and women combined) were just as likely to drop the class as either endorsers or non-endorsers. But “no opinion” women were the least prone of the three endorsement categories to drop the class. Of the 42 “no opinion” women, only four dropped the class. “No opinion” women showed this perseverance whether in a class taught by a male (2 dropped; 18 completed; χ²(1, n = 20) = 12.800, p < 0.001) or female instructor (2 dropped; 20 completed; χ²(1, n = 22) = 14.727, p < 0.001).

Discussion and conclusion

The significant differences found in this study between endorsing and non-endorsing women can be summarized as follows: endorsers were more anxious than non-endorsers on some measures, earned higher UPI scores on the first examination, and tended to drop the class more often, especially in classes taught by the male instructor. Anxiety, in particular, anxiety-about-the-specific-class, and mathematics anxiety, was more evident in endorsers than non-endorsers and are probably an indicator of stereotype threat. This finding is in agreement with Miller and Bichsel (2004) and Cherney and Cooney (2005) but contrary to Hunsley’s (1987) results. Most importantly, the stereotype threat demonstrated by endorsing women’s underperformance on the first examination was reduced by the end of the semester when they had female instructors. This is of note because the females in the male instructor’s classes had slightly higher overall GPAs than those in the female instructors’ classes.

Women more often agreed with the stereotype statement and underperformed more on the first examination if they had a female rather than a male instructor. These women may have initially made an upward social comparison in which their own concerns about statistics led them to feel less competent than and different from their instructor. By the end of the semester, perhaps as a result of frequent exposure, any such negative self-comparison may have been reduced. The possibility that upward social comparisons might have been involved in the results is supported by the findings of the Parks-Stamm et al.’s (2008) study relating social comparison to mathematics performance. Instead of being a positive role model, the female instructors may have initially sensitized women to the stereotype, increasing their underperformance. This supposition is also supported by the higher levels of anxiety-about-the-specific-class in women having a female rather than the male instructor. But, congruent with the findings of Stout et al. (2011), by the end of the semester, female instructors appeared to indeed be functioning as positive role models. The decrease in women’s underperformance is particularly impressive since the first examination is usually the easiest, and the final grade incorporates more difficult material. Furthermore, women with female instructors were the only group whose UPI scores decreased by the end of the class. Among endorsing women, the presence of a female instructor also lessened the likelihood of dropping the class.

High GPA women did not show higher UPI scores on the first examination compared with all the other women as had been predicted. But for this group only, there was a correlation between strength of endorsement and improvement in performance over the semester. That high-performing women were not initially affected by stereotype threat but showed improvement later may be explained by the fact that the first test is usually the easiest. As noted by O’Brien and Crandall (2003), when presented with a task that is quite easy, high-performing women turn stereotype threat to their advantage, using their motivation to disprove the stereotype and earn even higher scores.

The “no opinion” option group is worthy of further study. The fact that women chose the “no opinion” option less often than men suggests that women had probably either thought more about the question and/or were more sensitive to it. “No opinion” women were somehow less sensitive to or not engaged with the gender/mathematics issue than other women and, therefore, less likely to be affected by stereotype threat. One characteristic which might distinguish them is the amount of importance they place on gender identity (see Schmader, 2002, who found that gender salience moderated women’s but not men’s mathematics performance under stereotype threat conditions).

A female role model’s ability to reduce anxiety, which in turn may mediate stereotype threat, cannot be directly demonstrated in this study because anxiety was measured only at the beginning of the semester. Researchers exploring this issue in the future should examine changes in anxiety, especially mathematics anxiety, over time and also determine whether there are changes in the endorsement of the mathematics stereotype. Including gender identity salience as a variable in gender/mathematics stereotype studies would help clarify the factors affecting performance outcomes. The major limitation of this study was that there was only one male instructor. Any replication should include equal numbers of male and female instructors. These results were obtained from one US college and involved only an introductory-level class. Because of its quasi-experimental design, its results should be considered with some caution as the participants were not randomly assigned to instructors and other potentially confounding variables could not be controlled.

Our findings have important implications for teaching. In the classroom, instructors, particularly female mathematics and statistics instructors, should be aware of the potential long-term positive effects of their being role models. Even if women students do not initially do well on examinations, they should be encouraged to continue their efforts. Male and female instructors should be aware of and alert to factors such as stereotype threat that differentially affect female students. For example, their language and tone should be free of any suggestion of differing expectations of male and female students. Even the use of positive reinforcement for accomplishment should be offered in such a way that female students are not praised because they exceeded a lower expectation of success. This may require instructors to reflect on their assumptions and interactions with male and female students to be sure that they do not inadvertently reinforce the stereotype.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author biographies

Susan Kapitanoff is an Associate Professor of Psychology. Her research interests span the study of methods of improving the academic performance of college students, especially in statistics, and prejudice reduction in adolescents. Address: Department of Psychology, American Jewish University, 15600 Mulholland Drive, Los Angeles, CA 90077, USA. [email: skapitanoff@aju.edu]

Carol Pandey is a Professor Emeritus of Psychology. Her recent research has been focused on facilitating student performance in statistics and how students are affected by their perceptions of their instructors. Address: Los Angeles Pierce College, Woodland Hills, Los Angeles, CA, USA. [email: pandeycj@piercecollege.edu]

References

Ambady

Shih

Kim

et al . (2001) Stereotype susceptibility in children: Effects of identity activation on quantitative performance. Psychological Science 12(5): 385–90.

Amelink

(2010) Gender differences in math performance. In: Bogue

Cady

(eds) Assessing Women and Men in Engineering. University Park, PA: Pennsylvania State University, pp. 67–83.

Aronson

Lustina

Good

et al . (1999) When white men can’t do math: Necessary and sufficient factors in stereotype threat. Journal of Experimental Social Psychology 35(1): 25–46.

Asgari

Dasgupta

Stout

(2012) When do counterstereotypic ingroup members inspire vs. deflate? The effect of successful professional women on young women’s leadership self-concept. Personality and Social Psychology Bulletin 38(3): 370–83.

Blanton

Christie

Dye

(2002) Social identity versus reference frame comparisons: The moderating role of stereotype endorsement. Journal of Experimental Social Psychology 38(3): 253–67.

Bond

Perkins

Ramirez

(2012) Students’ perceptions of statistics: An exploration of attitudes, conceptualizations, and content knowledge of statistics. Statistics Education Research Journal 11(2): 6–25.

Ceci

Williams

Barnett

(2009) Women’s underrepresentation in science: Sociocultural and biological considerations. Psychological Bulletin 135(2): 172–210.

Cherney

Cooney

(2005) Predicting student performance in a statistics course using the Mathematics and Statistics Perception Scale (MSPS). Transactions of the Nebraska Academy of Science 30: 1–8. Available at: http://digitalcommons.unl.edu/tnas/41

Danaher

Crandall

(2008) Stereotype threat in applied settings re-examined. Journal of Applied Social Psychology 38(6): 1639–55.

10.

George-Jackson

(2011) STEM switching: Examining departures of undergraduate women in STEM fields. Journal of Women and Minorities in Science and Engineering 17(2): 149–71.

11.

Goff

Steele

Davies

(2008) The space between us: Stereotype threat and distance in interracial contexts. Journal of Personality and Social Psychology 94(1): 91–107.

12.

Hopko

Mahadevan

Bare

et al . (2003) The Abbreviated Math Anxiety Scale (AMAS): Construction, validity, and reliability. Assessment 10(2): 178–82.

13.

Hoyt

Simon

(2011) Female leaders: Injurious or inspiring role models for women? Psychology of Women Quarterly 35(1): 143–57.

14.

Hunsley

(1987) Cognitive processes in mathematics anxiety and test anxiety: The role of appraisals, internal dialogue and attributions. Journal of Educational Psychology 79(4): 388–92.

15.

Hyde

Lindberg

Linn

et al . (2008) Gender similarities characterize math performance. Science 32: 494–95.

16.

Keller

(2002) Blatant stereotype threat and women’s math performance: Self-handicapping as a strategic means to cope with obtrusive negative performance expectations. Sex Roles 47(3): 193–8.

17.

Kitzinger

Haran

Chimba

M, et al.

et al (2008) Role models in the media: An exploration of the visual experiences of women in science, engineering, and technology. Research Report, Series for UKRC No. 1, March. Available at: http://cf.ac.uk/jomec/resources/Kitzinger_Report_1.pdf

18.

Latu

Mast

Lammers

et al . (2013) Successful female leaders empower women’s behavior in leadership tasks. Journal of Experimental Social Psychology 49(3): 444–8.

19.

Lockwood

(2006) Someone like me can be successful: Do college students need same—gender role models? Psychology of Women Quarterly 30(1): 36–46.

20.

Lummis

Stevenson

(1990) Gender differences in beliefs and achievement: A cross-cultural study. Developmental Psychology 26: 254–63.

21.

Marx

Roman

(2002) Female role models: Protecting women’s math test performance. Personality and Social Psychology Bulletin 28(9): 1183–93.

22.

Massey

Charles

Lundy

et al . (2002) The Source of the River: The Social Origins of Freshmen at America’s Selective Colleges and Universities. Princeton, NJ: Princeton University Press.

23.

Miller

Bichsel

(2004) Anxiety, working memory, gender and math performance. Personality and Individual Differences 37(3): 591–606.

24.

Muzzatti

Agnoli

(2007) Gender and mathematics: Attitudes and stereotype threat vulnerability in Italian children. Developmental Psychology 43(3): 747–59.

25.

Nosek

Smyth

Sriram

et al . (2009) National differences in gender-science stereotypes predict national sex differences in science and math achievement. Proceedings of the National Academy of Sciences of the United States of America 106: 10593–7.

26.

O’Brien

Crandall

(2003) Stereotype threat and arousal: Effects on women’s math performance. Personality and Social Psychology Bulletin 29: 782–9.

27.

Onwuegbuzie

Wilson

(2003) Statistics anxiety: Nature, etiology, antecedents, effects, and treatment—A comprehensive review of the literature. Teaching in Higher Education 8(2): 195–209.

28.

Parks-Stamm

Heilman

Hearns

(2008) Motivated to penalize: Women’s strategic rejection of successful women. Personality and Social Psychology Bulletin 34(2): 237–47.

29.

Pillsbury

(2010) Women in math: A complex problem. The Chicago Maroon, 19 November. Available at: http://chicagomaroon.com/2010/11/19/women-in-math-a-complex-problem/

30.

Schmader

(2002) Gender identification moderates stereotype threat effects on women’s math performance. Journal of Experimental Social Psychology 38(2): 194–201.

31.

Schmader

Johns

(2003) Converging evidence that stereotype threat reduces working memory capacity. Journal of Personality and Social Psychology 85(3): 440–52.

32.

Schmader

Johns

Barquissau

(2004) The costs of accepting gender differences: The role of stereotype endorsement in women’s experience in the math domain. Sex Roles 50(11–12): 835–50.

33.

Smith

Sansone

White

(2007) The stereotyped task engagement process: The role of interest and achievement motivation. Journal of Educational Psychology 99(1): 99–114.

34.

Spencer

Steele

Quinn

(1999) Stereotype threat and women’s math performance. Journal of Experimental Social Psychology 35(1): 4–28.

35.

Spielberger

Gonzalez

Taylor

et al . (1980) Test Anxiety Inventory: Sample Set, Manual, Test Booklet, and Scoring Key. Redwood City, CA: Mind Garden.

36.

Spielberger

Gorsuch

Lushene

et al . (1983) State-trait Anxiety Inventory for Adults: Sample Set, Manual, Test, Scoring Key. Redwood City, CA: Mind Garden.

37.

Steele

(2010) Whistling Vivaldi: How Stereotypes Affect Us and What We Can Do. New York; London: W. W. Norton & Company.

38.

Steele

Aronson

(1995) Stereotype threat and the intellectual test performance of African Americans. Journal of Personality and Social Psychology 69(5): 797–811.

39.

Steele

Spencer

Aronson

(2002) Contending with group image: The psychology of stereotype and social identity threat. In: Zanna

(ed.) Advances in Experimental Social Psychology. San Diego, CA: Academic Press, pp. 377–440.

40.

Stout

Dasgupta

Hunsinger

et al . (2011) STEMing the tide: Using in group experts to inoculate women’s self-concept in science, technology, engineering, and mathematics (STEM). Journal of Personality and Social Psychology 100(2): 255–70.

41.

Zeidner

(1991) Statistics and mathematics anxiety in social science students: Some interesting parallels. British Journal of Educational Psychology 61(3): 319–28.