Female early-career scientists have conducted less interdisciplinary research in the past six decades: evidence from doctoral theses

Interdisciplinary research (IDR) is increasingly central to addressing complex societal challenges (e.g., public health, climate change, biodiversity) and is promoted by science policy and funding agencies. Early-career scientists (ECSs) face pressures to publish, specialize, and establish independence, making IDR a potentially risky strategy due to higher time costs, identity ambiguity, publication and recognition challenges, and possible short-term productivity penalties. Gendered systemic barriers in STEM—reduced access to funding, stereotypes, collaboration and leadership challenges, and caregiving burdens—may further discourage female ECSs from riskier IDR paths. This study focuses on five hard-science domains (behavioral sciences, biological sciences, engineering, health and medical sciences, mathematical and physical sciences) using U.S. doctoral theses (1950–2016) to examine ECSs’ engagement in IDR. Research questions: RQ1—What is the prevalence of IDR among ECSs over time, by domain and by university research intensity? RQ2—How does gender influence ECSs’ engagement in IDR across time, domains, and institution types? RQ3—Is advisors’ gender correlated with students’ participation in IDR, and does student–advisor gender pairing relate to gender disparities in IDR?

The literature indicates that IDR can yield long-term benefits (greater impact, novel combinations of knowledge) but may incur short-term citation or productivity penalties and recognition challenges. Evidence on gender and IDR is mixed. Some theories and studies suggest women may be inclined to integrative thinking and IDR; others emphasize cultural and systemic barriers (stereotypes, masculine STEM culture, Matilda effect) that can deter women from riskier strategies like IDR, especially early in careers. Mentorship is pivotal for thesis topic selection and research strategies; advisors influence students’ research framing and IDR engagement. Differences in mentoring styles by advisor gender and potential benefits of same-gender mentorship are discussed in prior work, but evidence is mixed and limited regarding how advisor or student–advisor gender pairing affects IDR. Interdisciplinarity measurement approaches include citation-based methods and top-down subject classification-based co-occurrence. For theses (often lacking citation data), subject co-occurrence and distances across subjects are suitable proxies. The study identifies gaps concerning ECSs’ historical IDR participation, gender effects on IDR across time, disciplines and institutions, and the role of advisor gender and gender pairing.

Data: ProQuest Dissertations & Theses (PQDT) Sciences and Engineering Collection covering U.S. doctoral theses in five domains (behavioral sciences, biological sciences, engineering, health and medical sciences, mathematical and physical sciences), 1950–2016. Initial dataset contained 1,109,491 theses; focusing on the five domains yields 920,619 theses. Final analytic sample retains theses with students’ gender confidently inferred as male or female (via first name), resulting in 675,135 theses from 747 universities. Advisors’ names are available mainly from 1980 onward; advisor-related analyses use records with advisor names. Universities are categorized by Carnegie Classification (R1: very high research activity, R2: high research activity, Others). Variables: - Dependent variable: Interdisciplinarity indicator per thesis (Distance_t), defined as the average pairwise cognitive distance between all PQDT secondary subjects assigned to the thesis. PQDT includes 552 secondary subjects across 22 broader categories. A co-occurrence matrix of subjects is constructed over all theses; cosine similarity between subjects yields pairwise distances (distance = 1 − cosine similarity). For a thesis with n>1 subjects, Distance_t is the average of pairwise distances among its subjects. Theses with a single subject receive Distance_t = 0 (non-interdisciplinary by definition). - Key independent variables: Student gender (binary female vs male from Gender-Guesser; only “female” or “male” retained). Advisor gender (binary female vs male where available). Interaction term: female student × female advisor to test gender-pairing effects. Controls and design: Descriptive analyses for RQ1 examine time trends, domain differences, and institution-type distributions of IDR (share of multi-subject theses and average Distance_t), with years 2000–2004 excluded in certain trend fits to mitigate bias from the introduction of an “interdisciplinary” subject category in 2000. For RQ2 and RQ3, OLS regressions estimate Distance_t with year fixed effects, university fixed effects, and domain fixed effects. Average VIF = 2.03 (below threshold 5). Robustness checks: (1) Alternative gender inference via Genderize.io using probability ≥0.9; (2) Including “mostly male/female” names reclassified into binary categories; (3) Alternative interdisciplinarity metric using doc2vec-based semantic subject vectors (PV-DM, 100-dim, learning rate 0.025, window 3) derived from thesis titles aggregated per PQDT subject, computing cosine distances between subjects and re-deriving Distance_t; (4) Advisor fixed effects to compare male vs female students under the same advisor to absorb advisor-specific, time-invariant attributes.

- Growth of interdisciplinary theses: The share of theses assigned multiple subjects increased from 32.1% before 2000 to 59.5% during 2000–2016. The average subject-distance (interdisciplinarity indicator) also rose over time, indicating integration of increasingly cognitively distant subjects. - Domain and institution differences: Mathematical and physical sciences have the lowest interdisciplinarity levels; biological and health/medical sciences are higher. Universities with higher research intensity show higher interdisciplinarity (notably R2 vs Others). - Gender disparities (multivariate): After controlling for year, university, and domain fixed effects, male-authored theses exhibit significantly higher interdisciplinarity than female-authored theses. The average difference is about 0.012 (p<0.01), approximately 6% of the overall sample mean Distance_t. This pattern holds across periods, institution types, and most domains. - Temporal pattern of gender gap: The gender disparity is larger in 2000–2016 than in 1950–1999 (female student coefficient around −0.015 without interaction in 2000–2016 vs around −0.005 in 1950–1999; with interaction, roughly −0.013 vs −0.006, respectively). - Domain-specific gender differences: Significant male>female interdisciplinarity in behavioral sciences, biological sciences, and engineering; differences are smaller and not very significant in health/medical sciences and in mathematical/physical sciences. - Institution-type gradient: The male–female gap is more pronounced in R1 universities than in R2 or Other institutions. - Advisor gender main effect: Advisor gender alone is not significantly associated with students’ interdisciplinarity levels. - Student–advisor gender pairing: The interaction female student × female advisor is significantly negative overall and in 2000–2016, indicating that female students supervised by female advisors show a lower interdisciplinarity level relative to female students supervised by male advisors, thereby amplifying the gender disparity. The interaction is not significant in 1950–1999. - Robustness: Results persist when (a) using Genderize.io for gender inference, (b) including “mostly male/female” names into binary categories, (c) measuring interdisciplinarity via doc2vec-based subject distances, and (d) adding advisor fixed effects (male students still show higher interdisciplinarity than female students under the same advisor).

The study addresses RQ1 by showing a clear, decades-long increase in interdisciplinary engagement in doctoral theses across the five hard-science domains and across universities of differing research intensity, with IDR becoming dominant since the 1990s. Regarding RQ2, multivariate analyses reveal a persistent gender gap with male-authored theses exhibiting higher interdisciplinarity. Although simple unadjusted comparisons suggested slightly higher interdisciplinarity for female-authored theses, controlling for year, university, domain, and advisor gender reverses this, indicating that contextual factors account for the unadjusted pattern. The gender gap widens over time and is stronger in top-tier institutions (R1) and in specific domains (behavioral sciences, biological sciences, engineering), consistent with the idea that systemic barriers and risk aversion pressures contribute to women’s lower engagement in IDR during early career stages. For RQ3, advisor gender alone shows no main effect; however, the student–advisor gender pairing matters—female students mentored by female advisors exhibit particularly lower interdisciplinarity, intensifying the gender gap, especially post-2000. This could reflect heightened awareness of risks and systemic barriers influencing strategic choices, leading to more conservative topic selection among female students under female mentors. Collectively, the findings underscore that structural and mentoring-context factors shape ECSs’ IDR engagement and that gendered dynamics contribute to differential participation in IDR with implications for career trajectories.

This study contributes large-scale, long-term evidence on ECSs’ engagement in IDR using 675,135 U.S. doctoral theses (1950–2016) across five hard-science domains. It documents: (1) a strong rise in interdisciplinary doctoral work and increasing cognitive distance among combined subjects; (2) a persistent gender disparity—with male-authored theses more interdisciplinary than female-authored ones, particularly in recent decades, in R1 institutions, and in several domains; and (3) the importance of mentorship configuration, where female student–female advisor pairings are associated with lower interdisciplinarity for female students. Policy implications include addressing systemic gender biases in hiring, promotion, and funding; providing targeted support for female ECSs engaging in IDR; and supporting female faculty to enhance their capacity and resources, which may influence future cohorts. Future research should extend to additional fields (including humanities and social sciences), other outputs beyond theses (e.g., journal articles, conference papers), and incorporate richer measures of mentorship quality, advisor IDR participation, students’ attitudes and personalities, and direct measures of social gender via surveys.

- Mentorship measured only via the named primary advisor; informal mentoring and full committee influences are not captured. - Analyses limited to five hard-science domains; generalizability to humanities and social sciences is uncertain. - Doctoral theses may not fully reflect students’ broader research strategies or other outputs (e.g., journal articles, conference papers). - Potential omitted variables (e.g., students’ personalities, attitudes toward IDR, advisor mentoring styles, advisor preferences for IDR) may influence results beyond included fixed effects. - Gender inference is based on first names and thus approximates sex assigned at birth; it does not directly capture social gender or non-binary identities. - Advisor names are incomplete before 1980; advisor-related analyses necessarily focus on records with available advisor information.