The association between early career informal mentorship in academic collaborations and junior author performance

The study investigates informal mentorship in academic collaborations, focusing on junior scientists who receive support from one or more senior collaborators without formal supervisory roles. Leveraging large-scale publication data, the authors aim to assess whether the quality of mentorship predicts protégés’ later scientific impact when they publish independently from their mentors. The work situates mentorship as a key mechanism for transmitting cultural capital, skills, and networks, potentially mitigating barriers for underrepresented groups. The authors highlight advantages over prior studies: broadening beyond thesis advisors, avoiding recall bias by using observed impact, spanning many journals and disciplines, and constructing careful matched comparisons across millions of mentor–protégé pairs. They contrast their focus on mentors (senior collaborators) with prior work on coauthorship with top scientists and address the role of mentors’ social capital versus their scientific impact. They also motivate a gender-focused analysis, examining how mentor and protégé gender relates to outcomes.

The paper draws on the Science of Science literature analyzing innovation, diversity, productivity, team assembly, and success via publication and citation data. It references mentorship research highlighting career benefits, organizational continuity, and developmental networks. In gender equity, prior studies document disparities and the potential benefits of mentorship for women, though often comparing those with female mentors to those without mentors rather than to those with male mentors. The authors differentiate their approach from Li et al. (2019), who study early coauthorship with top scientists irrespective of seniority, and from Ma et al. (2020), who analyze formal advisor–advisee relationships including coauthored outcomes; here, mentorship is informal, may involve multiple seniors, and post-mentorship outcomes exclude coauthored papers with mentors. The study also engages with debates on whether mentors’ scientific impact (big-shot) or network position (hub) is more predictive of protégés’ success.

Data: The study analyzes the Microsoft Academic Graph (MAG) up to December 31, 2019, which includes publication metadata and a citation network. The authors report analyzing 215 million scientists and 222 million papers. They address name disambiguation and derive scientist disciplines (majority rule: at least 50% of a scientist’s papers in one discipline), scientific impact (from citation networks), gender (via Genderize.io), and institutional rank (Shanghai ARWU). Mentor–protégé identification: Academic age is years since first publication. Junior years are the first 7 years; senior years are thereafter. A junior coauthoring with a senior constitutes a protégé–mentor pairing, provided they share discipline and a US-based affiliation, and have coauthored at least one paper with 20 or fewer authors. The mentorship period spans from the protégé’s first publication year to the year they become senior. Inclusion criteria require: (i) the protégé has at least one senior-year publication without mentors; (ii) protégé’s affiliations are in the US throughout mentorship; (iii) mentor and protégé share discipline; (iv) at least one shared affiliation on a publication; (v) at least one jointly authored paper with ≤20 authors during mentorship; (vi) no 5-year publication gap for the protégé. Ten disciplines with most pairs are analyzed (Biology, Chemistry, Computer Science, Economics, Engineering, Geology, Materials Science, Medicine, Physics, Psychology), covering over 97% of identified pairs. Measures: Mentorship quality is operationalized as (1) big-shot experience: mentors’ average citations per annum up to their first coauthored year with the protégé, averaged across mentors; (2) hub experience: mentors’ average degree in the collaboration network up to that year, averaged across mentors. Outcome is the protégé’s post-mentorship impact measured as the average c5 (citations accrued within 5 years) of papers authored during the protégé’s senior years that exclude all identified mentors. Survey validation: 2,000 identified protégés were invited; 167 completed a survey probing whether identified senior collaborators provided advice (writing, study design, data analysis, addressing reviewer comments, venue selection) and broader career support (grant writing advice, recommendation letters, career planning, introductions). High proportions reported receiving such mentorship (e.g., 72–85% agreed for skills; 95% agreed to at least one skill; ~80% affirmed at least one form of broader support), including among those for whom the mentor was not an advisor or committee member and across disciplines, providing face validity that the identified relationships involved mentorship. Analysis: Coarsened Exact Matching (CEM) is used to study associations between mentorship quality and outcomes. For each independent variable (big-shot and hub experience), protégés are divided into quintiles Q1–Q5. For i = 1..4, treatment is Q_{i+1} and control is Q_i. Matches control for protégé characteristics: number of mentors, first mentored publication year (cohort), discipline, gender, affiliation rank on first mentored paper, years active post mentorship, and average academic age of mentors; when testing big-shot, matching also balances hub experience, and vice versa. The authors estimate the relative increase in average post-mentorship impact for treatment versus control, report t-tests, and bootstrap 95% confidence intervals. Robustness checks include alternative outcome window (c10), alternative mentorship quality aggregations (max, median, sum), alternative junior/senior thresholds (≤6/≥9 and ≤5/≥10), and subgroup analyses. A total of 204 CEMs support Figure 2 and additional supplementary analyses; 32 matchings support Figure 3.

• Mentorship quality and protégé outcomes: Increases in mentors’ big-shot experience are significantly associated with higher protégé post-mentorship impact, with gains up to 35% between adjacent quintiles. Hub experience is also positively associated but with smaller gains, not exceeding 13%. Results are robust across alternative specifications and age thresholds, and not driven by differences in protégés’ innate ability per supplementary analyses. • Big-shot vs hub: The mentors’ scientific impact (big-shot experience) is a stronger predictor of protégé outcomes than mentors’ network degree (hub experience), implying that mentors’ impact matters more than their number of collaborators. • Robustness across contexts: The association between big-shot experience and post-mentorship impact persists across disciplines, affiliation ranks, number of mentors, average mentor age, protégé gender, and protégé cohort (first publication year). • Gender composition of mentorship: Increasing the proportion of female mentors is associated with a decrease in protégés’ post-mentorship impact for both male and female protégés; the decrease can reach as high as 35%, depending on the number of mentors and proportion female (Figure 3a,b), after controlling for big-shot experience and other covariates. • Mentor gains by protégé gender: Female mentors experience an average 18% loss in citations on mentored papers when mentoring female versus matched male protégés; male mentors’ gains do not significantly differ by protégé gender (Figure 3c).

The findings suggest that informal mentorship in academic collaborations is associated with protégés’ later independent impact, with mentors’ prior scientific impact (big-shot experience) playing a particularly important role relative to network connectedness (hub experience). This indicates that the quality and prestige of mentors may transfer resources, reputation, and skills that enhance protégé outcomes. The relationships hold broadly across fields and institutional contexts. The gender analysis indicates that a higher fraction of female mentors correlates with lower post-mentorship impact for protégés and reduced gains for female mentors when mentoring women, raising questions about mechanisms such as differential service burdens, resource access, topic selection, or systemic biases. While the matching design strengthens comparisons, the study is observational and cannot definitively establish causality or mechanisms. Nonetheless, the results inform debates on mentorship structures and diversity policies by highlighting potential unintended consequences and the importance of cross-gender mentorship for women’s long-term impact.

This study identifies 3 million informal mentor–protégé pairs in academia and triangulates large-scale publication data with a survey to validate that identified senior collaborators provided mentorship. It introduces two measures of mentorship quality—big-shot and hub experience—and shows that mentorship quality, especially mentors’ scientific impact, is associated with higher protégé impact on papers authored independently post mentorship. The relationship is robust across disciplines and institutional ranks. The gender analysis suggests that increasing the proportion of female mentors is associated with lower post-mentorship impact for protégés and reduced citation gains for female mentors when mentoring women, implying that opposite-gender mentorships may, on average, be linked to higher long-term impact for women who remain in academia. The authors call for future research to uncover mechanisms, including comparing newcomer vs. incumbent mentors, tracing whether mentors’ networks cite protégés, and analyzing topic and skill transfer during and after mentorship. Policymakers should consider potential second-order effects when designing diversity initiatives and ensure that advancing gender equity is a collective responsibility across the scientific community.

The study relies on observational data and matching methods (CEM), which cannot establish causality; unobserved confounding and selection effects may remain. Mentor–protégé identification depends on publication-based rules (shared discipline and affiliation, coauthorship thresholds) and name disambiguation, which may introduce misclassification. The analysis focuses on protégés with US affiliations during mentorship and excludes those without post-mentorship publications, potentially limiting generalizability. Measures of mentorship quality (big-shot and hub experience) and outcomes (c5) may be sensitive to field-specific citation practices and outliers despite bootstrapping. Gender inference via first names may incur errors, and the survey sample (n=167) is relatively small. Societal and institutional factors (resources, service loads, biases) are not captured, leaving mechanisms underlying gender-related associations unresolved.