Interdisciplinary research (IDR) is crucial for transformative science and national competitiveness. While IDR's importance is widely recognized, and many policies promote it, little is known about how early-career scientists (ECSs) engage in IDR, particularly regarding gender differences. ECSs face pressure to publish frequently, creating a tension between specializing for career security and pursuing potentially riskier, less immediately rewarding IDR. IDR presents various challenges, including increased effort, potential lower-quality outcomes, ambiguous identity, and less recognition. These risks are amplified for female ECSs, who already face underrepresentation, funding disparities, stereotypes, collaboration challenges, and childcare responsibilities. Women might be more risk-averse, leading to a preference for safer research strategies and less engagement in IDR. This study focuses on doctoral theses, crucial research outputs for junior scientists, to examine IDR engagement among ECSs, considering the influence of gender and mentorship. The study specifically investigates five scientific domains (behavioral, biological, engineering, health and medical, and mathematical and physical sciences) in the US, given its leading scientific role, strong IDR policies, and notable gender disparities in academia. The study aims to understand the prevalence of IDR among ECSs, the influence of gender on IDR engagement, and the correlation between advisor gender and students' IDR participation.

Existing literature highlights IDR's benefits for long-term career advancement and innovative breakthroughs. However, it also points to potential short-term drawbacks like lower productivity and challenges in securing recognition. Studies show mixed results regarding the relationship between IDR and scientific impact, with some suggesting a citation advantage for IDR while others report a potential citation penalty. The literature on gender and IDR is also inconsistent. Some suggest women's better ability to integrate diverse information and feminist arguments about less constrained thought processes may lead to more IDR. Others argue that cultural factors, stereotypes, a masculine STEM culture, and systematic barriers within the research system can hinder women's participation in IDR. These barriers include unequal access to funding, stereotype threats, challenges in collaboration and leadership, and difficulties balancing work and family responsibilities. A few studies have explored the relationship between mentorship and IDR, indicating that advisors' disciplinary backgrounds influence students' interdisciplinarity. However, the role of advisor gender and the interaction between student and advisor gender remains largely unexplored. Measuring IDR is also complex. Qualitative methods lack objectivity and scalability, while citation-based quantitative methods rely on unavailable data for doctoral theses. The study proposes using subject co-occurrence matrices from the ProQuest database as a suitable measure.

This study used data from the ProQuest Dissertations & Theses Database (PQDT), focusing on 675,135 doctoral theses from US universities (1950-2016) across five scientific domains. Since PQDT doesn't provide gender information, the Gender-Guesser package predicted student and advisor gender based on first names. Only theses with definitively predicted genders were included. Interdisciplinarity was measured by calculating the average distance between subjects assigned to each thesis using a co-occurrence matrix of subjects from PQDT. A larger distance indicates higher interdisciplinarity. To address research questions, descriptive analyses examined the prevalence of IDR over time, across domains, and among universities with varying research intensity. Multivariate regression models investigated gender differences in interdisciplinarity, controlling for year, university, domain, and advisor gender. An interaction term between student and advisor gender explored the potential amplifying effect of female advisors on gender disparities. Four robustness checks were performed. First, gender was re-predicted using Genderize.io. Second, 'mostly male' and 'mostly female' categories were reclassified. Third, interdisciplinarity was assessed using doc2vec on thesis titles. Finally, advisor fixed effects were added to the regression model.

The study found a significant increase in interdisciplinary doctoral theses over time, across all domains and university types. The average distance between subjects assigned to theses also increased, indicating greater integration of cognitively distant knowledge. However, persistent gender disparities emerged, with male-authored theses demonstrating significantly higher interdisciplinarity than female-authored theses (approximately 0.012 difference, about 6% of the mean). This difference was consistent across time periods, university types, and most scientific domains, though the magnitude varied. Advisor gender alone was not a significant predictor of interdisciplinarity. However, a significant negative interaction term between student and advisor gender was found. This means that the gender difference in interdisciplinarity was more pronounced when female students were supervised by female advisors, suggesting that female mentorship may exacerbate existing gender gaps in IDR. Robustness checks using alternative gender prediction methods, different interdisciplinarity measures, and the inclusion of advisor fixed effects consistently supported these findings.

The findings address the research questions by demonstrating a growing trend in IDR but also highlighting persistent and pervasive gender disparities in IDR engagement among ECSs. The gender gap is not merely a matter of overall numbers but also of the type of interdisciplinary work undertaken. Males appear to be undertaking more cognitively distant interdisciplinary work than females. The lack of a significant direct effect of advisor gender suggests that broader systemic issues, rather than individual advisor biases, are more influential. The interaction effect, however, highlights a concerning potential interaction effect where female mentorship may inadvertently discourage female students from pursuing more ambitious interdisciplinary projects. These results are consistent with the literature on gender bias in science and the challenges faced by women in STEM. The lower level of interdisciplinarity among female ECSs could have long-term consequences for their career advancement and impact.

This study reveals a persistent gender gap in interdisciplinary research among early-career scientists, despite a growing overall trend toward interdisciplinarity. The significant interaction between student and advisor gender suggests that providing support to female faculty is crucial not only for their individual success but also for fostering future generations of female scientists engaged in IDR. Further research should investigate informal mentorship, explore other scientific domains, and consider additional factors influencing ECS preferences for IDR. Addressing systemic gender biases and providing targeted support are critical for closing the gender gap in interdisciplinary research.

This study has several limitations. It focuses on five hard science domains and may not generalize to other fields. The reliance on first names for gender prediction may introduce some inaccuracies, though robustness checks mitigated this. The study uses doctoral theses as a proxy for IDR, which might not fully represent all research activities. The relatively low adjusted R-squared values in the regression models suggest that other factors influence interdisciplinarity levels beyond those considered. Further research should explore informal mentorship and the broader context of research choices made by early-career scientists.