New Findings Challenge Assumptions About What Makes STEM Mentoring Work
Ramos, H. V., & Vu, K.-P. L. (2026). Research, mentoring and science self-efficacy: A longitudinal study of a STEM intervention at CSULB. Journal of STEM Education, 27(2), 28–33.
Introduction
The case for undergraduate research as a driver of persistence in science, technology, engineering, and math fields is well established. Hurtado et al. (2011) found that students who take part in faculty-guided research report a stronger sense of connection to their intended major, and Cartrette and Melroe-Lehrman (2012) showed that research training builds resiliency and problem-solving skills that carry over well beyond the lab. What has remained murkier is why these programs work. Undergraduate research is one of the most consistently cited levers for boosting persistence in STEM, but few studies have pulled apart which part of a research-training program actually has an impact: the research experience itself, or the mentor guiding it. Ramos and Vu (2026) take up that question at California State University, Long Beach (CSULB), examining students in the NIH-funded Building Infrastructure Leading to Diversity (BUILD) Scholars Program, an upper-division, two-year research-intensive initiative.
Methods
Drawing on four years of longitudinal survey data (2019–2022) collected through UCLA’s Student Annual Follow-up Survey, the authors compared 74 BUILD scholars against 969 matched non-BUILD peers, using propensity score matching to balance the groups on science identity, race, gender, and degree aspirations. Science self-efficacy served as the outcome, measured through a validated six-item scale capturing confidence in tasks like designing studies and interpreting data. Stepwise multiple regression then tested how much of the BUILD effect could be explained by two components: having a research mentor and participating in research.
Results
BUILD scholars started out with notably higher science self-efficacy than their peers, but once research participation entered the model, that gap shrank by nearly 50%, and the mentoring variable stopped predicting self-efficacy altogether. Mentoring’s apparent contribution to science self-efficacy appears to run through research participation rather than standing on its own. At CSULB, faculty mentors are also the ones delivering research training, so once a student’s actual research involvement is accounted for, the mentoring relationship no longer adds separate predictive power. By the final model, with all controls included, the BUILD effect had fallen by nearly 70% from its original size (from 6.13 to 1.95) yet remained statistically significant throughout every specification. Across every model, BUILD participation remained a significant predictor, though its strength dropped by roughly 70% once demographic and program covariates were fully accounted for.
Discussion
The authors read their results as evidence that mentoring and research participation are not interchangeable components but are also not fully separable in a program like BUILD, where the same faculty member typically supplies both. That structural overlap makes it hard to say whether mentoring would matter more in a program where research supervision and mentoring were delivered by different people. The authors are candid about the limitations this creates: without a true baseline survey (since BUILD only recruits juniors), the design remains vulnerable to selection and maturation effects, even after propensity matching.
Implications for Mentoring Programs
For practitioners designing peer or faculty mentoring components within STEM interventions, the practical lesson is not that mentoring doesn’t matter, but that its benefit may be conditional on whether it is paired with real research access. A mentor who offers encouragement and psychosocial support without also facilitating hands-on research tasks may not move the self-efficacy needle in the way the literature often assumes. Programs might get more return on investment by explicitly structuring mentoring relationships around research participation, rather than treating mentoring and research training as separable, optional add-ons.
Read the full article here
