PaperlayerView exact revision on arXiv

Source-linked AI summary

Men Set Their Own Cites High: Gender and Self-citation across Fields and over Time

Molly M. King, Carl T. Bergstrom, Shelley J. Correll, Jennifer Jacquet, Jevin D. West

arXiv:1607.00376v2physics.soc-phcs.DL

TL;DR

The paper examines how common self-citation is and whether it varies by gender. Using author-to-author citation analysis of JSTOR publications, it finds that self-citations comprise an important share of references and that men self-cite more than women, while noting limits in author disambiguation and name-change tracking.

  • Problem

    The study asks how common self-citation is in scholarly publication and whether self-citation patterns differ by gender.

  • Method

    The authors analyze 1.5 million papers and author-to-author citations in JSTOR, constructing citation networks and counting each author-to-author combination.

  • Results

    9.4 percent of all citations are self-citations; men’s authorships self-cite at 31.4 percent versus 21.2 percent for women, and men self-cited at 1.7 times women’s rate in the last two decades.

  • Takeaways & Limitations

    Self-citation can make up an important fraction of citations to authors’ work, with substantial variation across fields and a persistent gender difference in self-citation rates.

  • Takeaways & Limitations

    The analysis cannot determine whether the self-citation gap causes or results from gender imbalances or productivity differences, and name changes may cause slight downward bias in women’s estimated self-citation rates.

Abstract

from arXiv · show

How common is self-citation in scholarly publication, and does the practice vary by gender? Using novel methods and a data set of 1.5 million research papers in the scholarly database JSTOR published between 1779 and 2011, the authors find that nearly 10 percent of references are self-citations by a paper's authors. The findings also show that between 1779 and 2011, men cited their own papers 56 percent more than did women. In the last two decades of data, men self-cited 70 percent more than women. Women are also more than 10 percentage points more likely than men to not cite their own previous work at all. While these patterns could result from differences in the number of papers that men and women authors have published rather than gender-specific patterns of self-citation behavior, this gender gap in self-citation rates has remained stable over the last 50 years, despite increased representation of women in academia. The authors break down self-citation patterns by academic field and number of authors and comment on potential mechanisms behind these observations. These findings have important implications for scholarly visibility and cumulative advantage in academic careers.

Self-citations: An Author-to-author Approach

The paper defines self-citation through author-to-author links and addresses name-matching and authorship-counting challenges. Its approach preserves broad longitudinal and cross-field coverage while limiting individual-level controls.

  • Identification: Because JSTOR authors are not disambiguated, citations to the same full name are treated as self-citations under the assumption that most such matches identify the same person.The authors acknowledge that identical names can belong to different individuals.
  • Scope and trade-offs: The analysis cannot control for career stage or individual productivity because authors cannot be tracked over time.Differences in the number of papers available to cite could therefore generate a gender gap even if self-citation behavior were identical.
  • Operationalization: A citation between papers with four and three authors produces 12 author-to-author pairs, with only overlapping author names counted as self-citations.This author-to-author approach can yield a lower self-citation fraction than paper-level counting.
  • Scope and trade-offs: The study prioritizes longitudinal and cross-disciplinary analysis over full author disambiguation, which was feasible only for a very small hand-coded sample.This design choice enables analysis across many fields and an unprecedented time span.
  • Operationalization: The self-citation rate is defined as mean self-citations per authorship within a specified group.The rate divides total self-citations by total authorships, including comparisons across years and genders.
  • Operationalization: Self-citation is identified by matching cited and citing authors’ full first and last names, while disregarding middle initials.The citation year is assigned according to the citing paper’s publication year.

The JSTOR “Network Data Set”

The study constructs large JSTOR network and analytic data sets, assigns fields through citation-network structure, and estimates self-citation rates with bootstrap confidence intervals. Name-based gender assignment yields a substantial but incomplete analytic sample.

  • The JSTOR “Network Data Set”: The JSTOR network data set combines citation and full-text publication data to calculate self-citation rates, assign author gender, and construct a citation network.The collection contains over 8 million documents, including 1.8 million citation-linked research articles.
  • The JSTOR “Network Data Set”: Approximately 1.5 million papers published from 1779 through 2011 form the analytical data set.Earlier periods are sometimes omitted when sample sizes are too small for meaningful conclusions.
  • Field classification: Hierarchical classification uses citation relations to identify nested fields, subfields, and finer research-topic partitions.The method applies the hierarchical map equation to citation-network flows.
  • Gender assignment: Gender is assigned from first names using Social Security Administration records and restricted to names among the 1,000 most popular in at least one year from 1879 to 2012.Authors with gender-ambiguous names or only first initials are excluded.
  • Gender assignment: 2.8 million of 3.6 million authorships have full first names, and gender is confidently assigned to 73.3 percent of those.The final analytic data set includes papers for which the gender of at least one author is known.
  • Uncertainty estimation: Bootstrap methods estimate confidence intervals by resampling papers within the relevant year, field, or year-and-field sample.Ratios are calculated from self-citation rates in each bootstrap sample.

How Common Is Self-citation?

Self-citation is common across scholarly publication, accounting for about 1 in 10 references overall, but its prevalence varies across fields, authorship structures, and gender. Men self-cite more than women on average, and this difference appears across the distribution of self-citation counts rather than only in extreme cases.

  • 9.4 percent of 8.2 million references were self-citations, or about 1 in 10 references across all fields and years.
  • Molecular biology had the highest self-citation rate per reference, while classical studies had the lowest.
  • A paper can accumulate many authorship-to-authorship self-citations by citing a few papers with many overlapping authors, as in a 175-author Science report.That paper cited only four previous papers written by its authors, but the overlapping authorship links produced 220 self-citations.
  • A sole-authored paper can also have many self-citations, as illustrated by an American Economic Review article with 70 self-citations among 130 references.The authors describe this as tracing the author’s career rather than excessive self-citation.
  • 21 percent of references in papers containing self-citations were self-citations.
  • Women’s authorships were more likely than men’s to contain zero self-citations, with 78.8 percent versus 68.6 percent reporting none.Women cited themselves one or more times less often than men, while differences were visible across common self-citation counts below five.

Self-citation Rates by Field

Men’s self-citation rates exceed women’s across fields, although rates vary widely among fields and subfields. Within some subfields, women self-cite more than men, and field gender composition shows no obvious relationship with relative self-citation rates.

  • Self-citation behavior varies widely across fields and subfields despite the overall male advantage.
  • Men’s self-citation rates are higher than women’s in every major academic field examined.
  • History and classical studies have the lowest women’s self-citation rates per authorship, while ecology and evolution, sociology, and molecular and cell biology have the highest.
  • There is no obvious relationship between the proportion of women in a field and the relative self-citation rates of women and men.
  • Some subfields have a self-citation ratio above 1.0, indicating that women self-cite more than men there.

Self-citation Rates by Field over Time

Within fields, men consistently self-cite more than women over time. The analysis focuses on the most recent 40 years because smaller field samples make earlier ratio confidence intervals more sensitive to sample size.

  • Men consistently self-cite more than women within each field over time.
  • Men’s self-citation rates are generally higher than women’s across time in the 16 largest fields.
  • Confidence intervals overlap more in fields and periods with fewer papers, including mathematics and years before 1970.

Self-citation Rates by Size of Author Team

The authors examine whether author-team size contributes to gender differences in self-citation by comparing mean self-citations per authorship across papers with 1 to 20 authors. They find no interaction with gender.

  • The analysis compares mean self-citations per authorship across papers with 1 to 20 authors.
  • There were no interactions with gender in the relationship between author-team size and self-citation.
  • Figure 10 reports mean self-citations per authorship by number of authors, truncating results at 20 authors because larger samples become excessively noisy.
  • A value of 1.0 on the vertical axis means that each author cites one previous paper on average, with shaded intervals showing 95 percent bootstrap confidence limits.

Discussion

Using a large JSTOR corpus, the study identifies self-citation as a substantial component of scholarly citation and finds a persistent gender difference. Men self-cite more often than women, with substantial variation across fields and no obvious relationship to field gender composition.

  • The study examines 1.5 million papers, 39.4 million author-to-author citations, and more than 1 million self-citations in JSTOR.
  • About 9.4 percent of all citations are self-citations referencing previous work by one or more current-paper authors.
  • Men are more than 10 percentage points more likely than women to self-cite: 31.4 percent versus 21.2 percent of authorships.
  • In the last two decades of the data, men cited themselves at 1.7 times the rate of women.
  • Self-citation varies widely across fields and subfields, with no obvious relationship between field gender composition and relative self-citation rates.

Potential Mechanisms

The paper considers behavioral and structural explanations for the gender gap in self-citation, while noting that its data cannot conclusively distinguish among them. It also reports that the gap persists over time and may affect cumulative scholarly advantage and evaluation.

  • Potential Mechanisms: The authors caution that their JSTOR data lack variables needed to determine the causes of the observed patterns.They also note that encouraging women to self-cite more may have unintended consequences because of backlash against women’s self-promotion.
  • Potential Mechanisms: Five proposed mechanisms include self-assessment, social penalties for self-promotion, specialization, publication volume, and paper type.The authors regard these mechanisms as potentially complementary rather than mutually exclusive.
  • Potential Mechanisms: Men may publish more papers, particularly earlier in their careers, creating more available targets for self-citation.Higher productivity can also generate more highly cited papers.
  • Potential Mechanisms: A productivity difference alone could produce a self-citation gap even when men and women self-cite at identical rates.The paper’s simplified example gives men three papers and women two, producing different average self-citation rates.
  • Potential Mechanisms: The gender self-citation gap did not move toward equality over the past 50 years despite increased representation of women in senior academic positions.This observation does not establish whether the gap is caused by productivity differences or self-citation behavior.
  • Potential Mechanisms: Differences in paper types may matter because women are underrepresented in single-authored papers and prestigious first- or last-author positions.The authors suggest these papers may be more likely to lie within scholars’ core research areas and be self-cited.

Appendix A

Appendix A tests the JSTOR findings in a separately constructed SSRN data set. The alternative data show a similarly sized gender gap, but do not indicate greater self-citation among men and women with equal paper counts.

  • Appendix A: The SSRN data set contains 426,412 papers, including preprints, from 99,465 authors.Authors were carefully disambiguated so distinct people with the same name could be separated.
  • Appendix A: Men accounted for 73 percent of identifiable authors, while women accounted for 27 percent.Gender assignment followed the same procedure used for the JSTOR data.
  • Appendix A: The SSRN data showed a gender self-citation gap equivalent in magnitude to the JSTOR gap.Men represented 73 percent of authorships but 87 percent of self-citations.
  • Appendix A: Men with n papers did not appear to self-cite appreciably more than women with n papers in SSRN.The gap could instead reflect differences in citation targets or selective participation in the archive.
  • Appendix A: SSRN differs from JSTOR because it is smaller, covers fewer disciplines, and is a non-peer-reviewed archive used voluntarily by only some authors.These differences may make database selection relevant to interpreting the results.

Appendix B

Appendix B examines whether field composition produces threshold or continuous effects in self-citation. The analysis finds no clear threshold and only a slight, inconclusive association between women’s self-citation and women’s representation in a field.

  • Appendix B: No cutoff produced an apparent threshold effect in self-citation at the proposed 15 percent level.The 15 percent point falls between law and economics in the authorship data.
  • Appendix B: Figures B1 and B2 order fields from the highest to lowest proportion of women authorships.This ordering is used as a proxy for women’s representation across fields.
  • Appendix B: Figure B3 reports self-citation rates separately for men and women authorships by the gender composition of fields.The rate is normalized by references per authorship to account for differential citation practices between fields.
  • Appendix B: Women’s self-citation showed a slight upward linear association with the proportion of women in a field, but the available data were too sparse to establish a clear effect.The figure focuses on the top fields.

Supplemental Data

The supplemental data documentation explains how JSTOR data support the paper’s field classification and gender-authorship analyses. It also identifies the authors and their institutional and research backgrounds.

  • Supplemental Data: The JSTOR data were provided under a data-use license with the University of Washington, with openness limited by that agreement.The authors state that they aim to make the data as open as permitted.
  • Supplemental Data: The study uses JSTOR to hierarchically classify roughly 2 million papers into fields and subfields and to determine gender patterns of authorship.The categorization follows the authors’ previous PLoS ONE work.
  • Supplemental Data: Molly M. King’s research concerns knowledge ownership and how class, gender, and race structure inequality.
  • Supplemental Data: Carl T. Bergstrom studies information in biological and social systems, including communication, complex networks, and citation-related systems.
  • Supplemental Data: Shelley J. Correll studies how status beliefs, organizational policies, and legal mandates shape workplace gender inequalities.
  • Supplemental Data: Jennifer Jacquet studies large-scale cooperation dilemmas, including overfishing, climate change, and the illegal wildlife trade.
Loading 1607.00376v2…