Source-linked AI summary
The extent and drivers of gender imbalance in neuroscience reference lists
Jordan D. Dworkin, Kristin A. Linn, Erin G. Teich, Perry Zurn, Russell T. Shinohara, Danielle S. Bassett
TL;DR
The paper asks whether neuroscience citation practices exhibit gender bias and what drives that bias, addressing citation behavior rather than citation counts alone. Using articles from five top neuroscience journals, it finds overcitation of men and undercitation of women, driven largely by men’s citation practices, including when men’s social networks are representative.
Problem
The paper examines whether neuroscience reference lists are gender-biased, an important question because citation engagement can affect scholars’ perceived centrality and downstream career outcomes.
Method
The study analyzes authors and reference lists from articles published in five top neuroscience journals since 1995, using gender estimates, citation links, self-citation removal, and co-authorship networks.
Results
Neuroscience reference lists overcite men and undercite women relative to expectation, with men-led papers overciting man/man papers by 8% and underciting woman/woman papers by 23%.
Takeaways & Limitations
Men tend to overcite men even when their social networks are representative, making citation practices a target for addressing neuroscience’s gender inequities.
Takeaways & Limitations
The study’s binary gender assignments do not accommodate intersex, transgender, or non-binary identities, and it does not examine intersecting biases such as race or ethnicity.
Abstract
from arXiv · showhide
Like many scientific disciplines, neuroscience has increasingly attempted to confront pervasive gender imbalances within the field. While much of the conversation has centered around publishing and conference participation, recent research in other fields has called attention to the prevalence of gender bias in citation practices. Because of the downstream effects that citations can have on visibility and career advancement, understanding and eliminating gender bias in citation practices is vital for addressing inequity in a scientific community. In this study, we sought to determine whether there is evidence of gender bias in the citation practices of neuroscientists. Using data from five top neuroscience journals, we find that reference lists tend to include more papers with men as first and last author than would be expected if gender were not a factor in referencing. Importantly, we show that this overcitation of men and undercitation of women is driven largely by the citation practices of men, and is increasing over time as the field becomes more diverse. We develop a co-authorship network to assess homophily in researchers' social networks, and we find that men tend to overcite men even when their social networks are representative. We discuss possible mechanisms and consider how individual researchers might address these findings in their own practices.
Introduction
Neuroscience is confronting persistent gender inequities, but citation practices remain understudied despite evidence from other fields that women-led research is cited less often. This study therefore examines whether neuroscience reference lists show gender bias and what drives it.
- Background: Persistent gender imbalances affect women’s participation and academic success across scientific fields.Prior research identifies inequalities in compensation, grant funding, and other measures of academic inclusion.
- Neuroscience context: Neuroscience has increased attention to gender and diversity through inclusion initiatives, society discussions, and efforts to balance journal editors and reviewers.
- Citation bias: Women-led research receives fewer citations than comparable men-led research in astronomy, international relations, and political science.Related work also finds broad undercitation of marginalized groups in communications and philosophy.
- Possible mechanism: A proposed Matilda effect suggests that men’s contributions are perceived as more central and therefore sought out and evaluated more highly.Women-led work could consequently remain underdiscussed and perceived as marginal to men-led work.
- Study rationale: Citation behavior matters because inequitable engagement can affect scholars’ perceived centrality and produce harmful downstream effects.Unlike keynote representation, citation-list equity can be pursued by researchers during paper writing.
- Study aim: The study investigates gender bias in neuroscience citations by relating authors’ gender to the gender composition of their reference lists.It examines articles from five top neuroscience journals and distinguishes direct citation behavior from passive citation-count consequences.
Results
Across five top neuroscience journals, women’s authorship increased, yet reference lists overrepresented MM papers and underrepresented women-led papers. This imbalance was driven largely by MM citing teams, increased over time, and persisted after accounting for relevant paper characteristics and local co-authorship networks.
- Results: Gender assignment was probabilistic rather than a direct measure of authors’ sex or self-identified gender.The analysis used the probability of an author receiving a man or woman label from available sources.
- Results: The proportion of articles with a woman as first or last author increased from 36% in 1995 to 50% in 2018.Across the five journals, the overall increase was roughly 0.60% per year.
- Results: 61.7% of citations went to MM papers versus an expected 55.3%, while WM, MW, and WW papers were cited less than expected.MM papers were cited 11.6% more than expected, whereas WM, MW, and WW papers were cited 10.1%, 12.5%, and 30.2% less than expected, respectively.
- Results: After accounting for publication date, journal, author count, review status, and author seniority, MM papers remained 5.2% overcited and WW papers 13.9% undercited.The relevant-characteristics model compared observed citations with expected proportions among similar cited papers.
- Results: MM citing teams drove most of the imbalance: they overcited MM papers by 8.0%, whereas W∪W teams did so by only 2.5%.Differences between MM and W∪W reference lists were significant, and women’s leadership on citing teams corresponded to more representative citation proportions.
- Results: MM overcitation increased by 0.54 percentage points per year in MM reference lists and 0.29 percentage points per year in W∪W reference lists.Observed MM citation proportions remained relatively stable while expected proportions decreased as the field became more diverse.
- Results: The median MM team’s co-authorship network contained 8.2% more men than the field base rate, compared with 3.8% for the median WW team.Mixed-gender teams fell between these values, with 6.4% and 5.7% more men for WM and MW teams, respectively.
Discussion
Neuroscience reference lists show gender imbalance largely driven by men’s citation practices, even when men’s social networks are representative. The discussion considers possible mechanisms, practical responses, unresolved citation ethics, and limits on generalization.
- Reference lists overcite papers with men as first and last authors, and this imbalance is driven largely by men’s citation practices.The authors frame citation behavior as an individual-level contributor to persistent field inequities.
- Men tend to overcite other men even when their social networks represent the field, indicating that network homophily alone does not account for the citation imbalance.The authors describe homophily as a possible contributor but report a remaining difference under representative networks.
- Men’s citation behavior may reflect explicit or implicit evaluative bias toward women-led work, although the paper presents this as one possible mechanism.The proposed explanation is consistent with reported evaluative bias in graduate admissions, hiring, funding, and promotion.
- Researchers can use responsible-citation guidance, diversity-measurement tools, and organizational resources to create more representative reference lists.The discussion also highlights editorial transparency, reviewer guidance, and graduate education as possible interventions.
- The ethics of citation distribution remain unsettled because equality-, equity-, and difference-based models imply different standards for allocating citations.The paper notes that distributive models emphasize parity, while difference models may support reparative approaches.
- The findings warrant attention because present citation patterns can affect the future of neuroscience, but generalization is limited by the focus on five top journals.Institutional prestige may also confound the observed relationship between gender and citation behavior.
- The study’s gender-determination methods use binary man/woman assignments and do not accommodate intersex, transgender, or non-binary identities.Future work could examine intersecting biases and use self-identification or pronoun-based methods.
- Future research could analyze collaboration-network composition and longitudinal within-author citation behavior to clarify mechanisms and inform individualized recommendations.These extensions could examine how co-authors and network characteristics relate to changes in citation practices.
Methods
The study combines bibliometric data, author-identity processing, gender assignment, self-citation removal, and network-based statistical analyses to examine gendered citation behavior in neuroscience.
- Data collection: Articles from five neuroscience journals were downloaded from Web of Science, matched through DOIs, and filtered to include research, review, and proceedings papers with DOIs.The journal selection used the five highest Eigenfactor scores among neuroscience journals.
- Author identification: Author names were disambiguated across papers by matching name variants and initials, while unresolved or conflicting matches were left unassigned.Incorrect disambiguation could omit papers, retain some self-citations, or remove valid citations, although sensitivity analyses suggested the latter effects were small.
- Gender determination: Gender assignments used probabilistic name-based sources, with labels assigned when the probability of belonging to a man or woman exceeded 0.70.The method combined Social Security Administration and Gender API information; automated assignments were accurate for 0.96 of a random sample of 200 authors.
- Citation processing: Primary citation analyses removed self-citations defined by overlap between cited and citing papers’ first or last authors, with broader definitions examined in supplementary analyses.This restrictive definition was chosen because the cited author’s gender was necessarily linked to the citing author’s gender in those cases.
- Statistical analysis: Observed frequencies of MM, WM, MW, and WW cited papers were compared with expected frequencies from a generalized additive model accounting for publication timing and other characteristics.Confidence intervals used article-level bootstrapping, while p-values used a randomization-based null model.
- Network analysis: A temporal co-authorship network and weighted median quantile regressions tested whether local social-network composition was associated with citation behavior.Network measures compared the gender composition of each paper’s local author and paper neighborhoods with the overall network, and weights reflected candidate citations per reference list.
Supplementary Information for “The extent and drivers of gender imbalance in neuroscience reference lists”
The supplementary analyses validate the gender-assignment procedure, test missing-data and weighting choices, and examine self-citation, citation-distribution, null-model, and network-related robustness. Across these checks, the reported citation-imbalance patterns were assessed under alternative assumptions and subgroup definitions.
- Validation: Automated gender assignments were validated against manual assessments in samples of 200 authors and 100 papers.The supplementary tables report author-level and paper-level accuracy assessments.
- Missing data: Primary analyses used the 88% of papers with reliably assigned first- and last-author genders, while imputation and bootstrapping assessed the full dataset.Missing genders were modeled using publication year, author count, seniority, journal, and review status.
- Weighting: Article weighting depended on the analysis: citation totals weighted papers by candidate citations, whereas network analyses used active weighting schemes.An unweighted article-level analysis was also reported as a sensitivity check.
- Self-citation: MM and WM teams self-cited more relative to reference-list length than MW and WW teams, but rates were relatively similar when scaled by potential self-citations.Self-citations were excluded from primary analyses to isolate more comparable citation behavior.
- Null model: The graph-preserving null model randomized cited-paper gender categories while retaining citation-network structure and other paper characteristics.This framework tested whether observed citation patterns could be explained by graph structure and related paper attributes.
- Citation distribution: The citation-imbalance pattern was similar above and below the median citation count, despite the upper half accounting for three-quarters of citations.Relative to expectation, below-median MM/WM/MW/WW papers were +4.8%, -2.2%, -6.6%, and -14.6%; above-median papers were +5.3%, -7.6%, -4.1%, and -13.8%.
- Research networks: More productive teams showed subtle network-composition differences, while their higher- and lower-productivity citation patterns were similar.Network overrepresentation varied slightly with team productivity and last-author gender.