Source-linked AI summary
Quantitative Evaluation of Gender Bias in Astronomical Publications from Citation Counts
Neven Caplar, Sandro Tacchella, Simon Birrer
TL;DR
The paper asks whether first-author gender is associated with differences in astronomy citations, publishing frequency, and self-citation. Using a large five-journal dataset and controls for non-gender-specific paper properties, it finds fewer citations for female first-author papers, while the self-citation difference disappears after control.
Problem
The study addresses limited systematic evidence on how first-author gender relates to citations, publishing frequency, and self-citation in astronomy.
Method
The authors analyze over 200,000 astronomy publications from 1950–2015, assign first-author gender where feasible, and use random forests to control for non-gender-specific properties.
Results
10.4 ± 0.9% fewer citations were received by female-authored papers than expected for papers with the same non-gender-specific characteristics written by male authors.
Takeaways & Limitations
The measured citation difference persists after matching paper characteristics, whereas the observed self-citation effect disappears when non-gender-specific variables are controlled.
Takeaways & Limitations
Gender assignments are unavailable for many authors using initials and may contain estimation bias, potentially restricting the analyzed sample.
Abstract
from arXiv · showhide
We analyze the role of first (leading) author gender on the number of citations that a paper receives, on the publishing frequency and on the self-citing tendency. We consider a complete sample of over 200,000 publications from 1950 to 2015 from five major astronomy journals. We determine the gender of the first author for over 70% of all publications. The fraction of papers which have a female first author has increased from less than 5% in the 1960s to about 25% today. We find that the increase of the fraction of papers authored by females is slowest in the most prestigious journals such as Science and Nature. Furthermore, female authors write 19$\pm$7% fewer papers in seven years following their first paper than their male colleagues. At all times papers with male first authors receive more citations than papers with female first authors. This difference has been decreasing with time and amounts to $\sim$6% measured over the last 30 years. To account for the fact that the properties of female and male first author papers differ intrinsically, we use a random forest algorithm to control for the non-gender specific properties of these papers which include seniority of the first author, number of references, total number of authors, year of publication, publication journal, field of study and region of the first author's institution. We show that papers authored by females receive 10.4$\pm$0.9% fewer citations than what would be expected if the papers with the same non-gender specific properties were written by the male authors. Finally, we also find that female authors in our sample tend to self-cite more, but that this effect disappears when controlled for non-gender specific variables.
1. INTRODUCTION
The paper examines whether gender affects citation counts, publishing frequency, and self-citation among astronomy first authors. It defines gender bias as citation differences between matched papers with the same non-gender properties.
- Motivation: Gender inequality persists in science despite increasing numbers of women earning doctorates and evidence of gendered evaluation in peer review.Prior work also found increased female authorship after ecology journals adopted double-blind review.
- Prior evidence: Women remain underrepresented in prestigious publications and astronomy-related academic settings, including conferences and telescope-time proposals.Studies cited in the paper report fewer questions from women and lower proposal success for female principal investigators.
- Research aim: The study measures whether first-author gender is associated with citation counts in astronomy.The analysis labels citation differences between matched male- and female-first-author papers as gender bias.
- Research scope: The authors also investigate publishing frequency and self-citing tendency using a large, homogeneous astronomy dataset.The dataset is intended to control spurious citation dependencies and identify factors associated with differences between male- and female-first-author papers.
- Scope: The analysis refers to deduced first-author gender rather than gender assignments known with absolute certainty.The authors note possible estimation bias and restrict interpretation to the constraints of their analysis.
2. DATA
The study compiles and cleans astronomy publications from five journals, enriches them with paper and author characteristics, and determines first-author gender where feasible. The resulting dataset supports comparisons of publication patterns and gender-related differences.
- Data collection: 1950–2015 publications were collected from Astronomy & Astrophysics, Astrophysical Journal, MNRAS, Nature, and Science.The journals were selected for their established status and long historical records.
- Paper characteristics: ArXiv records supplied field classifications and, when available, TeX sources for measuring paper length, equations, floats, and mathematical expressions.TeXcount measurements below 500 words or affected by multiple source files were excluded.
- Field classification: Random-forest classification correctly assigned about 80% of papers to a field when arXiv classifications were unavailable.Misclassifications were often between similar subject categories.
- Institutional information: First-author institutions were assigned to North America, Europe, or Other after country identification from affiliation strings.Institutional information appeared in 85% of papers, and 97% of papers with affiliations were assigned uniquely to a country.
- Author information: Author identities were linked across initials, full names, and surname changes to recover publication histories and define seniority from the first first-author publication.Seniority is the number of years since that initial first-author publication; ambiguous cases received no seniority assignment.
- Sample characteristics: From the 1960s to 2015, annual publication output increased by more than one order of magnitude, while female first-author papers rose from less than 5% to about 25%.Gender was recognized for roughly 60–80% of papers from 1980 onward.
- Gender identification: Gender was assigned by matching recovered first names against three name databases.The sources included SexMachine, United States and United Kingdom official name data, and Gender API.
- Data cleaning: 208,577 ADS entries were downloaded, and 58,836 were removed, leaving a final dataset of 149,741 papers.Removed records included papers with zero citations or zero references and other incomplete or unsuitable entries.
3. CHARACTERISTICS OF PUBLICATIONS IN ASTRONOMY
The sample shows rising publication activity and persistent gender differences in representation, seniority, citations, and career publishing frequency. Female first-author representation reached about 25% in several major journals by 2015 but remained lower in Nature and Science.
- Global trends: About 10 to 60 references per paper: the average increased roughly 500% from the 1960s to today.Female first-author papers contained 7 ± 3% more references than male first-author papers from 1980 onward.
- Global trends: Before 1978, seniority estimates are incomplete; by 1978, seniority data exceeded 90% completeness.Average seniority was about seven years in 1980 for both female and male first-author papers.
- Global trends: ∼25%: female first-author papers reached this share in A&A, ApJ, and MNRAS by 2015, compared with ∼17% in Nature and Science.For 2000 papers, Science and Nature received twice the citations of A&A and MNRAS, while ApJ received about 40% more.
- Publishing frequency: Nearly twice as many papers per year: publication frequency for both genders increased by roughly this factor from the 1980s to 2015.For males, frequency rose from 0.6 to 0.9; for females, it rose from 0.4 to 0.8.
- Publishing frequency: 19 ± 7% fewer papers: female authors published this much over seven years after their first publication compared with males.The female-to-male publishing-frequency ratio declined to 0.81 seven years after first publication.
4. GENDER DIFFERENCE
Male first-author papers received more citations than female first-author papers, although the unadjusted difference narrowed over time. Matching seniority did not remove the gap, and random-forest adjustment estimated a larger intrinsic citation difference.
- Gender difference: 50% to 100% more citations: male authors exceeded female authors by this range in the early sample years, when uncertainties were large.The difference decreased over time and appeared to saturate at approximately 5%.
- Gender difference: 1.056 ± 0.010: the fitted gender-difference value using 1985 as the cutoff year.Using 2000 instead yielded 1.046 ± 0.009, without significantly changing the result.
- Controlling for seniority: About 5% more citations: male first-author papers received this excess at all career stages, with no significant trend across seniority bins.The difference was present even among authors at the lowest seniority.
- Correcting for non-gender properties: Random forest: the analysis predicted female-paper citations from male papers using seniority, references, authors, year, journal, field, and institutional region.Testing and training subsets from male papers were used to check that the model reproduced citation values without introducing a gender difference.
- Correcting for non-gender properties: 10% fewer citations: female-authored papers received approximately this much less than expected from their non-gender-specific properties.The fitted ratio was bff′ = 0.896 ± 0.009.
- Correcting for non-gender properties: 0.958 ± 0.008: replacing female papers’ measured citations with predicted values produced this male-to-female difference estimate.This consistency check compared the adjusted result with the uncorrected difference.
6. SELF-CITATION
Self-citation comparisons depend strongly on how self-citations are defined and on differences in publication history. Using a recent-paper definition, females self-cited more before adjustment, but the gender difference disappeared after controlling for paper properties.
- Defining self-citation: 56% more self-citations: a cited whole-JSTOR study reported this male excess, but the paper links it to differing publication counts and self-citation definitions.The authors note that authors with more published papers are expected to accumulate more self-citations.
- Defining self-citation: Immediately prior paper: a paper counted as a self-citation when it cited the same first author’s most recently published paper.First publications were excluded, and the definition did not depend on an author’s total self-citation count.
- Unadjusted comparison: 0.91 ± 0.02: the male-to-female self-citation ratio after 1985 before controlling for non-gender-specific properties.Males therefore self-cited less than females under this measure.
- Adjusted comparison: 1.015 ± 0.011: the adjusted female self-citation ratio after 1985, indicating little difference from the male-based expectation.Starting the fit in 2000 gave 1.004 ± 0.011.
7. DISCUSSION
The analysis is constrained by incomplete and potentially biased gender identification, especially for authors using initials or names from cultures underrepresented in the classification data. The authors therefore caution that the reported citation difference may be underestimated and is not definitively measured gender bias.
- Gender identification: Gender identification may miss authors who use initials, disproportionately excluding younger and less frequent female authors.This can leave the sample more representative of established female authors and may affect measured paper characteristics and citation expectations.
- Gender identification: Name changes after marriage may aggravate gender-recognition problems and lead to underestimation of seniority for some female authors.Misidentifying established authors as new arrivals could contribute to the observed citation difference.
- Gender identification: The classifier is less likely to recognize names from cultures outside Europe and North America, particularly as astronomy becomes more globalized.The authors report that gender bias was largely independent of host-institution region, so they do not expect a strong effect on the analysis.
- Scope of interpretation: The authors cannot claim to have actually measured gender bias because other parameters could also need to be matched.They present the results as their best effort using the available data and encourage further dataset development.
8. CONCLUSIONS
Using more than 200,000 astronomy papers and gender assignments for about 70% of first authors, the study examines participation, citation differences, controlled citation outcomes, and self-citation. Female participation rises over time, but female first-author papers receive fewer citations, while the observed self-citation difference disappears after controlling for paper characteristics.
- Study scope: Over 200,000 papers from five major astronomy journals were analyzed, with first-author gender assigned for about 70% of publications.Most unresolved cases involved authors who used initials throughout their publishing careers.
- Participation: About 25% of recent papers had female first authors, rising more slowly in Nature and Science, where the fraction was 17%.Female participation has increased consistently over time.
- Citations: 5.6 ± 1.0% more citations favored male first-author papers when measured from 1985 onward.The difference changed little when later starting years were used because it decreased very slowly or not at all.
- Controlled citation comparison: 10.4 ± 0.9% fewer citations were received by female-authored papers than expected for comparable papers written by male authors.The comparison controlled for differences between male and female first-author papers using machine-learning techniques.
- Self-citation: Female authors were 9 ± 2% more likely to cite their previous work, but controlled analysis found no significant intrinsic difference in self-citation propensity.The self-citation metric was the probability of citing the author's previous paper.
- Limitations and data access: The conclusions are limited by the inability to determine the gender of all paper authors, and the authors believe this likely strengthens the inferred gender-bias conclusion.The dataset is publicly available for further research.
A. FURTHER CHARACTERISTICS OF OUR SAMPLE
The appendix notes that the dataset is rich and can support investigation of additional characteristics beyond the main analyses.
- Further characteristics: The authors present further details because the dataset supports investigation of many additional aspects.This section introduces supplementary analysis of the sample's characteristics.
A.1. Leaving the field
The study measures field departure by tracking whether authors publish another first-author paper within eight years of their first publication. About 20–30% leave after the first year, but no male–female difference is detected within measurement errors.
- Method: Field departure was measured as the fraction of authors whose final first-author paper occurred in each year after their first publication.The analysis tracks subsequent publication behavior in the five journals considered.
- Method: The analysis followed authors for up to 8 years after their first publication.The data were organized by years after the first publication and seniority.
- Results: 20–30% of male and female authors published no further papers after their first-year paper, falling to about 5% in the second year and 5–10% thereafter.The data were stacked in 5-year intervals for the figure.
- Results: No difference between males and females, or across time from 1980 to the present, was found within measurement errors.Figure A.1 compares the fraction leaving the field as a function of years after first publication.
A.2. Length of papers
Male-authored papers tend to be about 5% shorter than female-authored papers, consistent with female papers containing more references. However, the analysis begins only in 2007 and may be affected by gender-identification bias.
- 5% shorter: Male-authored papers tend to be around 5% shorter than female-authored papers.The analysis covers papers from 2007 onward because accessible *.tex files begin that year.
- More references: Female-authored papers contain more references than male-authored papers across all sample years.
- Uncertain effect: The authors cannot distinguish a real length difference from bias against recognizing female authors who write shorter papers.
B. COMPLETENESS OF SENIORITY
Seniority is measured from an author's initial first-author publication, but early-time estimates may be incomplete because the sample begins in 1950. The analysis reaches at least 90% completeness from 1978.
- Seniority: Defined as the number of years since the author’s initial first-author publication.
- 90% completeness: By 1978, at least 90% of papers have first authors whose seniority can be determined from publications after 1965.
- Early estimates: Seniority may be inaccurate at early times because publications before 1950 are excluded.