COMMENT

SPACE Astrophysics’ long HISTORY At last, a PSYCHOLOGY How a ECOLOGY Pathogen could wipe
relationship with the definitive biography of questionable personality out New Zealand’s oldest tree
military p.173 Helmholtz p.175 quiz went global p.176 species p.177
ILLUSTRATION BY DAVID PARKINS

The scientists who publish

a paper every five days
To highlight uncertain norms in authorship, John P. A. Ioannidis, Richard Klavans
and Kevin W. Boyack identified the most prolific scientists of recent years.

A
uthorship is the coin of scholar- that this could be a useful exercise in under- physics, projects are done by large inter-
ship — and some researchers are standing what scientific authorship means. national teams that can have upwards of
minting a lot. We searched Scopus We must be clear: we have no evidence 1,000 members. All participants are listed
for authors who had published more than that these authors are doing anything inap- as authors as a mark of membership of the
72 papers (the equivalent of one paper propriate. Some scientists who are members team, not for writing or revising the papers.
every 5 days) in any one calendar year of large consortia could meet the criteria for We therefore excluded authors in physics.
between 2000 and 2016, a figure that many authorship on a very high volume of papers. Of what remained, 909 author records
would consider implausibly prolific1. We Our findings suggest that some fields or were Chinese or Korean names. Because
found more than 9,000 individuals, and research teams have operationalized their Scopus disambiguates Chinese and Korean
made every effort to count only ‘full papers’ own definitions of what authorship means. names imperfectly, these may have wrongly
— articles, conference papers, substantive The vast majority of hyperprolific authors combined distinct individuals. For 2016
comments and reviews — not editorials, (7,888 author records, 86%) published (when disambiguation had improved for
letters to the editor and the like. We hoped in physics. In high-energy and particle Chinese and Korean names), at least

1 3 S E P T E M B E R 2 0 1 8 | VO L 5 6 1 | NAT U R E | 1 6 7
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 COMMENT

12, and possibly more than 20, authors similar to its share of published science. Ger- the European Prospective Investigation on
based in China were hyperprolific, the larg- many and Japan are over-represented. There Cancer and Nutrition; other large epidemio-
est number from any country that year. We were disproportionally more hyperprolific logical studies were also represented. Hyper-
believe that this could be connected to Chi- authors in Malaysia (n = 13) and Saudi Ara- prolific authors were also concentrated in
nese policies that reward publication with bia (n = 7), countries both known to incentiv- cardiology and crystallography.
cash or to possible corruption2,3. ize publication with cash rewards5. These biological and medical disciplines
Because of the disambiguation issues, we Hyperprolific authors also tended to cluster with many hyperprolific authors exhibit dif-
excluded these names from further analysis, in particular institutions, often as part of a ferent patterns from those found in particle
as well as group names and cases in which common study. For example, Erasmus Uni- and high-energy physics. Papers with hun-
we found errors (such as journalistic news versity Rotterdam in the Netherlands had dreds to thousands of authors are the norm
items misclassified as full articles), duplicate nine hyperprolific across a community of many thousands
entries, or conference papers misassigned to authors, more than “Whether and of scientists working in projects based at
an organizer. any other institu- how authorship CERN, Europe’s particle-physics laboratory
This left 265 authors (see Supplementary tion. Seven of them is justified near Geneva, Switzerland. In crystallogra-
Information). The number of hyperprolific co-authored mostly unavoidably phy, papers tend to have few co-authors. In
authors (after our exclusions) grew about papers related to the varies for each epidemiology and cardiology, long lists of
20-fold between 2001 and 2014, and then Rotterdam study, author and authors appear only in relationship to spe-
levelled off (see ‘Hyperprolific authors pro- a nearly 30-year- each paper, and cific research teams that seem to have a tra-
liferate’). Over the same period, the total old epidemiological norms differ by dition of extensive authorship lists.
number of authors increased by 2.5-fold. project, or its suc- field.” This raises the question of what
We e-mailed all 265 authors asking for cessor Generation authorship entails. The US National Insti-
their insights about how they reached this R study, which have followed multiple health tutes of Health, for example, has guidelines
extremely productive class. The 81 replies parameters in thousands of older adults and on the activities that qualify: actively super-
are provided in the Supplementary Informa- yielded thousands of publications. Five hyper- vising, designing and doing experiments,
tion. Common themes were: hard work; love prolific investigators from Harvard Univer- and data acquisition and analysis outside
of research; mentorship of very many young sity in Cambridge, Massachusetts, also often “very basic” work plus drafting the manu-
researchers; leadership of a research team, or co-authored papers related to cohort studies. script. Collecting funds or distant mentor-
even of many teams; extensive collaboration; Eleven hyperprolific authors across different ship do not qualify. Most of the 6,000 authors
working on multiple research areas or in core institutions were on one large cohort study, in a recent survey across many geographical
services; availability of suitable extensive regions and disciplines felt that drafting a
resources and data; culmination of a large paper, interpreting results and analysing data
project; personal values such as generosity S UR V E Y should qualify for authorship, but attitudes
and sharing; experiences growing up; and varied by region and field6.
sleeping only a few hours per day. Criteria fulfilled?
About half of the hyperprolific authors AUTHORSHIP CRITERIA
were in medical and life sciences (medicine One-third of the 81 authors identified Perhaps the most widely established
n = 101, health sciences n = 11, brain n = 17, as hyperprolific in 2016 replied requirements for authorship are the
biology n = 6, infectious diseases n = 3). When when asked how often they met Vancouver criteria established by the Inter­
we excluded conference papers, almost two- each of 4 criteria established for national Committee of Medical Journal
thirds belonged to medical and life sciences authorship of medical studies. Of the Editors in 1988. These specify that authors
(86/131). Among the 265, 154 authors pro- 27 responders, 19 admitted they had must do all of four things to qualify: play a
duced more than the equivalent of one paper not met at least 1 criterion more than part in designing or conducting experiments
every 5 days for 2 or more calendar years; 69 25% of the time. Eleven wrote that or processing results; help to write or revise
did so for 4 or more calendar years. Papers they had not met two or more criteria the manuscript; approve the published ver-
with 10–100 authors are common in these upwards of 25% of the time. sion; and take responsibility for the article’s
CVs, especially in medical and life sciences, contents.
but papers with the hundreds of authors seen ●●Substantial contributions to the The International Committee of Medical
in particle physics are uncommon. conception or design of the work; Journal Editors does not count supervision,
Materials scientist Akihisa Inoue, former or the acquisition, analysis or mentoring or obtaining funding as sufficient
president of Tohoku University in Japan and interpretation of the data for the work for authorship. We did observe that some
a member of multiple prestigious academies, (9 of 27 met this criterion in less than authors seemed to become hyperprolific on
holds the record. He met our definition of 75% of their papers). becoming full professors, department chairs
being hyperprolific for 12 calendar years or both. It is common and perhaps expected
between 2000 and 2016. Since 1976, his ●●Drafting the work or revising it for scientists who assume leadership roles in
name appears on 2,566 full papers indexed critically for important intellectual large centres to accelerate their productivity.
in Scopus. He has also retracted seven papers content (9 of 27 met this criterion in For example, clinical cardiologists publish
found to be self-duplications of previously less than 75% of their papers). more papers after they assume director roles
published work4. We searched for news (despite heavy clinical and administrative
articles in Google detailing retractions for ●●Final approval of the version to be duties). Occasionally, the acceleration is
the next 20 most hyperprolific authors and published (3 out of 27 met this criterion stunning: at the peak of their productivity,
found only one other author (Jeroen Bax) to in less than 75% of their papers). some cardiologists publish 10 to 80 times
have one retracted paper. more papers in one year compared with their
The 265 hyperprolific authors worked in ●●Agreement to be accountable for all average annual productivity when they were
37 countries, with the highest number in the aspects of the work (14 out of 27 met 35–42 years old. There was also often a sharp
United States (n = 50), followed by Germany this criterion in less than 75% of their decrease after passing the chair to a succes-
(n = 28) and Japan (n = 27). The propor- papers). sor. Another study noted similar patterns
tion from the United States (19%) is roughly two decades ago7.

1 6 8 | NAT U R E | VO L 5 6 1 | 1 3 S E P T E M B E R 2 0 1 8
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 COMMENT

One unexpected result was that some HYPERPROLIFIC 20 authors, another might give credit only
SOURCE: J. P. A. IOANNIDIS, R. KLAVANS & K. W. BOYACK

hyperprolific authors placed many publica- to 3 people or none. For example, genome-
tions in a single journal. Prominent in this
AUTHORS PROLIFERATE wide studies typically include many dozens
Numbers of authors with more than 72 papers
regard were Acta Crystallographica Section E: a year increased dramatically over time. of authors. As a dramatic counter-example,
Structure Reports Online (relaunched in 2014 one recent publication of a genome-wide
2002
as Section E: Crystallographic Communica- study had only one author9, and apparently
tions, with brief structural data reports now that researcher did the same amount of work
published in IuCrData) and Zeitschrift für = 1 author for which perhaps dozens would get author-
Kristallographie New Crystal Structures. Three 4 ship credit in similar papers spearheaded by
authors have each published more than 600 different teams. Some evidence suggests
articles in the former (Hoong-Kun Fun, Seik 2008 that the increase in the average number of
Weng Ng and Edward Tiekink); three authors authors per paper does not reflect so much
have each published more than 400 papers in the genuine needs of team science as the
the latter (Karl Peters, Eva Maria Peters and pressure to ‘publish or perish’10.
Edward Tiekink). Three other authors (Anne Widely used citation and impact metrics
Marie Api, Charlene Letizia, Sneha Bhatia) should be adjusted accordingly. For instance,
published many papers in single supplement if adding more authors diminished the credit
issues of Food and Chemistry Toxicology each author received, unwarranted multi-
focused on reviews of fragrance materials. 39 authorship might go down. We found that
Journals indexed in Scopus are generally the 30 hyperprolific authors who seemed to
2016
considered to be quality journals. The cita- benefit the most from co-authorship num-
tion impact of hyperprolific authors was bered 6 cardiologists and 24 epidemiologists
usually high, but there was large variability: (including those working on population
with a median of 19,805 citations per author genetics studies). (For these scientists, the
(range: 380 to 200,439). The median num- ratio of their Hirsch H index to their co-
ber of full papers per hyperprolific author in authorship-adjusted Schreiber Hm index was
2000–2016 was 677; across all hyperprolific higher; see Supplementary Information.)
authors, last author positions accounted for Overall, hyperprolific authors might
42.5%, first author positions for 7.1%, and include some of the most energetic and excel-
single authorships for 1.4%. Across the years, lent scientists. However, such modes of pub-
the median proportion of papers with mid- lishing might also reflect idiosyncratic field
81
dle author positions (that is, not a single, norms, to say the least. Loose definitions of
first or last author) was 51%, but varied from authorship, and an unfortunate tendency
2.1% to 98.5% for individual authors. meeting Vancouver criteria. to reduce assessments to counting papers,
Our work to identify hyperprolific authors Not all authors had approved the final muddy how credit is assigned. One still needs
is admittedly crude. It is mainly intended to versions of their own papers, but all consid- to see the total publishing output of each sci-
raise the larger question of what authorship ered approval of the final version necessary entist, benchmarked against norms for their
entails. Whether and how authorship is jus- for authorship. Fifty-nine per cent (16 of 27) field. And of course, there is no substitute for
tified unavoidably varies for each author said that they had contributed more than reading the papers and trying to understand
and each paper, and norms differ by field. any other listed author for 25 or more of the what the authors have done. ■
It is likely that sometimes authorship can be papers they authored in 2016.
gamed, secured through coercion or pro- Responses to the question “What, in John P. A. Ioannidis is a professor of
vided as a favour. We could not assess these your own words, do you think should be medicine at the Meta-Research Innovation
patterns in our data. We did not examine required for authorship?” generally reflected Center at Stanford (METRICS), Stanford
contributorship statements8, which are not a requirement for “significant contributions”, University, California, USA. Richard
archived in Scopus. Nevertheless, even con- but also dissatisfaction with how authorship Klavans and Kevin W. Boyack are
tributorship statements can be gamed and was assessed. One scientist said, “I person- researchers at SciTech Strategies in
might not be accurate. ally don’t count them as ‘my papers’ and Philadelphia, Pennsylvania, and
Further work is needed to explore how don’t have them on my CV as such, as there is Albuquerque, New Mexico, USA.
to best normalize these data and what is the a distinction between being a ‘named author’ e-mail: [email protected]
optimal level of normalization: for example, versus a ‘consortium member’ authorship.”
1. Wager, E., Singhvi, S. & Kleinert, S. PeerJ 3, e1154
adjusting for wide discipline, relatively nar- Another observed that authorship was often (2015).
row field and/or highly specific research team. awarded for seniority, and another that bet- 2. Quan, W., Chen, B. & Shu, F. Preprint at https://
ter distinctions were essential. “I think there arxiv.org/abs/1707.01162 (2017).
3. Hvistendahl, M. Science 342, 1035–1039 (2013).
WHAT AUTHORS SAY should be levels of authorship — and not 4. Nature 483, 246 (2012).
To better understand authorship norms, we those implied by order!” It will be interest- 5. Abritis, A., McCook, A. & Retraction Watch.
e-mailed a survey to the 81 hyperprolific ing to monitor how innovations in assign- Science 357, 541 (2017).
6. Patience, G. S., Galli, F., Patience, P. A. &
authors of 2016 (see Supplementary Infor- ing credit, such as data citation or formal Boffito, D. C. Preprint at bioRxiv https://doi.
mation). We asked whether they fulfilled author contribution taxonomies, could alter org/10.1101/323519 (2018).
all four Vancouver criteria. Of the 27 who authorship conventions. Authorship norms 7. Drenth, J. J. Am. Med. Assoc. 280, 219–221
(1998).
completed the survey, most said they did can vary within each field and even within 8. Sauermann, H, & Haeussler, C. Sci. Adv. 3,
not (see ‘Survey’). Almost all the responders each team. For example, some teams in epi- e1700404 (2017).
were from US and European institutions. demiology and cardiology apparently offer 9. Kim, S. K. PLoS ONE 13, e0200785 (2018).
The only two responders from elsewhere authorship more generously; others stick to 10. Papatheodorou, S. I., Trikalinos, T. A. & Ioannidis,
J. P. J. Clin. Epidemiol. 61, 546–551 (2008).
stated that they failed Vancouver criteria stricter (and probably more appropriate)
in most of their papers. It is likely that the authorship criteria. For a similar task and Supplementary Information accompanies this
survey underestimates the proportion not contribution, one cohort study might credit Comment online (see go.nature.com/2phadg2).

1 3 S E P T E M B E R 2 0 1 8 | VO L 5 6 1 | NAT U R E | 1 6 9
©
2
0
1
8
S
p
r
i
n
g
e
r
N
a
t
u
r
e
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.