Mentorship resources and citation-elite journal publication trajectories after training: Evidence from bioscience mentor-mentee networks

• Prior research leaves the long-term consequences of mentorship unresolved, with potential benefits and constraints suggesting outcomes may not be uniform.The introduction notes concerns about unequal academic capital, dependence within supervisory relationships, and constraints on distinctive research agendas and intellectual autonomy.Source: INTRODUCTION · p. 3
• The introduction presents research capabilities as gradually developed and potentially not observable at a single post-graduation point.It describes the transformation of tacit knowledge into cognitive schemas, strategic repertoires, and relational skills.Source: INTRODUCTION · p. 3
• Prior studies have focused on mentor-provided resources and limited post-graduation windows, leaving sustained scholarly development and long-term productivity insufficiently explained.Source: INTRODUCTION · p. 3
• The paper uses descriptive heterogeneity analyses, propensity score matching, and regression analyses in a multi-stage empirical strategy.The strategy examines supervisory experiences across multiple career windows.Source: INTRODUCTION · p. 3
• The study asks how mentorship-period citation-elite publishing, resource-related factors, and post-graduation continuity relate to later publication across career stages.Its three questions compare mentorship-period publishing with post-graduation outcomes and examine resource and continuity relationships across training and post-graduation periods.Source: INTRODUCTION · pp. 3–4

Mentorship and scholarly development

• Mentorship can build research skills, networks, and productivity, but outcomes vary and relationships may combine support with harmful or ineffective experiences (Howard et al., 2006; Steiner et al., 2002, 2004).Reported problems include mentor absence, misuse of power, incompatibility, inadequate support, misaligned expectations, and unequal treatment.Source: Mentorship and scholarly development · pp. 4–5

Mentor guidance and capability formation

• The study distinguishes mentorship resources from later capabilities, including research autonomy, thematic mobility, and network agency, to assess sustained independent trajectories beyond graduation.These capabilities involve formulating original questions, navigating research contexts, and building scholarly ties beyond one’s doctoral lineage.Source: Mentorship and scholarly development · p. 5, Mentor guidance and capability formation · p. 6
• Mentorship-related coauthorship and joint research can provide mentees mentors’ expertise, reputation, networks, and publication-practice knowledge, helping them enter high-impact venues and generating persistent advantages in later academic careers (W. Li et al., 2019).Long-term benefits may depend on whether mentees develop intellectual independence.Source: Mentor guidance and capability formation · pp. 5–6

Peer interactions and laboratory structure

• Feldon et al. (2019) and Ynalvez et al. (2017) associate senior-peer interactions and stimulating laboratory climates with skill development and doctoral publication productivity.Source: Peer interactions and laboratory… · p. 7
• Larger mentoring groups may benefit academic performance among those remaining in academia, yet constrained advisor attention can intensify competition and lower long-term academic survival (Xing et al., 2025; Luckhaupt et al., 2005).Source: Peer interactions and laboratory… · p. 7

• The study examines how direct supervision and peer-group contexts within mentorship affect mentees’ citation-elite journal publication.Source: Peer interactions and laboratory… · p. 7

Data and sample

• The biosciences mentorship dataset covers 743,176 mentor–mentee relationships; 239,931 mentor–mentee pairs had at least one mentor-linked publication, and records span 1800 to 2020.Source: METHODOLOGY · pp. 7–8

Observation periods

• The study extends mentorship through three years after graduation and contrasts years 1–3 continuity with later outcomes in years 4–10 and 6–20.It also assesses outcomes over 5-, 10-, and 20-year windows and the full observed post-graduation span.Source: METHODOLOGY · p. 8

Analytical strategy

• Analyses progress from descriptive and trajectory comparisons to propensity-score matching and regression, then non-overlapping landmark analyses; these are observational, not causal, designs.Matching and regression adjust for measured mentor characteristics, prior mentee performance, and supervision environments.Source: METHODOLOGY · p. 9

Outcome definition

• The reported source-selection rule uses global 95th-percentile thresholds of 1.6667, 52.0, and 16,239.6, respectively; it identifies 6,847 journals, or 2.63%.Source: Outcome variable · p. 9

Trajectory groups

• Graduates are classified into four mutually exclusive groups by changes in annual citation-elite publication rates; 11.8% published in neither period.The outcome is average annual post-graduation citation-elite publication count, avoiding direct dependence of cumulative counts on observed career length.Source: Outcome variable · p. 10

Measures

• Training context is measured by the share of mentee publications coauthored with the mentor and the number of overlapping co-mentees; post-training continuity tracks recurring topics and mentor-linked coauthors.Net_Consistency captures the retained share of collaborators from mentor–mentee joint publications, not the mentee’s full network-building capability.Source: Outcome variable · pp. 10–12

Propensity score matching

• PSM0 links pre-graduation citation-elite publishing to mentoring-period resources, while PSM1 compares post-graduation publishers with nonpublishers.The focal mentoring resources are direct mentee–mentor coauthorship intensity and peer group size.Source: Control variables · pp. 13–14
• PSM2 compares publication intensity above and below each focal variable’s median among graduates with positive post-graduation output, without a causal ATT interpretation.The analysis conditions on observed post-graduation publication, and focal variables are not externally assigned.Source: Control variables · pp. 14–15
• Matching uses logistic propensity models with linear and squared covariates, common-support restrictions, one-to-one matching with replacement, and mentee-clustered inference.The three designs are interpreted as adjusted observational comparisons.Source: Control variables · pp. 14–15

Regression analyses

• For complete-case pair samples, PPML models zero-heavy, right-skewed publication counts; annualized-rate OLS is a benchmark, and VIF checks found no problematic multicollinearity.The principal PPML specification includes an exposure offset and mentee-clustered standard errors.Source: Regression · pp. 15–16

Two-part model

• A two-part robustness model estimates the probability of any citation-elite publication and, among positive outcomes, publication intensity using a Gamma model with a log link.Both parts use standard errors clustered at the mentee level.Source: Regression · p. 16

Landmark design

• The non-overlapping landmark design measures topic and network continuity before two subsequent outcome periods, re-estimates the models, and retains an observational interpretation.The design separates continuity measurement from later publication outcomes but does not make the estimates causal.Source: Landmark design · p. 17

Citation-elite journal publication across two periods

• Mentorship-period citation-elite and overall publication rates correlate more strongly with post-graduation citation-elite rates than most explanatory variables do.Most explanatory variables have absolute correlation coefficients generally below 0.2; aggregate correlations cannot establish whether persistence reflects mentor-related resources, cumulative advantage, or internalized research capability.Source: Citation-elite journal publication… · p. 17
• Pre-graduation citation-elite publishers have higher Direct_Percent and PeerCount_log, while G1 shows lower Topic_Continuity and Direct_Percent than G2.G1 is associated with mentors with higher h-index values and larger mentored collaboration networks; Net_Consistency varies by subgroup and pre-graduation publication status.Source: Citation-elite journal publication… · p. 18
• Pre-graduation and post-graduation citation-elite publishing are positively associated but incompletely persistent across career stages.Some mentees without pre-graduation citation-elite publications publish after graduation and resemble sustained performers on Topic_Continuity and Net_Consistency.Source: Citation-elite journal publication… · p. 19

Resource-related factors and citation-elite journal publication

• Across 5-, 10-, and 20-year post-graduation windows, Direct_Percent and PeerCount_log are generally negatively associated with outcomes; higher Direct_Percent also corresponds to lower estimated probability and intensity.For a 0.1-unit increase in Direct_Percent, estimated probability differences are -0.11, -0.09, and -0.08 percentage points across the 5-, 10-, and 20-year windows, and intensity outcomes are -4.0%, -3.6%, and -2.3%.Source: Resource-related factors and… · p. 20
• After matching, mentorship-period citation-elite publishers show higher Direct_Percent and PeerCount_log than matched nonpublishers.The PSM0 standardized mean differences are +0.47 for Direct_Percent and +0.13 for PeerCount_log, consistent with stronger direct mentor involvement and larger peer groups during training.Source: Resource-related factors and… · p. 20, Figure 3 · p. 21

• Early topic retention may lower transition costs, but longer-window negative estimates do not establish intentional agenda broadening because continuity and output are measured contemporaneously.Source: Capability-related factors and… · p. 22
• Retained collaborator share is more consistently associated with publication intensity, although the measure captures observed continuity rather than autonomous network agency.Existing ties may facilitate coordination and access to complementary expertise as topics change (Reagans & McEvily, 2003; Tripodi et al., 2020).Source: Capability-related factors and… · p. 22
• Topic_Continuity estimates shift from +0.35 short term to -0.29 medium term and -1.03 long term, while Net_Consistency remains positive at +0.30, +0.34, and +0.51.Source: Capability-related factors and… · p. 22, Figure 4 · p. 23

DISCUSSION AND CONCLUSION

• Mentees who publish in citation-elite journals during mentorship are more likely to publish in them after graduation, though the observational study cannot distinguish selection, cumulative advantage, or training processes.The persistence pattern is robust across the analyses, but it does not identify which mechanism accounts for it.Source: DISCUSSION AND CONCLUSION · p. 24
• Some mentees without citation-elite publishing during mentorship later publish by mobilizing established knowledge and collaboration networks, so early publication status does not fully determine trajectories.The authors suggest early disadvantages associated with resource-dependent training need not persist when mentees effectively use and extend accumulated resources after graduation.Source: DISCUSSION AND CONCLUSION · pp. 24–26
• Greater mentor-collaboration concentration and larger co-mentee groups are associated with training-period publishing, but larger groups—and mentor-linked advantages—do not necessarily remain favorable after graduation.The authors interpret the co-mentee pattern as a difference in observable publishing trajectories, not academic survival; employment status and reasons for leaving publication were not observed.Source: DISCUSSION AND CONCLUSION · p. 25
• Topic continuity can aid the transition to independent research, whereas retained collaborative ties remain useful as agendas evolve and longer-term development may require new directions.The authors suggest balancing continuity with adjustment: prior knowledge can ease entry into an independent career, while collaborations can be redeployed across changing research agendas.Source: DISCUSSION AND CONCLUSION · p. 26
• The authors caution that journal-level indicators do not measure individual quality and recommend portable support—not reduced mentoring or smaller groups—including multiple mentors, clear credit, agenda ownership, external networks, and transition support.They frame these practices as ways to pair resource-rich training with mechanisms that make support portable after training.Source: DISCUSSION AND CONCLUSION · p. 26
• All estimates are observational and do not establish causation; analyses also condition on successful linkage, mentor-linked activity, window eligibility, and complete data, while longer windows disproportionately represent earlier cohorts.The Academic Family Tree is crowdsourced and includes multiple mentorship types.Source: DISCUSSION AND CONCLUSION · pp. 26–27
• Topic retention, retained-collaborator share, coauthorship concentration, and overlapping co-mentee counts are proxies rather than direct measures of support, peer interaction, autonomy, or capability.Source: DISCUSSION AND CONCLUSION · p. 27
• The fixed citation-elite classification uses globally pooled source indicators that may reflect field, journal-age, and snapshot-time differences; venue placement is not article-level quality.Source: DISCUSSION AND CONCLUSION · p. 27
• The data do not identify career retention, employment sector, or whether associations differ by gender, geography, institutional prestige, citizenship, caregiving, or other structural conditions.Future work should test field- and year-normalized outcomes, cohort heterogeneity, and qualitative mechanisms.Source: DISCUSSION AND CONCLUSION · p. 27

• ChatGPT was used to improve the manuscript’s language, clarity, and readability.Source: GENERATIVE AI STATEMENTS · p. 27
• The authors reviewed and revised all AI-assisted text.Source: GENERATIVE AI STATEMENTS · p. 27
• The authors take full responsibility for the manuscript’s accuracy and final content.Source: GENERATIVE AI STATEMENTS · p. 27

• The code can be found at the authors’ GitHub repository.Source: CODE AND DATA AVAILABILITY · p. 27
• The data are open to the public.Source: CODE AND DATA AVAILABILITY · p. 27

• Across matched comparisons, topic-continuity associations vary by window, while network consistency is positive and direct mentor collaboration and peer count are generally negative.Source: SUPPLEMENTARY INFORMATION · p. 51
• The maximum variance inflation factor is 3.651, below the conventional threshold of 5.Source: SUPPLEMENTARY INFORMATION · p. 52
• The two-part model separates publication probability from positive-output intensity, while PPML models counts with log exposure as an offset.Source: SUPPLEMENTARY INFORMATION · pp. 55, 57
• Landmark analyses measure capabilities in post-graduation years 1–3 and outcomes in non-overlapping years 4–10 or 6–20, strengthening temporal ordering without establishing causality.Source: SUPPLEMENTARY INFORMATION · p. 59
• Long-horizon network consistency and peer-count results are the most stable across the probability and positive-output intensity margins.Source: SUPPLEMENTARY INFORMATION · p. 60