Effects of mentor–mentee discordance on Latinx undergraduates’ intent to pursue graduate school and research productivity

Abstract

Owing to an undergraduate demographic transition in STEM, Latinx students are likely to work with faculty from different backgrounds when participating in undergraduate research experiences (UREs). However, the effects of mentor–mentee demographic discordance on student STEM development are unclear. This paper examines how mentoring discordance affects Latinx students’ intentions to pursue graduate school and research productivity. We collected data from participants in a multiyear, multi-institutional URE program (n = 171 dyads), which we analyzed using multivariable generalized estimating equations. Results indicate that compared with gender concordance, gender discordance was associated with a 17% increase in intent to pursue graduate school for Latina students. Compared with racial/ethnic concordance, racial/ethnic discordance was associated with a 38% increase in intent to pursue graduate school for Latino students. When paired with gender discordant mentors, Latina students were 70% less likely to present their URE projects at a professional conference. When faculty mentors were culturally competent and interacted with mentees frequently, Latinx students became more interested in pursuing graduate school. Because mentors from different demographic backgrounds contribute to the development of Latinx STEM students in varied ways, URE program directors should integrate opportunities for multiple mentorships.

Introduction

As is the case for other underrepresented minority groups, there are disproportionately few Latinx professionals in the science, technology, engineering, and math (STEM) workforce in the United States.¹ Many educational interventions at different levels have been instituted to address this diversity crisis in STEM.^2–4 Among them, faculty-mentored undergraduate research experiences (UREs) have emerged as a particularly impactful practice in higher education.^5,6 Myriad benefits of UREs have been documented, including improved critical thinking, increased academic achievement and retention, persistence to STEM degree completion, clarification of career plans, and improved preparedness or desire for graduate study.^5,7–10 Particularly for Latinx undergraduate students, UREs are effective in arousing and sustaining their STEM career interests.¹¹

Faculty mentorship is a central element of UREs, and demographically discordant mentoring relationships—wherein students and mentors do not share the same gender or race/ethnicity—are an experiential reality for many undergraduate students pursuing STEM research training. Over the past 20 years, a growing number of programs have received millions of dollars of support from the National Science Foundation (NSF), the National Institutes of Health (NIH), other government agencies, and private foundations to provide faculty-mentored UREs to college students. Many programs are designed to support students from traditionally underrepresented backgrounds. Thanks in part to these programs, the number of women and minority undergraduate students earning STEM degrees has increased. Among those who obtained STEM bachelor’s degrees in the United States in 2019, half were women and 23% were underrepresented racial/ethnic minorities, including 14% Latinxs.¹ However, increases in the number of women and minority faculty in STEM have not kept pace with growth at the bachelor’s level. In 2019, 36% of current STEM faculty were women, 9% were underrepresented racial/ethnic minorities, and only 5% were Latinx.¹ Thus, despite the shift toward more diverse participation of undergraduate students, the majority of faculty in STEM fields continue to be White men.

While we know that UREs are generally beneficial for Latinx students, it is unclear how mentor–mentee demographic discordance affects their specific outcomes. To date, mentor–mentee discordance has been mainly studied in terms of gender or racial/ethnic similarity between mentors and mentees separately, and findings from those studies have been mixed.^12,13 To advance knowledge, we developed a novel conceptual framework to guide an analysis of the effects of demographic discordance in URE mentoring relationships on two outcomes of Latinx students—intentions to pursue graduate school and research productivity—which are important for students’ STEM development. This study was conducted in the context of a comprehensive multi-institutional and multiyear URE program hosted by a Hispanic Serving Institution (HSI) in a U.S.–Mexico border state. HSIs in border states are important to the education of Latinx college students nationally. About 65% of all Latinx college students in the United States attend HSIs,¹⁴ and the U.S.–Mexico border states house the majority of HSIs.¹⁵

In what follows, we first ground our expectations in the relevant literature, introduce the conceptual model and methods, and describe our results. We then discuss the implications of our findings for theory building regarding mentor–mentee relationships as well as practical efforts to enhance the diversity of the STEM workforce.

Background

Latinx undergraduate students in STEM

There are multiple explanations for why Latinx groups are underrepresented in STEM. Latinx young adults are less likely to attend 4-year colleges than are Whites.^16,17 When pursuing STEM degrees, Latinx students tend to have less family support¹⁸ and are more likely to experience discrimination^19,20 than White students. Particularly, when Latinx students apply to graduate school, they face specific barriers, such as financial concerns and inadequate support systems.²¹ It can also be difficult for Latinx students to develop professional networks and find role models in STEM.¹⁸ Herrera and Hurtado²² asserted that, in order to address the underrepresentation of Latinx groups, it is essential to promote Latinx students’ interests in pursuing STEM research careers.

Research reveals that UREs can encourage Latinx STEM students to pursue graduate school. For example, Villarejo et al.¹¹ found that underrepresented minority students, including Latinx students, started thinking about graduate school for the first time after participating in UREs. Other studies show that underrepresented minority students who participate in UREs, including Latinx students, are more likely to report intentions to enter STEM-related graduate programs than underrepresented minority students who have not participated in UREs.^23–25

UREs also provide some students opportunities to engage with the STEM community through national/regional conference presentations, which are a key productivity metric at this career stage. Conference presentations are an important element of the acculturation of undergraduates into the STEM community because they provide students an opportunity to disseminate research findings and can strengthen their commitment to pursue a STEM career.²⁶ More importantly, as evidence of research productivity, presenting at conferences is often used as a selection criterion by STEM graduate programs.²⁶ UREs and faculty mentors are critical in encouraging more underrepresented minority students, including Latinx students, to present their research at professional conferences.²⁷ URE program participants have described their faculty mentors as “door openers” for offering deep knowledge of presenting research at conferences and applying to graduate school.²⁷ However, no prior research has examined the effects of demographic discordance between faculty mentors and undergraduate mentees on these two student outcomes.

Mentor–mentee demographic concordance

The role of mentor–mentee demographic concordance in mentees’ outcomes is the subject of debate in the mentoring literature. Findings have varied depending on the population examined, an outcome predicted, and an analytical approach. Some studies have found demographic concordance to be beneficial to mentees, while others have found discordance to be beneficial. Still, others have found no effect.

Demographic concordance can be beneficial for mentees because mentees feel comforted by guidance from those who have successfully confronted the challenges they themselves are facing, and it may easier to trust “one’s own” than someone who resembles “the other.”^13,28,29 An early study reported gender matching between graduate students and their mentors was associated with more publications.³⁰ Since then, the majority of research on mentor–mentee concordance has focused primarily on women and racial/ethnic minority mentees.^31–34 Lockwood³⁵ found that women students were more inspired by women role models than men role models. Women students also reported greater comfort³⁶ and more psychosocial support^37,38 when they had women versus men mentors.

In terms of racial/ethnic concordance, minority mentors may be particularly important to minority students because they represent role models and prototypes that enable students to gain academic self-efficacy.³⁹ When mentor and mentee are both racial/ethnic minorities, they tend to have better communication and develop closer and more effective relationships than mentor–mentees in racial/ethnic discordant dyads.³¹ Racial/ethnic minority students also tend to receive more psychosocial and instrumental (e.g., research and career) support from minority mentors than White mentors.^40–43 A study focused on Latinx undergraduates found that those with Latinx mentors (versus non-Latinx mentors) perceived their mentors to be significantly more supportive in furthering their personal and career development.⁴⁴ To summarize, the majority of studies on mentor–mentee concordance assert that mentees benefit from demographically concordant mentoring relationships because mentors with similar backgrounds serve as positive role models. We refer to this as the supportive role model effect of mentor–mentee concordance.

Some scholars engaged in this debate found that women were more satisfied with mixed-gender mentoring and racial/ethnic minority mentees were more satisfied with mixed-race mentoring relationships than same-gender/race mentoring relationships.^{12,13,45–47} Scholars thus argue that this satisfaction might be attributable to men or White mentors having more resources and/or more developed professional networks than women or minority mentors. As a result, they may be better positioned to provide career development support, such as buffering mentees from adverse experiences and opening avenues for mentees’ advancement.^29,48–51 For example, a study focusing on sociology graduate students found that for minority students, having a White man mentor in graduate school was positively related to employment at Research I universities.⁵² We refer to this as the privilege crossover effect of mentor–mentee discordance.

A third group engaged in this debate has found demographic concordance to be unimportant. A number of studies suggest that gender concordance in a mentoring relationship is not influential on mentees’ performance or outcomes.^53–55 Others have found no differences in student outcomes for same- versus cross-race mentoring relationships.^56–60 In an examination of a national sample of Black undergraduates majoring in STEM disciplines, Hernandez et al.⁵⁸ found that demographic similarity between mentors and mentees was not associated with mentees’ greater commitment to STEM careers. We refer to this as the null effect of demographic concordance between mentors and mentees.

In sum, prior studies support three competing hypotheses: among Latinx students, H^a demographic concordance with faculty mentors will positively affect two student outcomes (intention to pursue graduate school and research productivity); H^b demographic discordance with faculty mentors will positively affect the two student outcomes; and for H^c the two student outcomes will not be affected by demographic concordence/discordance between faculty mentors and undergraduate mentees.

Conceptual model

To test the three hypotheses, we developed a novel conceptual model (see Fig. 1). When characterizing a mentoring dyad, previous research has focused on the quantity and quality of mentor–mentee interactions.^61–63 In our model, we add demographic discordance as the third crucial characteristic of a mentoring dyad, which includes two dimensions (gender and race/ethnicity). The conceptual model conceives of dyads in context, which, in this case, includes features of the intervention and the URE host institution. This is because social cognitive career theory posits that contextual factors can affect student career interests.^64–66 Students’ URE outcomes are also likely affected by the research environments at the institutions where they conduct research (e.g., research- versus teaching-intensive institutions) and certain characteristics of the intervention (e.g., financial support). As such, we include institutional and intervention-specific contextual factors in our conceptual model.

Figure 1.

The conceptual model.

To summarize, we conceptualize Latinx students’ URE outcomes as being influenced by the three elements of their mentoring relationships and by contextual factors shaping their UREs. Since we focus on disentangling the effects of mentor–mentee discordance on student outcomes, we include additional measures as control variables in our conceptual (and statistical) model.

Methods

Participants

Participants were trainees in a comprehensive URE program that ran during four consecutive summers. To protect confidentiality, we henceforth refer to this program as “THRIVE.” THRIVE’s goal is to encourage more undergraduate students, especially those from underrepresented backgrounds, to pursue STEM research career paths. THRIVE is housed at a public HSI (which we call the primary institution), but the program is open to students from other regional 2- or 4-year partner institutions, most of which are also HSIs. Since 2016, the THRIVE program has sponsored undergraduate students to conduct full-time 10-week summer research at research-intensive partner universities and the primary institutions.

The structure of the THRIVE program as relevant to this analysis is depicted in Figure 2. Two different types of interventions (long- versus short-term) were designed and implemented by THRIVE. The long-term intervention involves a 12-month scholarship for at least 2 years, including at least one summer URE, and an intensive training plan that combines summer and academic-year research opportunities, course-based UREs, professional development training, and funding support for research supplies and attending professional conferences. The majority of trainees from the primary institution receive a long-term intervention. We refer to them as academic-year scholars. Academic-year scholars worked with faculty mentors at the primary institution during academic years, and most conducted summer research with different mentors at research partner institutions. A few stayed at the primary institution. The application and selection process of THRIVE is similar to other URE programs. Program directors review student applications, conduct interviews, and select THRIVE academic-year scholars based on their academic performance and interests in pursuing STEM research careers.

Figure 2.

The THRIVE program structure.

The short-term intervention is a 10-week summer URE, including a summer scholarship that provides a student stipend, research supplies, funding for attending a professional conference, and a training plan during 10 weeks. All students from regional partner institutions and some from the primary institution received the short-term intervention. We denote those students as summer scholars. Summer scholars were selected following a similar application process as academic-year scholars. The majority of summer scholars conducted summer research at the primary institution, and a few of them went to research-intensive partner universities. THRIVE participants were not affiliated with any other URE programs while participating in THRIVE.

THRIVE participants are ideal for this study for two reasons. First, THRIVE is generally representative of multi-institutional and multiyear URE programs offering financial support, faculty-mentored research opportunities, and professional development training, such as Minority Access to Research Careers (MARC) programs, Research Initiative for Scientific Enhancement (RISE) programs, or Building Infrastructure Leading to Diversity (BUILD) programs. Second, despite different interventions, both academic-year and summer scholars share important similarities: they were selected following a similar application process; they were all matched with faculty mentors by the same program coordinators based on research interests; they were invited to participate in summer research multiple times; and they expressed interest in pursuing STEM graduate school.

Data collection and research design

At the conclusion of each summer URE, THRIVE administered a mentor–mentee paired survey to all students and faculty mentors. The Mentor and Mentee surveys were open from August to September, and follow-up e-mails were sent to nonrespondents weekly for 4 weeks. The survey was approved by the Institutional Review Board at the primary institution and was administered using Qualtrics survey software. Both Mentor and Mentee surveys included sociodemographic questions and URE-related instruments (e.g., mentor competency and research career interests). In total, the Mentor and Mentee surveys included 35 and 45 questions, respectively; the 3-year average response rate was 63.3% for the Mentor survey and 92.7% for the Mentee survey.

From Summer 2017 to Summer 2019, 131 different THRIVE students and their faculty mentors took part in the survey. We used data pertaining to Latinx students (n = 119) for this study. As shown in Table 1, 66% of all Latinx students were women, 36% were first-generation students, and 90% were born in the United States. Among the 119 Latinx students, 72% were from the primary institution, a Hispanic majority HSI, and 28% were from other regional partner institutions.

Table 1.

Demographic information for participating students (n = 119 Latinx students)

Variables	Frequency	% missing	Mean	SD
Agea		24.4	20.32	2.37
Gender:
Man student	41	0.0
Woman student	78
First-generation college student status:
First-generation student	36	16.0
Continuing-generation student	64
Nativity:
U.S.-born student	89	16.8
Foreign-born student	10
Classification:a
Freshman	35	0.0
Sophomore	51
Junior	23
Senior	10
Number of mentoring relationships:
One	76	0.0
Two	34
Three	9
Home institution:
Primary institution	86	0.0
Regional partner institutions	33

^aFor students who participated in summer research more than once, we reported their ages and classifications when they participated in summer research for the first time.

Our data also have a clustered structure: among the 119 Latinx students who took the survey, 76 participated in summer research once, 34 participated twice, and 9 participated three times. Thus, some participated in summer research more than once and took the survey each time. Our sample includes 171 faculty–undergraduate dyads composed of Latinx students and their mentors. Different from previous mentoring studies that rely on student-level analyses,^44,45 we used each mentoring relationship as the analysis unit. We borrowed this approach from family science scholars who study intergenerational relationships^67,68 because it was appropriate for our data structure. More importantly, for those students who had multiple mentoring relationships, relationship-level analyses allowed us to detect variation in outcomes within each student and examine how they were influenced by characteristics of each of their mentoring relationships.

Measures

URE student outcomes.

Two dependent variables were included in the study. We first assessed students’ intentions to pursue graduate school after conducting summer UREs using a Likert-type item (1 = “not more likely” to 5 = “extremely more likely”) from the Mentee survey: “Compared to your intentions before the summer program, how likely are you now to enroll in a PhD program in STEM?”

This variable is appropriate for assessing the influences of UREs and mentorships on students’ intentions because it gauges the change in students’ intentions from before to after interventions. Among the 171 dyads in our sample, about 20% involved the mentee stating that after the summer research program, she/he became extremely more likely to intend to pursue a PhD program in STEM, while 18% included the mentee who did not become more intent on pursuing a PhD program in STEM after the summer research program.

The Mentee survey also asked whether the student presented her/his summer URE project at a national or regional professional conference. We used the responses to this question to create the second dependent variable measuring research productivity (0 = the student did not present at a professional conference; and 1 = the student presented at a professional conference). As a direct and short-term outcome of summer research, conference presentation is a suitable measurement of undergraduate research productivity, and it has been used as an outcome of URE productivity in previous studies.²⁶ About 26% of the 171 dyads produced a conference presentation for the undergraduate mentee. All THRIVE participants were required to present their summer projects at a program-specific research symposium, and this dependent variable did not include those required presentations.

Mentoring relationship characteristics.

In terms of gender discordance, both Mentor and Mentee surveys asked: “What is your gender?” The response options included “Woman” and “Man.” We used those terms in the survey and in this paper because they constitute socially constructed gender categories and refer to people’s identities, as opposed to anatomical differences, which are captured with the terms female and male. In our sample, 66% of dyads included a woman mentee, and 34% included a man mentee. In terms of the faculty mentor, 46% of dyads involved a woman, and 54% a man. On the basis of faculty and student gender, we created the first four dummy variables (each coded as 0 = no; and 1 = yes) to measure gender discordance, including “a Latina student and a woman mentor” (n = 47, reference), “a Latina student and a man mentor” (n = 52), “a Latino student and a man mentor” (n = 32, reference), and “a Latino student and a woman mentor” (n = 25).

In our sample, 34% of dyads involved a Latinx faculty mentor. To measure racial/ethnic discordance, we also developed four dummy variables, and each variable was coded as 0 = no; and 1 = yes. They are “a Latina student and a Latinx mentor” (n = 24, reference), “a Latina student and a non-Latinx mentor” (n = 75), “a Latino student and a Latinx mentor” (n = 19, reference), and “a Latino student and a non-Latinx mentor” (n = 38).

In terms of the quantity of mentor–mentee interaction, in the Mentee survey, students were asked to rate the amount of time they spent with their mentors using a Likert-type item (from 1 = “poor” to 4 = “excellent”). About 44% of dyads in our sample involved mentees rating the amount of time they spent with mentors as excellent and 26% of dyads had mentees rating it as poor or fair.

For the quality of mentor–mentee interaction, we focused on mentor cultural competency, defined as the mentor’s ability to respect and value diverse cultural backgrounds. Mentor cultural competency can enhance mutuality, trust, and empathy within the mentoring relationship and is particularly important for mentees from traditionally underrepresented backgrounds.⁶⁹ Studies show that mentees’ perceptions of their mentors’ cultural competence are linked to better ratings of relationship quality.⁷⁰ In the Mentee survey, students rated their faculty mentors’ cultural competency in three areas, that is, discussing diversity, considering their own cultural background as well as yours, and valuing and respecting cultural differences, using the same 3-point Likert-scale in which 1 = “not at all skilled” and 3 = “very skilled.” We created a composite measure by taking the mean response of the three items. The average mentor cultural competency score of the 171 dyads was 2.60, and it ranged from 1 to 3.

Contextual factors.

We constructed the intervention context variable to measure the two different interventions in THRIVE, that is, 0 = short-term intervention/summer scholar; and 1 = long-term intervention/academic-year scholar. Among the 171 dyads, 83% included an academic-year scholar. For institutional context, we used a URE location variable, which indicates where the students conducted their summer research (0 = research partner institution; and 1 = primary institution). We constructed the variable in this way for two reasons: first, the count for each research-intensive partner university was too small to be examined separately; thus, we grouped them into one category; second, given that all THRIVE partners were chosen because of their national and international research reputations, the experience of conducting summer research at research partner institutions might be fundamentally different than conducting research at the primary institution. Therefore, we separated the primary and partner institutions into two categories. Approximately 68% of dyads were at research partner universities. Finally, we controlled for students’ previous research experiences. Approximately 55% of dyads included students who had never participated in summer UREs. Table 2 reports descriptive statistics for all analysis variables.

Table 2.

Descriptive statistics for analysis variables (original data, 171 dyads)

Variables	Frequency	% missing	Mean	SD	Min	Max
Dependent variable
Compared with your intentions before the summer program, how likely are you now to enroll in a PhD program in STEM?		4.7	3.22	1.36	1	5
Did you present your summer URE project at a professional conference?
Yes	126	0.0
No	45
Mentee gender:
Man student	59	0.0
Woman student	112
Previous research experiences:
Never participated in summer UREs before this summer	84	11.1
Have participated in summer UREs before this summer	68
Contextual factors
Institutional context of summer URE:
At research partner institutions	117	0.0
At primary institution	54
Intervention context:
Short-term intervention (summer scholars)	29	0.0
Long-term intervention (academic-year scholars)	142
Mentoring relationship characteristics
Student satisfaction with mentor-mentee interaction time		0.6	3.12	0.93	1	4
Mentor cultural competency		11.7	2.60	0.55	1	3
Gender discordance:
Woman mentor + Latina student	47	8.8
Man mentor + Latina student	52
Man mentor + Latino student	32
Woman mentor + Latino student	25
Race/ethnicity discordance:
Latinx mentor + Latina student	24	8.8
Non-Latinx mentor + Latina student	75
Latinx mentor + Latino student	19
Non-Latinx mentor + Latino student	38

Missing values.

To address potential bias associated with missing values of the analysis variables, we used multiple imputation (MI). MI involves creating multiple sets of values for missing observations using a regression-based approach and is appropriate for survey data.⁷¹ We used the dependent variables, all independent variables, and additional relevant variables (e.g., students’ research self-efficacy) in the estimation process. The MI process fits a model using all other variables in the model as predictors, then imputes missing values for each analysis variable. The method continued until 200 iterations were reached, and the imputed values at the maximum iteration were saved to the imputed dataset.⁷¹ This process ran 20 times such that 20 complete multiply imputed datasets were created (n = 171 in each dataset).

We treated originally ordinal measures (e.g., mentor cultural competency), including one of the dependent variables (intentions to pursue graduate school), as continuous variables in the models, because it is a best practice when using imputed data.⁷¹ Rounding-off imputed values based on discrete categorical specifications produces more biased parameter estimates than treating them as continuous.⁷¹

Statistical modeling.

We analyzed the 20 MI datasets using generalized estimating equations (GEEs) and report results from pooled analyses. GEEs provide a general method for analyzing clustered variables by relaxing several assumptions of traditional regression models.^72–74 Our data have a clustered structure: faculty–undergraduate dyads are clustered within students, and the number of dyads per student ranges from one to three. Thus, GEEs are appropriate for this study. More importantly, because our GEEs statistically controlled for the clustering of student-level factors (e.g., interest in research, upbringing, and personality) as a nuisance,^73,74 they allowed us to isolate the direct effects of mentor–mentee discordance on the two student outcomes.

In total, we estimated four GEE models. Both Models 1 and 2 predicted Latinx students’ intentions to pursue graduate school. Models 3 and 4 predicted Latinx students’ research productivity. In Models 1 and 3, gender discordance variables were focal-independent variables, while racial/ethnic discordance variables were focal-independent variables for Models 2 and 4.

Since the dependent variable for Models 1 and 2 was continuous after MI, we tested normal, gamma, and inverse Gaussian distributions with logarithmic (log) and identity link functions to select the best fitting specifications.⁷⁵ The normal distribution with a log link function was best fitting for both models. For Models 3 and 4, we selected the binomial distribution with the logit link function because the dependent variable was binary.⁷⁵ We selected those specifications as best fitting because they yielded the lowest quasi-likelihood under the independence model criterion (QIC) values. GEEs also require the specification of an intracluster dependency correlation matrix.^73,74 In this study, we specified the unstructured correlation matrix, as it fits better than other applicable specifications (i.e., independent and exchangeable) for the four GEE models. Finally, on the basis of variance inflation factor, tolerance, and condition index criteria, inferences from the GEEs were not affected by multicollinearity. We standardized all continuous independent variables before including them in the models.

Results

Table 3 reports pooled GEE results for the first two models predicting Latinx students’ intentions to pursue graduate school (Models 1 and 2). The exponentiated coefficient values shown in Table 3 are interpretable as the percent change in the dependent variable (after the exponentiated coefficient is subtracted from 1 and multiplied by 100) associated with a unit change in each independent variable. The results from Model 1 show that gender discordance significantly influenced women students: as compared with Latina students who had women mentors, Latina students who had men mentors reported 17% higher levels of intention to pursue a STEM PhD program after their summer UREs (P = 0.019). Gender discordance did not significantly affect Latino (men) students’ graduate school intentions. However, results from Model 2 demonstrate that racial/ethnic discordance affected Latino students’ intentions to pursue graduate school. Specifically, Latino students had 38% higher levels of intention to attend a STEM PhD program when they worked with non-Latinx (versus Latinx) mentors during the summer (P = 0.018). Racial/ethnic discordance was not a significant predictor of Latina students’ intentions to pursue graduate school.

Table 3.

Results of Models 1 and 2 predicting Latinx students’ intentions to pursue STEM graduate school (n = 171 dyads)

Variables	Model 1				Model 2
	B	Exp(B)	CI	P	B	Exp(B)	CI	P
Intercept	1.03**	2.80**	(0.73 to 1.33)	<0.001	0.74**	2.10**	(0.37 to 1.12)	<0.001
Mentee gender:
Man student	ref	ref	ref	ref	ref	ref	ref	ref
Woman student	−0.06	0.94	(−0.25 to 0.13)	0.529	0.26	1.30	(−0.02 to 0.53)	0.064
Previous research experiences:
Never participated in summer UREs	ref	ref	ref	ref	ref	ref	ref	ref
Have participated in summer UREs	−0.05	0.95	(−0.19 to 0.09)	0.461	−0.02	0.98	(−0.17 to 0.12)	0.751
Contextual factors
Institutional context:
Research partner institutions	ref	ref	ref	ref	ref	ref	ref	ref
Primary institution	−0.002	0.99	(−0.18 to 0.17)	0.979	0.07	1.07	(−0.11 to 0.25)	0.447
Intervention context:
Summerscholars	ref	ref	ref	ref	ref	ref	ref	ref
Academic-yearscholars	0.17	1.19	(−0.08 to 0.43)	0.177	0.20	1.22	(−0.05 to 0.45)	0.112
Mentoring relationship characteristics
Mentor-mentee interaction time	0.10*	1.11*	(0.02 to 0.18)	0.013	0.10*	1.11*	(0.03 to 0.18)	0.009
Mentor cultural competency	0.09*	1.09*	(0.01 to 0.17)	0.026	0.11*	1.12*	(0.03 to 0.19)	0.008
Gender discordance:
Woman mentor + Latina student	ref	ref	ref	ref	–	–	–	–
Man mentor + Latina student	0.16*	1.17*	(0.03 to 0.30)	0.019	–	–	–	–
Man mentor + Latino student	ref	ref	ref	ref	–	–	–	–
Woman mentor + Latino student	−0.03	0.97	(−0.26 to 0.21)	0.824	–	–	–	–
Race/ethnicity discordance:
Latinx mentor + Latina student	–		–	–	ref	ref	ref	ref
Non-Latinx mentor + Latina student	–		–	–	−0.03	0.97	(−0.18 to 0.11)	0.667
Latinx mentor + Latino student	–		–	–	ref	ref	ref	ref
Non-Latinx mentor + Latino student	–		–	–	0.32*	1.38*	(0.06 to 0.59)	0.018

^*P < 0.05.

^**P < 0.001.

The other two characteristics of mentoring relationships also influenced Latinx students’ intentions to pursue graduate school. Both Models 1 and 2 suggest that during the summer UREs, if faculty mentors interacted with the mentees more frequently (P = 0.013; P = 0.009) and were more culturally competent (P = 0.026; P = 0.008), undergraduate mentees had greater intentions to pursue graduate school.

Table 4 reports pooled GEE results from Models 3 and 4, which predict whether Latinx students presented their URE projects at national or regional professional conferences. Model 3 indicates that gender discordance was associated with Latina students’ research productivity. When they had a man mentor, they were 70% less likely to present their URE projects at professional conferences than when they had a woman mentor (P = 0.020). Model 4 results show that associations between racial/ethnic discordance and research productivity were not statistically significant. In contrast to intentions to pursue graduate school, research productivity was not associated with the quality and quantity of mentor–mentee interactions.

Table 4.

Results of Models 3 and 4 predicting Latinx students’ research productivity (n = 171 dyads)

Variables	Model 3				Model 4
	B	Exp(B)	CI	P	B	Exp(B)	CI	P
Intercept	−0.18	0.84	(−1.51 to 1.16)	0.795	−0.10	0.90	(−1.63 to 1.43)	0.899
Mentee gender:
Man student	ref	ref	ref	ref	Ref	ref	ref	ref
Woman student	−0.36	0.70	(−1.20 to 0.48)	0.398	−0.44	0.64	(−1.52 to 0.65)	0.431
Previous research experiences:
Never participated in summer UREs	ref	ref	ref	ref	ref	ref	ref	ref
Have participated in summer UREs	0.59	1.80	(−0.26 to 1.43)	0.175	0.50	1.65	(−0.31 to 1.32)	0.226
Contextual factors
Institutional context:
Research partner institutions	ref	ref	ref	ref	Ref	ref	ref	ref
Primary institution	0.40	1.49	(−0.56 to 1.36)	0.415	0.21	1.23	(−0.75 to 1.17)	0.669
Intervention context:
Summer scholars	ref	ref	ref	ref	ref	ref	ref	ref
Academic-year scholars	−0.73	0.48	(−1.79 to 0.32)	0.174	−0.85	0.43	(−1.93 to 0.24)	0.125
Mentoring relationship characteristics
Mentor-mentee interaction time	0.07	1.07	(−0.34 to 0.47)	0.744	0.04	1.04	(−0.37 to 0.45)	0.843
Mentor cultural competency	0.11	1.12	(−0.32 to 0.54)	0.628	0.12	1.13	(−0.30 to 0.53)	0.584
Gender discordance:
Woman mentor + Latina student	ref	ref	ref	ref	–	–	–	–
Man mentor + Latina student	−1.23*	0.29*	(−2.26 to −0.20)	0.020	–	–	–	–
Man mentor + Latino student	ref	ref	ref	ref	–	–	–	–
Woman mentor + Latino student	−0.35	0.70	(−1.32 to 0.62)	0.482	–	–	–	–
Race/ethnicity discordance:
Latinx mentor + Latina student	–	–	–	–	ref	ref	ref	ref
Non-Latinx mentor + Latina student	–	–	–	–	−0.44	0.64	(−1.36 to 0.47)	0.344
Latinx mentor + Latino student	–	–	–	–	ref	ref	ref	ref
Non-Latinx mentor + Latino student	–	–	–	–	−0.07	0.93	(−1.23 to 1.08)	0.902

^*P < 0.05.

Discussion

Currently, there is limited systematic knowledge of the effect of demographic discordance between mentors and mentees on the STEM career trajectories of underrepresented minority students. This knowledge gap exists, despite the reality that many women students and students of color are underrepresented in their STEM fields and currently experience discordant mentoring relationships when participating in UREs. To begin to fill this gap, we systematically tested competing hypotheses about the effects of mentor–mentee discordance on Latinx students’ URE outcomes in one multisite and multiyear URE program, using the faculty–undergraduate dyad as the analysis unit and controlling for other factors that characterize URE mentoring relationships and are known to affect student outcomes. Depending on the dimension of mentor–mentee discordance (gender versus race/ethnicity), student gender, and student outcome, we found differing effects of demographic discordance within URE mentoring relationships on Latinx students.

In terms of changing intentions to pursue graduate school due to URE participation, Latina students reported stronger intentions when they were mentored by men rather than women, and Latino students reported greater intentions when having non-Latinx mentors than Latinx mentors. There were no significant differences between Latina and Latino students. In other words, gender and racial/ethnic discordant mentoring dyads were associated with increased intentions to pursue graduate school among these Latinx students, which aligns with H^b.

Some scholars have observed the privilege crossover effect of cross-gender or -race mentoring relationships on women and minority mentees. They assert that as STEM fields remain White male–dominated, White men mentors typically have more social capital (e.g., connections to power in their disciplines) than women or minority mentors. Their social capital may facilitate their mentees’ career development, for example, by sponsoring mentees in pursuit of high-ranking positions.^29,49–52 In the context of UREs, it may be that men or White mentors’ social capital is particularly valuable for women or racial/ethnic minority students. For example, as compared with other types of mentors, men or White mentors might be able to provide their undergraduate mentees more opportunities to network with other scholars in STEM, which could elicit their mentees’ interests in pursuing STEM graduate programs. It is also possible that students perceive White or men mentors as powerful figures who can enable their career success, so acceptance and support from those mentors increase their interests and confidence in applying for graduate school. Results may also be explained by the marginalization of women and minority faculty in White male–dominated STEM fields. They experience productivity-reducing stress,⁷⁶ receive fewer resources due to bias in their institutional and scientific communities,⁷⁷ and have greater service demands.⁷⁸ As a result, the structural underprivileging of women and minority faculty might make it more difficult for them to inspire women and minority students to pursue graduate school.

These explanations are currently speculative. Future research needs to test these relationships in other populations. An important caveat here is that previous research suggested that faculty members who placed greater value on increasing diversity in STEM were more motivated to engage in URE programs as mentors.⁷⁹ The THRIVE program has a mission to provide research experiences to students from underrepresented and disadvantaged backgrounds, and faculty mentors included in this study voluntarily signed up to mentor THRIVE students. It is possible that we captured a specific group of White and men mentors who understood the diversity crisis in STEM and were committed to supporting women and minority students. Future URE programs should consider faculty’s dedication to diversity issues in STEM when recruiting mentors and make efforts to provide training to help mentors become more aware of and committed to students from underrepresented backgrounds.

We noted different patterns when examining the second student outcome, research productivity. Results suggest that Latina students were more productive in terms of conference presentations when having gender-concordant versus -discordant mentoring relationships, which aligned with H^a. This finding might be related to the “diversity–innovation paradox,” identified in a prior study.⁸⁰ By analyzing data from nearly all U.S. PhD recipients across three decades, researchers found that more demographically diverse research groups innovated at higher rates, but women and racial/ethnic minorities’ novel contributions were often discounted and less likely to earn them academic positions.⁸⁰ Another explanation might be that some Latina students hesitated to disseminate their URE projects at conferences because they felt anxious about public speaking, or they (or their family) did not feel comfortable with them traveling by themselves or with men mentors. Compared with men mentors, women mentors might be better able to help Latina students overcome performance anxiety. Latina students may also feel more comfortable traveling with women mentors to conferences than they would with men mentors. These dynamics may be why we found that Latina students who worked with women mentors were more likely to give conference presentations than Latina students who worked with men mentors. However, for Latino students, gender and racial/ethnic discordance had null effects on their research productivity (aligning with H^c).

Our analyses partially revealed a parallel paradox for Latina students in UREs, who are doubly underrepresented in STEM. On the one hand, they were inspired by their men mentors to pursue STEM graduate school, but on the other hand, when they worked with men mentors, they were less productive in terms of conference presentations. This matters because conference presentations are an expected scholarly activity that can help undergraduates gain acceptance into graduate programs.⁸¹ While men mentors might inspire Latinas to attend graduate school due to the privilege crossover effect, women mentors could actually help them become more competitive for STEM graduate programs by increasing their conference presentations.

With regard to other elements of mentoring relationships, studies have documented that the quality and quantity (frequency) of mentor–mentee interactions were key influences on students’ research satisfaction and gains and STEM persistence.^61–63 Our results showed that when faculty mentors were culturally competent and interacted with mentees frequently, Latinx student mentees were more interested in pursuing graduate school. These findings underscore the importance of cultural competence training for URE faculty mentors as well as ensuring that they have adequate time to provide consistent mentorship to their undergraduate mentees.

An important contribution of this study is that we examined mentor–mentee discordance at the relationship level, while most previous studies were conducted at the student level.^31,41,52 This approach was appropriate for our data structure, allowed us to control for student-level factors, and will be valuable for future studies on URE mentoring relationships. With the growth in URE programs, increasing numbers of students will experience multiple mentoring relationships during their undergraduate careers. Relationship-level analyses are perfectly suited to investigate data with this structure, that is, when a student has multiple faculty mentor relationships. Therefore, we call for scholars to conduct more URE mentorship studies at the relationship level, using statistical techniques such as GEEs that appropriately adjust for within-student clustering of multiple mentoring relationships.

There are several limitations of this study, which suggest future research directions. First, to measure different institutional contexts, we only distinguished the primary institution from research-intensive partner institutions, which might explain why the institutional context variable was not significant. With a larger sample size, future studies should consider more contextual variables at the institutional level. Second, future studies should include other options for gender identity, such as trans man, trans woman, or genderqueer/gender nonconforming, beyond just woman or man, and expand beyond our focus on Latinx students. Our conceptual model, statistical approach, and findings should be tested with other student populations. Third, most students included in this study (including all academic-year scholars) were from HSIs. This context might limit the generalizability of findings to Latinx students at majority-White institutions. Fourth, there were fewer dyads for certain groups (e.g., summer scholars, Latinx mentors, and Latino students with Latinx mentors). Larger samples would benefit future studies. Fifth, presenting at a professional conference is one of several productivity metrics applicable to undergraduate researchers. The variable “presenting at a conference” also does not differentiate students who applied to present but were rejected from those who did not apply to begin with. Future research should seek to improve this measure and explore additional measures of productivity (e.g., journal publication). Sixth, the measure of intentions to pursue graduate school we used in this study gauged only post-URE changes in intentions relative to pre-URE intentions. It does not assess the level of intention nor actual enrollment in graduate school, which future studies should investigate. Seventh, we realize that for many STEM research teams or labs, graduate students and postdoctoral researchers often serve as postgraduate mentors and interact frequently with undergraduate students. While we focused on faculty–undergraduate dyads, future research should examine the effects of mentor–mentee discordance in the context of faculty mentor–postgraduate mentor–undergraduate mentee triads. Finally, future researchers could also explore other more nuanced dimensions of mentoring relationships (e.g., social class discordance), interactions between the three mentoring relationship characteristics, mentee’s implicit bias with respect to the mentor, and student identities using qualitative or mixed-method approaches.

Conclusions

Over the past 20 years, many URE programs like THRIVE have been implementing various interventions to encourage students from underrepresented backgrounds to pursue STEM career paths. Findings from this study provide university administrators, program directors, and future analysts some initial insights on how different aspects of those URE interventions might affect Latinx students’ outcomes. When participating in UREs, many Latinx students engage in demographically discordant mentoring relationships due to the ongoing demographic transition in STEM. It is, therefore, likely that students’ URE faculty mentors will be White men.

Our findings from THRIVE suggest that mentor–mentee discordance may be affecting Latinx students in complex ways, depending on the outcomes of focus. On the one hand, those results are encouraging, as they demonstrate that mentors from different demographic background are contributing to the STEM development of these Latinx students by encouraging them to pursue STEM graduate programs. On the other hand, however, we found that Latina students were less likely to present at conferences when working with men mentors.

One way for program directors to address the potential challenges and maximize the opportunities provided by mentor–mentee discordance/concordance is to ensure that students are offered the opportunity to work in meaningful ways with multiple diverse mentors. This is particularly relevant in multiyear UREs. Other researchers have argued for the United States of multiple mentors to increase diversity in STEM as it provides students with exposure to more diverse perspectives.⁸² We also believe that if URE students have the opportunity to be mentored by multiple faculty mentors from different demographic backgrounds, they will benefit from both discordant and concordant mentoring relationships. Therefore, universities should develop more structured and multiyear interventions (like THRIVE), and URE programs should make greater efforts to create a large and diverse pool of faculty mentors so that more undergraduate students can benefit from multiple research experiences with diverse mentors. Finally, the STEM community must invest more into improving conditions for faculty from underrepresented backgrounds, such that they are positioned to most effectively serve as the role models for students from underrepresented backgrounds.