Source-linked AI summary
Historical comparison of gender inequality in scientific careers across countries and disciplines
Junming Huang, Alexander J. Gates, Roberta Sinatra, Albert-Laszlo Barabasi
TL;DR
Evidence on gender differences in academic careers has been extensive but fragmented. Using longitudinal bibliometric data across countries and disciplines, the study finds that men and women have similar annual productivity and career impact when compared at equal output, while career length and dropout differences account for much of observed career gaps.
Problem
Evidence on gender differences in academic careers is extensive but fragmented, limiting a comprehensive longitudinal comparison across countries and disciplines.
Method
The study reconstructs publication careers for 1,523,002 authors across 13 disciplines and 83 countries using bibliometric publication data.
Results
With total productivity matched, female scientists receive an average of 1.9% more citations, while annual productivity and citation distributions are otherwise very similar.
Takeaways & Limitations
Matching career length brings productivity and impact gaps close to zero, indicating that career duration and dropout differences explain much of reported gender inequality.
Abstract
from arXiv · showhide
There is extensive, yet fragmented, evidence of gender differences in academia suggesting that women are under-represented in most scientific disciplines, publish fewer articles throughout a career, and their work acquires fewer citations. Here, we offer a comprehensive picture of longitudinal gender discrepancies in performance through a bibliometric analysis of academic careers by reconstructing the complete publication history of over 1.5 million gender-identified authors whose publishing career ended between 1955 and 2010, covering 83 countries and 13 disciplines. We find that, paradoxically, the increase of participation of women in science over the past 60 years was accompanied by an increase of gender differences in both productivity and impact. Most surprisingly though, we uncover two gender invariants, finding that men and women publish at a comparable annual rate and have equivalent career-wise impact for the same size body of work. Finally, we demonstrate that differences in dropout rates and career length explain a large portion of the reported career-wise differences in productivity and impact. This comprehensive picture of gender inequality in academia can help rephrase the conversation around the sustainability of women's careers in academia, with important consequences for institutions and policy makers.
Supplementary Materials
The supplementary materials provide figures and tables documenting gender imbalance, gender gaps in scientific publishing careers, and analyses controlling for career length.
- Gender imbalance: Figure 1 tracks gender imbalance since 1955 across active authors, disciplines, countries, and scientists’ temporal publication records.The figure distinguishes female and male authors and includes publication dates and 10-year citation counts.
- Gender gaps in publishing careers: Figure 2 quantifies statistically significant gender differences in total productivity across percentiles, disciplines, countries, affiliation ranks, and decades.The caption defines the gap as the relative difference between mean male and female authors’ productivity and reports p-values below 10^-4 unless otherwise stated.
- Controlling for career length: Figure 3 examines career-length effects by correlating gender gaps in career length with productivity gaps across disciplines and countries.The reported Pearson correlations are 0.80 across disciplines and 0.56 across countries.
- Controlling for career length: Figure 3 also describes a matching experiment pairing each female author with a male author of identical discipline, country, and career length.The figure compares average productivity under matched career lengths with population averages.
C D E
Controlling for age-dependent dropout changes the observed gender gaps in total productivity and impact. In a sample matched on total productivity, the total impact gap is eliminated.
- C: Controlling for age-dependent dropout affects the gender gap in total productivity.
- D: Controlling for age-dependent dropout affects the gender gap in impact.
- E: The total impact gap is eliminated in a sample matched on total productivity.
S1 Data sets · S1.1 Web of Science
The study uses Clarivate Analytics’ Web of Science Core Collection to reconstruct publication and citation histories, covering millions of authors, publications, authorships, and citation relationships. Publications are classified across 153 disciplines and regrouped into a coarser disciplinary partition for analysis.
- S1.1 Web of Science: The primary publication source is Clarivate Analytics’ Web of Science Core Collection, covering the Science Citation Index Expanded and Social Sciences Citation Index.
- S1.1 Web of Science: 1900–2016: the dataset includes articles, reviews, and letters while excluding other document types such as editorials and book reviews.
- S1.1 Web of Science: 7,863,861 authors contributed 101,961,318 authorships to 53,788,499 publications.
- S1.1 Web of Science: 694,439,758 citation relationships were obtained by extracting citation histories for all publications.
- S1.1 Web of Science: Each article is assigned to at least one discipline within a three-layer hierarchy containing 153 disciplines.Assignments primarily use journal information, while selected multidisciplinary journals provide article-specific categories.
- S1.1 Web of Science: The 153 leaf-layer disciplines were grouped into a coarser partition because the leaf layer was too fine-grained and the upper layers insufficiently detailed.The grouping follows the partition described in Section S2.7.
S1.2 Microsoft Academic Graph
The study used the Microsoft Academic Graph, a comprehensive index of journal and conference publications, downloaded through its authorized API in November 2017.
- S1.2 Microsoft Academic Graph: 77,642,549 publications were downloaded through Microsoft Research’s authorized API in November 2017.The Microsoft Academic Graph indexes scientific publications from both journals and conferences.
- S1.2 Microsoft Academic Graph: 88,223,538 authors contributed a total of 211,897,481 publications.
S1.3 DBLP
The study uses the DBLP Computer Science Bibliography, comprising 4,181,940 publications downloaded on June 5, 2018. It analyzes peer-reviewed publication types from 1970–2010, produced by 2,129,492 authors across 12,090,783 authorships.
- DBLP: 4,181,940 publications were drawn from the DBLP Computer Science Bibliography, downloaded June 5th, 2018.The database covers computer science journals and conference proceedings.
- DBLP: 1970–2010 publications included articles, review articles, proceedings, book chapters, and dissertations.Other document types, including webpages and notes, were excluded because they are generally not peer-reviewed.
- DBLP: 2,129,492 authors contributed 12,090,783 authorships to the publication corpus.
S2 Data pre-processing · S2.1 Identifying scientific careers
The study addresses name-disambiguation and database bias by comparing three bibliometric databases with independent author-identification procedures. Because disambiguation and gender inference are less reliable for Asian names, researchers from several Asian countries were excluded, while DBLP was regarded as especially reliable.
- S2.1 Identifying scientific careers: Three databases—WoS, MAG, and DBLP—were used to replicate the analysis with different name-disambiguation procedures.The replication was intended to test robustness to database bias and author-disambiguation errors.
- S2.1 Identifying scientific careers: WoS and MAG use proprietary algorithms, whereas DBLP assigns authors unique identifiers when manuscripts are submitted to registered Computer Science venues.MAG also uses online CVs, Wikipedia profiles, and ORCID career profiles in author-paper association.
- S2.1 Identifying scientific careers: DBLP was considered to have arguably the most reliable name disambiguation available in a bibliometric database.The database has also been used in peer-reviewed studies of scientific careers.
- S2.1 Identifying scientific careers: Name-disambiguation algorithms often reconstruct careers for authors with European names but struggle more with authors with Asian names.The paper also notes known difficulties inferring the gender of Asian names.
- S2.1 Identifying scientific careers: Researchers from China, the Democratic People’s Republic of Korea, Japan, Malaysia, the Republic of Korea, and Singapore were excluded.This conservative choice was motivated by name-disambiguation and gender-inference problems involving Asian names.
- S2.1 Identifying scientific careers: Replication across three databases with independent disambiguation methods led the authors to argue that errors from misappropriated or missing publications are negligible.The robustness claim rests on comparing databases that use different procedures.
S2.2 Career selection criteria
The analysis selects authors with sufficiently comprehensive publication careers by applying minimum career-history and publication-rate criteria and requiring their last article by December 31, 2010. The main conclusions remain unchanged under more stringent or modified selection filters.
- S2.2 Career selection criteria: Authors must have authored at least two papers.
- S2.2 Career selection criteria: Their publication careers must span more than one year, defined as 365 days.
- S2.2 Career selection criteria: Authors must average fewer than 20 papers per year and publish their last article on or before Dec 31st, 2010.
- S2.2 Career selection criteria: The main conclusions do not change under more stringent selection criteria or modified filters.
S2.3 Country label
Authors receive a single country label based on their most frequently occurring affiliation country, yielding labels for 1,876,950 authors. The country-specific analysis includes 83 countries after excluding those with fewer than 100 male or female authors.
- Country-label assignment: 1,876,950 authors receive a country label using their most frequently occurring affiliation country.The method retains each author’s known affiliation countries and selects the most frequent one.
- Country-label assignment: 3.12% of authors receive different labels under earliest-affiliation assignment, which does not qualitatively affect results.The alternative earliest-country method disagrees with the frequency-based approach for 58,576 authors.
- Country-specific analysis: 83 countries are included in country-specific analysis after excluding countries with fewer than 100 male or 100 female authors.The exclusion criterion is intended to ensure sufficiently reliable statistics.
S2.4 Affiliation rank
The study assigns each author the rank of their highest-ranked affiliated institute using Times Higher Education World University Rankings 2019 and publication affiliations. Of 1,876,950 authors with recorded affiliations, 1,296,995 were aligned to an institute rank.
- Ranking assignment: The study uses the Times Higher Education World University Rankings 2019, which indexes more than 1,250 universities.Ranking information was collected from this global university ranking.
- Ranking assignment: Authors are associated with universities by examining affiliations in their publications and disambiguating university-name variations.The procedure queried affiliation and ranking data with Google Maps to resolve spelling variants into unique university names.
- Ranking assignment: Each author receives the rank of the highest-ranked institute to which they were affiliated during their career.This career-level assignment uses the highest-ranked affiliation observed across the author’s publications.
- Coverage: 1,296,995 of 1,876,950 authors with at least one recorded affiliation were aligned to an institute rank.The aligned authors represent the subset for whom an institute rank was assigned.
S2.5 Gender assignment … S7 Tables and Figures
The study infers gender from country-specific names, aligns incomplete authorship metadata across databases, and analyzes career performance using normalized, self-citation-free measures across disciplines, countries, and career cohorts. Matching experiments and replications show that career length, dropout, and total productivity account for much of the observed gender gaps, while annual productivity and impact per equal body of work are similar or favor women.
- S2.5 Gender assignment: Gender labels were inferred from country-specific first names using Genderize.io, while researchers from several Asian countries and Brazil were excluded because of low assignment accuracy.The database maps names to binary gender labels using publicly available census statistics.
- S3 Indicators; S3.1 Characterizing the scientific career; S3.2 Characterizing the scientific population: Career indicators comprise total and annual productivity, career length, total normalized impact, academic age, and dropout, with gender gaps defined relative to male means.Career length spans first to last publication, while annual productivity is productivity divided by the days between those publications.