Mentorship quality, not quantity, boosts STEM students’ identity and well-being

When it comes to mentorship in science, technology, engineering, and mathematics, a new longitudinal study suggests that what matters most is not how many mentors a student has, but how good those mentoring relationships are. The findings, published in the International Journal of STEM Education, come from a team led by Rachelle Pedersen of Texas Tech University, together with colleagues at Texas A&M University and Claremont Graduate University, and draw on more than two years of data from hundreds of STEM undergraduates enrolled at twelve public Hispanic-Serving Institutions in the western United States. Using a social network approach that treats mentorship as a web of relationships rather than a single faculty-student pairing, the researchers found that the average quality of a student’s mentor network—measured through emotional and psychosocial support, career guidance, role modeling, and relationship satisfaction—predicted gains in both STEM domain identity and psychological well-being over time. The sheer number of mentors, and even the structural diversity of the network, mattered far less.

The study is grounded in Developmental Network Theory, a framework that reconceives mentorship as a constellation of supportive relationships spanning faculty, graduate students, peers, family members, and professionals outside the university. Where classic mentoring research has tended to zoom in on dyadic relationships—a lone professor guiding a lone student—the developmental network perspective asks how the composition, structure, and content of an entire support system shape a young scientist’s trajectory. Within this framework, mentoring support is traditionally divided into three processes: psychosocial support such as encouragement and counseling, career support such as coaching and sponsorship, and role modeling, in which mentors demonstrate that success in the field is achievable and provide a concrete pathway toward it. Each of these processes was assessed for every mentor a student named, allowing the team to compute a composite quality score for each student’s entire network.

The data come from the My College Pathways project, a longitudinal study launched in Fall 2019 that recruited 1,310 White and Hispanic/Latino(a) STEM juniors and seniors across twelve universities. The present analysis focused on the 372 students who reported having at least one mentor in Spring 2021, the third semester of the study, when the mentor network questionnaire was administered. Students listed up to five mentors by name or initials and answered structured questions about each: the mentor’s gender, race or ethnicity, and career stage, along with ratings of the psychosocial support, career support, and role modeling each mentor provided, all on seven-point scales. Crucially, the questionnaire also captured whether mentors knew one another, enabling the researchers to calculate each network’s “effective size”—a social network metric reflecting the number of unique, non-redundant connections and thus the diversity of information, resources, and opportunities flowing through the network.

To test whether mentor networks predicted student outcomes, the team employed pre-registered longitudinal structural equation modeling, a statistical technique that can estimate the effect of network characteristics on later outcomes while statistically controlling for earlier levels of those same outcomes. This longitudinal control is critical: because a student who already feels like a scientist is likely to attract and sustain better mentoring, any credible analysis must account for baseline identity, well-being, and grade point average before attributing change to the mentor network. The model fit the data well, with confirmatory indices within accepted thresholds, and the researchers applied Benjamini-Hochberg corrections to guard against false positives across the many paths tested. Missing data, screened with Little’s test and found to be missing completely at random, were handled through full information maximum likelihood estimation, and standard errors were cluster-robusted to account for students nested within universities.

The results were strikingly consistent. Average network mentorship quality uniquely and positively predicted later domain identity—measured with an adapted science identity centrality scale asking students, for example, whether they had come to think of themselves as a “scientist” in their particular major—and it also uniquely predicted later psychological well-being, assessed with Ryff’s multidimensional model of eudaimonic well-being covering purpose in life, personal growth, environmental mastery, and self-acceptance. The standardized effects, while modest at approximately 0.16 for identity and 0.13 for well-being, were obtained over and above the substantial influence of prior levels of each outcome. By contrast, none of the structural or compositional features of the network—the effective size of the mentor set, the proportion of women, the proportion of faculty, the presence of near-peer or off-campus mentors—predicted any of the three outcomes. Even more surprisingly, no mentor network characteristic predicted cumulative GPA once prior GPA was controlled, suggesting that mentoring’s influence operates primarily through motivational and psychological channels rather than grades directly. A sensitivity analysis confirmed that replacing the effective-size metric with a simple one-mentor-versus-many distinction produced no appreciable differences, reinforcing the conclusion that quality, not quantity, is the operative ingredient.

One exploratory finding added an important nuance. Because Hispanic/Latino(a) students in the sample tended to have a larger share of Hispanic/Latino(a) mentors in their networks than their White peers, the researchers ran a multiple-groups analysis testing whether the pathways from network characteristics to outcomes differed by ethnicity. For most outcomes, the constrained model fit equally well, indicating similar patterns across groups. But for well-being, the freely estimated model fit better, revealing that the proportion of Hispanic/Latino(a) mentors in the network was uniquely and positively associated with well-being for Hispanic/Latino(a) students—a homophily effect that did not appear for White students with respect to White mentors. The authors suggest that demographically similar mentors may supply culturally resonant emotional support and strategies for buffering negative stereotypes, and note that many such mentors come from outside the institution: family members, coaches, peers, and community figures who already play a central role in many Hispanic/Latino(a) students’ decisions to pursue STEM. Programs like Familias por el Exito en STEM, which formally integrate family and community support into a student’s scientific journey, exemplify how institutions might build on this resource.

The study’s context deserves attention as well. Data collection spanned the COVID-19 pandemic, beginning in Spring 2020, and the researchers observed that Hispanic/Latino(a) students reported significantly lower well-being than their White counterparts at the Spring 2021 time point, consistent with evidence that students from historically underrepresented backgrounds shouldered compounding responsibilities of family, work, and school during the crisis. Against this backdrop, the finding that high-quality mentoring buffered well-being is particularly notable. It is also telling that of the original 1,310 participants, more than 500 could not identify a single mentor at all when surveyed—numbers the authors interpret as a symptom of the pandemic’s disruption of mentoring access, and more broadly as evidence that access itself remains a first-order barrier for historically underrepresented students. The researchers argue that universities should expand pathways into mentoring relationships through internships, research experiences, summer bridge programs, and professional organizations, while also reforming institutional policies that disproportionately burden women and minority faculty with mentoring labor, including recognition of mentoring in promotion and review practices.

For practitioners, the message is actionable in two directions. Students, the authors suggest, can be trained not merely to find mentors but to map the kinds of support they are receiving across their networks and identify gaps—a strategy used in reflection-based interventions that have already shown promise in prior studies of STEM students. Mentors and programs, meanwhile, can take deceptively simple steps to raise relationship quality: opening lab meetings with personal check-ins, using question prompts that invite self-disclosure, and other practices that foster what psychologists call positivity resonance and psychological similarity. Prior work has shown that such low-cost techniques measurably improve mentoring relationship quality, and the present findings suggest those improvements translate into downstream gains in how students see themselves as scientists and how well they sustain a sense of purpose and meaning through the inevitable adversity of a scientific career.

The authors are careful to note limitations. The sample comprised only White and Hispanic/Latino(a) students at Hispanic-Serving Institutions, so the patterns should not be generalized to all demographic groups; gender categories used at the time of data collection were not fully inclusive; mentor networks were measured at a single time point even though networks are hypothesized to evolve dynamically; and the ego-centric network design limited the number of questions that could be asked per nominated mentor. Future work, they write, should track how network composition shifts across a student’s undergraduate tenure and examine whether structural features begin to matter at different developmental stages. Still, the central conclusion stands as a clear signal for higher education policy: efforts to broaden participation in STEM will get further by investing in the depth of mentoring relationships—ensuring every mentor a student encounters delivers real emotional support, career guidance, and credible role modeling—than by simply maximizing headcounts.