Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Scientometrics (2020) 125:1–28

https://doi.org/10.1007/s11192-020-03615-1

Gender, seniority, and self‑citation practices in political


science

Michelle L. Dion1   · Sara McLaughlin Mitchell2   · Jane L. Sumner3 

Received: 29 July 2019 / Published online: 20 July 2020


© Akadémiai Kiadó, Budapest, Hungary 2020

Abstract
Many studies in political science and other disciplines show that published research by
women is cited less often than research by male peers in the same discipline. While pre-
vious studies have suggested that self-citation practices may explain the gender citation
gap in political science, few studies have evaluated whether men and women self-cite at
different rates. Our article examines the relationship between author gender, author expe-
rience and seniority, and authors’ decisions to include self-citations using a new dataset
that includes all articles published in 22 political science journals between 2007 and 2016.
Contrary to our expectations, we fail to reject the null hypothesis that men are more likely
cite their previous work than women, whether writing alone or co-authoring with others
of the same sex. Mixed gender author teams are significantly less likely to self-cite. We
also observe lower rates of self-citation in general field journals and Comparative/Interna-
tional Relations subfield journals. The results imply that the relationship between gender
and self-citation depends on several factors such as collaboration and the typical seniority
and experience of authors on the team.

Keywords  Citations · Gender · Political science · Sociology of science

Mathematical Subject Classification  91C99 · 91F10 · 91D99

JEL Classification  A11 · A14 · B54 · J44

Previous versions of this paper were presented at the 2019 European Conference on Politics and
Gender (ECPG), Amsterdam, Netherlands and the 2019 Midwest Political Science Association
(MPSA) Annual Conference, Chicago, IL. We are grateful to Kathy Dolan and Kim Fridkin for
comments.

* Michelle L. Dion
dionm@mcmaster.ca
1
Department of Political Science, McMaster University, Kenneth Taylor Hall 527,1280 Main Street
West, Hamilton, ON L8S 4M4, Canada
2
Department of Political Science, University of Iowa, 341 Schaeffer Hall, Iowa City, IA 52242,
USA
3
Department of Political Science, University of Minnesota, 1472 Social Sciences Building, 267
19th Ave S, Minneapolis, MN 55455, USA

13
Vol.:(0123456789)
2 Scientometrics (2020) 125:1–28

Many studies in political science and other disciplines show that published research by
women is cited less often than research by male peers in the same discipline.1 Achiev-
ing a critical mass of female scholars in research areas helps to improve the overall num-
ber of citations to women’s work, but the gender gap remains, as male authors cite sig-
nificantly fewer works by women, even in fields where women constitute the majority of
scholars (Dion et al 2018). Beyond the opportunities for citing work by women (based on
the density of female scholars), questions also emerge about whether gender citation gaps
are driven by differences in self-citation practices. Self-citation involves the authors of an
article citing work in the paper’s bibliography that was written by the authors. For solo-
authored work, this involves a citation to the single author him/herself. For coauthored
work, this involves citation to at least one of the authors’ previous works. In other words,
men might have a higher willingness to cite their own work than women, which could
account for observed citation gaps (Hutson 2006; Maliniak et al 2013; Ghiasi et al 2016).
This is important because Fowler and Aksnes’ (2007) suggest that self-citations accrue
new citations over time. If men cite their own work more frequently than women early in
their careers, these early cites help to increase the number of later career citations.
In some ways, the discussion of self-citation practices reflects a broader discussion in
society about whether gender biases can be addressed by women “leaning in” or by chang-
ing structural conditions that create the biases. Concluding that women have fewer citations
because they cite their own research less often than men is a version of the “lean in” men-
tality. If women just cited themselves more, the problem would be fixed! Yet the research
on “Matthew effects” in academia shows this solution is likely to be inadequate because
work by men can still be viewed as the most important or central research to cite, even
in fields that have higher densities of female scholars (Dion et  al 2018).2 Implicit biases
would not be reduced by a “lean in” strategy of women being more careful about citing
their own research. It also ignores the fact that co-authorship emerges in academic envi-
ronments that are hierarchically constructed. If junior women coauthor with senior men,
for example, they may be reluctant to advocate for citations of their work.3 Authors from
similar academic generations may be more confident in adding citations to their research
in coauthored papers. In short, sociological factors influence how publishing teams emerge
and how they construct bibliographies and we need to understand these structural condi-
tions more fully.4
In this paper, we examine whether published articles include one or more self-citations
in 8715 articles in 22 political science journals between 2007 and 2016. We answer the
question, are women authors less likely to cite their previous work than men, and do cer-
tain other characteristics of authors, such as experience and seniority, reduce these differ-
ences? In addition to presenting descriptive data across journals, we estimate multivariate
self-citation models that compare solo-authored and multi-authored works and we examine

1
 In political science, see Maliniak et  al (2013), Mitchell et  al (2013), Roberts et  al (2018), Dion et  al
(2018); in other disciplines, see Ferber (1988). Håkanson (2005), Leahey et al (2008), Aksnes et al (2011),
Ferber and Brun (2011), Cameron et al (2016), and Beaudry and Lariviere (Beaudry and Larivière 2016).
2
  Interestingly, King et al’s (2017: 15) network analysis of 1.5 million JSTOR articles finds no significant
correlation between the average number of self-citations and the percentage of male authors in a field.
3
  This fits with a broader pattern of women being “punished” for self-promoting their citations (King et al
2018).
4
  Our paper focuses on political science, but there are some differences established across disciplines for
self-citations, such as natural science papers having a higher average number of self-citations (Snyder and
Bonzi 1998; Hyland 2003a, b).

13
Scientometrics (2020) 125:1–28 3

interactions with author seniority and experience. We hypothesize that female authors are
less likely to cite their own research than male authors and that higher levels of seniority
and experience increase self-citations for men more so than women authors. We control
for a variety of factors that influence the likelihood of self-citation including number of
authors, number of references, and the highest number of previous citations accrued by
authors.
Our empirical analyses suggest several interesting things. For solo-authored work, our
results lead us to conclude that there is no significant difference in the likelihood of self-
citation between male and female authors, which indicates that women in political science
are in fact “leaning in” and citing their research as often as men. We also fail to reject the
null hypothesis of no differences in self-citation practices between all male author teams
and all female author teams. However, mixed author gender teams are significantly less
likely to cite previous work by one of the authors than other author groups and the inter-
active effects of seniority and experience are distinct for this group. We also find higher
probabilities of self-citation in general field journals and lower probabilities in Compara-
tive/ International Relations (IR) subfield journals. These results imply that the relationship
between gender and the decision to self-cite depends on several factors, such as collabora-
tion and the typical seniority and experience of authors on the team.
Our paper briefly reviews research on self-citations. We then describe our dataset and
coding rules used to indicate the presence and number of self-citations in each article,
which is our outcome of interest. This is followed by a section reviewing our empirical
results and a discussion of the implications of our findings for scholarly practices in pub-
lishing and proper attribution of female scholars’ research.

Literature review

We begin by describing how often scholars cite their own research in journal articles and
we describe several factors that increase self-citations. An analysis by King et  al (2018)
of 1.5 million research papers in JSTOR published between 1779-2011 found that about
10% of references are self-citations by the authors. Political science references contain 9%
self-citations (King et al 2018: 7), which is higher than some fields (e.g. classics, history)
but lower than others (e.g. math, molecular biology). Furthermore, the study showed that
men cite their own research 56% more often than women do and that this gap has remained
steady or even gotten larger over time (King et  al 2018: 1). The authors discuss several
possibilities for the self-citation gender gap including the idea that women may be pun-
ished more for self-promotion (Deschacht and Maes 2017), that men publish more articles
(Aksnes et al 2011; Hesli and Lee 2011; Williams et al 2015) and thus have more oppor-
tunities to self-cite, that men have higher views of the quality of their research, or that
women work in research areas with fewer overall citations. The study hints at the potential
impact of authorship order as well.
Research published by teams of authors is also more likely to involve self-citations,
especially if members of the team have published highly cited research in the past and are
more senior scholars (Hutson 2006). In a study of over 42,000 Norwegian publications
from 1981 to 1996, Aksnes (2003) finds that 21% of the total 640,710 citations are self-
citations to the authors’ previously published research. The author points out that although
self-citations are natural to establish credibility and link research to prior work, the scien-
tific community may view excessive self-citation as problematic (Hyland 2003a, b), which

13
4 Scientometrics (2020) 125:1–28

is why many citation indices report versions with and without self-citations (e.g. h-index).
Analyses (Aksnes 2003, 239-243) suggest that while self-citations are higher on average
for articles with more citations overall, the highest rate of self-citations occurs in the least
cited papers. Aksnes (2003) also finds that self-citation rates are higher for papers with a
higher number of authors and that they decline as years since publication increase. Mishra
et al (2018) find however that the gender gap in self-citations disappears once prior publi-
cations are controlled for in large N analyses.
Some studies suggest that women scholars cite their own work with less frequency than
male scholars. Maliniak et al (2013) examine 2541 articles published in international rela-
tions and show that women self-cite their research significantly less often than men in solo-
authored or coauthored articles (with other women; mixed teams are not different from all
male teams). Furthermore, the overall gender gap in citations is not eliminated by remov-
ing self-citations from the citation measure. Cameron el at (2016) find that women ecolo-
gists self-cite 19% less often than men in 2011 articles in the field’s top journals, a behav-
ioral practice that inflates male h-indices by 2+ points. Deschacht and Maes (2017) analyze
1346 articles in business and management and show that men engage in more frequent
self-citations than women across all cultural contexts (e.g. individualist vs. collectivist).
Using several thousand articles published in Denmark, Nielsen (2016) shows no signifi-
cant differences between men and women in self-citations, but in some areas such as medi-
cal research, men cite their own work more frequently than women. Nielsen also shows
that women tend to write solo-authored articles more often than male peers and given that
coauthored teams accrue greater citations (and self-citations), this could help explain why
we observe lower self-citation averages for articles authored by women.5
One factor that remains less explored in the self-citation literature is the interaction of
gender, seniority, and author order in coauthored publications. While several studies sug-
gest that mixed gender citation practices mimic the overall citation behavior of all male
author teams (e.g. Dion et al 2018), we know less about how author seniority and alpha-
betical order influences citations. Women tend to be younger as first and last authors, for
example, creating fewer opportunities for self-citation (Mishra et al 2018). If women exit
academic disciplines at each level through the leaky pipeline, then we might observe more
self-citations by men when working in coauthored teams because they are on average more
senior and well-cited.

Research hypotheses

Based on the preceding discussion of the literature, we test several hypotheses. First,
we test whether men who publish in political science journals self-cite more often than
women publishing in those journals. This expectation is based on our own research
(Dion et  al 2018), published studies in our discipline on self-citations (Maliniak et  al
2013), and the findings discussed in our literature review (e.g. Deschacht and Maes
2017; King et al 2018). Our analyses draw upon a much larger sample of journals and
articles than any previous studies in the discipline, thus we can also see how the pat-
terns vary across subfields (e.g. American politics, International Relations, Compara-
tive politics). We also compare solo-authored articles with coauthored articles to see if
gendered self-citations vary based on the number and mix of authors. Citations to work

5
  Women also tend to work more in interdisciplinary areas which can increase citations but depress the
total number of publications, reducing the long run ability to accrue citations (Leahey et al 2017).

13
Scientometrics (2020) 125:1–28 5

by women are less likely for male authors in all male or mixed gender author teams than
for female authors (Dion et al 2018). As extended to self-citations, if women work with
men and tend to be younger as the lead authors (Mishra et al 2018), they may be reluc-
tant to reference their own work compared with their male junior colleagues working in
all male teams.

H1 Articles authored by men will self-cite more than those by either mixed gender or
female author teams.

H1A Solo-male authored articles will self-cite more than solo-female authored articles.
H1B Articles by all-male author teams will self-cite more than mixed-gender or all-
female author teams.

Second, we evaluate whether more senior and experienced scholars self-cite their
research more frequently than junior scholars (Phelan 1999). Seniority is captured by time
since first publication, while experience records an author’s total citations. More senior
scholars have a higher probability of citing their own research primarily because they have
a larger corpus of research to draw upon (Hutson 2006). More experienced scholars with a
higher overall number of citations also self-cite more often (Aksnes 2003), with the rela-
tionship between “foreign citations” and self-citations following a square root law (Glänzel
et al 2006). Scholars who publish a higher number of articles are more likely to have arti-
cles reach the threshold of the top 1% most cited articles in their field (Larivière and Costas
2016). We include the seniority and experience for the most senior author in our empirical
models since studies show that gender gaps in self-citations may be weaker once publica-
tion counts and other measures of productivity or experience are captured (Mishra et  al
2018).

H2  Articles with more senior authors will self-cite more.

H3  Articles with more experienced authors will self-cite more.

Third, we interact the effects of gender with seniority and experience. Our previous find-
ings that show differences between all male and mixed gender teams in citation patterns
(Dion et  al 2018) produce an expectation that younger, less cited female authors will be
less comfortable engaging in self-promotion of their research than their male peers, espe-
cially when working with more senior male authors. However, as women become more
accomplished scholars, we anticipate that these gender interaction effects will diminish.
But in general, the existing findings in political science suggest that self-citation patterns
for mixed gender teams may be different than self-citation in all male or all female teams.

H4  The effects of seniority and experience will vary according to author team gender.

H4A At lower levels of seniority and experience, female authors (solo or all-female
teams) will self-cite less than comparable male or mixed gender author teams.
H4B The effects of seniority and experience will be diminished in mixed gender teams.

We also include several other variables that help explain variance in self-citation prac-
tices. Other article or journal characteristics may also be associated with self-citation. For

13
6 Scientometrics (2020) 125:1–28

example, as the number of authors increases, we expect the likelihood of self-citation to


increase because each additional author may increase the potential supply of available pub-
lications to cite (Fowler and Aknaes 2007). In addition, articles with more citations over-
all may be more likely to include self-citations because such articles generally use more
expansive criteria for citations (Fowler and Aknaes 2007). Similarly, certain subfields
or topics, like political theory or American political behavior, may be more likely to cite
older or “classic” literature, while other articles may be situated in newer areas or address
recent events. When the research generally references more recent publications, we expect
a higher likelihood of self-citation. Journal characteristics may also shape authors’ self-
citation practices. Journals with larger average number of references may impose fewer
constraints on length, including citations, creating more opportunities for self-citation. Fur-
ther, the average number of authors in a journal may also represent differences across sub-
fields that create incentives for self-citation. Finally, authors may be more likely to include
self-citations in articles appearing in higher impact journals to enhance the audience for
self-citations. In addition, when authors publish in higher impact journals, they may have
already published related work previously, increasing the supply of potential self-citations.
We include indicators to control for all of these alternative explanations and control vari-
ables in the analysis that follows.

Data description

We analyze the self-citation patterns in articles published between 2007 and 2016 in 22
political science journals, including general and field-specific journals (see Table 1). Politi-
cal science is a good discipline for studying self-citation patterns because it has an estab-
lished gender gap citation problem (Maliniak et  al 2013; Mitchell et  al 2013; Dion et  al
2018). The original dataset of articles including cited references was collected from the
Web of Science. Only partial information for the references is included in the Web of Sci-
ence: first author name, journal or monograph title, date of publication, and Document
Object Identifier (DOI, if available). To collect all reference author names, we used the
available information to query either the Crossref API (Crossref 2018) using the rcrossref
package (Chamberlain et al. 2017) or the Google Books API (Google Books 2018) using
the jsonlite R package (Ooms et  al. 2017). The result is a dataset of 8717 articles with
470,934 cited references. Of these, we successfully retrieved complete author information
for 229,712 (48.8%) references. Missing references include journal articles without valid
DOIs reported, monographs with organizational authors (e.g., World Bank), newspaper
articles, unpublished conference papers, and references with failed lookup queries (e.g.,
errors in original reference, such as incorrect publication year).
We define a self-citation as any reference that was written by one (or more) of the arti-
cle’s author(s). Because self-citation decisions are made within the context of a specific
article, our unit of analysis is each article. According to this definition, 10,335 (or 4.5%)
of the 229,712 references for which we have complete author information are self-cita-
tions. This is somewhat lower than studies that coded political science articles in JSTOR
over a longer time period which reported 9% self-citation rates (King et al 2017).6 We use
three measures of self-citations within an article as the dependent variable, including a

6
  However, self-citations as a percentage of total citations are declining in many disciplines (Hyland and
Jiang 2018). Given that our sample focuses on articles published in more recent years than King et  al
(2017), it is not surprising that the rate of self-citation in political science is lower in our study.

13
Table 1  Journal article sample frequencies (2007-2016)
Journal Field 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Total

American Journal of Political Science Gen 60 57 60 59 63 65 64 69 66 70 633


American Political Science Review Gen 51 34 33 44 45 45 51 53 53 55 464
British Journal of Political Science Gen 35 31 33 44 39 40 38 38 32 48 378
Scientometrics (2020) 125:1–28

Comparative Political Studies CP/IR 59 52 52 55 57 53 57 70 57 59 571


Comparative Politics CP/IR 20 20 21 24 21 20 22 22 24 27 221
International Organization CP/IR 27 24 19 23 24 25 28 32 32 24 258
International Political Science Review Gen 19 20 21 27 30 28 29 34 36 47 291
International Studies Quarterly CP/IR 41 31 49 51 52 61 65 67 62 66 545
Journal of Conflict Resolution CP/IR 37 40 38 37 41 43 43 57 57 54 447
Journal of Politics Gen 77 79 96 83 89 79 80 76 81 85 825
Legislative Studies Quarterly Sub 25 24 22 21 21 22 21 21 24 34 235
Party Politics Sub 32 33 33 34 37 46 46 74 66 65 466
Perspectives on Politics Gen 36 41 53 22 28 36 36 39 57 348
Political Analysis Sub 25 26 25 26 32 28 31 28 35 40 296
Political Behavior Sub 22 24 25 28 30 32 36 39 41 40 317
Political Research Quarterly Gen 59 66 62 69 70 68 68 74 63 67 666
Politics & Gender Sub 0 33 30 34 40 39 43 37 46 38 340
Politics & Society Gen 14 19 16 24 20 23 22 21 22 24 205
Public Opinion Quarterly Sub 41 52 50 41 52 43 50 51 41 42 463
Review of International Political Economy CP/IR 31 28 39 36 30 36 43 46 41 36 366
State Politics & Policy Quarterly Sub 21 22 18 20 21 22 25 22 22 19 212
World Politics CP/IR 12 11 15 18 17 20 18 18 20 21 170
Total 708 762 798 851 853 866 915 985 960 1017 8717
7

13
8 Scientometrics (2020) 125:1–28

dichotomous indicator for the presence of at least one self-citation; an ordinal count of
self-citations ranging from zero to five (or more) self-citations; and the total count of self-
citations in an article. Of all articles in our estimation sample, 53.5% have at least one
self-citation. The mean number of self-citations per article is 1.19. The distribution of the
self-citation count is positively skewed and leptokurtic (minimum = 0, and maximum =
20, standard deviation = 1.69, skewness = 2.45, and kurtosis = 12.29).
We code article authors’ genders as all female, mixed gender, or all male, and we expect
male author teams to have higher self-citations per article than mixed gender or all female
author teams (H1B). We code the probable author gender of the first five authors of each
article by querying the genderize.io API (genderize.io 2018) using the genderizeR pack-
age (Wais et  al. 2016). The genderize.io API uses social media data sources to estimate
the probability that a given name is female, and has been found to have an error rate of
only two percent when over 1000 political science articles were hand-coded for author
gender (Teele and Thelen 2017).7 We only assign male or female genders to names when
the probability that a name is female is less than 0.20 or greater than 0.80 respectively,
thereby omitting the relatively small number of names that generate ambiguous probabili-
ties. When we have coded genders for all of the first five authors and they are all male or all
female, we code the author team as male or female, respectively. When we have coded at
least one male and one female author in the first five authors (even if other author genders
are missing), we code the team as mixed gender.
We also expect articles with more experienced and more senior authors to be more
likely to have self-citations. To measure these concepts, we query the Crossref API for
each author’s name in the Crossref categories “Sociology and Political Science” and
“Political Science and International Relations.” If the query returned empty, we assumed
that the article in our dataset was an author’s first publication. Otherwise, we filtered the
query responses by the author’s given and family names in the author’s field, excluding all
publications before 1975 (to eliminate possible false positives), and extracting the earliest
publication year as well as the sum of Crossref’s citation count for all returned articles. To
measure seniority of each author, we use the earliest publication year, which is coded into
an indicator for no previous publishing (0), 1–7 years since first publication (1), 8–12 years
since first publication (2), 13–19 years since first publication (3), and 20–40 years since
first publication (4). The total citation count of the author with the largest number of cita-
tions is recoded into an indicator that ranges from zero citations (0), one-100 citations (1),
101–500 citations (2), and more than 500 citations (3). We then assign to the article the
maximum value for both total citations and years since first publication as an indicator of
the most experienced and senior author of an author team.
We include seven control variables in our analyses following previous literature on self-
citation practices. We code the number of article authors (to a maximum of 5) and the
total number of references (including those with organizational authors, newspapers, etc.),
which are both expected to be associated with higher numbers of self-citations.8 In our

7
  Teele and Thelen (2017) do not exclude ambiguous probabilities. Also, genderize.io fails to generate a
prediction for some names for which it lacks sufficient information to make a reliable prediction, including
many East and Southeast Asian given names. Yet handing coding of missing values does not alter results
for gender citation gaps in political science (Dion et al 2018).
8
  Only 68 (0.78%) articles have more than five authors, and the results do not change if we use the total
number of authors.

13
Scientometrics (2020) 125:1–28 9

sample, the mean number of article authors is 1.8. The mean number of references per arti-
cle is 51. In general, authors tend to reference more recent publications or sources, rather
than older works, though this varies by subfield or topic. We calculated the median age of
each article’s references. The journal with the lowest average median age of references per
article is the Review of International Political Economy (7.6 years), while Legislative Stud-
ies Quarterly has the oldest average of the median reference age per article (11.6). Overall,
the average age of a reference, as measured by the median age of references per article, is
10 years. Similarly, some subfields have higher rates of collaboration, which has increased
in political science recently, so we include a control variable capturing the average number
of article authors in a journal that year. Because the number of references may be con-
strained by journal editorial policies (e.g.., references may be selectively trimmed to mini-
mize article lengths), we also include a control for the average number of references in the
year of the journal in which the article appears. Finally, we also include the Social Science
Citation Index Impact Factor for each journal for each year because self-citations may be
more likely when articles are published in higher visibility journals. Table 2 includes the
mean and standard deviation of all indicators for the full sample and by author team gen-
ders. The table also includes the results of F tests for comparison of the means of the indi-
cators across the different team types. All indicators, except the number of references, vary
significantly across team types. Articles with male authors (solo, all male, mixed) have
higher levels of author experience, seniority, impact factor scores, and number of authors,
and they tend to cite older studies, The next section reports the results of our multivariate
tests of the research hypotheses.
We classify eight out of 22 journals as “general” or cross-disciplinary journals because
the journals’ aims and scope indicate that these journals publish works across all subfields
of political science, including political theory and political methodology as well as other
substantive empirical fields, such as American politics, comparative politics, international
relations, or public administration.9 The remaining fourteen journals are classified as either
comparative politics/international relations (CP/IR, seven journals) or other subfield jour-
nals (other, seven journals). Previous research indicates that female scholars are less likely
to work in certain subfields or publish in certain journals (Maliniak et  al 2013; Theele
and Thelen 2017). However, in the aggregate, these groupings – general, CP/IR, and other
subfields – include a balance of fields or subfields that have more or fewer women overall.
For example, women are more likely to work in comparative politics and men in inter-
national relations, but several journals publish work in either field or at their intersection
(e.g., International Studies Quarterly or World Politics). Likewise, the other subfield cat-
egory includes journals from subfields dominated by women (e.g., Politics & Gender) and
men (e.g., Political Analysis). Table 3 includes the distribution of author genders and solo
vs. team authors by journal field. Overall, 60.6% of articles were written by men, 18.8%
by women, and 20.6% by mixed gender teams, while 48.6% were written by solo authors
and 51.4% by teams.10 In general, CP/IR journals have more male authors and more solo

9
  While narrower subfields or topics exist, these are the largest fields within political science around which
teaching and research activities are often organized. See Table 1 for a list of which journals are coded into
which categories.
10
  37.5% of the members of the American Political Science Association are women, although women’s rep-
resentation in political science journals is much lower than expected (compared with female membership in
sections who sponsor journals) for most journals affiliated with the organization (Dion and Mitchell2020).

13
10
Table 2  Mean descriptive statistics by author teams
Author team Total Solo male Solo female Team male Team female Team mixed ANOVA
Variable Mean Mean Mean Mean Mean Mean F test

13
SD SD SD SD SD SD p Value

Article includes self-citation (0/1) 0.52 0.40 0.40 0.67 0.55 0.62 86.71
0.50 0.49 0.49 0.47 0.50 0.49 0.00
Self-cite count (0–4) 1.02 0.66 0.62 1.49 1.08 1.28 135.57
1.25 1.00 0.91 1.41 1.25 1.34 0.00
Self-citation count 1.14 0.69 0.65 1.70 1.18 1.47 114.49
1.65 1.13 1.14 1.99 1.56 1.90 0.00
Max author experience (0–3) 1.51 1.18 0.96 1.99 1.33 1.88 338.62
0.93 0.84 0.75 0.86 0.73 0.87 0.00
Max author seniority (0–4) 2.30 1.84 1.56 2.90 2.20 2.83 269.00
1.36 1.35 1.26 1.17 1.14 1.18 0.00
N references 50.74 50.87 50.35 50.07 50.70 51.70 0.78
24.48 25.14 27.77 23.00 25.30 22.49 0.54
Median age of article refs 10.34 11.16 9.82 10.17 8.94 9.84 21.43
5.10 6.36 4.80 4.13 3.68 4.09 0.00
Mean N of refs in Journal/Year 52.82 53.88 51.68 52.91 50.91 52.16 9.79
10.68 10.29 12.22 10.06 11.39 10.60 0.00
Mean N of authors in Journal/Year 1.84 1.80 1.69 1.92 1.81 1.93 104.04
0.31 0.30 0.30 0.29 0.31 0.32 0.00
Journal SSCI impact factor 1.79 1.73 1.63 1.94 1.61 1.83 21.73
0.89 0.86 0.83 0.99 0.83 0.85 0.00
N of authors (1–5+) 1.73 2.34 2.13 2.58 61.35
0.86 0.63 0.37 0.77 0.00
Diff in experience (0–3) 0.91 1.84 1.23 1.80 48.19
1.09 0.88 0.68 0.88 0.00
Scientometrics (2020) 125:1–28
Table 2  (continued)
Author team Total Solo male Solo female Team male Team female Team mixed ANOVA
Variable Mean Mean Mean Mean Mean Mean F test
SD SD SD SD SD SD p Value

Diff in seniority (0–4) 1.14 2.21 1.57 2.36 33.27


1.46 1.33 1.09 1.31 0.00
Most senior author female (0/1) 0.25 0.31
Scientometrics (2020) 125:1–28

0.43 0.46
Most experienced author female (0/1) 0.25 0.32
0.43 0.47
Observations 5482 1851 813 1473 216 1129

Includes observations in estimation samples


11

13
12 Scientometrics (2020) 125:1–28

Table 3  Journal fields and author characteristics


Journal field Male authors Female authors Mixed gender Solo authors Author team Total

CP or IR journals 884 264 269 757 660 1417


62.4% 18.6% 19.0% 53.4% 46.6% 100%
General journals 1552 396 487 1148 1287 2435
63.7% 16.3% 20.0% 47.1% 52.9% 100%
Other subfields 889 369 373 759 872 1631
54.5% 22.6% 22.9% 46.5% 53.5% 100%
Total 3325 1029 1129 2664 2819 5483
60.6% 18.8% 20.6% 48.6% 51.4% 100%

authors than the overall sample, while other subfield journals have a higher rate of female
authors and co-authored articles.

Data analysis and results

We begin by evaluating our first hypotheses (H1, H1A, H1B) with models of whether an
article includes at least one self-citation; these results are presented in Table  4. Because
the dependent variable is dichotomous, we estimate logistic regression models with robust
standard errors clustered by journal. In Model 1, we find mixed evidence in support of our
first three hypotheses. Contrary to our expectations, we fail to reject the null hypothesis
and find that solo male and solo female authors do not differ significantly in the likelihood
that an article includes at least one self-citation (H1A). Similarly, all-male and all-female
author teams appear to have similar likelihoods of including a self-citation (H1B). How-
ever, mixed gender author teams are less likely than all male teams to include one self-
citation, supporting H1B. When women coauthor with men, the articles are less likely to
include citations to the authors’ research.
Figure 1 plots the predicted probability (with 95% confidence intervals) that an article
includes at least one self-citation by author team gender.11 The authors with the greatest
probability of including at least one self-citation are all male teams followed by all female
teams. This suggests that in publications with more equal distribution of work among simi-
lar genders, there is a greater likelihood that the team will include one or more self-cita-
tions compared to solo authors or mixed gender teams.
Our results support our expectation that articles with a higher number of authors have
more self-citations, although we show interesting interactions based on the gender compo-
sition of the team. There is no significant difference in the likelihood of including a self-
citation between solo male or female authors which shows, contrary to other analyses in
the political science literature (Maliniak et al 2013), that women are in fact leaning in and
citing their own work as solo authors. Overall, gender alone does not seem to explain the

11
 Predicted probabilities with larger confidence intervals (e.g., solo female authors or non-alphabetical
order all female teams) reflect smaller numbers of authors with these characteristics in the dataset.

13
Scientometrics (2020) 125:1–28 13

Table 4  Self-citation logistic regression models


Models (1) (2) (3)
Specification Full model Gender x experience Gender x seniority
Dependent variable Self-citation (0/1) Self-citation (0/1) Self-citation (0/1)

Solo female 0.120 0.271 0.194


(0.124) (0.255) (0.167)
Male team 0.342*** 0.430* 0.474*
(0.125) (0.169) (0.201)
Female team 0.291** 0.210 0.400
(0.139) (0.269) (0.309)
Mixed team 0.113 0.601** 0.486*
(0.119) (0.199) (0.213)
Max author experience (0–3) 0.543*** 0.663*** 0.547***
(0.053) (0.061) (0.053)
Solo female X max experience − 0.121
(0.164)
Male team X max experience − 0.103
(0.080)
Female team X max experience 0.038
(0.159)
Mixed team X max experience − 0.330**
(0.106)
Max author seniority (0–4) 0.236*** 0.233*** 0.292***
(0.026) (0.026) (0.034)
Solo female X max seniority − 0.035
(0.056)
Male team X max seniority − 0.078
(0.063)
Female team X max seniority − 0.071
(0.111)
Mixed team X max seniority − 0.170*
(0.082)
N of authors (1–5+) 0.120 0.146* 0.148
(0.076) (0.073) (0.076)
N references 0.013*** 0.013*** 0.013***
(0.002) (0.002) (0.002)
Median age of article references − 0.042*** − 0.042*** − 0.042***
(0.007) (0.007) (0.007)
Mean N of references in Journal/Year 0.006 0.006 0.006
(0.005) (0.005) (0.005)
Mean N of authors in Journal/Year − 0.144 − 0.133 − 0.134
(0.182) (0.184) (0.181)
Journal SSCI impact factor 0.091* 0.089 0.089
(0.048) (0.047) (0.049)
CP or IR − 0.144** − 0.146* − 0.140*
(0.060) (0.063) (0.060)

13
14 Scientometrics (2020) 125:1–28

Table 4  (continued)
Models (1) (2) (3)
Specification Full model Gender x experience Gender x seniority
Dependent variable Self-citation (0/1) Self-citation (0/1) Self-citation (0/1)

Other subfield 0.031 0.024 0.031


(0.107) (0.110) (0.109)
Constant − 1.990*** − 2.178*** − 2.149***
(0.614) (0.585) (0.611)
Observations 5483 5483 5483
Pseudo ­R2 0.124 0.126 0.125
Null log likelihood − 3793.47 − 3793.47 − 3793.47
Log likelihood − 3321.62 − 3315.81 − 3318.34

***p < 0.001, **p < 0.01, *p < 0.05 Robust standard errors in parentheses, clustered by journal

Fig. 1  Predicted probability (with 95% confidence intervals) of any self-citation by author team (based on
Table 4, Model 1)

probability of self-citation, except that mixed gender teams are less likely to self-cite than
gender homogenous teams. Given the similar averages of the control variables between the
all male and mixed author groups (Table 2), this suggests that the mixed teams are more
frequently senior males working with more junior female scholars. For example, all male

13
Scientometrics (2020) 125:1–28 15

Fig. 2  Marginal effect of experience and seniority by author team type [based on Table 4, Model 2 (experi-
ence) and Model 3 (seniority)]

and mixed gender teams have more author experience and seniority than all female author
teams. Citation studies find that mixed gender teams mimic the behavior of all male teams
(Mitchell et al 2013; Dion et al 2018), citing work by women less often, and our study sug-
gests that self-citation practices are also similar between all male and mixed author teams.
With regards to our second (seniority, H2) and third (experience, H3) hypotheses, in
both instances we reject the null hypotheses and conclude that both factors are signifi-
cantly associated with a higher probability of self-citation (Table 4, Models 2 and 3). Sub-
stantively, each increase in seniority (age since first publication) increases the odds of
self-citation by 27%, and each increase in experience (total number of citations for most
cited author) increases the odds of self-citation by 72%. These results confirm previous
self-citation studies in political science and other academic disciplines and suggest that
more accomplished scholars regularly work in coauthor teams and cite their own previ-
ous research. Political science norms for tenure typically include expectations for junior
scholars to publish solo-authored work, which could explain why we see fewer overall self-
citations in solo-authored pieces.12 Yet it is interesting, though, that these solo-authored
pieces produce no gender gaps in self-citation rates.

12
  Self-citation rates are generally lower in humanities fields compared with science fields. This is because
these fields often have expectations for scholars to publish as “lone wolves” much like junior faculty in
political science (Snyder and Bonzi 1998).

13
16 Scientometrics (2020) 125:1–28

In Table 4, Models 2 and 3, we include interactions between the indicators for author
genders and our indicators for experience and seniority to evaluate our final hypotheses
(H4, H4A, H4B) that the relationships between experience and seniority vary based on
author genders. Figure 2 plots the marginal effects of experience and seniority by author
gender based on Models 2 and 3 in Table  4. Experience has a significant positive asso-
ciation with self-citation across all team genders, while seniority has a significant posi-
tive association with self-citation across all team genders except mixed gender teams, for
whom the association is positive but not statistically significant.13 In both estimations,
mixed gender teams have much weaker associations between experience and seniority and
self-citation compared to other authors. Mixed gender teams also have higher median and
mean scores for experience (209 median and 469 mean citations) and seniority (16 median
and 17 mean years since first publication) compared to all female (mean seniority = 9 and
mean experience = 103) and all male (mean seniority = 14 and mean experience = 344)
teams. Similarly, mixed gender teams led by men have an average of 19 years’ experience
and 550 citations, compared to 15 years’ experience and 305 citations for mixed gender
teams led by women. Otherwise, similar to the results in Model 1, the marginal effects
of experience and seniority on self-citation are similar for male and female solo authors
and also for all-male and all female teams. Our hypotheses receive partial support. While
we find that the effects of seniority and experience are weaker for mixed gender teams,
consistent with rejecting the null hypothesis for H4B, we do not find that females (solo
or teams) self-cite less often when they have lower levels of seniority or experience. The
mixed gender environment generates different citation practices.
In Table 4, the estimates for the control variables are generally consistent with expec-
tations. As the numbers of authors increase, so too does the likelihood of self-citation,
though collinearity with author gender (solo or team dimension) may explain why the asso-
ciation is not statistically significant in some models. Articles with more references have a
significantly higher likelihood of self-citation, while articles with older references have a
significantly lower likelihood of self-citation. Journal characteristics are not significantly
associated with self-citation practices, though self-citations are slightly more likely in jour-
nals with more average number of references or higher impact and less likely in journals
with higher average number of authors per article. Articles in specialized subfield journals
are slightly more likely to have self-citations than those in general interest journals, and
articles in the subfields of comparative politics or international relations are significantly
less likely than general interest journals to include self-citations. We return to differences
across political science subfields below.
We evaluate the robustness of our results in several ways. First, we estimate models
using an alternative self-citation measures that capture an ordered scale and the total count
of self-citations. In Table 5, Model 1, the outcome is the ordered count of self-citations,
ranging from zero to five (or more) self-citations, while in Model 2, the outcome is the total
self-citation count for each article. We estimate these models as ordered logistic regres-
sions and negative binomial regressions, respectively. In Table 5, Model 3, we estimate a
zero-inflated negative binomial (ZINB) regression model to account for the large propor-
tion (47.6%) of articles that have no self-citation. ZINB estimation involves two models,

13
  Using the testparm command (null hypothesis is base and interaction terms all simultaneously = 0) in
STATA, we find that the variables are jointly significant for each model that includes interaction terms
(Model 2: 2 = 471.12 with 9 degrees of freedom and p < 0.0001; Model 3: 2 = 312.85 with 9 degrees of
freedom and p <0.0001).

13
Scientometrics (2020) 125:1–28 17

Table 5  Self-citation alternative dependent variable indicators & estimation strategies


Model (1) (2) (3) (3)
Estimation Ordered logit Negative binomial Zero-inflated negative binomial

Dependent variable 0–5 Count Zero Count

Solo female 0.112 0.078 − 0.486 0.051


(0.124) (0.095) (0.424) (0.096)
Team male 0.323** 0.242*** − 0.430 0.196***
(0.102) (0.052) (1.179) (0.049)
Team female 0.303* 0.235** 1.085 0.231***
(0.128) (0.074) (1.357) (0.063)
Team mixed 0.038 0.093 − 0.392 0.044
(0.120) (0.074) (1.552) (0.074)
Max author seniority (0–4) 0.233*** 0.163*** − 3.695*** 0.080***
(0.026) (0.016) (0.640) (0.022)
Max author experience (0–3) 0.638*** 0.426*** 0.264 0.418***
(0.046) (0.027) (0.357) (0.028)
N of authors (1–5+) 0.199** 0.119*** 0.359 0.134***
(0.062) (0.030) (0.952) (0.029)
N references 0.015*** 0.010*** − 0.007 0.010***
(0.002) (0.001) (0.009) (0.001)
Median age of article refs − 0.041*** − 0.029*** 0.181*** − 0.022***
(0.007) (0.005) (0.048) (0.006)
Mean N of refs in Journal/Year 0.007 0.004 − 0.011 0.003
(0.005) (0.003) (0.020) (0.003)
Mean N authors in Journal/Year − 0.100 − 0.029 0.340 − 0.016
(0.172) (0.112) (0.913) (0.108)
Journal SSCI impact factor 0.067 0.023 0.038 0.024
(0.053) (0.028) (0.226) (0.025)
CP or IR − 0.152 − 0.100 − 0.280 − 0.106
(0.093) (0.070) (0.490) (0.076)
Other subfield 0.041 0.031 − 0.270 0.021
(0.123) (0.075) (0.795) (0.075)
Constant (cut1) 2.419*** − 1.861*** − 0.506 − 1.575***
(0.552) (0.328) (2.364) (0.295)
Constant cut2 3.706***
(0.555)
Constant cut3 4.650***
(0.551)
Constant cut4 5.481***
(0.569)
Observations 5482 5482 5482
Ln(alpha) − 0.685*** − 0.789***
(0.078) (0.082)
Alpha 0.504*** 0.454***
(0.040) (0.037)
Pseudo ­R2 0.0945 0.0922
Null likelihood − 7382.81 − 8090.51 − 8090.51
Log likelihood − 6685.41 − 7344.57 -7274.63

13
18 Scientometrics (2020) 125:1–28

Table 5  (continued)
***p < 0.001, **p < 0.01, *p < 0.05 Robust standard errors in parentheses, clustered by journal

Fig. 3  Odds-ratios (with 95% confidence intervals) by measure of citations and model estimation (Authors’
calculations based on Table 4, Model 1 and Table 5, Models 1–3)

one that explains the likelihood of an article having zero self-citations, and a second that
captures the self-citation count for those articles with at least one self-citation.
Figure 3 plots the odds-ratios of all these models to facilitate comparisons across meas-
ures of self-citations and estimation strategies. The results across these models are highly
consistent with those of Table 4, Model 1. For example, author teams self-cite more than
solo authors, except for mixed gender teams, who have citation practices like solo authors.
Seniority and experience also increase the self-citation counts in these models. Finally,
across these models, the number of authors and references are associated with significantly
more self-citations, while articles with older references are significantly associated with
fewer self-citations. Because the logistic regression model has the most conservative (wid-
est) standard errors and is the easiest to interpret, we use this model as the primary model
for evaluating alternative specifications and estimation strategies.
Second, we estimate Model 1 in Table 4 and Model 1 in Table 5 using bootstrapping
and jackknife procedures to calculate the standard errors because the number of clusters
(journals) in our model is relatively small (22). We find very similar results overall, with

13
Scientometrics (2020) 125:1–28 19

Table 6  Alternative specifications of team characteristics


(1) (2) (3)
Specification Most senior or Unequal seniority Alpha order by team gender
experience female or experience

Sample Co-authored Co-authored Co-authored


Dependent variable Self-citation (0/1) Self-citation (0/1) Self-citation (0/1)

Team female − 0.148 − 0.072


(0.134) (0.137)
Team mixed − 0.268*** − 0.219***
(0.074) (0.066)
Most experienced author female − 0.025
(0.116)
Most senior author female 0.110
(0.103)
Diff in seniority (0–4) − 0.215***
(0.050)
Diff in experience (0–3) 0.013
(0.146)
Alpha, male team − 0.053
(0.167)
Alpha, mixed gender team − 0.246
(0.148)
Other, female team -0.333
(0.325)
Other, male team 0.009
(0.178)
Other, mixed gender team − 0.321
(0.196)
Max author citations (0–3) 0.515*** 0.494** 0.512***
(0.063) (0.154) (0.064)
Max author pub age (0–4) 0.198*** 0.378*** 0.197***
(0.044) (0.055) (0.043)
N of authors (1–5+) 0.158* 0.221** 0.161*
(0.077) (0.078) (0.075)
N references 0.015*** 0.016*** 0.016***
(0.003) (0.003) (0.003)
Median age of article refs -0.039*** − 0.041*** − 0.039***
(0.011) (0.011) (0.011)
Mean N of refs in Journal/Year 0.012* 0.011* 0.011*
(0.005) (0.005) (0.005)
Mean N of authors in Journal/Year − 0.082 − 0.060 − 0.085
(0.172) (0.176) (0.169)
Journal SSCI impact factor 0.094* 0.096* 0.094
(0.046) (0.047) (0.049)
CP or IR − 0.185* − 0.188* − 0.193*
(0.089) (0.089) (0.091)

13
20 Scientometrics (2020) 125:1–28

Table 6  (continued)
(1) (2) (3)
Specification Most senior or Unequal seniority Alpha order by team gender
experience female or experience

Sample Co-authored Co-authored Co-authored


Dependent variable Self-citation (0/1) Self-citation (0/1) Self-citation (0/1)

Other subfield 0.011 0.000 0.008


(0.110) (0.108) (0.112)
Constant − 2.154** − 2.353*** − 2.112**
(0.667) (0.653) (0.696)
Observations 2819 2819 2819
Pseudo ­R2 0.0855 0.0894 0.0858
Null log likelihood − 1842 − 1842 − 1842
Log likelihood − 1685 − 1677 − 1684

***p < 0.001, **p < 0.01, *p < 0.05 Robust standard errors clustered by article in parentheses

the team female variable being significant at the 95% level and all other results being sta-
ble.14 (See Appendix.)
Third, we also evaluate several other alternative propositions that might explain
underlying differences in self-citation practices across different types of co-authorship
arrangements. These results are presented in Table 6. In Model 1, we include an indi-
cator for whether the most experienced or most senior author is a woman. This tests
the proposition that when a woman is the author with the most experience or seniority,
self-citation practices will differ. We find no statistically significant difference. In Model
2, we measure the maximum differences between the most and least experienced and
most senior and junior author as an indicator for team diversity on these two dimen-
sions. We find that differences in seniority are significantly associated with lower like-
lihoods of self-citation. This makes sense because junior team members may be less
likely to have a body of literature to cite or these teams may include senior and junior
scholars working together outside of their own primary topics (e.g., a research assis-
tant or complementary skillsets). However, differences in experiences, as measured by
total citations, has no similar relationship with self-citations. Finally, Table  6, Model
3 looks at the issue of the order of author names. In political science, convention dic-
tates that the order of authors’ names reflects the relative contribution of each author
to the final product. When author teams want to signal that all authors contributed
equally to a paper, they will often choose to use alphabetical order based on authors’
family names.15 While these are general conventions of the political science discipline,

14
  We thank a reviewer for this suggestion. These alternate models are presented in the online appendix.
For bootstrapped standard errors in the logit and event count models, we use the vce (bootstrap) option clus-
tered by journal in STATA with 300 replications. For jackknife standard errors, we use the vce (jackknife)
option clustered by journal.
15
  Sometimes, when teams have a series of papers, they alternate alphabetical and reverse alphabetical or
some other rotation of names, even when authorship is equal. While sometimes such decisions are noted in
the paper, they are often left unstated.

13
Scientometrics (2020) 125:1–28 21

Table 7  Self-citation logistic regression models by journal field


Model (1) (2) (3)
Sample General journals CP/IR journals Other subfield
Dependent variable Self-citation (0/1) Self-citation (0/1) Self-citation (0/1)

Solo female 0.073 − 0.265* 0.131


(0.142) (0.119) (0.142)
Male team 0.543*** 0.402* − 0.013
(0.122) (0.185) (0.255)
Female team 0.358* 0.170 0.011
(0.179) (0.443) (0.323)
Mixed team 0.177 0.235 − 0.058
(0.145) (0.198) (0.288)
Max author experience (0–3) 0.585*** 0.683*** 0.566***
(0.070) (0.066) (0.089)
Max author seniority (0–4) 0.261*** 0.183*** 0.219***
(0.032) (0.027) (0.053)
N of authors (1–5+) − 0.023 − 0.013 0.316**
(0.110) (0.093) (0.098)
N references 0.010*** 0.015*** 0.016***
(0.002) (0.002) (0.004)
Median age of article references − 0.053*** − 0.050*** − 0.027
(0.012) (0.009) (0.015)
Mean N of references in Journal/Year 0.008 0.034*** − 0.005
(0.005) (0.009) (0.010)
Mean N of authors in Journal/Year 0.393 0.747* 0.085
(0.383) (0.339) (0.057)
Journal SSCI impact factor 0.283*** 0.002 − 0.028
(0.054) (0.099) (0.131)
Constant − 3.379*** − 5.064*** − 2.350***
(0.952) (0.784) (0.559)
Observations 2435 1417 1631
Pseudo ­R2 0.142 0.134 0.123
Null log likelihood − 1677.16 − 980.70 − 1130.39
Log likelihood − 1439.21 − 848.85 − 991.01

***p < 0.001, **p < 0.01, *p < 0.05 Includes journal fixed effects (not shown) with robust standard errors
in parentheses, clustered by journal

there are often exceptions that have no explicit explanation. Nevertheless, we consider
whether the self-citation practices across author teams differ when the authors’ family
names are ordered alphabetically or not by team gender. We find no significant differ-
ences when author family names are alphabetical or in another order for different team
genders. Together, these results suggest that, with the exception of teams with larger
differences in seniority, other author team characteristics are not significantly associated
with particular self-citation practices.

13
22 Scientometrics (2020) 125:1–28

Fig. 4  Odds-ratios (with 95% confidence intervals) for author team genders by journal field (Reference cat-
egory is solo male author. Authors’ calculations based on Table 4, Model 1 and Table 7, Models 1–3)

Finally, we consider whether self-citation varies across different subfield areas of politi-
cal science. As mentioned above, the baseline probability of self-citation varies signifi-
cantly across different types of journals. To further examine differences across different
types of political science journals, we estimate our main logit model with fixed effects for
journals. These results are in Table 7, and the odds-ratios for the author genders are plotted
in Figure 4. The results for the general political science journals mirror those of the full
sample, with significantly higher likelihood of self-citations by author teams and as author
experience and seniority increases. In contrast, in Comparative Politics/International Rela-
tions journals, solo female authors are significantly less likely to self-cite than their solo
male author peers,16 and female and mixed gender author teams are not significantly more
likely to self-cite, unlike their all-male author team peers, who are significantly more likely
to self-cite than solo male authors. This suggests that self-citation norms may vary across
research areas of the discipline and that this could be related to differences in female rep-
resentation across areas of political science (Dion et  al 2018). In general, in the sample
of other subfield journals, the likelihood of self-citation does not vary significantly across

16
 This helps us understand why Maliniak et  al (2013) found different results from ours, namely that
women self-cite less often. Their sample contains a larger number of IR journals relative to the total sample
than ours.

13
Scientometrics (2020) 125:1–28 23

author genders. Author experience and seniority are significantly and positively associated
with higher likelihood of self-citations, as expected.

Discussion and conclusions

Scholars often cite their own research when publishing journal articles to give proper attri-
bution to their previous work and help create connections across studies. Yet some view
these self-citations more cynically as a form of self-promotion, with the suggestion that
some types of academic or cultural characteristics will produce differences in how fre-
quently self-citation occurs. Many previous studies suggested that women cite their own
work less frequently than their male peers, a gap that could be remedied by encouraging
women scholars to “lean in” with respect to self-citations. However, using a sample of
8717 political science articles, our analyses show that the self-citation gender gap is mini-
mal. Women publishing research as solo authors cite their own work at a rate similar to
male solo authors. Similarly, when women work together in teams in political science, their
self-citation practices are much like their male team peers. Mixed gender teams though
have fewer overall self-citations which may reflect differences in seniority or experience
among coauthors (e.g. senior males writing with junior females). We show instead that
the number of article authors and the seniority of the authors (age and total citations) help
account for the variance in self-citations among political scientists.
There are many avenues available for extending our analyses to address additional ques-
tions. We utilized the article as the unit of analysis and assume independence across obser-
vations. Yet this fails to capture dynamics effects that could occur over time (e.g. authors
are more likely to cite their own work when they are more senior) which could be assessed
more easily with a dataset that treats the author as the unit of analysis. We also do not
consider how scholars make choices about publishing research on their own versus work-
ing in same sex or mixed gender teams. Given that the distribution of women scholars
varies quite a bit in political science across research areas (Key and Sumner 2019; Dion
and Mitchell 2020), the propensity to write with male or female authors could change as
well. The critical mass of women in a field alters the propensity for authors to cite work by
women and this could also influence self-citation practices, something we do not explore
in this paper. We do not find significant differences between multi-authored pieces with
alphabetical or non-alphabetical author ordering. Yet some areas of the discipline have
fewer self-citations, such as international relations and comparative politics. Future analy-
ses should explore differences across substantive areas of political science in more detail.

Appendix

See Tables 8, 9, 10.

13
24 Scientometrics (2020) 125:1–28

Table 8  Measures of self-citations by journal (2007-2016)


Journal Total N Articles w/ Self-citations/ Mean N of
self-citation references w/ self-citations per
(%) complete author article
information
% Total N

American Journal of Political Science 633 64.1% 4.28% 22,811 1.54


American Political Science Review 464 57.1% 3.10% 19,014 1.27
British Journal of Political Science 378 58.5% 4.02% 12,889 1.37
Comparative Political Studies 571 53.1% 3.42% 18,875 1.13
Comparative Politics 221 43.4% 2.76% 5395 0.67
International Organization 258 62.4% 3.23% 11,037 1.38
International Political Science Review 291 42.3% 2.63% 8814 0.80
International Studies Quarterly 545 53.0% 3.17% 20,499 1.19
Journal of Conflict Resolution 447 56.6% 4.21% 16,093 1.52
Journal of Politics 825 57.2% 4.08% 27,145 1.34
Legislative Studies Quarterly 235 48.1% 3.28% 7312 1.02
Party Politics 466 50.0% 4.27% 12,456 1.14
Perspectives on Politics 348 55.2% 3.54% 12,246 1.24
Political Analysis 296 52.4% 4.80% 7475 1.22
Political Behavior 317 51.7% 3.04% 12,119 1.16
Political Research Quarterly 666 49.5% 3.24% 20,383 0.99
Politics & Gender 340 51.5% 4.14% 7180 0.87
Politics & Society 205 51.2% 4.06% 5295 1.05
Public Opinion Quarterly 463 49.5% 5.16% 10,150 1.13
Review of International Political Economy 366 51.6% 3.45% 11,160 1.05
State Politics & Policy Quarterly 212 49.1% 3.75% 6030 1.07
World Politics 170 50.6% 2.61% 7131 1.09
Total 8717 53.50% 3.67% 281,509 1.19

13
Scientometrics (2020) 125:1–28 25

Table 9  Other article characteristics by journal (2007-2016)


Journal Mean N Mean N Mean of median age of Mean Journal
authors per references per references per article impact factor
article article (SSCI)

American Journal of Political 2.07 54.46 10.92 3.09


Science
American Political Science 1.78 65.27 10.86 3.23
Review
British Journal of Political 1.95 55.68 10.81 1.89
Science
Comparative Political Studies 1.77 56.09 9.62 1.78
Comparative Politics 1.44 54.32 9.74 0.94
International Organization 1.67 69.24 9.67 3.41
International Political Science 1.57 55.65 9.80 0.83
Review
International Studies Quarterly 1.68 63.67 10.18 1.58
Journal of Conflict Resolution 1.96 56.22 9.60 1.88
Journal of Politics 1.93 50.69 10.91 1.71
Legislative Studies Quarterly 1.82 48.34 11.56 0.89
Party Politics 1.72 45.62 11.20 1.32
Perspectives on Politics 1.49 62.54 9.93 2.32
Political Analysis 2.08 39.08 8.98 3.17
Political Behavior 2.04 56.07 11.06 1.44
Political Research Quarterly 1.82 49.09 11.05 0.95
Politics & Gender 1.44 36.13 8.62 1.36
Politics & Society 1.42 60.62 10.35 1.48
Public Opinion Quarterly 2.51 35.78 9.71 1.80
Review of International Politi- 1.40 65.95 7.61 1.60
cal Economy
State Politics & Policy Quar- 1.96 46.96 11.42 0.70
terly
World Politics 1.54 80.00 10.01 4.03
Total 1.82 54.02 10.2 1.79

13
26 Scientometrics (2020) 125:1–28

Table 10  Models with alternative calculations of standard errors


(1) (2) (3) (4)
Paper specification Table 4, Model 1 Table 4, Model 1 Table 5, Model 1 Table 5, Model 1

Dependent variable 0/1 0/1 0–5 0–5

Regression method Logistic Logistic Ordered logistic Ordered logistic

SE method Bootstrapped Jackknifed Bootstrapped Jackknifed

Solo female 0.120 0.120 0.112 0.112


(0.123) (0.161) (0.117) (0.160)
Team male 0.342* 0.342* 0.323** 0.323**
(0.133) (0.131) (0.099) (0.105)
Team female 0.291* 0.291 0.303* 0.303*
(0.147) (0.145) (0.122) (0.134)
Team mixed 0.113 0.113 0.038 0.038
(0.122) (0.124) (0.123) (0.124)
Max author seniority (0–4) 0.236*** 0.236*** 0.233*** 0.233***
(0.025) (0.030) (0.025) (0.030)
Max author experience (0–3) 0.543*** 0.543*** 0.638*** 0.638***
(0.052) (0.063) (0.046) (0.054)
N of authors (1–5+) 0.120 0.120 0.199** 0.199**
(0.074) (0.081) (0.061) (0.066)
N references 0.013*** 0.013*** 0.015*** 0.015***
(0.001) (0.002) (0.002) (0.002)
Median age of article refs − 0.042*** − 0.042*** − 0.041*** − 0.041***
(0.008) (0.008) (0.007) (0.008)
Mean N of refs in journal/year 0.006 0.006 0.007 0.007
(0.006) (0.008) (0.005) (0.007)
Mean N of authors in journal/year − 0.144 − 0.144 − 0.100 − 0.100
(0.200) (0.279) (0.199) (0.269)
Journal SSCI impact factor 0.091 0.091 0.067 0.067
(0.053) (0.069) (0.060) (0.079)
CP or IR journal − 0.144 − 0.144 − 0.152 − 0.152
(0.075) (0.080) (0.111) (0.123)
Other subfield journal 0.031 0.031 0.041 0.041
(0.139) (0.149) (0.169) (0.170)
Constant/Cut 1 − 1.990** − 1.990* 2.419*** 2.419**
(0.615) (0.857) (0.568) (0.756)
Cut 2 3.706*** 3.706***
(0.570) (0.759)
Cut 3 4.650*** 4.650***
(0.568) (0.755)
Cut 4 5.481*** 5.481***
(0.584) (0.772)
Observations 5483 5483 5483 5483
Pseudo ­R2 0.124 0.124 0.0945 0.0945
Null log likelihood − 3794 − 3794 − 7383 − 7383
Log likelihood − 3322 − 3322 − 6685 − 6685

Clustered standard errors by journal in parentheses. ***p < 0.001, **p < 0.01, *p < 0.05

13
Scientometrics (2020) 125:1–28 27

References
Aksnes, D. W. (2003). A macro study of self-citation. Scientometrics, 56(2), 235–246. https​://doi.
org/10.1023/A:10219​19228​368.
Aksnes, D. W., Rorstad, K., Piro, F., & Sivertsen, G. (2011). Are female researchers less cited? A large-
scale study of Norwegian scientists. Journal of the American Society for Information Science and
Technology, 62(4), 628–636. https​://doi.org/10.1002/asi.21486​.
Beaudry, C., & Larivière, V. (2016). Which gender gap? Factors affecting researchers’ scientific impact
in science and medicine. Research Policy, 45(9), 1790–1817. https​://doi.org/10.1016/j.respo​
l.2016.05.009.
Cameron, E. Z., White, A. M., & Gray, M. E. (2016). Solving the Productivity and Impact Puzzle:
Do Men Outperform Women, or are Metrics Biased? BioScience, 66(3), 245–252. https​://doi.
org/10.1093/biosc​i/biv17​3.
Chamberlain, S., Boettiger, C., Hart, T., & Ram, K. (2017). rcrossref: Client for Various “CrossRef”
“APIs”. https​://cran.r-proje​ct.org/web/packa​ges/rcros​sref/index​.html. Accessed 4 December 2017
Crossref. 2018. Crossref REST API. https​://api.cross​ref.org/
Deschacht, N., & Maes, B. (2017). Cross-cultural differences in self-promotion: A study of self-citations
in management journals. Journal of Occupational and Organizational Psychology, 90(1), 77–94.
https​://doi.org/10.1111/joop.12162​.
Dion, M., & Mitchell, S. M. (2020). How many citations to women in “Enough”? Estimates of gen-
der representation in political science. PS: Political Science & Politics, 53(1), 107–113. doi.
org/10.1017/S1049096519001173.
Dion, M. L., Sumner, J. L., & Mitchell, S. M. (2018). Gendered Citation Patterns across Political Sci-
ence and Social Science Methodology Fields. Political Analysis, 26(3), 312–327. https​://doi.
org/10.1017/pan.2018.12.
Ferber, M. A. (1988). Citations and Networking. Gender & Society, 2(1), 82–89. https​://doi.
org/10.1177/08912​43880​02001​006.
Ferber, M. A., & Brün, M. (2011). The Gender Gap in Citations: Does It Persist? Feminist Economics,
17(1), 151–158. https​://doi.org/10.1080/13545​701.2010.54185​7.
Fowler, J. H., & Aksnes, D. W. (2007). Does self-citation pay? Scientometrics, 72(3), 427–437. https​://
doi.org/10.1007/s1119​2-007-1777-2.
genderize. (2018). genderize. https​://gende​r ize.io/.
Ghiasi, G., Larivière, V., & Sugimoto, C. R. (2016). Gender differences in synchronous and diachronous
self-citations. In STI conference, 8. Valencia.
Glänzel, W., Debackere, K., Thijs, B., Schubert, A. (2006). A concise review on the role of author self-
citations in information science, bibliometrics and science policy Scientometrics, 67(2), 263-–77.
Google Books. (2018). Google Books API. https​://devel​opers​.googl​e.com/books​/
Håkanson, M. (2005). The Impact of Gender on Citations: An Analysis of College & Research Libraries,
Journal of Academic Librarianship, and Library Quarterly | Håkanson | College & Research Librar-
ies. https​://doi.org/10.5860/crl.66.4.312
Hesli, V. L., & Lee, J. M. (2011). Faculty Research Productivity: Why Do Some of Our Colleagues Publish
More than Others? PS: Political Science & Politics, 44(2), 393–408. https​://doi.org/10.1017/S1049​
09651​10002​42
Hutson, S. R. (2006). Self-Citation in Archaeology: Age, Gender, Prestige, and the Self. Journal of Archae-
ological Method and Theory, 13(1), 1–18. https​://doi.org/10.1007/s1081​6-006-9001-5.
Hyland, K. (2003a). Self-citation and self-reference: Credibility and promotion in academic publication.
Journal of the American Society for Information Science and Technology, 54(3), 251–259. https​://doi.
org/10.1002/asi.10204​.
Hyland, K. (2003b). Self-citation and self-reference: Credibility and promotion in academic publication.
Journal of the American Society for Information Science and Technology, 54(3), 251–259. https​://doi.
org/10.1002/asi.10204​.
Hyland, K., & Jiang, F. (2018). Changing patterns of self-citation: cumulative inquiry or self-promotion?
Text & Talk, 38(3), 365–387. https​://doi.org/10.1515/text-2018-0004.
Key, E.M., & Jane Lawrence Sumner. 2019. “You Research Like a Girl: Gendered Research Agendas and
Their Implications.” PS: Political Science & Politics 52 (4): 663–68. https​://doi.org/10.1017/S1049​
09651​90009​45.
King, M. M., Bergstrom, C. T., Correll, S. J., Jacquet, J., & West, J. D. (2017). Men Set Their Own
Cites High: Gender and Self-citation across Fields and over Time. Socius, 3, 1–22. https​://doi.
org/10.1177/23780​23117​73890​3.

13
28 Scientometrics (2020) 125:1–28

Larivière, V., & Costas, R. (2016). How Many Is Too Many? On the Relationship between Research Pro-
ductivity and Impact. PLOS ONE, 11(9), e0162709. https​://doi.org/10.1371/journ​al.pone.01627​09.
Leahey, E. (2008). Methodological Memes and Mores: Toward a Sociology of Social Research. Annual
Review of Sociology, 34(1), 33–53. https​://doi.org/10.1146/annur​ev.soc.34.04050​7.13473​1.
Leahey, E., Beckman, C. M., & Stanko, T. L. (2017). Prominent but Less Productive: The Impact of Inter-
disciplinarity on Scientists’ Research*. Administrative Science Quarterly, 62(1), 105–139. https​://doi.
org/10.1177/00018​39216​66536​4.
Maliniak, D., Powers, R., & Walter, B. F. (2013). The Gender Citation Gap in International Relations. Inter-
national Organization, 67(04), 889–922. https​://doi.org/10.1017/S0020​81831​30002​09.
Mishra, S., Fegley, B. D., Diesner, J., & Torvik, V. I. (2018). Self-citation is the hallmark of productive
authors, of any gender. PLoS ONE, 13(9), e0195773. https​://doi.org/10.1371/journ​al.pone.01957​73.
Mitchell, S. M., Lange, S., & Brus, H. (2013). Gendered Citation Patterns in International Relations Jour-
nals. International Studies Perspectives, 14(4), 485–492. https​://doi.org/10.1111/insp.12026​.
Nielsen, M. W. (2016). Gender inequality and research performance: Moving beyond individual-merito-
cratic explanations of academic advancement. Studies in Higher Education, 41(11), 2044–2060. https​
://doi.org/10.1080/03075​079.2015.10079​45.
Ooms, J., Lang, D. T., & Hilaiel, L. (2017). jsonlite: A Robust, High Performance JSON Parser and Gen-
erator for R. https​://cran.r-proje​ct.org/web/packa​ges/jsonl​ite/index​.html. Accessed 4 December 2017
Snyder, H., & Bonzi, S. (1998). Patterns of self-citation across disciplines (1980–1989). Journal of Informa-
tion Science, 24(6), 431–435. https​://doi.org/10.1177/01655​51598​02400​606.
Teele, D. L., & Thelen, K. (2017). Gender in the Journals: Publication Patterns in Political Science. PS:
Political Science &amp; Politics, 50(2), 433–447. https​://doi.org/10.1017/s1049​09651​60029​85
Wais, K., VanHoudnos, N., & Ramey, J. (2016). genderizeR: Gender Prediction Based on First Names.
https​://cran.r-proje​ct.org/web/packa​ges/gende​rizeR​/index​.html. Accessed 4 December 2017.
Williams, H., Bates, S., Jenkins, L., Luke, D., & Rogers, K. (2015). Gender and Journal Authorship: An
Assessment of Articles Published by Women in Three Top British Political Science and International
Relations Journals. European Political Science, 14(2), 116–130. https​://doi.org/10.1057/eps.2015.8.

13

You might also like