Female peer mentors early in college increase women’s positive academic experiences and retention in engineering

Significance

The scarcity of women in the American science and engineering workforce is a well-recognized problem. However, field-tested interventions outside artificial laboratory settings are few. We provide evidence from a multiyear field experiment demonstrating that women in engineering who were assigned a female (but not male) peer mentor experienced more belonging, motivation, and confidence in engineering, better retention in engineering majors, and greater engineering career aspirations. Female mentors promoted aspirations to pursue engineering careers by protecting women’s belonging and confidence. Greater belonging and confidence were also associated with more engineering retention. Notably, grades were not associated with year 1 retention. The benefits of mentoring endured beyond the intervention, for 2 y of college, the time of greatest attrition from science, technology, engineering, and mathematics (STEM) majors.

Abstract

Scientific and engineering innovation is vital for American competitiveness, quality of life, and national security. However, too few American students, especially women, pursue these fields. Although this problem has attracted enormous attention, rigorously tested interventions outside artificial laboratory settings are quite rare. To address this gap, we conducted a longitudinal field experiment investigating the effect of peer mentoring on women’s experiences and retention in engineering during college transition, assessing its impact for 1 y while mentoring was active, and an additional 1 y after mentoring had ended. Incoming women engineering students (n = 150) were randomly assigned to female or male peer mentors or no mentors for 1 y. Their experiences were assessed multiple times during the intervention year and 1-y postintervention. Female (but not male) mentors protected women’s belonging in engineering, self-efficacy, motivation, retention in engineering majors, and postcollege engineering aspirations. Counter to common assumptions, better engineering grades were not associated with more retention or career aspirations in engineering in the first year of college. Notably, increased belonging and self-efficacy were significantly associated with more retention and career aspirations. The benefits of peer mentoring endured long after the intervention had ended, inoculating women for the first 2 y of college—the window of greatest attrition from science, technology, engineering, and mathematics (STEM) majors. Thus, same-gender peer mentoring for a short period during developmental transition points promotes women’s success and retention in engineering, yielding dividends over time.

The odds do not favor women in most physical sciences, engineering, and computing. Despite educational advances, women, who constitute 56% of university students in the United States (1), hold only 13–33% of bachelor’s and master’s degrees and 11–21% of doctoral degrees in these fields (2). Even among degree holders in engineering, computing, and physical sciences, women are less likely than men to hold jobs related to science, technology, engineering, and mathematics (STEM) degrees (2). Overall, the proportion of women in physical sciences, engineering, and computer science is very small relative to men and gets smaller still with every level of advancement (3). Engineering is notable for having one of the lowest proportions of women among all sciences (2) and is the focus of our research.

Attempts to explain the relative scarcity of women engineers as due to women’s “free choice” to pursue alternate career paths (4), or lower aptitude and intrinsic motivation (5), neglect widespread structural and psychological contributors to this phenomenon (6, 7). Many engineering environments are subtly unfriendly or sometimes overtly hostile for women (8, 9). The numeric scarcity of women (10, 11), nonverbal behavior from male colleagues that excludes women from professional conversations (12), use of masculine pronouns to refer to all scientists and engineers (12, 13), and the prevalence of sexist jokes (14) all signal to women that they are outsiders who do not belong in engineering (6, 15, 16). Even in organizations that prioritize diversity, the ideal engineer is implicitly assumed to be male (17), eroding women’s belonging and self-efficacy, leading to burnout and attrition (18).

A number of interventions aim to counter negative effects of STEM stereotypes on women (19), but few have been tested in naturally existing field settings (cf. refs. 20 and 21). One real-world exception, aimed at increasing diversity, is mentoring, which is in widespread use in the academy (22), government (23), and industry (24), and commonly assumed to work because it is correlated with positive health, attitudes, motivation, and behaviors (25). Despite its popularity, however, evidence supporting mentoring is shaky because serious methodological flaws make it impossible to separate benefits of mentoring from confounds (see refs. 25–28 for meta-analyses). Most studies used correlational surveys, case studies, pretest–posttest studies of a single mentored group with no comparison group or nonequivalent comparison groups. Participants opted in knowing these studies were on mentoring and self-reported how mentors affected them, raising concerns about sampling bias and self-report bias, which could have inflated positive results. Mentees and mentors often chose each other, raising doubts as to whether mentoring in general, or a unique connection between mentor–mentee, produced the benefits. Randomized controlled experiments are rare in mentoring research, making it impossible to determine whether having a mentor (vs. no mentor) produced any benefits.

Another source of ambiguity in mentoring research comes from not knowing whether ingroup vs. outgroup mentors are most beneficial. Some research suggests that women reap more benefits from male mentors in professional settings because men, being advantaged, confer organizational legitimacy on their mentees and provide resources required for success (29, 30). However, other studies argue that for women who are a small minority in achievement settings, female mentors enhance social belonging in otherwise alienating environments (6, 31, 32). Role model research also suggests that mere exposure to successful ingroup (vs. outgroup) members enhances motivation and aspirations among negatively stereotyped individuals (31, 33–36). Applied to women in STEM, past research supports three possible predictions: male mentors may be best; female mentors may be best; or any mentor regardless of gender may be better than no mentor. Unfortunately, the absence of controlled experiments comparing the effect of mentor gender on women in STEM makes it impossible to adjudicate this issue.

Our goal is to investigate these unresolved questions by assessing whether mentoring—presumed to be beneficial—has any real causal benefits for women in engineering, using a longitudinal randomized controlled field experiment. Second, we sought to test whether mentors’ group membership has any impact on mentee outcomes, and if so, why. Third, if mentoring is beneficial, we sought to investigate how long that benefit endures after the mentor is gone. Finally, whereas mentoring is usually a hierarchical relationship between experts and novices (25), we focused on peer mentoring (37) between advanced students and first-years because this is easy to scale up without placing undue burden on women faculty in STEM if same-gender mentors are necessary. These goals were informed by the Stereotype Inoculation Model (6), which predicts that—analogous to a biomedical vaccine that inoculates one’s physical body against noxious bacteria—exposure to ingroup experts and peers serves as a “social vaccine” that inoculates one’s mind against noxious stereotypes, and is especially effective during developmental transitions when individuals experience most self-doubt.

Current Study

We conducted a multiyear longitudinal field experiment investigating whether a peer mentoring intervention, with advanced students as mentors, would increase the success of women who are beginners in engineering. We predicted that beginning women students paired with female peer mentors would have better experiences in engineering than women without mentors, in terms of belonging in the major, self-efficacy, less anxiety (threat), and more motivation (challenge). Second, we expected that women with female mentors would have higher retention in engineering and more intentions to pursue advanced engineering degrees than controls. Third, we proposed that positive everyday experiences in engineering (such as feelings of belonging) would likely mediate any effects of having a female mentor on women’s future intentions to pursue engineering after college. Fourth, the benefits of same-gender mentoring were expected to endure long after mentoring ended. Finally, we had competing predictions regarding the impact of male mentors. Although some past research suggests that women with any mentor regardless of gender fare better than those without mentors (38), other research suggests that women with male mentors fare best in fields where men dominate (29, 30), whereas research on stereotype inoculation and role modeling suggests that women with female mentors fare best (6, 31).

To test these predictions, we recruited 150 female students, all incoming majors in engineering at a public university, by sending mass emails to all women in each entering class. Participants were randomly assigned to one of three conditions: one-third was assigned to female peer mentors, one-third to male peer mentors, and the rest had no mentor (control group). Mentor–mentees met in person roughly once a month and mentors kept track of their interactions using online surveys. All were blind to experimental hypotheses (see details in SI Materials and Methods). Mentoring relationships lasted for 1 y. We surveyed mentees’ experiences in engineering at three time points during year 1: before mentor assignment at the beginning of the year, and then at the middle and end of the academic year when mentoring relationships were active. A fourth survey was administered 1 y after mentoring had ended (year 2). We measured participants’ belonging in engineering, self-efficacy, feelings of threat and challenge, career aspirations, and global appraisals of engineering. College transcripts, obtained from the university registrar with students’ consent, provided grades and retention information in engineering majors.

SI Materials and Methods

Measures.

Psychological experiences in engineering.

Our primary measures focused on participants’ experiences in engineering contexts measured at multiple time points. These measures are listed below.

Social belonging.

Four items assessed the degree to which participants felt they belonged in an engineering major. The items were “I felt connected to my peers in engineering,” “I felt accepted by my peers in engineering,” “I felt like an outsider among my peers in engineering” (reverse coded), and “I felt invisible among my peers in engineering” (reverse coded). Participants rated the degree to which they agreed or disagreed with these items from 1 (not at all true) to 7 (very true). These items, adapted from prior research (45), were aggregated into a composite index of social belonging (values of α = 0.78–0.88).

Threat and challenge.

We used 10 items to assess the degree to which participants experienced their engineering contexts as threatening (anxiety-provoking) vs. challenging (motivating), adapted from past research (46–50). The five threat items were “My engineering related classes this year are likely to be difficult,” “I feel worried about my engineering related classes this year,” “I feel stressed about my engineering related classes this year,” “I feel unsure about my engineering related classes this year,” and “I feel anxious about my engineering related classes this year.” The five challenge items were “I have the basic skills and abilities to be successful in my engineering related classes this year,” “I will be able to overcome any difficulties I experience in my engineering related classes this year,” “I have what it takes to handle my engineering related classes this year,” “I am prepared to deal with my engineering related classes this year,” and “I feel confident about my engineering related classes this year.” Participants indicated their agreement or disagreement with each item on scales ranging from 1 (not at all true) to 7 (very true). All items appeared in random order and were combined into two separate indices of threat (values of α = 0.83–0.88) and challenge (values of α = 0.85–0.92). At data analysis, threat and challenge were treated as separate variables and also analyzed as a ratio of threat to challenge in keeping with past research (46–50). The ratio measure assesses the degree to which participants’ anxieties about engineering (perceived threat) are offset by their motivation to overcome these difficulties with their existing skills and ability (perceived challenge). This ratio measure is directly related to Lazarus and Folkman’s theory of stress and other theories that refer to stress as a relative balance between demands and resources (51, 52). Although the ratio measure is a popular metric of threat–challenge balance, some research has found that threat and challenge may also have independent effects on psychological outcomes and behavior (49, 50). As such, in our research threat and challenge were treated as a ratio as well as separate variables.

Engineering self-efficacy.

Two questions (31) were used to assess women’s appraisals of their talent and confidence in engineering: “Do you think you have a talent for engineering?” and “How confident do you feel about your engineering ability?” Women evaluated themselves on scales ranging from 1 (not at all) to 7 (very much so). These items were combined into a single index for each time point because they were highly correlated (values of r = 0.64–0.81, values of P < 0.001).

Evaluations of mentor–mentee relationship.

Women in the two mentor conditions were asked eight questions about their relationship with their mentors at time 2 and time 3: (i) “How much do you identify with your peer mentor?” (ii) “How similar do you feel to your peer mentor?” (iii) “Do you feel personally connected to your peer mentor?” (iv) “Do you feel your mentor–mentee relationship has good chemistry?” (v) “How much do you admire your peer mentor?” (vi) “How much support have you been getting from your peer mentor?” (vii) “How much has your peer mentor been available to you?” (viii) “Can you imagine yourself achieving a similar level of success in engineering as your peer mentor in the future?” Women responded to each question on a scale from 1 (not at all) to 7 (very much). These items, adapted from past research by Dasgupta and colleagues (31, 36, 53), were analyzed separately and also combined into a single index (values of α = 0.94–0.96).

Perceived mentor–self overlap.

One item, adapted from The Inclusion of Other in the Self Scale (54) assessed the degree to which women felt close to their mentors and thus included their mentors within their own self-concept. At time 2 and time 3, women with mentors were asked, “Which of the pictures below best describes your mentor–mentee relationship? Please type the number that best corresponds to your response.” Below the text were seven pairs of circles showing increasingly overlapping Venn diagrams; each pair of circles was assigned a number from 1 to 7, in ascending order. The first pair of circles was nonoverlapping, whereas the seventh pair of circles represented near-complete overlap. The Inclusion of Other in the Self Scale was originally designed to assess the degree to which significant others’ identities are included in one’s own self-concept. According to the self-expansion model, when a close other makes his or her resources available to a relationship partner (not necessarily a romantic partner), the relationship partner will perceive those resources as their own, leading to a greater perceived degree of overlap between the other and the self (55). If women feel close to their mentors, this should be reflected in the degree of overlap they perceive with their mentors, and might indicate a perception of sharing resources with the mentor.

Future outcomes in engineering.

Another set of primary measures involved student retention in the major and their future aspirations.

Retention in engineering majors.

Retention was calculated using women’s official major as listed on their college transcript obtained from the university registrar. At time 1, all participants were officially registered in one of six engineering majors at the university or were registered as a pre-engineering major working on fulfilling the prerequisites. If women’s majors changed to nonengineering majors in a subsequent semester, we coded them as having dropped out of engineering.

Frequency of thoughts about switching majors.

The frequency of women’s thoughts about changing their major was evaluated via the single item: “How often do you think about changing your major?” Women rated the frequency from 1 (not at all) to 7 (a lot).

Intentions to pursue advanced degrees in engineering.

Using an item from prior published research (10), participants were asked: “How likely are you to pursue graduate study in engineering (master’s or PhD)?” Response options ranged from 1 (not at all likely) to 7 (very likely).

Intentions to pursue engineering careers.

To assess women’s intentions to pursue careers in engineering, we used the following item from prior research (10, 31): “How likely are you to pursue a professional job in engineering?” Women rated their likelihood from 1 (not at all likely) to 7 (very likely).

Global appraisals of engineering.

A third set of measures borrowed from prior research (10, 31) assessed participants’ global appraisals of engineering including their overall attitudes toward engineering, identification with the field, and stereotypes of engineering. These were assessed using self-report items and also unobtrusively using Implicit Association Tests (IAT) (56). Global appraisals of engineering were only measured in year 1 and were not included in the year 2 follow-up survey because they were less important to the goals of this study than students’ everyday experiences in engineering environments described above.

Implicit stereotypes about engineering.

One IAT measured how strongly participants associated engineering compared with English with masculine vs. feminine concepts by assessing how quickly gender “popped into mind” when people thought of each discipline. This measure uses participants’ reaction time as an unobtrusive measure of stereotyping (56). One feature of the IAT is that preference for one discipline (e.g., engineering) is assessed in relation to a second discipline (e.g., English). Participants saw four types of words appear on a computer screen in rapid succession; words related to the following: (i) engineering (e.g., equation, computation); (ii) English (e.g., literature, poetry); (iii) masculine pronouns (e.g., he, him); and (iv) feminine pronouns (e.g., she, her). Participants classified each word as quickly as possible using one of two response keys. For some blocks they were instructed to use the same key to classify engineering and feminine words and a different key to classify English and masculine words (abbreviated as engineering + feminine, English + masculine). In other blocks, response key assignment was reversed (i.e., engineering + masculine, English + feminine). The order in which these two blocks were completed was counterbalanced between participants. If participants implicitly stereotype engineering as masculine, they ought to be faster at grouping engineering + masculine and English + feminine and comparatively slower at grouping engineering + feminine and English + masculine. However, if they have no gender stereotypes about disciplines, they should be equally fast at both blocks. The difference in participants’ response latency for each type of block converted into effect size was a measure of implicit stereotypes of engineering. Participants completed a total of seven blocks of trials in the IAT of which three were practice bocks and four were data collection blocks.

Explicit stereotypes of engineering.

Participants also completed four items assessing explicit gender stereotypes about engineering. They were asked to indicate on a 7-point scale whether they thought of “mostly men” (anchored at 1), “equal numbers of men and women” (anchored at 4), or “mostly women” (anchored at 7) when they thought of people who are very good at engineering, and people who have careers in engineering. Two virtually identical items assessed what participants thought “other people” believe. All four items were combined into a single index (values of α = 0.66–0.78).

Implicit attitudes toward engineering.

A second IAT assessed participants’ implicit attitudes toward engineering compared with English by measuring how quickly they associated engineering compared with English with good concepts (e.g., love) vs. bad concepts (e.g., hate). The basic procedure of this IAT was identical to the one described above. If participants implicitly prefer engineering to English, they ought to be faster at grouping together “engineering + good” and “English + bad,” and slower at grouping “engineering + bad” and “English + good.” However, if they prefer English to engineering, they ought to be faster at the opposite groupings.

Explicit attitudes toward engineering.

Explicit attitudes toward engineering were measured using 7-point scales anchored by dislike–like, hate–love, boring–fun, and bad–good (31). Ratings on the four items were combined into a single index (values of α = 0.86–0.88).

Implicit identification with engineering.

A third IAT assessed participants’ implicit identification with engineering vs. English by assessing how quickly they associated engineering compared with English with first-person pronouns (e.g., I, me) vs. third-person pronouns (e.g., they, them) (31). The basic procedure of this IAT was identical to the one above. If participants implicitly identify more strongly with engineering than English, they ought to be faster at grouping together “engineering + me” and “English + they” than vice versa.

Explicit identification with engineering.

Explicit identification with engineering was measured with three items: How important is engineering to you? How useful is engineering to you? How much do you care about doing well in engineering? Participants responded on 7-point scales anchored by “not at all” to “very much” (values of α = 0.68–0.94).

Procedure.

Participants.

Four cohorts of female students intending to major in engineering were recruited to participate in our experiment (n = 158) in academic years 2011–2012, 2012–2013, 2013–2014, and 2014–2015. Of the total sample, 80% were first-years and 20% were sophomores or transfers recruited from new student orientations, first-year seminars, and via emails to all women entering engineering majors. Participants were paid $20 for the first assessment in early fall of their first year (time 1), $30 for the midyear assessment (time 2), and $35 for the end-of-year assessment (time 3), and were given a $10 gift card if they completed a brief survey at the end of their second year (time 4). Of the original 158 women recruited in the fall semester of their first year, eight (5.1%) dropped out of the experiment within a few weeks after the baseline (time 1) assessment and had no contact with mentors. We report results for the remaining 150 women who completed at least one assessment after assignment to mentor condition.

Female student recruits were told that our experiment aimed at investigating “factors related to success and development in engineering majors.” Participants were unaware that the experiment had anything to do with mentoring. They completed a baseline assessment (time 1) in August or September of their starting academic year, a midyear assessment (time 2) in January or February, and a year-end assessment (time 3) in April or May. A time 4 survey was administered at the end of year 2 in late spring or summer. A female experimenter administered the first three assessments at a quiet location using laptop computers. The order of all measures was counterbalanced between participants with demographics always administered last. With participants’ consent, we obtained their transcripts from the university registrar each year until their graduation. Following the time 1 assessment, all participants were randomly assigned to a female mentor (n = 52), male mentor (n = 51), or given no mentor (control group; n = 47). Mentors were matched with mentees who were in the same engineering major as they. A few women, who entered the study without a specific engineering major (i.e., they were generic engineering premajors), were matched to an available mentor if they had been randomly assigned to a mentor condition.

Peer mentors.

Mentors (N = 58) were recruited via email based on recommendations from engineering faculty before the start of the academic year. They were mostly seniors and some juniors who were student leaders of engineering organizations on-campus and high performers in engineering. Mentors were all declared majors in one of four departments in the College of Engineering: Chemical Engineering, Civil and Environmental Engineering, Electrical and Computer Systems Engineering, and Mechanical and Industrial Engineering. Thirty-two mentors were female, and twenty-six were male. Nine of these mentors participated for 2 consecutive years (when the mentor was both a junior and a senior). Each mentor had between 1 and 5 mentees (28 mentors had 1 mentee each, 19 mentors had 2 mentees each, 9 mentors had 3 mentees each, and 2 mentors had 5 mentees each). Mentors attended a half-day training session with the researchers at the beginning of the academic year.

The training had three important components. First, we asked peer mentors to reflect back on their own early experiences in college and asked them to respond to four questions: What difficulties (if any) did they experience in their first 2 y of college? Who or what helped them through these difficulties? What types of support did they wish they had had in their early years of college, but that were missing? Are there any issues related to engineering education or college in general that they wish someone had told them about that they think new students should know? This focus group discussion yielded many common themes that were summarized visually on a white board. We used the discussion to emphasize the importance of understanding what factors keep students interested in engineering majors and motivated to pursue careers in engineering; and what other factors make students drop out of engineering to pursue another major. We emphasized the importance of their role as mentors in the lives of students with whom they would work.

Second, we delved in the structure and content of the mentoring work they would do. Mentors were asked to meet with their mentee individually once a month for the entire academic year. We provided mentors with a lot of ideas about what to discuss during their get-togethers with mentees—specifically, using that time to (i) develop a strong personal connection with mentees (e.g., by initiating joint social activities, helping mentees develop a social network on-campus, and sharing their own early college experiences with mentees); (ii) give academic advice [e.g., on coursework, balancing course load, developing good work habits, forming good relationships with faculty and teaching assistants (TAs), using all campus academic resources including faculty and TA office hours, tutoring services, and campus workshops, alerting mentees to expect critical feedback and failure, and sharing personal stories of how mentors dealt with similar failure in their own lives]; (iii) provide occasional tutoring if needed; (iv) help mentees develop long-term college plans [e.g., what courses to take when; and when and how to get involved in student societies, seek out research assistantships in faculty laboratories, and apply for summer Research Experience for Undergraduates (REU) programs and internships, etc.]; and (v) help mentees develop postcollege career plans (e.g., when and how to seek out summer REU programs, off-campus internships, how to explore different types of career opportunities, etc.). These ideas were summarized in a handout that all mentors received during training. We encouraged mentors to tailor their mentoring activities in a way that met their mentees’ needs. We emphasized that they did not have to use all of the suggestions in the handout, only those that were relevant to their mentees’ needs.

Third, we discussed logistics of the program. Mentors were told that they would be paired with one to three new students in the same major as they in September. They were asked to get in touch with their mentees right away and make plans to get together one-on-one once a month for the entire academic year (ideally three times in the fall and spring). We recommended that each get-together be at least 1 h long. We asked mentors to complete a brief online survey after each meeting telling us what they did together and the content of their conversations. In each diary entry, mentors recorded their name, their mentee’s name, the date and time of the meeting, and responded to the following open-ended questions: “Briefly describe what you did together,” “List the issues or topics that came up in your conversation,” “How much informal contact have you had this past month with your mentee outside of your official meetings? Do you email, text, talk on the phone, Skype, or chat? If so, could you briefly describe what you do and how frequently you are in touch?” A member of our research team kept track of survey completions each month, contacted mentors if they had not completed a survey for the month to remind them to meet with their mentees, and brainstormed obstacles if necessary. At the beginning of the spring semester, we organized a pizza party for all mentors that year to meet and discuss how they were doing, celebrate successes, and acknowledge and troubleshoot difficulties. During this session, mentors often gave each other advice on how to handle scheduling difficulties and other obstacles they might have been facing.

Coding students’ interactions with mentors.

The open-ended diary entries for each mentor–mentee meeting were deidentified, and two research assistants categorized each meeting according to the content of the activities and discussions. Meetings categorized as “primarily social” included activities and conversations related to mentor and mentees’ social and personal lives, friendship formation, and conversations about fit or belonging in college or in the major. In contrast, meetings categorized as “primarily career-related” included activities and conversations related to engineering coursework, homework, internships, career plans, study skills or studying, or accessing engineering-related resources. Research assistants also made frequency judgments to track the amount of electronic contact between mentors and mentees (e.g., texting, online messaging, social media, email, Skype, etc.) based on the mentors’ open-ended responses. Interrater agreement was good (intraclass correlation coefficients, 0.76–0.94), and a third research assistant “broke the tie” in the case of any disagreements.

Results

Mentoring Quality.

Male and female mentors did not differ in the quality or quantity of their interactions with mentees. Participants perceived their mentors to be equally supportive regardless of mentor gender; they admired and felt connected to all mentors regardless of gender; and they met equally frequently regardless of mentor gender, all indicating that male and female mentors were equally conscientious (Tables S1 and S2). The only advantage for female mentors was that women mentees felt somewhat closer and more similar to female mentors than male mentors.

Table S1.

Mentees' evaluations of mentoring relationships

Mentee reports (7-point scales: 1 = not at all, 7 = very much)	Male mentors		Female mentors
	Mean	SE	Mean	SE	P value
Support received from mentor	4.69	0.28	4.92	0.25	0.55
Availability of mentor	5.04	0.25	4.96	0.26	0.83
Personal connection with mentor	4.35	0.28	4.69	0.22	0.33
Chemistry with mentor	4.86	0.27	5.12	0.23	0.46
Admire mentor	4.84	0.26	5.31	0.23	0.18
Can attain same level of success as mentor	4.86	0.27	5.40	0.22	0.12
Feel similar to mentor	4.24	0.26	4.83	0.22	0.09
Identify with mentor	4.51	0.27	5.10	0.21	0.09
Feel close to mentor (mentor–self overlap)	4.27	0.23	4.90	0.23	0.05

Table S2.

Mentors' reports of mentoring relationships

Mentor reports	Male mentors		Female mentors
	Median	Range	Median	Range	P value
Face-to-face meetings (total no.)	4.00	0–7	4.00	0–7	1.0
% of meetings with career content	50	0–100	50	0–69	0.58
% of meetings with social content	33	0–88	37	0–100	0.70
How career-oriented vs. social were meetings: 1 (career) to 6 (social)	1.92	1–4	2.00	1–6	0.32
Other types of contact (text, social media, email): 1 (not at all) to 6 (very frequent)	2.04	1–6	2.11	1–5	0.33

Analytic Strategy.

Multilevel modeling was used to analyze whether random assignment to mentor condition changed mentees’ experiences over 1 and 2 y, using participant experiences before mentor assignment as the baseline. Below, we first describe how women’s engineering outcomes change over time separately within each mentor condition. We then compare whether change trajectories differed significantly across conditions.

Female Mentors Protect Positive Academic Experiences in Engineering.

In terms of belonging in engineering, women with no mentors and those with male mentors reported steep declines in feelings of belonging in engineering from the beginning to end of the first year (B = −0.45, SE = 0.17, P = 0.007, and B = −0.42, SE = 0.18, P = 0.02, respectively), whereas women with female mentors maintained positive belonging that did not change across the first year of college (B = 0.13, SE = 0.18, P = 0.46). Comparing change trajectories between conditions, women with female mentors reported more stable belonging than those without mentors (B = 0.58, SE = 0.25, P = 0.03) or with male mentors (B = 0.58, SE = 0.25, P = 0.024). Women with male mentors did not differ from those without mentors (B = −0.04, SE = 0.26, P = 0.89) (Fig. 1).

Fig. 1.

Effect of mentor condition on women’s belonging in engineering. The y-axis values are difference scores from time 1, before mentor assignment. Deviations from zero show a relative increase or decrease from time 1. Statistical analyses were conducted using actual responses, not difference scores.

We next examined the impact of mentoring on students’ self-efficacy in engineering. Women without mentors showed steep declines in self-efficacy across the first year (B = −0.63, SE = 0.17, P < 0.001), as did those with male mentors (B = −0.29, SE = 0.17, P = 0.08). In contrast, women with female mentors maintained positive self-efficacy that did not change (B = 0.03, SE = 0.17, P = 0.862). Comparing change trajectories between conditions, students with female mentors reported more stable self-efficacy than those with no mentors (B = 0.66, SE = 0.24, P = 0.007). Male mentors fell in-between and did not differ from either group (SI Results) (Fig. 2).

Fig. 2.

Effect of mentor condition on women’s self-efficacy in engineering. The y-axis values are difference scores from time 1, before mentor assignment. Deviations from zero show a relative increase or decrease from time 1. Statistical analyses were conducted using actual responses, not difference scores.

Female mentors also affected the degree to which students’ anxiety about engineering (threat) was offset by their belief that they possessed skills to overcome academic difficulties (challenge). This was measured as the ratio of threat vs. challenge. A threat/challenge ratio greater than 1 indicates that women’s anxiety overwhelmed their perceived skills; a ratio less than 1 indicates that perceived skills outweighed anxiety, and a ratio of 1 indicates an equal standoff between anxiety and perceived skills. Women with no mentors felt increasingly threatened more than challenged as the first year progressed (B = 0.32, SE = 0.13, P < 0.001), as did those with male mentors (B = 0.17, SE = 0.09, P = 0.059). In contrast, women with female mentors did not show any change in threat vs. challenge across the year (B = 0.07, SE = 0.09, P = 0.445). Comparing change trajectories between conditions, women with female mentors exhibited significantly less rise in threat vs. challenge than those with no mentors (B = −0.25, SE = 0.13, P = 0.047). Students with male mentors fell between the other two conditions, nonsignificantly different from both (Fig. 3). These results are specific to threat/challenge ratio; threat and challenge, considered separately, did not yield group differences (details in SI Results).

Fig. 3.

Effect of mentor condition on women’s feelings of threat vs. challenge in engineering. The y-axis values are difference scores from time 1, before mentor assignment. Deviations from zero show a relative increase or decrease from time 1. Statistical analyses were conducted using actual responses, not difference scores.

Female Mentors Protect Retention and Postdegree Aspirations in Engineering.

To examine whether mentoring would reduce women’s attrition from engineering, we examined the frequency with which women thought about switching majors, their objective retention rates in engineering majors, and their intentions to pursue advanced degrees in engineering after college. Women without mentors increasingly thought about switching to another major over time (B = 0.94, SE = 0.28, P < 0.001), whereas for those with female and male mentors, thoughts about switching majors did not change over time (B = 0.27, SE = 0.27, P = 0.31, and B = 0.19, SE = 0.26, P = 0.47 respectively). Comparing change trajectories between conditions, women without mentors reported a marginally greater increase in thoughts of switching majors than those with female or male mentors (B =0.74, SE = 0.38, P = 0.055; B =0.66, SE = 0.39, P = 0.089, respectively) (Fig. S1).

Fig. S1.

Effect of mentor condition on changes in thoughts of switching majors. All y-axis values are difference scores from time 1, before mentor assignment. Deviations from zero represent a relative increase or decrease from time 1. Statistical analyses were conducted using actual responses, not difference scores.

Although thoughts about switching majors looked similar for the two mentor conditions, when it came to actual decisions to stay or leave, female mentors were more beneficial: 100% of women with female mentors remained in engineering majors at the end of year 1 compared with 82% with male mentors, and 89% without mentors (χ² = 8.19, P < 0.01, Cohen’s d = 0.48) (Fig. 4).

Fig. 4.

Effect of mentor condition on women’s retention in engineering majors at end of year 1.

In terms of after-college aspirations, women with no mentors and male mentors showed declining intentions to pursue advanced degrees in engineering (B = −1.06, SE = 0.25, P < 0.001, and B = −0.71, SE = 0.24, P = 0.003, respectively), whereas those with female mentors maintained consistent intentions to pursue advanced degrees in engineering over time (B = −0.06, SE = 0.23, P = 0.806). Comparing trends across conditions, there was a significant drop in advanced degree intentions for women without mentors compared with those with female mentors (B = −1.01, SE = 0.34, P = 0.004), and a marginal drop compared with male mentors (B = −0.65, SE = 0.33, P = 0.054). Trends in advanced degree intentions did not differ significantly between the two mentor conditions (B = 0.36, SE = 0.34, P = 0.298) (Fig. 5). In short, women with female mentors consistently intended to pursue advanced engineering degrees after college as the year progressed; this trajectory was significantly better than the trend for those without mentors, whereas women with male mentors fell in between.

Fig. 5.

Effect of mentor condition on women’s intentions to pursue advanced degrees in engineering. The y-axis values are difference scores from time 1, before mentor assignment. Deviations from zero show a relative increase or decrease from time 1. Statistical analyses were conducted using actual responses, not difference scores.

Social Belonging and Self-Efficacy Mediate the Relation Between Mentor Condition and Engineering Career Aspirations.

We used multilevel structural equation modeling to test whether change over time in women’s belonging, self-efficacy, or threat relative to challenge (treated as multiple mediators) would mediate the effects of mentor condition on changes in engineering career aspirations during the first year of college. Our analyses revealed that women with female mentors (vs. no mentors) reported more stable feelings of belonging and self-efficacy in engineering over time, both of which in turn predicted increased intentions to pursue future careers in engineering [B = 0.24, SE = 0.12, P = 0.041, lower-level confidence interval (LLCI) = 0.03, upper-level confidence interval (ULCI) = 0.49, and B = 2.58, SE = 0.98, P = 0.008, LLCI = 0.65, ULCI = 4.49, respectively]. Although having a female mentor also protected feelings of threat vs. challenge, this variable did not significantly mediate the relation between mentor condition and engineering career aspirations (B = −1.77, SE = 0.93, P = 0.058, LLCI = −3.59, ULCI = 0.05). These results suggest that having a female peer mentor (vs. no mentor) protects women’s career aspirations in engineering by preserving belonging and feelings of self-efficacy. See Table S3 and SI Results for more details.

Table S3.

Social belonging and engineering self-efficacy mediate the relation between mentor condition and engineering career aspirations

Independent variable (X)	Multiple mediators (M)	Dependent variable (Y)	a paths	b paths	c path	a*b path indirect effects for ΔM1, ΔM2, ΔM3	a*b 95% CI using ΔM1, ΔM2, ΔM3
			X → ΔM1	ΔM1 → ΔY	X → ΔY
			X → ΔM2	ΔM2 → ΔY			LLCI	ULCI
			X → ΔM3	ΔM3 → ΔY
Female mentor vs. no mentor	Belonging (M1)	Engineering career aspirations	0.58 (0.27)*	0.41 (0.07)***	−1.19 (0.98)	0.24 (0.12)*	0.03	0.49
	Self-efficacy (M2)		0.65 (0.25)**	3.95 (0.12)***		2.58 (0.98)**	0.65	4.49
	Threat/challenge (M3)		−0.26 (0.14)†	6.94 (0.18)***		−1.77 (0.93)†	−3.59	0.05
Female vs. male mentor	Belonging (M1)	Engineering career aspirations	0.55 (0.26)*	0.41 (0.08)***	−0.87 (0.96)	0.22 (0.11)*	0.02	0.46
	Self-efficacy (M2)		0.32 (0.24)	3.95 (0.12)***		1.27 (0.96)	−0.61	3.14
	Threat/challenge (M3)		−0.10 (0.13)	6.94 (0.18)***		−0.71 (0.91)	−2.50	1.08
Male mentor vs. no mentor	Belonging (M1)	Engineering career aspirations	0.03 (0.27)	0.41 (0.07)***	−0.32 (0.22)	0.01 (0.11)	−0.21	0.24
	Self-efficacy (M2)		0.33 (0.25)	3.95 (0.12)***		1.31 (0.98)	−0.61	3.24
	Threat/challenge (M3)		−0.15 (0.13)	6.94 (0.18)***		−1.06 (0.93)	−2.90	0.75

Note: ^†P < 0.10, *P < 0.05, **P < 0.01, and ***P < 0.001; values for a, b, c, and a*b paths are unstandardized slopes with SEs in parentheses.

Comparing women with female vs. male mentors, only social belonging emerged as a significant mediator. Specifically, women with female mentors reported more stable feelings of belonging in engineering over time than others with male mentors; belonging in turn predicted increased intentions to pursue engineering careers (B = 0.22, SE = 0.11, P = 0.046, LLCI = 0.02, ULCI = 0.46). Analyses comparing the male-mentor vs. no-mentor conditions were nonsignificant. See Table S3 and SI Results. Taken together, these results suggest that greater belonging and self-efficacy may serve as underlying psychological processes that explain why female peer mentors, who are slightly more advanced in college, would promote engineering career aspirations among women who are new to engineering.

One Year Later: Long-Term Effects of Female Mentors on Student Outcomes.

We used a subsample of women who had completed their second year in college (n = 78) to examine the long-term effects of peer mentoring 1 y after the mentors were gone. The results are promising but preliminary because of the smaller sample size. Results showed that women who had male mentors in year 1 showed a consistent decline in belonging through the end of year 2 (B = −0.03, SE = 0.01, P = 0.007), whereas those with female mentors showed stable belonging through the end of year 2 even after their mentors had graduated (B = 0.02, SE = 0.01, P = 0.209). Students without mentors exhibited a nonsignificant decline in belonging through the end of year 2 (B = −0.01, SE = 0.01, P = 0.184). Upon comparing conditions, the belonging trajectory for women with male mentors was significantly more negative than that for female mentors (B = −0.03, SE = 0.02, P = 0.024), with the no-mentor condition falling between, not differing from either (Fig. S2A).

Fig. S2.

One year later: Effects of mentor condition on women’s (A) belonging in engineering, (B) perceived threat, and (C) intentions to pursue advanced engineering degrees. The y-axis values are difference scores from time 1 before mentor assignment. Deviations from zero show a relative increase or decrease from time 1. Statistical analyses were conducted on actual responses, not difference scores.

For feelings of threat, women with male mentors in year 1 displayed a sharp increase in threat through the end of year 2 (B = 0.06, SE = 0.01, P < 0.001) as did those with female mentors (B = 0.02, SE = 0.01, P = 0.016) and no mentors (B = 0.03, SE = 0.01, P < 0.001). A comparison of trends across conditions showed that participants with female mentors displayed significantly less increase in threat than others with male mentors (B = −0.03, SE = 0.01, P = 0.038); the control condition fell between, not significantly different from either (Fig. S2B). This pattern was only true for threat; relative threat vs. challenge through the end of year 2 did not differ by mentor condition. See SI Results.

Declining interest in advanced engineering degrees persisted through the end of year 2 for women who had male mentors or no mentors in year 1 (B = −0.08, SE = 0.02, P < 0.001; and B = −0.05, SE = 0.02, P = 0.006), whereas those with female mentors maintained stable interest in advanced engineering degrees through the end of year 2 (B = −0.02, SE = 0.02, P = 0.248) (Fig. S2C). Comparing trajectories across condition, participants without mentors showed significantly greater decline in their advanced degree intentions compared with those with female mentors (B = −0.07, SE = 0.03, P = 0.011); participants with male mentors fell in between, not significantly different from either. In sum, the results of this 1-y follow-up are consistent with data from year 1 when mentoring was active, suggesting that the benefits of having a female mentor persisted through 2 y, extending after mentors were gone, and were evident across multiple outcomes. With that said, we recommend interpreting these results with some caution because this follow-up subsample is smaller than the original sample.

Male Peer Mentors Provide Limited Benefits.

Although the effects of male mentors sometimes mimicked those of female mentors, women’s outcomes in the male-mentor condition tended to be weaker and no different from the control condition, with one exception. Women with male mentors showed stable engineering grade point averages (GPAs) across 2 y (B = −0.0004, SE = 0.007, P = 0.952), whereas women with female mentors and no mentors showed typical GPA declines as coursework became more advanced (B = −0.014, SE = 0.006, P = 0.038, and B = −0.02, SE = 0.006, P = 0.003, respectively) (39) (Fig. S3). A comparison of the GPA trajectory across 2 y showed a significant difference between male-mentor vs. control conditions (B = 0.02, SE = 0.009, P = 0.043), but no difference between male- vs. female-mentor conditions (B = 0.01, SE = 0.009, P = 0.158), suggesting that male mentors did not protect grades any more than female mentors.

Fig. S3.

Effect of mentor condition on engineering grade point average (GPA) at the end of year 2.

Several findings suggest that the stable GPA advantage for women with male mentors is not a good predictor of women’s retention and career aspirations in engineering. Rather, subjective feelings of belonging and self-efficacy in engineering are strongly implicated in retention and persistence in engineering (40). First, year 1 GPA was not significantly associated with women’s retention in engineering majors (Wald χ² = 0.37, P = 0.542), whereas social belonging and self-efficacy at the end of year 1 were both significantly associated with retention in engineering in year 1 (Wald χ² = 4.65, P = 0.031, and Wald χ² = 16.35, P < 0.001 respectively). Second, recall that engineering retention for women with male mentors was significantly lower (82%) than for those with female mentors (100%) and no different from controls (89%). Third, GPA for students with male mentors did not correlate with their feelings of belonging in engineering (r = −0.04, P = 0.82), thoughts of switching majors (r = −0.03, P = 0.87), interest in pursuing engineering careers (r = −0.11, P = 0.57), or advanced degrees (r = −0.06, P = 0.76) (Table S4). Fourth, although second-year GPA was significantly associated with engineering retention aggregated across all conditions (Wald χ² = 7.21, P = 0.007), by this time women with male mentors [mean (M) = 3.14, SE = 0.12] and female mentors (M = 3.08, SE = 0.12) had similar GPAs [t₍₆₃₎ = 0.38, P = 0.708]. In sum, the stable GPA advantage for women with male mentors does not translate to better retention and career aspirations for women in engineering. See SI Results.

Table S4.

Correlations between women’s engineering GPA at the end of year 2 with their everyday experiences and future intentions related to engineering

Correlations with engineering GPA at the end of year 2	End of year 1 (n = 101)			End of year 2 (n = 72)			All conditions
	Control/no mentor	Male mentor	Female mentor	Control/no mentor	Male mentor	Female mentor	End of year 1	End of year 2
Belonging	0.170	−0.044	0.086	0.324	0.046	0.257	0.091	0.229†
Engineering self-efficacy	0.150	0.237	0.263	0.334†	0.254	0.307	0.216*	0.316**
Threat/challenge ratio	−0.219	−0.356†	−0.214	−0.323	0.030	−0.369	−0.262**	−0.279*
Thoughts of switching majors	−0.271	−0.032	−0.269	−0.379†	−0.165	−0.129	−0.227*	−0.214†
Intentions to pursue engineering careers	0.021	−0.107	0.239	0.301	−0.033	−0.089	0.088	0.097
Intentions to pursue advanced engineering degrees	0.361*	−0.059	0.102	0.335	−0.009	−0.015	0.183†	0.138

Note: ^†P < 0.10, *P < 0.05, and **P < 0.01.

SI Results

Quantity and Quality of Mentor–Mentee Meetings from Mentors’ Diary Entries.

As shown in Tables S1 and S2, the median number of face-to-face meetings per mentor–mentee pair was 4.0, with a range of 0–7. Four mentor–mentee pairs had no face-to-face meetings (two male mentors and two female mentors), but they did have electronic contact with mentees via email, texting, etc. We included these pairs as mentored participants even though they did not have the ideal degree of contact with mentors so as to preserve the initial random assignment and provide a rigorous test of our hypotheses.

The content and activities comprising mentor–mentee meetings were more career-focused than personal or social; this was true for both female and male mentor pairs. The frequency of electronic contact outside of in-person meetings was relatively low for both female and male mentor pairs, with the caveat that mentors’ diary entries did not always contain sufficient detail to code the frequency of electronic contact. Coders were also instructed to discount electronic contact that was primarily for the purpose of scheduling meetings. All told, female and male mentors did not differ in the frequency of contact or quality of contact with their mentees (Tables S1 and S2), indicating that the nature of what the mentors do with their mentees is unlikely to explain differences in student outcomes by mentor gender.

Evaluations of Mentor Relationship.

Mentees also evaluated their relationship with their mentors. Six of the nine items assessing mentees’ subjective evaluations of their relationship with their mentors—in terms of support received, mentor availability, personal connection and chemistry with mentors, and admiration—showed that women felt equally favorable toward their female and male mentors (Table S1). Only participants’ ratings of similarity with mentors, identification with mentors, and relationship closeness showed gender differences favoring female mentors. Specifically, women felt marginally closer to their mentor (greater self–other overlap) when the mentor was female (M = 4.90, SE = 0.23) than when the mentor was male [M = 4.27, SE = 0.23; t₍₉₉₎ = 1.97, P = 0.051]. They also felt marginally more similar to female mentors (M = 4.83, SE = 0.22) than to male mentors [M = 4.24, SE = 0.26; t₍₉₉₎ = 1.72, P = 0.09], and identified marginally more with female mentors (M = 5.10, SE = 0.21) than male mentors [M = 4.51, SE = 0.27, t₍₉₉₎ = 1.72, P = 0.09]. Taken together, these findings suggest that shared identity mattered to women students, making them feel closer to female mentors. However, at the same time, women students felt equally positive about their relationships with male and female mentors and met with them just as frequently regardless of mentor gender.

Effect of Mentor Condition on Psychological Experiences in Engineering Contexts.

We used multilevel modeling [HLM software (57)] to test whether mentor condition would impact female students’ appraisals of their experiences in engineering, as well as their performance, retention, and future career goals in engineering, over the course of the academic year. We can summarize this mixed model using the following scalar expression:

$$Y_{ti}={\mathrm{\beta }}_{00}+{\mathrm{\beta }}_{01}\text{FemaleMentor}+{\mathrm{\beta }}_{02}\text{MaleMentor}+{\mathrm{\beta }}_{10}\text{Time}+{\mathrm{\beta }}_{11}\text{FemaleMentor}\ast \text{Time}+{\mathrm{\beta }}_{12}\text{MaleMentor}\ast \text{Time}+r_{0i}+r_{1\mathrm{i}}\text{Time}+e_{ti}.$$

In the above model, Y_ti represents each dependent variable, β₀₀ is the intercept for the dependent variable when the value of both mentor conditions and time is zero, β₀₁ is the difference in intercept between participants with female mentors vs. those in the control group (the reference group), and β₀₂ is the difference between participants with male mentors vs. those in the control group. To assess change over time, we centered time at time 3, or the end of the academic year. β₁₀ represents the change over time (slope) of the dependent variable for one unit of time (i.e., the difference between time 2 and time 3), β₁₁ represents the effect of having a female mentor (vs. no mentor) on the change in the dependent variable over time (the effect of having a female mentor on the slope of the line), β₁₂ represents the effect of having a male mentor (vs. no mentor) on change in the dependent variable over time (the effect of having a male mentor on the slope of the line), r_0i is the error term for the intercept, whereas r_1i is the error term for the slope, and e_ti is the within-subjects error term. Although women in the no-mentor (control) group were the natural statistical reference group, we analyzed the data for every dependent variable using each group as the reference group to test the relative effects of each mentor condition. We report the slopes for each group when they are the reference group in the main text. (For slopes relative to the female-mentor condition as the reference group, see Tables S5 and S6.)

Table S5.

Summary of mixed models analyzed using multilevel modeling showing effects of mentor condition on all dependent variables in year 1 (reference group = female mentor)

Predictor	Belonging	Threat/challenge ratio	Engineering self-efficacy	Thoughts about switching majors	Intentions to pursue advanced engineering degrees	Intentions to pursue engineering careers
β00 Reference intercept	5.22 (0.19)***	0.91 (0.08)***	5.01 (0.17)***	3.00 (0.26)***	4.96 (0.24)***	5.94 (0.24)***
β01 Male mentor	−0.26 (0.27)	0.00 (0.11)	−0.12 (0.24)	0.14 (0.37)	−0.17 (0.34)	−0.38 (0.34)
β02 No mentor/control	−0.22 (0.28)	0.24 (0.11)*	−0.44 (0.25)^	0.73 (0.37)^	−0.13 (0.35)	−0.24 (0.35)
β10 Reference slope	0.12 (0.17)	0.06 (0.08)	0.02 (0.16)	0.19 (0.26)	−0.05 (0.23)	−0.19 (0.25)
β11 Male mentor	−0.54 (0.24)*	0.10 (0.12)	−0.32 (0.23)†	0.08 (0.37)	−0.64 (0.33)^	−0.57 (0.36)†
β12 No mentor/control	−0.58 (0.25)*	0.25 (0.12)*	−0.66 (0.23)**	0.74 (0.38)^	1.00 (0.34)**	−0.44 (0.36)

Note: ^†P < 0.20, ^P < 0.10, *P < 0.05, **P < 0.01, and ***P < 0.001. SEs are in parentheses. β₁₀ is the intercept for the female mentor group at time 3. β₀₁ and β₀₂ represent the relative effects of having a male mentor and no mentor, respectively. The absolute intercepts for each group can be derived by computing the difference from the female mentor group (β₀₀). Negative values for β₀₁ or β₀₂ indicate that women in the male-mentor condition or no-mentor condition have a lower mean at time 3 than women with female mentors. β₁₀ represents change in the dependent variable over time for the female mentor group. β₁₁ and β₁₂ are the relative differences in change over time for women with male mentors and no mentors, respectively. The absolute change over time (slope) for each group can be derived by computing the difference from the female mentor group (β₁₀). Negative values for β₁₀, β₁₁, and β₁₂ indicate decreases in the dependent variables over time; positive values indicate increases in the dependent variables over time. Significance tests for the reference indicate whether the intercept and slope for the female mentor group (β₀₀ and β₁₀) differ from zero. Significance tests for the male- and no-mentor conditions indicate differences from the female-mentor condition.

Table S6.

Summary of mixed models analyzed using multilevel modeling showing effects of mentor condition on all dependent variables in year 2 (reference group = female mentor)

Predictor	Belonging	Threat	Intentions to pursue advanced engineering degrees
β00 Reference intercept	5.24 (0.23)***	4.86 (0.18)***	4.65 (0.33)***
β01 Male mentor	−0.5 (0.32)†	0.11 (0.26)	−0.27 (0.47)
β02 No mentor/control	−0.2 (0.32)	0.35 (0.26)†	−0.59 (0.46)
β10 Reference slope	0.005 (0.01)	0.02 (0.01)*	−0.02 (0.01)
β11 Male mentor	−0.03 (0.01)*	0.03 (0.01)*	−0.02 (0.02)
β12 No mentor/control	−0.02 (0.01)	0.007 (0.01)	−0.07 (0.02)*

Note: ^†P < 0.20, ^P < 0.10, *P < 0.05, **P < 0.01, and ***P < 0.001. SEs are in parentheses. β₀₀ is the intercept for the female-mentor group at time 4 (19 mo). β₀₁ and β₀₂ represent the relative effects of having a male mentor and no mentor, respectively. The absolute intercepts for each groups can be derived by computing the difference from the female-mentor group (β₀₀). Negative values for β₀₁ and β₀₂ indicate that women in the male-mentor condition or no-mentor condition have a lower mean at time 4 than women with female mentors; positive values indicate a higher mean at time 4. β₁₀ represents change in the dependent variable over time for women with female mentors. β₁₁ and β₁₂ are the relative differences in change over time for women with male mentors and no mentors, respectively. The absolute change over time (slope) for each group can be derived by computing the difference from the female-mentor group (β₁₀). Negative values for β₁₀, β₁₁, and β₁₂ indicate decreases in the dependent variable over time; positive values indicate increases in the dependent variable over time. Significance tests for the reference indicate whether the intercept and slope for the female mentor group (β₀₀ and β₁₀) differ from zero. Significance tests for the male- and no-mentor conditions indicate differences from the female-mentor condition.

Social belonging.

As reported in the main text and Table S5, having a female mentor (vs. male mentor or no mentor) preserved women’s belonging in engineering over the course of the academic year (Fig. 1). In other words, a comparison of slopes between conditions showed a significantly greater decline in belonging among participants with no mentors and others with male mentors compared with participants with female mentors. Although the slopes varied significantly, there was no mean difference in belonging at the end of year 1 between participants with female mentors vs. the other two conditions.

Engineering self-efficacy.

As reported in the main text and Table S5, women without mentors showed steep declines in self-efficacy across the first year, as did those with male mentors. In contrast, women with female mentors maintained positive self-efficacy that did not change. Comparing change trajectories between conditions, students with female mentors reported stable self-efficacy significantly more so than those with no mentors, with male mentors falling in-between not differing from either group (Fig. 2).

Threat and challenge.

In our data, threat and challenge were negatively correlated at each time point. The magnitude of these correlations were moderate in year 1 (time 1: r = −0.38, P < 0.001; time 2: r = −0.36, P < 0.001; time 3: r = −0.32, P < 0.001) and large in year 2 (time 4: r = −0.676, P < 0.001). In keeping with past research (46–50), we analyzed threat and challenge as a ratio of students’ self-reported threat relative to challenge, which reflects the degree to which a participant’s anxiety about engineering (perceived threat) was offset by her motivation to overcome any difficulties using her existing skills (perceived challenge). If the ratio of threat/challenge is greater than 1, it indicates that women perceived engineering to be more threatening than challenging; values less than 1 indicate that women perceived engineering to be more challenging than threatening. A ratio of 1 would indicate that women found engineering to be equally threatening as challenging. As reported in Table S5 (also see Fig. 3), in the absence of any mentors, women felt more threatened than challenged in engineering as the year progressed. Women with female mentors were the exception to this trend—their feelings of threat relative to challenge did not increase over time. Women with male mentors fell in the middle, showing a marginal increase in threat relative to challenge, a trend that was statistically equivalent to the control condition as well as the female-mentor condition. These analyses focused on threat relative to positive challenge. Neither threat nor challenge alone was impacted by mentor condition over time. Women with female mentors and male mentors were no different from women without mentors in terms of their self-reported threat over time (B = −0.08, SE = 0.276, P = 0.755; B = −0.18, SE = 0.278, P = 0.521, respectively). The same was true of challenge: women with female and male mentors reported no differences in self-reported challenge from women without mentors (B = 0.17, SE = 0.202, P = 0.409; B = 0.06, SE = 0.204, P = 0.756, respectively).

Effect of Mentor Condition on Retention and Postcollege Aspirations.

Frequency of thoughts of changing majors and objective retention in the major.

Across all conditions, women reported thinking of changing their major somewhat frequently (M = 3.02, SE = 0.18), and for women in the control group, these thoughts of leaving increased over time (B = 0.94, SE = 0.279, P < 0.001). Having any mentor was protective against thoughts about changing majors (slope for any mentor vs. none: B = −0.70, SE = 0.336, P = 0.038) (Fig. S1). However, as reported in the main text, women’s thoughts about staying in engineering majors were more likely to translate into actual retention decisions if their mentor was female, whereas thoughts about staying in engineering did not always translate into actual retention when the mentor was male (Fig. 4).

Postcollege aspirations.

As reported in the main text and Table S5, participants with female mentors (but not male mentors or no mentor) maintained stable intentions to pursue advanced degrees in engineering throughout their first year in the major (Fig. 5). There was no effect of mentor condition on changes in women’s intentions to pursue professional jobs in engineering, however (F < 1). We believe the phrasing of the item assessing future careers in engineering may have failed to capture women who intended to apply their engineering degree to related careers that do not conventionally fall within the boundary of engineering, such as construction and building sciences, biomedical applications of engineering, etc. Indeed, several participants mentioned these types of careers as their future career goal. Despite the lack of direct effects of mentor condition on engineering career aspirations, there were indirect effects of mentor condition on career aspirations through social belonging and engineering self-efficacy (see below).

Social Belonging and Self-Efficacy Mediate the Relation Between Mentor Condition and Engineering Career Aspirations.

We conducted multiple mediation analyses using multilevel structural equation modeling in MPLUS (58) to compare our three theory-based mediators (social belonging, engineering self-efficacy, and threat vs. challenge) in a single mediation model. In our mediation model, the dependent variable (Y), and all three mediators (M1, M2, and M3) occurred at level 1 (within-subjects), whereas the mentor condition predictor (X) occurred at level 2 (between-subjects); this was a multiple mediator version of a “2-1-1 Mediation Model” (58). We regressed each level 1 variable on time to create random intercepts and slopes for the dependent variable and the three mediators. Time was centered at time 1 for the multiple mediation analyses to minimize multicollinearity. We then ran mediation analyses by regressing the random slope for the dependent variable on the random slopes for each of the three potential mediators, and regressing the random slopes of the dependent variable and mediators on mentor condition. Mentor condition was dummy-coded; we adapted the procedures for mediation and testing indirect effects with a multicategorical independent variable (59) to the multilevel path model.

We examined whether mentor condition would indirectly affect changes in women’s future aspirations (engineering career aspirations, graduate school intentions, thoughts about switching majors) through changes in women’s experiences in engineering [feelings of belonging, self-efficacy, and threat/challenge; potential mediators derived from theory (6)]. As reported in the main text, there were significant indirect effects of having a female mentor vs. no mentor on women’s engineering career aspirations through increased belonging and engineering self-efficacy over time. Threat vs. challenge did not significantly mediate the relation between mentor condition and engineering career aspirations. Furthermore, having a female mentor vs. a male mentor also indirectly enhanced women’s engineering career aspirations through belonging, but not through engineering self-efficacy or threat/challenge, suggesting that changes in social belonging may be better able to explain differences in the benefits of having a female peer mentor relative to a male peer mentor or no peer mentor. These variables did not significantly mediate the relation between mentoring condition and graduate school intentions or thoughts about switching majors.

We point out three potential caveats in interpreting these findings. The first caveat is in regard to the relative magnitude of the indirect effects of belonging vs. engineering self-efficacy. Although social belonging seems to uniquely disambiguate the benefits of female vs. male mentors, the coefficients for both indirect effects for belonging are of smaller magnitude and estimated with somewhat less precision than the coefficient for the indirect effect for engineering self-efficacy (Table S3). Although this may be due to the fact that mentor condition explained the majority of the variance in the slope of belonging, the relative contributions of belonging and self-efficacy should be further examined in future research. The second caveat is due to the nature of random slopes (our indices of change over time). Because our interest was in changes in women’s experiences and aspirations rather than absolute levels, (given that women may enter engineering with a wide range of experiences, appraisals, and motivations), our mediators do not have clear temporal precedence over the outcome. Thus, although our theory-derived predictions propose that belonging and self-efficacy ought to predict engineering career aspirations and not vice versa (6), we caution against strong causal claims based on these data. Finally, and relatedly, these analyses represent an initial foray into understanding when and why ingroup peer mentors might protect women’s experiences and aspirations in engineering. We urge other researchers to seek to replicate this research and investigate additional mechanisms underlying effective mentoring for members of underrepresented groups. See Table S3 for the multiple mediation results.

One Year Later: Long-Term Effects of the Peer Mentoring Intervention on Student Outcomes.

We invited participants to complete a follow-up survey at the end of their second year of the engineering major (time 4 assessment). Of the original sample of 150 participants, 135 women still remained in engineering majors at the end of the first year and 30 of these women had not yet advanced to their second year at the time of data analysis, leaving a subsample of 105 women eligible for the end-of-second-year assessment (time 4). Out of this subsample of 105 women, 78 women completed the time 4 survey (n = 27, 23, and 28 in the female-, male-, and no-mentor conditions, respectively; 74% of the eligible sample). Multilevel model analyses were conducted again with this subsample, with the Time variable updated to include time 4 as a fourth point of assessment. To account for variability in the dates on which the follow-up survey was completed relative to the first baseline assessment, Time was computed as total months from each participant’s baseline (time 1) assessment. See Table S6.

Social belonging.

As reported in the main text and in Table S6, women who had male mentors in year 1 showed a consistent decline in belonging through the end of year 2, whereas those with female mentors showed stable belonging through year 2 end even after their mentors had graduated. Students without mentors exhibited a nonsignificant decline in belonging through year 2 end. A comparison of conditions showed that the belonging trajectory for women with male mentors was significantly more negative than that for others with female mentors, with the no-mentor condition falling between, not differing from either (Fig. S2A).

Threat.

Although all women, on average, showed an increase in threat over 2 y as their coursework became more difficult (B = 0.04, SE = 0.006, P < 0.001), as reported in the main text and shown in Table S6, women who originally had male mentors displayed the steepest rise in threat by the end of year 2, followed by those without mentors, and those with female mentors. A comparison of change trajectories across conditions showed that the rise in threat was significantly sharper for participants with male mentors compared with female mentors, with the no-mentor condition falling in between, not significantly different from the other two conditions (Fig. S2B). Furthermore, although the rise in threat was significantly sharper for women with male mentors compared with female mentors, these condition differences were specific to threat alone; relative threat/challenge did not vary significantly across mentor conditions at the end of year 2. Similarly, the trajectory of challenge did not vary significantly across mentor conditions at the end of year 2.

Intentions to pursue advanced degrees in engineering.

As reported in the main text and Table S6, declining interest in advanced engineering degrees persisted through the end of year 2 for women who had male mentors or no mentors in year 1, whereas those with female mentors maintained stable interest in advanced engineering degrees through the end of year 2. A comparison of trajectories across conditions showed that participants without mentors showed significantly greater decline in their advanced degree intentions compared with those with female mentors; participants with male mentors fell in between, not significantly different from either (Fig. S2C).

Correlations between GPA and psychological experiences in engineering.

We explored whether women’s performance in engineering (GPA in year 2) was associated with their psychological experiences in the major at the end of year 1 and year 2 and retention in engineering majors at the end of each year. Although higher grades were significantly correlated with more self-efficacy and less threat relative to challenge (aggregated across all conditions), there were no consistent correlations between women’s year 2 GPA in engineering with their feelings of belonging in engineering, future intentions to pursue advanced degrees in engineering, or career aspirations in engineering (Table S4). This held true upon examining correlations for each mentor condition separately.

In terms of GPA and retention, women’s GPA was not significantly associated with retention in engineering majors at the end of the first year (Wald χ² = 0.37, P = 0.542) but was significantly related at the end of the second year (Wald χ² = 7.21, P = 0.007). The consistent absence of correlations between GPA with social belonging and future intentions, and the changing relation between GPA and retention, suggest that, for women in engineering, GPA may not be the best predictor of retention or career aspirations in engineering especially in the first year of college, which is consistent with past research (40). Objectively strong performance does not ensure that women will stay in engineering. Notably, our data suggest that subjective experiences (social belonging and self-efficacy) play a big role in women’s retention in engineering and aspirations to pursue engineering careers in the future. These findings caution against overemphasizing test performance and grades, which traditionally have been the focus of stereotype threat research (60, 61), and underscores the importance of looking at other psychological predictors of retention and attrition that may be more powerful than tests and grades.

Correlations of mentor closeness and similarity with psychological experiences in engineering.

Greater closeness and perceived similarity with female mentors positively correlated with stronger self-efficacy in engineering, intentions to pursue advanced degrees and careers in engineering, and feeling positively challenged across the first year. Parallel correlations for male mentors were mixed: whereas closeness and similarity with male mentors correlated significantly with belonging in engineering, they did not correlate with self-efficacy, challenge, intentions to pursue advanced degrees in engineering or careers in engineering (Table S7).

Table S7.

Correlations between mentor closeness and similarity with changes in women's experiences and future intentions in engineering in year 1

Changes in women’s experiences in engineering (Δ)	Closeness		Similarity
	Male mentor	Female mentor	Male mentor	Female mentor
Δ Engineering self-efficacy	0.110	0.325*	0.037	0.274*
Δ Challenge	0.164	0.215	0.093	0.264†
Δ Belonging	0.266†	0.095	0.350*	0.198
Δ Intentions to pursue advanced engineering degrees	0.159	0.302*	−0.051	0.408***
Δ Intentions to pursue engineering careers	0.201	0.294*	0.174	0.495***

Note: ^†P < 0.10, *P < 0.05, and **P < 0.01.

Our finding that same-gender peer mentors are important for women in engineering, and that felt similarity with female (more so than male) mentors is significantly associated with positive outcomes for women seems at odds with Cheryan’s research on women in computer science, a field in which women are similarly scarce as engineering. Cheryan and colleagues found that gender of peer role models had no effect on women’s interest and anticipated success in computer science, but role models’ stereotypicality mattered a great deal because it affected how similar women felt to those role models (62). In contrast, we found that the gender of peer mentors mattered significantly to women in engineering, and similarity with female (more than male) peer mentors was an added benefit. Two reasons are likely responsible for these differing findings. First, although our participants were women who were already in engineering majors, Cheryan and colleagues’ participants were women who were not enrolled in computer science majors nor had they expressed any interest in computing before the experiment. It is plausible that interacting with same-gender peer role models matters more once women are already in STEM majors like engineering and computer science and know first-hand the experience of being a tiny minority. In contrast, interacting with counterstereotypic role models of any gender may matter more for the recruitment of women who are not yet in the major because such role models may help to dispel stereotypic preconceptions about the field for people with little direct experience (63). A second reason for these different findings may be the duration of the relationship between women participants and their peer mentors. Our longitudinal study allowed mentor–mentee relationship to evolve over a year; we tested whether development of a bond with female vs. male mentors over time would differentially influence women’s experiences in engineering. This is different from the experience of having a brief conversation with a peer in a laboratory setting about a predetermined topic and not seeing that person again, which was the experimental protocol used by Cheryan and colleagues. The short interaction and constrained conversation topic may have drawn participants’ attention to the stereotypical dimension (as was intended by their manipulation) more than other commonalities they might have shared. These different findings raise an interesting point that brief cross-sectional vs. longitudinal research designs focusing on the same topic may sometimes show different results because some phenomena (e.g., the formation of peer relationship bonds) take time to unfold.

Global appraisals of engineering.

As reported in the main manuscript, mentor condition had significant effects on women’s dynamic experiences and appraisals in engineering in terms of social belonging, threat relative to challenge, self-efficacy, motivation to pursue advanced degrees in engineering, and ultimately also affected student retention in the major. However, the mentoring intervention did not significantly affect global stereotypes of, attitudes toward, and identification with engineering at an implicit or explicit level.

Implicit and explicit stereotypes about engineering.

On average, women reported that they explicitly viewed engineering as more masculine than feminine (M = 2.92, SE = 0.076; on a 7-point scale where 1 was masculine and 7 was feminine), but these explicit stereotypes were not impacted by mentor condition [F_(2,149) = 1.29, P = 0.278]. Similarly, on the IAT, women implicitly stereotyped engineering as more masculine than feminine [M = 0.16, SE = 0.022, t₍₁₄₉₎ = 7.02, P < 0.001]. However, this also was not significantly affected by mentor condition, although it trended in the expected direction [F_(2,149) = 2.30, P = 0.104].

Implicit and explicit attitudes toward engineering.

On average, women reported positive explicit attitudes toward engineering (M = 5.56, SE = 0.108; on a 7-point scale), but these attitudes were not affected by mentor condition (F < 1). In contrast, on the IAT, women displayed less positive implicit attitudes toward engineering and relative preference for languages [M = −0.15, SE = 0.037, t₍₁₄₉₎ = 3.54, P = 0.001]. These implicit attitudes did not vary significantly by mentor condition (female mentor vs. control: P = 0.997; female vs. male mentor: P = 0.274), but they did change over time. Specifically, when examining the change over time within condition, women with no mentors and those with male mentors exhibited less positive attitudes toward engineering over time (B = −0.45, SE = 0.152, P = 0.003; and B = −0.42, SE = 0.146, P = 0.004, respectively), whereas women with female mentors showed stable implicit attitudes toward engineering that did not change over time (B = −0.115, SE = 0.145, P = 0.426). These change trajectories (slopes) were somewhat different between the female-mentor condition compared with the control and male-mentor conditions, but the differences did not reach conventional levels of significance (B = −0.34, SE = 0.210, P = 0.111; and B = −0.31, SE = 0.206, P = 0.139, respectively).

Implicit and explicit identification with engineering.

On average, women reported that engineering was important to them and that they identified with the field (M = 6.11, SE = 0.067; 7-point scale), but explicit identification with engineering did not vary as a function of mentor condition (F < 1). Similarly, on the IAT, women implicitly identified with engineering more so than with languages [M = 0.18, SE = 0.022, t₍₁₄₉₎ = 8.23, P < 0.001]. Implicit identification with engineering did not vary significantly by mentor condition (female mentor vs. control: P = 0.52; female vs. male mentor: P = 0.53), but they did change over time in a manner consistent with the attitude IAT above. Specifically, women with no mentors and those with male mentors showed declining implicit identification with engineering over time (B = −0.35, SE = 0.187, P = 0.06; and B = −0.45, SE = 0.179, P = 0.013, respectively), whereas women with female mentors showed stable implicit identification that did not change over time (B = 0.006, SE = 0.178, P = 0.971). The slope difference between the female vs. male mentor condition was marginally significant (B = −0.46, SE = 0.252, P = 0.072), and the slope difference between the female mentor vs. control condition trended in the same direction, but did not reach significance (B = −0.36, SE = 0.258, P = 0.164).

The fact that same-gender peer mentors inoculated women’s experiences in engineering by preserving feelings of belonging despite negative stereotypes, but had weak or negligible effects on their global attitudes and beliefs about engineering, is consistent with our prior findings on peer influences as “social vaccines” for women in STEM (10). In both the prior study and the present one, the presence of female peers had a profound impact on women’s everyday experiences in their local academic environments but did not necessarily generalize to their global perceptions of the field. This may have happened for two reasons. First, changing global perceptions of stereotypically masculine fields may require something different—exposure to highly successful and advanced female experts who are leaders in that field and who function as aspirational role models. To that point, our past research shows that exposure to such experts does indeed positively influence women’s global attitudes toward, and identification with, STEM and other traditionally masculine professions (35, 40, 49). This would suggest that same-gender experts, but not peers, may have more impact in changing women’s global mental representations regarding who is a prototypical member of a field. In comparison, same-gender peers may be more successful at shaping women’s everyday experiences in STEM as suggested by our present findings. A second explanation for the implicit findings is different: given that our data show predictable trends in implicit attitudes and identification with engineering in ways that fit with other outcomes measuring women’s everyday experiences, and given that these trends are marginal or close, it is possible that a larger sample size might have been needed to capture changes in implicit reactions in response to a peer intervention.

Conclusion.

In closing, we provide evidence from a multiyear field experiment demonstrating that women in engineering who were given a female (relative to male) peer mentor experienced more belonging in engineering, motivation, confidence, retention in engineering majors, and postcollege engineering aspirations. For those with female peer mentors, increased belonging and confidence across the first year of college were significantly associated with more aspirations to pursue engineering careers after graduation. Better engineering grades were not associated with women’s retention or engineering career aspirations in the first year of college. The benefits of mentoring endured well after the intervention had ended, for 2 y of college, the time of greatest attrition from STEM majors. See Table S8 for a visual summary of the results and Table S9 for means of the primary dependent variables by mentor condition.

Table S8.

Summary of results by mentor condition for years 1 and 2

Dependent variables	Significant change over time within conditions (slopes)			Comparing change over time (slopes) between conditions
	Control/no mentor	Male mentor	Female mentor	Female vs. control	Male vs. control	Female vs. male
Year 1
Belonging	Decrease	Decrease	No change	Significant	Nonsignificant	Significant
Threat/challenge ratio	Increase	Increase	No change	Significant	Nonsignificant	Nonsignificant
Engineering self-efficacy	Decrease	Decrease	No change	Significant	Nonsignificant	Nonsignificant
Thoughts of switching majors	Increase	No change	No change	Significant	Significant	Nonsignificant
Intentions to pursue advanced engineering degrees	Decrease	Decrease	No change	Significant	Significant	Nonsignificant
Year 2
Belonging	No change	Decrease	No change	Nonsignificant	Nonsignificant	Significant
Threat	Increase	Increase	Increase	Nonsignificant	Nonsignificant	Significant
Intentions to pursue advanced engineering degrees	Decrease	Decrease	No change	Significant	Nonsignificant	Nonsignificant
Retention in engineering majors, % retained	89%	82%	100%

Table S9.

Means for primary dependent variables by mentor condition

	No mentor			Male mentor			Female mentor
Dependent variables	Time 2	Time 3	Time 4	Time 2	Time 3	Time 4	Time 2	Time 3	Time 4
Belonging	5.359 (0.249)	5.107 (0.240)	5.224 (0.246)	5.447 (0.273)	4.959 (0.264)	4.914 (0.270)	5.094 (0.257)	5.449 (0.248)	5.382 (0.254)
Threat	4.704 (0.199)	4.914 (0.195)	5.216 (0.201)	4.054 (0.223)	4.294 (0.218)	5.421 (0.225)	4.610 (0.209)	4.671 (0.204)	4.980 (0.210)
Threat/challenge ratio	0.893 (0.054)	1.190 (0.137)	1.138 (0.103)	0.785 (0.061)	0.953 (0.153)	1.266 (0.115)	0.852 (0.057)	0.906 (0.143)	1.131 (0.108)
Engineering self-efficacy	4.859 (0.173)	4.742 (0.202)	4.514 (0.239)	5.313 (0.191)	5.151 (0.223)	4.914 (0.264)	5.106 (0.182)	4.798 (0.213)	4.60 (0.252)
Thoughts of switching majors	3.008 (0.295)	3.031 (0.308)	2.691 (0.368)	2.521 (0.329)	2.710 (0.344)	2.924 (0.411)	2.605 (0.300)	2.599 (0.313)	2.820 (0.374)
Intentions to pursue advanced engineering degrees	5.714 (0.258)	5.279 (0.227)	4.557 (0.342)	5.451 (0.280)	5.343 (0.246)	5.017 (0.370)	5.297 (0.268)	5.410 (0.235)	4.753 (0.354)
Intentions to pursue engineering careers	6.547 (0.170)	6.381 (0.203)	5.822 (0.281)	6.260 (0.183)	6.160 (0.219)	6.338 (0.304)	6.246 (0.172)	6.202 (0.206)	6.023 (0.286)

Note: SEs are in parentheses. These means are based on analysis of covariance (ANCOVA) in which time 1 means for each outcome variable were used as the covariate in separate ANCOVAs. Covariates appearing in these ANCOVAs are evaluated at the following values: time 1 belonging = 5.406, time 1 threat = 4.247, time 1 threat/challenge ratio = 0.819, time 1 self-efficacy = 5.123, time 1 switching majors = 2.621, time 1 engineering advanced degree intentions = 5.552, and time 1 engineering career intentions = 6.284.

Discussion

In conclusion, same-gender peer mentoring during the transition to college appears to be an effective intervention to increase belonging, confidence, motivation, and ultimately retention of women in engineering. Our findings make four contributions that advance knowledge about how best to increase and sustain gender diversity in STEM. First, our data show not all peer mentors are equally effective even though the objective content and frequency of mentor–mentee interactions may be similar. Shared identity matters for retention and other engineering outcomes. Second, female mentors protect women’s feelings of belonging and connection to other peers in engineering during their first year in college, when they are most vulnerable to self-doubt; greater belonging in turn protects women’s aspirations to pursue careers in engineering after college. Third, contrary to common wisdom, better performance in engineering courses (higher GPA) does not necessarily correspond to stronger feelings of belonging or more intentions to pursue engineering careers and advanced degrees. Instead, women’s subjective experiences in engineering—notably their feelings of belonging and self-efficacy—predict retention in engineering majors and engineering career intentions. Fourth, the benefits of same-gender peer mentors endured long after mentoring had ended, inoculating women for 2 y of college, the window of greatest attrition from STEM majors (41).

Although female peer mentors had significantly more desirable effects on first-year women in engineering, this does not mean male mentors are unimportant. We expect that female mentors’ support will become less critical as women move beyond the college transition, at which point male and female mentors may be equally effective (42). This speculation is consistent with the Stereotype Inoculation Model (6, 10), which identifies developmental transitions, such as the beginning of college, as times of special vulnerability to negative stereotypes. Moreover, whereas our intervention focused on peer mentors, male faculty who are scientists and engineers likely play important roles as advisors and career sponsors.

Findings from this randomized longitudinal field experiment open the door to future tests of the generalizability and boundary conditions of mentoring interventions. First, future research should test whether our findings generalize to disciplines other than engineering where women are also negatively stereotyped and exist in tiny numbers well below critical mass (43). Second, future research should also explore whether female peer mentors would be similarly beneficial to women experiencing other transitions in academic or professional life (e.g., transition to graduate school). Third, our theory (6) affords the prediction that ingroup peer mentors may be beneficial for members of other groups that are underrepresented and negatively stereotyped in high achievement environments (e.g., African American and Latino students); the question of whether our findings would generalize to such other groups is an important one to pursue. Finally, we propose that the importance of ingroup peer mentors is likely to diminish as individuals become more advanced, as their numbers in a field increase, and as negative stereotypes about their ability fade; these are important boundary conditions to investigate. Our findings, and the future research avenues identified above, have important implications for scientists and engineers in higher education, university leaders, and policy makers searching for evidence-based interventions to increase equality in higher education and to diversify the American workforce.

Materials and Methods

Participants.

Female students majoring in engineering (n = 158) at the University of Massachusetts Amherst participated in our experiment after we obtained approval from the institutional review board. Women account for a tiny minority (16%) of all engineering students at this university, similar to the national average (44). They were paid $20 to $35 for each survey. Of the original recruits, eight (5.1%) dropped out of the experiment soon after the baseline assessment and before mentor contact. The final sample included 150 women. The mean age was 18.34 (SD = 1.34). The majority were White (67.3%); others were Asian (17.3%), multiracial (5.3%), Black (2.7%), Hispanic (2.7%), or indicated another ethnicity (2%). Most were American citizens (91.3%); 16% were first-generation college students and 28% had parents in engineering-related occupations. Mentors were mostly seniors and some juniors who were student leaders of engineering organizations and high performers in engineering. They received an honorarium of $100 for each mentee. See SI Materials and Methods.

Procedure.

Female recruits were unaware that the experiment had anything to do with mentoring. All provided written consent approved by the university institutional review board and completed a baseline assessment in early fall of their starting academic year (time 1), a midyear assessment in winter (time 2), and a year-end assessment in late spring (time 3). A time 4 survey was administered 1 y later in the spring. Grades were obtained from the university registrar each year. After time 1, participants were randomly assigned—within their engineering major—to a female mentor (n = 52), male mentor (n = 51), or not given a mentor (control group; n = 47) (for details, see SI Materials and Methods and SI Results).

Datasets.

All of the data reported in this manuscript are available in Dataset S1, which is accompanied by two coding guides explaining all variables (Datasets S2 and S3).