RESEARCH ARTICLE - Specialized Research AI at MIT

RESEARCH ARTICLE

All downhill from the PhD? The typical impact
trajectory of U.S. academic careers

Mike Thelwall

and Ruth Fairclough

Statistical Cybermetrics Research Group, University of Wolverhampton, UK

a n o p e n a c c e s s

j o u r n a l

Keywords: academic careers, career trajectory, citation analysis, MNLCS, United States

Citation: Thelwall, M., & Fairclough, R.
(2020). All downhill from the PhD? The
typical impact trajectory of U.S.
academic careers. Quantitative Science
Studies, 1(3), 1334–1348. https://doi.
org/10.1162/qss_a_00072

DOI:
https://doi.org/10.1162/qss_a_00072

Received: 03 February 2020
Accepted: 17 May 2020

Corresponding Author:
Mike Thelwall
m.thelwall@wlv.ac.uk

Handling Editor:
Ludo Waltman

Copyright: © 2020 Mike Thelwall and
Ruth Fairclough. Published under a
Creative Commons Attribution 4.0
International (CC BY 4.0) license.

The MIT Press

ABSTRACT

Within academia, mature researchers tend to be more senior, but do they also tend to write
higher impact articles? This article assesses long-term publishing (16+ years) United States (U.S.)
researchers, contrasting them with shorter-term publishing researchers (1, 6, or 10 years). A
long-term U.S. researcher is operationalized as having a first Scopus-indexed journal article in
exactly 2001 and one in 2016–2019, with U.S. main affiliations in their first and last articles.
Researchers publishing in large teams (11+ authors) were excluded. The average field and
year normalized citation impact of long- and shorter-term U.S. researchers’ journal articles
decreases over time relative to the national average, with especially large falls for the last
articles published, which may be at least partly due to a decline in self-citations. In many
cases researchers start by publishing above U.S. average citation impact research and end by
publishing below U.S. average citation impact research. Thus, research managers should not
assume that senior researchers will usually write the highest impact papers.

INTRODUCTION

U.S. university departments seem to be managed by, and partly populated by, experienced
tenured researchers, who, if they continue publishing, also have greater access to resources
(e.g., Levitt & Levitt, 2017). In this context, and in conjunction with gaining knowledge through
long-term participation in a field, it would be reasonable to expect that experienced U.S.
academics would produce higher citation impact research. Knowledge about the extent to which
this is true would be useful for research managers deciding on the optimal balance of junior and
senior researchers or the types of activities that would make the best use of senior researchers’ time.

Previous studies of academic careers have tended to be qualitative or cover individual fields,
investigating different factors and producing divergent findings (Sugimoto, Sugimoto, Tsou, et al.,
2016). For junior researchers, effective mentoring has been shown to be helpful for long-term
publishing prospects in neuroscience and biomedical science (Liénard, Achakulvisut, et al.,
2018) and the same is true for early collaborations with highly cited scientists in four disciplines
(Li, Aste, et al., 2019). A comparison of 100 junior and 200 senior physicists suggested that bad
luck might prematurely terminate careers (Petersen, Riccaboni, et al., 2012). A survey of 624 U.S.
plastic surgeons found that those publishing more during their training also published more
afterwards (DeLong, Hughes, et al., 2014), with the same found for medicine, science, and tech-
nology at one Swedish university (Lindahl, Colliander, & Danell, 2020). Early research grants
moderately associate with higher career impact for three social science fields in the Netherlands
(Van den Besselaar & Sandström, 2015).

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

It seems likely that there are substantial overall differences in the trajectories of careers, asso-
ciated with, for example, international differences in the financial support for and growth-rate of
higher education (Finkelstein, 2015), as well as disciplinary differences in publishing rates
(Larivière & Costas, 2016), output types (Verleysen & Ossenblok, 2017), collaboration types
(Lewis, Ross, & Holden, 2012), research group culture (Pull, Pferdmenges, & Backes-Gellner,
2016), mentoring (Ooms, Werker, & Hopp, 2019), and gender (Fox & Stephan, 2001;
Thelwall, Bailey, et al., 2019). Finally, senior transportation researchers produce higher citation
impact papers (Hanssen & Jørgensen, 2015).

Highly cited researchers have been subjected to special attention for career factors. Analyses
of 450 highly cited scientists suggested that their reputation boosted the citation rates of their later
papers (Petersen, Fortunato, et al., 2014; see also Petersen, Jung, et al., 2011), and a logical con-
sequence of this is that average citation rates would increase with academic age for highly cited
researchers. Nevertheless, an investigation of highly cited scientists in seven disciplines did not
find a temporal pattern in the citation impact of their work (Sinatra, Wang, et al., 2016).

A range of career influences on citation impact have also been investigated for other sets of
researchers. Sociology, politics, or political science elite U.S. institution faculty CVs (n = 1,002)
have been used to analyze academic careers, finding that productivity (average number of journal
articles) was relatively constant, the proportion of collaboratively authored articles increased, and
average (arithmetic mean) citation rates decreased over time (not statistically significant) (Sugimoto
et al., 2016). Nevertheless, an analysis of 6,388 Quebec professors found that the average (arith-
metic mean) field normalized citation impact of their papers was lowest at age 50, when their pro-
ductivity (articles per year) also peaked (Gingras, Larivière, et al., 2008). Similarly, for Spanish
research council scholars, those scoring highest on a composite indicator based on a range of pro-
ductivity and citation indicators tended to be younger than average (Costas, van Leeuwen, &
Bordons, 2010). In contrast, although with different methods, in Mexico total citations over a
4-year period peaked at age 56 for researchers (González-Brambila & Veloso, 2007).

Second-order factors, such as collaboration, gender, mobility, or productivity, may influence
citation rates differently over time. The most researched age-related factor seems to be produc-
tivity rather than citations, however (e.g., Kyvik & Olsen, 2008; Mishra & Smyth, 2013; Rørstad &
Aksnes, 2015). The number of collaborators of a computer scientist or physicist increases with
academic age (Wang, Yu, et al., 2017), and collaboration overall in academia associates with
more citations (Larivière, Gingras, et al., 2015). Based on 375 U.S. researchers, senior chemists
seemed to maintain high productivity at the expense of reduced average citation impact, but in
mechanical engineering higher productivity associated with higher average research impact
(Kolesnikov, Fukumoto, & Bozeman, 2018). In Japan, older researchers write fewer papers,
perhaps because they have less time for research (Kawaguchi, Kondo, & Saito, 2016), but it is
not known if this influences their average citation impact. In the United States, female researchers
can expect to attract marginally more citations for their research (Thelwall, 2018b). The average
citation rate for physicists moving to less prestigious institutions was found to decrease in another
study, but the reverse was not true (Deville, Wang, et al., 2014).

Productivity is a key second-order factor for citation impact. As mentioned above, there is
conflicting evidence about whether highly cited researchers may attract a late career citation
boost (Petersen et al., 2014; Sinatra et al., 2016). Nevertheless, the Matthew effect suggests that
“established researchers” have an advantage when attracting funding, citations, and credit
(Merton, 1968). They would presumably also tend to publish more, due to increased resources,
opportunities for collaboration, and motivation. If such researchers are more likely to continue
publishing than others, then the citation impact and productivity of longer-term researchers may

Quantitative Science Studies

1335

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

tend to be higher. It is not clear how this would affect career trajectories for citations, because a
researcher gaining from the Matthew effect is likely to have published highly cited early work to
become established, and their “extra” Matthew effect citations could target their early or later
work. If researchers can become established through high productivity rather than high-quality
work (Merton, 1988), then the Matthew effect would create an additional connection between
productivity and average citation rates. For 48,000 author disambiguated Swedish researchers
based on four publication years (2008–11) in the Web of Science ( WoS), higher productivity
was associated with a higher probability for each publication to be more cited for its field and
year (more specifically, to fall in a higher field normalized citation class: Sandström & van den
Besselaar, 2016), supporting a positive association between productivity and citation impact.
This relationship was also found for an international set of 28 million WoS authors over 34 years
(1980–2013): More publications per year were associated with a higher probability for each one
to be in the top 1% cited for its field and year in the Medical and Life Sciences, Social and
Behavioral Sciences, and Natural Sciences, but not Law, Arts, and Humanities (Larivière &
Costas, 2016).

Some studies have analyzed complete or complete to date academic publishing careers at the
national level. For three universities in Australia, on average, researchers at the end of their
careers publish in higher impact journals than researchers at the start of their careers (Gu &
Blackmore, 2017, Figure 18), but it is not clear whether this is because less successful early career
researchers leave academia or stop publishing. The average (arithmetic mean) citation impact is
highest midcareer (Gu & Blackmore, 2017, Table 4), but the results are not field normalized.
There is a relatively small difference in the likelihood of being a first author between career stages,
although junior and senior researchers are more likely to be first authors than midcareer
researchers (Gu & Blackmore, 2017, Figure 16). Combining this with the results reported in pre-
vious paragraphs (younger researchers more cited in Spain and Quebec, older researchers more
cited in Mexico, all using different methods and samples), there is not a clear relationship
between average citation impact and either (academic or physical) age or career stage. The
above mentioned international WoS study found that the likelihood of an article being in the
top 1% cited for its WoS field was higher for authors with longer publishing careers (Larivière
& Costas, 2016). This could be a second-order effect of international differences, however, for
example if U.S. researchers had longer careers and published higher impact work.

This article assesses changes over time in the average citation impact of U.S. researchers based
on their publishing career duration. Based on the above discussion it is not known whether, in
general, research impact per paper tends to change during (publishing) careers and whether any
change is influenced by the duration of those careers. The focus is on the United States for
methodological pragmatism: The most robust career publishing data is available for this country,
as described in the methods.

(cid:129) RQ1: For long-term U.S. researchers, how does the average citation impact of their work

vary during their careers?

(cid:129) RQ2: Does the answer to the above question change for shorter-term researchers?

2. METHODS

The research design was to identify separate sets of long-term U.S. researchers at a given start
date through their Scopus IDs and publication record in Scopus, identifying any citation impact
changes over time. The United States is a suitable initial case study because it is a large country
with a research-intensive culture (hence a large amount of data) and international citation indexes

Quantitative Science Studies

1336

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

seem to have relatively extensive coverage of English-language publications and U.S. journal
publishers (Mongeon & Paul-Hus, 2016); coverage for researchers in advanced non-English-
speaking countries seems likely to be lower due to language issues (e.g., Kulczycki, Guns,
et al., in press). In addition, as it is a rich nation, successful researchers seem less likely to leave
for a better funded position elsewhere. For example, successful long-term researchers from poorer
nations might be tempted to move to the United States for higher salaries or better research
support (although researchers can collaborate internationally instead, e.g., Kwiek, in press).
Less developed countries are, in general, less attractive targets for researcher mobility (IDEA
Consult, WIFO, & Technopolis, 2017), with the United States being a particularly attractive place
to research (Janger, Strauss, & Campbell, 2013). One study of 12,502 full professors at elite U.S.
institutions found that 88% had U.S. PhDs, although 22% had moved to the United States after
their undergraduate degree (Yuret, 2018). This occurs despite extensive geographic mobility
within the United States (He, Zhen, & Wu, 2019).

Scopus was chosen in preference to the WoS for more comprehensive coverage (Mongeon
& Paul-Hus, 2016) and in preference to Dimensions (Thelwall, 2018a), Google Scholar
(Harzing & Alakangas, 2016), and Microsoft Academic (Harzing & Alakangas, 2017) due to
having a standard classification scheme that is known to be reasonably consistent (Klavans &
Boyack, 2017) and substantial long-term coverage. Scopus records from 1996 to 2013 were
downloaded in late 2018 and Scopus records from 2014 to 2019 were downloaded in early
2020, so every citation count used is based on at least 3 full years of citations. An open citation
window was used to give the maximum statistical power to the results. Only documents of type
Journal Article in Scopus were included, because these are the primary research outputs in most
fields. The exclusion of review articles is a limitation, as these can make important contributions
(Yadav, 2010), but it does not seem possible to accurately field normalize their citation counts
due to their relative scarcity. Similarly, the omission of conference papers, monographs, and
edited books is a limitation because they are central to some fields, but there do not seem to be
scholarly databases that can be connected to Scopus (in terms of author IDs) with a substantial
fraction of academic books.

Researcher track records were traced from 1996, after an expansion of Scopus, and a researcher
was assumed to have started his or her career in 2001 if they had not published any coauthored
articles in Scopus 1996–2000 but published at least one in 2001. This is an oversimplification,
because a researcher may have had a period of inactivity or may have produced other forms of
outputs 1996–2000. Nevertheless, it seems reasonable as a way of selecting researchers that
are likely to have started publishing in 2001. The number of years since a first publication is
sometimes called academic age (e.g., Milojevic(cid:1), 2012).

Scopus researcher IDs were used to track individual researchers throughout the period. These
are imperfect but seem to be highly accurate overall (Aman, 2018; Kawashima & Tomizawa, 2015;
Strotmann & Zhao, 2012). Intuitively, they may be least reliable for the start of a career, when
researchers can move between institutions after their PhD, between postdoctoral positions, and
to a permanent post. Thus, it is possible that long-term researchers (defined as below) some-
times have Scopus track records omitting their earliest work. Similarly, some short-term
researchers in Scopus (defined as below) may instead represent the career starts of longer-term
researchers. Each publication listed with an author’s ID in Scopus was assigned to that author.
The assignment process is imperfect and seems to be limited to 100 authors per paper, but seems
likely to be accurate for almost all papers, except perhaps in astronomy and high-energy physics,
where collaborations with hundreds or thousands of authors are common. This problem has been
ignored, so the results may include researchers that should have been excluded for participating
in large team collaborative research. This seems unlikely to make much difference to the results,

Quantitative Science Studies

1337

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

Quantitative Science Studies

because very large team papers seem to be rare and may represent a different type of activity for
researchers who also produce smaller team research.

A long-term researcher was defined to be one with a first Scopus-indexed publication in 2001
(to give a common start year, and prior publications checked back to 1996) and at least one
Scopus-indexed publication in 2016–19, to ensure that they were active for at least 16 years.
A researcher was classed as being from the United States if the affiliations of their first and last
publications were within the United States. This includes non-U.S. researchers that moved to the
United States for a PhD and then stayed there (or left and returned). The U.S. affiliation could be
for any type of organization, rather than just universities. This seems to be the most relevant group
from a U.S. policy perspective. Researchers were excluded if any of their publications had more
than 10 authors. This step was designed to exclude highly coauthored research for which
coauthorship may not be a good indicator of contribution. This affects highly collaborative areas
of genomics, astrophysics, and high-energy physics, for which coauthorship seems to be proce-
dural. Researchers were also excluded if they had published fewer than five articles, as these
would be relatively inactive from a publishing perspective and would therefore provide little
midcareer publishing evidence.

To test for the accuracy of the above method to identify each researcher’s first publication year,
a random sample of 100 of the matching long-term U.S. researchers (selected with a random
number generator) were searched for in Scopus by ID and their earliest publication date checked.
In 21 cases there was a Scopus publication before 2001, with the earliest being 1967. Thus, about
a fifth of the sample of long-term researchers have an earlier initial publication date than found by
the method used here.

Medium- and short-term researchers were defined as above, except that the last publication
had to be on a specified date so that the exact duration of their publishing career was known.
Medium term was defined to be 10 publishing years and short term 6 publishing years. In both
cases multiple starting years are allowed, beginning with 2001 and ending not after 2016, giving
multiple cohorts. The years 6 and 10 were chosen as illustrative intermediate values between 1
and 19 rather than for any theoretical reason. Researchers publishing a single journal article were
also analyzed as very short-term researchers (Table 1). The team size restriction (excluding
researchers ever authoring in teams of 11+) excluded most of the 15,329 long-term researchers,
giving a final data set size of 5,825. Of these, only 771 had their most recent journal article
published in 2016 (the remainder had articles published in 2017, 2018, or 2019).

The average citation impact of the publications produced by the chosen researchers in each
field and year was calculated using the Mean Normalized Log Citation Score (MNLCS) (Thelwall,
2017) variant of the Mean Normalized Citation Score (MNCS) (Waltman, van Eck, et al., 2011).
The MNLCS calculation log transforms all citation counts with ln(1 + c) because citation data is

Table 1. Cohort sizes for the main and supplementary data sets analyzed. Unless specified, researchers
are excluded if any of their publications have 11+ authors

Group
Long-term researchers

Average researchers per cohort
5,825

Medium-term researchers

Short-term researchers

Single-paper researchers

814

1,042

70,204

Min
5,825

585

711

Max
5,825

1,155

1,691

51,524

94,980

1338

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

skewed, and the results would otherwise reflect highly cited papers to some extent rather than
average behavior. The formula divides each logged citation count with the world average logged
citation count for the field and year of publication (or the average of the world averages for papers
assigned to multiple fields). Scopus narrow fields were used for the normalization, with 330 separate
fields in each year, giving fine-grained field normalization. Researchers producing more than one
paper in a given year contributed the average impact of these papers, rather than each paper sep-
arately. Researchers not producing any papers in a year were ignored for that year. Confidence
intervals were calculated for MNLCS values using the normal distribution formula, which is ap-
propriate because the log transformation greatly reduces the skewing in the citation count data.
The formula is likely to be a little conservative because some of the data points are averages over
multiple papers by the same researcher, reducing variation. Articles can be counted multiple times
if they have multiple qualifying authors, because the focus is on the average per researcher rather
than per author. The resulting MNLCS is above 1 only if the research has more impact than the
world average and can fairly be compared between years.

3. RESULTS

The main results are introduced first, with factors potentially influencing them discussed afterwards.

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

3.1. Citation Impact Trends Over Careers

Long-term U.S. researchers (16+ years of journal article publishing) begin with citation impact
above the U.S. (and world) average, but the citation impact of their articles declines steadily until
it is below average for the United States (but above world average) (Figure 1). Thus, although long-
term U.S. researchers tend to be able to produce high citation impact research, this capability

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Figure 1. Average citation impact of publications coauthored by U.S. long-term researchers (first
Scopus publication in 2001; at least one publication 2016–19; first and last publications with a U.S.
affiliation, at least five publications, ignoring researchers ever coauthoring in teams of 11+; n =
5,825 researchers). Error bars show 95% confidence intervals. The reference set is researchers with
first and last publications with a U.S. affiliation, ignoring researchers ever coauthoring in teams of
11+ (n = 643,204 researchers).

Quantitative Science Studies

1339

All downhill from the PhD?

apparently declines with age unless other factors explain the decrease. The U.S. average here is the
MNLCS for all publications from U.S. researchers (defined as having first and last publications with
a U.S. affiliation), irrespective of first and last publication date, except excluding researchers that
have ever coauthored in teams of 11 or more. The U.S. average reference set is therefore calculated
from the publications of comparable researchers.

As with long-term researchers, medium-term researchers (exactly 10 years of journal article
publishing, according to Scopus) tend to initially publish above-average citation impact journal
articles but their last articles have substantially lower citation impact on average (Figure 2). The
pattern is broadly similar irrespective of starting year. Confidence intervals are not shown (avail-
able in the supplementary material) but are wider than for Figure 1, so the occasional increases in
the line heights could be statistical anomalies.

Short-term researchers (exactly 6 years of journal article publishing, according to Scopus)
tend to initially publish above average citation impact journal articles but with average citation
impact decreasing over time (Figure 3). For consistency, the five-publication minimum require-
ment applies to Figure 3 and Figure 2, so the short-term researchers in Figure 3 have a higher
productivity requirement and may therefore be more capable, on average, than short-term
researchers without this requirement.

U.S. authors publishing only a single article indexed in Scopus tend to produce work that has a
low citation impact for the United States (Figure 4). Thus, the early career high citation impact for
long, medium and short-term researchers does not apply in this case.

3.2. Factors Associating with Citation Impact

Team size influences citation impact. If the restriction on the maximum team size for researchers is
removed then the average impact of long-term researchers is much higher (Figure 5; compare with

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Figure 2. Average citation impact of publications coauthored by four cohorts (585 to 1,155 researchers
per cohort) of U.S. medium-term (10 years of Scopus journal article publishing) researchers (first and last
Scopus publications as specified in legend; first and last publications with a U.S. affiliation, at least five
publications, ignoring researchers ever coauthoring in teams of 11+).

Quantitative Science Studies

1340

All downhill from the PhD?

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

Figure 3. As Figure 2 for six cohorts of short-term researchers (6 years of Scopus journal article
publishing; 711 to 1,691 researchers per cohort).

Figure 1) but still has a consistent downward trend in average citation impact after 2009, especially
compared to the new reference set without a team size restriction (which also has a higher average
citation impact) in Figure 5. Recall that this is a less reasonable graph because it includes re-
searchers who have only made contributions to large team studies, perhaps occasionally in a minor
role. The MNLCS dip in the reference set for 2014 and 2015 is presumably due to a change in the
journals indexed in Scopus that particularly influences highly collaborative publications.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Figure 4. Average citation impact of publications coauthored by U.S. researchers only ever pub-
lishing a single article in Scopus (ignoring researchers coauthoring in teams of 11+).

Quantitative Science Studies

1341

All downhill from the PhD?

Figure 5. As Figure 1, but without any team size restriction. The reference set is as in Figure 1 but
ignoring the team size restriction.

If the restriction on the minimum number of papers is removed (i.e., reduced from 5 to 2, to
accommodate the initial and final publications) then this has no effect on the results (Figure 6).

The extent to which a long-term researcher collaborates may vary during their career. Any
such changes could explain changes in average citation impact because more collaborative
articles tend to be more cited (Larivière et al., 2015). The overall trend is for the rate of collabo-
ration for long-term researchers to increase over time at a slightly higher rate than the U.S. average
(Figure 7), which does not explain the decreasing impact in Figure 1. It is perhaps surprising that

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Figure 6. As Figure 1, but without any productivity restriction (minimum two publications, one in
2001, one on or after 2016).

Quantitative Science Studies

1342

All downhill from the PhD?

Figure 7. As Figure 1, but for the average (geometric mean) number of authors per paper.

long-term researchers collaborate less than average researchers in the United States (a lower line
in Figure 7). This might be due to nonresearchers being occasionally added to larger collabora-
tion teams for specialist services (e.g., medical doctors advising a survey; a technician making a
particularly useful piece of equipment). Another possibility (suggested by a reviewer of this paper)
is that early career researchers may need to collaborate initially as they learn, but then leave

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Figure 8. Average citation impact of publications coauthored by U.S. long-term researchers sub-
tracting each researcher’s average over the period (first Scopus publication in 2001; at least one
publication 2016–19; first and last publication with a U.S. affiliation, at least five publications, ignor-
ing researchers ever coauthoring in teams of 11+). Error bars show 95% confidence intervals. The
reference set is researchers with first and last publication with a U.S. affiliation, ignoring researchers
ever coauthoring in teams of 11+, with the average MNLCS subtracted from each year.

Quantitative Science Studies

1343

All downhill from the PhD?

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

Figure 9. Average citation impact of publications coauthored by four cohorts of U.S. medium term
researchers, subtracting each researcher’s average over the period (first and last Scopus publication
as specified in legend; first and last publication with a U.S. affiliation, at least five publications,
ignoring researchers ever coauthoring in teams of 11+).

academia if they fail to develop as independent researchers. The results also do not change sub-
stantially if researcher productivity is factored out (Figure 8).

For medium-term researchers, after factoring out productivity, there is a decrease in average
impact for all cohorts between the first and subsequent publication years (Figure 9), and
the same applies for short-term researchers (Figure 10). Thus, impact drops after the first

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Figure 10. As Figure 9 for six cohorts of short-term researchers.

Quantitative Science Studies

1344

All downhill from the PhD?

publication year and at the end of publishing careers seem to be universal for researchers with
at least short-term careers.

Confidence intervals for all figures as well as data for additional figures are available in the

online supplementary materials: https://doi.org/10.6084/m9.figshare.11791059.

4. DISCUSSION

The results are limited by the data source (Scopus), the accurate but imperfect author identifi-
cation procedure used by Scopus, the limitation to Scopus-indexed journal articles, the analysis
of articles using whole author counting, the restriction to authors that have never published in
large teams (11+ authors), and the aggregate cross-discipline reporting. For Figure 1 and all
results with data starting from 2001, as 21% of authors had earlier publications than found by
the method here (before 1996), the trends may be stronger than shown in the graphs. This is
because the first year of publication tends to have the highest citation impact, so including pub-
lications from authors not in their first publishing year would tend to reduce the average impact
of the set. It would add to the robustness of the results if they could be replicated with a citation
database that is at least as large, differentiates different publication types, and includes author
IDs or enough information to systematically disambiguate author names, but no other database
satisfies these criteria. Lacking this, and taking into account the exclusions of reviews, confer-
ence papers, and academic books, weakens the strength of the evidence. Nevertheless, for the
trends identified to be incorrect, there would need to be systematic academic age-related factors
related to these issues that run strongly counter to the results reported above, which seems un-
likely. The trends for long-term researchers may also vary between cohorts, which has not been
tested for. There are also likely to be disciplinary differences in the trends, which have also not
been tested for. The citation impact trends should not be used to judge the wider value of the
work of the researchers. This is because citations only reflect one type of research impact and
academics can make contributions in ways other than publishing journal articles, such as
through education, mentoring, and management. For the shorter career paths, trained
researchers can also make valuable contributions to society outside of the academic publishing
model. The findings should also not be extrapolated beyond the United States. There are many
reasons why these findings are unlikely to replicate to all countries. On an international scale,
the U.S. system seems to be relatively competitive (Angermuller, 2017), as well as being a large,
rich nation with long research track record and relatively comprehensive journal coverage in
scholarly databases.

The results show that the average field normalized citation impact of U.S. researchers declines
over the course of their career, and particularly towards the end. After factoring out the produc-
tivity effect, there is also a sharp impact drop after the first publishing year. This applies to long-,
medium-, and short-term researchers. It is not affected by the restrictions on team size and produc-
tivity chosen to make the default analysis more useful (results are per paper, not per academic, so
publishing multiple papers in a year is not an advantage for the figures reported). It is not a side-
effect of collaboration, because collaboration increases over career length, and is known to as-
sociate with higher citation impact (Larivière et al., 2015). The overall decline agrees with a study
of three social sciences in the U.S. (Sugimoto et al., 2016) but sharply contrasts with the increase
in field normalized average citations after age 50 found in Quebec, Canada (Gingras et al., 2008)
either because of country or language issues or the use of the arithmetic mean in the latter study.
The overall decline also contrasts with the midcareer average citation peak in Australia, although
this data was not field normalized (Gu & Blackmore, 2017). As a previous study has shown
different career citation patterns in some respects for different disciplines in the United States

Quantitative Science Studies

1345

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

(Kolesnikov, Fukumoto, & Bozeman, 2018), the overall decreasing trend found here may well not
apply to all U.S. fields.

There are multiple possible explanations for the trends in the graphs and the apparently coun-
terintuitive finding that average research impact per article decreases over careers, including the
following.

(cid:129) Citation impact influence on career: Producing low-impact research, perhaps by acci-
dent, seems likely to directly influence the probability of an academic ceasing publishing.
This could be due to failing tenure, failing to get a new job, moving to a teaching-oriented
role, or taking on administration or managerial roles.

(cid:129) Career influence on citation impact: Low-impact articles might be more applied in nature
and therefore less cited (e.g., Moed, 2010), with the application signaling willingness to
leave academia for a preferred outside career or spin-off company. Speculatively, senior
researchers might (a) pursue mature research areas that are less cited, (b) mentor less
capable PhD students, producing lower impact coauthored work, (c) devote less effort to
their publishing after achieving career goals, or (d) author higher risk research with more
chance of creating a highly cited paper even though most papers attract few citations.
(cid:129) Networking effects: Assuming that a researcher ceasing publishing retires or moves to a
nonacademic job, their last publication may be less promoted by them to colleagues or
friends.

(cid:129) Technical factors: The last publication written by a researcher is the least likely to be self-
cited (coauthors may self-cite), losing a source of citations. Given that papers typically
have at least four authors, this seems unlikely to be a major factor. Nevertheless, reduced
self-citations seem likely to make some decrease in average citation impact at the end of
careers. Follow-up studies using data without self-citations would be useful to test this
hypothesis. Moreover, if Scopus IDs tend to detach the early parts of careers (PhD, post-
doctoral positions) from later positions, for example due to institutional moves, then the
earliest publications of longer-term researchers may have lower impact than suggested by
the data.

5. CONCLUSIONS

The career-long decline in average citation impact per article for the United States overall is a key
new finding, although it may be partly due to fewer self-citations to a researcher’s last output (but
coauthors may self-cite it). This effect for short-term or medium-term researchers is not concerning,
because they leave academic publishing, perhaps because they struggle to produce higher impact
work. The long-term researcher results are more important, however. Of course, the declining
average citation impact per article finding is a statistical average phenomenon and a substantial
minority of U.S. academics will follow more positive career trajectories.

The tendency for average citation impact per article to decline over careers for academics with
first and last journal articles published from the United States is a potential issue for academic
decision-makers. As journal article publishing may be a minor part of the role of senior
academics, this is not a personnel concern. Nevertheless, it suggests that policy focusing on
creating high-impact work should consider prioritizing junior and perhaps midcareer academics.
This adds new support to concerns previously raised about the domination of biomedical funding
by senior academics (Levitt & Levitt, 2017), for example. It also confirms the need for funding
programs targeting early career researchers, such as one from the Department of Energy
(science.osti.gov/early-career).

Quantitative Science Studies

1346

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

As this is apparently the first study with this analysis approach (all researchers in a country,
tracking cohorts from first to last publication, separately by year), it is important to follow up
the results by identifying national and disciplinary differences. Most importantly, the major
causes of the fall in average citation impact need to be identified so that the results can be properly
interpreted in context.

AUTHOR CONTRIBUTIONS
Mike Thelwall: Conceptualization, Investigation, Methodology, Visualization, Writing—original
draft, Writing—review & editing. Ruth Fairclough: Writing—original draft, Writing—review &
editing.

COMPETING INTERESTS

The authors have no competing interests.

FUNDING INFORMATION

This research was not funded.

DATA AVAILABILITY

The processed data used to produce the tables and figures are available in the supplementary
material (https://doi.org/10.6084/m9.figshare.11791059). A subscription to Scopus is required
to replicate the research, except with updated citation counts, with the methods described above.

REFERENCES

Aman, V. (2018). Does the Scopus author ID suffice to track scientific
international mobility? A case study based on Leibniz laureates.
Scientometrics, 117(2), 705–720.

Angermuller, J. (2017). Academic careers and the valuation of
academics. A discursive perspective on status categories and
academic salaries in France as compared to the US, Germany
and Great Britain. Higher Education, 73(6), 963–980.

Costas, R., van Leeuwen, T. N., & Bordons, M. (2010). A biblio-
metric classificatory approach for the study and assessment of
research performance at the individual level: The effects of age
on productivity and impact. Journal of the American Society for
Information Science and Technology, 61(8), 1564–1581.

DeLong, M. R., Hughes, D. B., Tandon, V. J., Choi, B. D., & Zenn,
M. R. (2014). Factors influencing fellowship selection, career tra-
jectory, and academic productivity among plastic surgeons. Plastic
and Reconstructive Surgery, 133(3), 730–736.

Deville, P., Wang, D., Sinatra, R., Song, C., Blondel, V. D., &
Barabási, A. L. (2014). Career on the move: Geography, stratifi-
cation, and scientific impact. Scientific Reports, 4, 4770.

Finkelstein, M. J. (2015). How national contexts shape academic
careers: A preliminary analysis. In U. Teichler & W. K. Cummings
(Eds.) Forming, recruiting and managing the academic profession
(pp. 317–328). Cham: Springer.

Fox, M. F., & Stephan, P. E. (2001). Careers of young scientists:
Preferences, prospects and realities by gender and field. Social
Studies of Science, 31(1), 109–122.

Gingras, Y., Larivière, V., Macaluso, B., & Robitaille, J. P. (2008). The
effects of aging on researchers’ publication and citation patterns.
PLOS ONE, 3(12), e4048. https://doi.org/10.1371/journal.pone.
0004048

González-Brambila, C., & Veloso, F. M. (2007). The determinants
of research output and impact: A study of Mexican researchers.
Research Policy, 36(7), 1035–1051.

Gu, X., & Blackmore, K. (2017). Quantitative study on Australian

academic science. Scientometrics, 113(2), 1009–1035.

Hanssen, T. E. S., & Jørgensen, F. (2015). The value of experience

in research. Journal of Informetrics, 9(1), 16–24.

Harzing, A. W., & Alakangas, S. (2016). Google Scholar, Scopus
and the Web of Science: A longitudinal and cross-disciplinary
comparison. Scientometrics, 106(2), 787–804.

Harzing, A. W., & Alakangas, S. (2017). Microsoft Academic is one
year old: The Phoenix is ready to leave the nest. Scientometrics,
112(3), 1887–1894.

He, Z., Zhen, N., & Wu, C. (2019). Measuring and exploring the geo-
graphic mobility of American professors from graduating institutions:
Differences across disciplines, academic ranks, and genders. Journal
of Informetrics, 13(3), 771–784.

IDEA Consult, WIFO, & Technopolis. (2017). MORE3 study:
Support data collection and analysis concerning mobility pat-
terns and career paths of researchers. https://cdn1.euraxess.org/
sites/default/files/policy_library/final_report_2.pdf

Janger, J., Strauss, A., & Campbell, D. (2013). Academic careers: A
cross-country perspective (No. 37). WWWforEurope Working
Paper. https://www.econstor.eu/ bitstream/10419/125692/1/
WWWforEurope_WPS_no037_MS64.pdf

Kawaguchi, D., Kondo, A., & Saito, K. (2016). Researchers’ career
transitions over the life cycle. Scientometrics, 109(3), 1435–1454.
Kawashima, H., & Tomizawa, H. (2015). Accuracy evaluation of
Scopus Author ID based on the largest funding database in Japan.
Scientometrics, 103(3), 1061–1071.

Quantitative Science Studies

1347

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d

f
/

1
3
1
3
3
4
1
8
7
0
0
8
8
q
s
s
_
a
_
0
0
0
7
2
p
d

b
y
g
u
e
s
t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

All downhill from the PhD?

Klavans, R., & Boyack, K. W. (2017). Which type of citation analysis
generates the most accurate taxonomy of scientific and technical
knowledge? Journal of the Association for Information Science and
Technology, 68(4), 984–998.

Kolesnikov, S., Fukumoto, E., & Bozeman, B. (2018). Researchers’
risk-smoothing publication strategies: Is productivity the enemy
of impact? Scientometrics, 116(3), 1995–2017.

Kulczycki, E., Guns, R., Pölönen, J., Engels, T. C., Rozkosz, E. A.,
Zuccala, A. A., Bruun, K., Eskola, O., Starc(cid:3)ic(cid:3), A. I., Petr, M., &
Sivertsen, G. (in press). Multilingual publishing in the social sciences
and humanities: A seven-country European study. Journal of the
Association for Information Science and Technology. https://doi.
org/10.1002/asi.24336

Kwiek, M. (in press). Internationalists and locals: International research

collaboration in resource-poor systems. Scientometrics.

Kyvik, S., & Olsen, T. (2008). Does the aging of tenured academic
staff affect the research performance of universities? Scientometrics,
76(3), 439–455.

Larivière, V., & Costas, R. (2016). How many is too many? On the
relationship between research productivity and impact. PLOS ONE,
11(9), e0162709. https://doi.org/10.1371/journal.pone.0162709
Larivière, V., Gingras, Y., Sugimoto, C. R., & Tsou, A. (2015). Team
size matters: Collaboration and scientific impact since 1900.
Journal of the Association for Information Science and Technology,
66(7), 1323–1332.

Levitt, M., & Levitt, J. M. (2017). Future of fundamental discovery in
US biomedical research. Proceedings of the National Academy
of Sciences, 114(25), 6498–6503.

Lewis, J. M., Ross, S., & Holden, T. (2012). The how and why of
academic collaboration: Disciplinary differences and policy
implications. Higher Education, 64(5), 693–708.

Li, W., Aste, T., Caccioli, F., & Livan, G. (2019). Early coauthorship
with top scientists predicts success in academic careers. Nature
Communications, 10(1), 1–9.

Liénard, J. F., Achakulvisut, T., Acuna, D. E., & David, S. V. (2018).
Intellectual synthesis in mentorship determines success in academic
careers. Nature Communications, 9(1), 1–13.

Lindahl, J., Colliander, C., & Danell, R. (2020). Early career perfor-
mance and its correlation with gender and publication output
during doctoral education. Scientometrics, 122(1), 309–330.
Merton, R. K. (1968). The Matthew effect in science: The reward
and communication systems of science are considered. Science,
159(3810), 56–63.

Merton, R. K. (1988). The Matthew effect in science, II: Cumulative
advantage and the symbolism of intellectual property. ISIS, 79(4),
606–623.

Milojevic(cid:1), S. (2012). How are academic age, productivity and col-
laboration related to citing behavior of researchers? PLOS ONE,
7(11), e49176. https://doi.org/10.1371/journal.pone.0049176
Mishra, V., & Smyth, R. (2013). Are more senior academics really
more research productive than junior academics? Evidence from
Australian law schools. Scientometrics, 96(2), 411–425.

Moed, H. F. (2010). Measuring contextual citation impact of scientific

journals. Journal of Informetrics, 4(3), 265–277.

Mongeon, P., & Paul-Hus, A. (2016). The journal coverage of Web
of Science and Scopus: A comparative analysis. Scientometrics,
106(1), 213–228.

Ooms, W., Werker, C., & Hopp, C. (2019). Moving up the ladder:
Heterogeneity influencing academic careers through research
orientation, gender, and mentors. Studies in Higher Education,
44(7), 1268–1289.

Petersen, A. M., Fortunato, S., Pan, R. K., Kaski, K., Penner, O., Rungi,
A., & Pammolli, F. (2014). Reputation and impact in academic
careers. Proceedings of the National Academy of Sciences, 111(43),
15316–15321.

Petersen, A. M., Jung, W. S., Yang, J. S., & Stanley, H. E. (2011).
Quantitative and empirical demonstration of the Matthew effect
in a study of career longevity. Proceedings of the National Academy
of Sciences, 108(1), 18–23.

Petersen, A. M., Riccaboni, M., Stanley, H. E., & Pammolli, F. (2012).
Persistence and uncertainty in the academic career. Proceedings of
the National Academy of Sciences, 109(14), 5213–5218.

Pull, K., Pferdmenges, B., & Backes-Gellner, U. (2016). Composition
of junior research groups and PhD completion rate: Disciplinary
differences and policy implications. Studies in Higher Education,
41(11), 2061–2077.

Rørstad, K., & Aksnes, D. W. (2015). Publication rate expressed by
age, gender and academic position—A large-scale analysis of
Norwegian academic staff. Journal of Informetrics, 9(2), 317–333.
Sandström, U., & van den Besselaar, P. (2016). Quantity and/or quality?
The importance of publishing many papers. PLOS ONE, 11(11),
e0166149. https://doi.org/10.1371/journal.pone.0166149

Sinatra, R., Wang, D., Deville, P., Song, C., & Barabási, A. L.
(2016). Quantifying the evolution of individual scientific impact.
Science, 354(6312), aaf5239.

Strotmann, A., & Zhao, D. (2012). Author name disambiguation:
What difference does it make in author-based citation analysis?
Journal of the American Society for Information Science and
Technology, 63(9), 1820–1833.

Sugimoto, C. R., Sugimoto, T. J., Tsou, A., Milojevic(cid:1), S., & Larivière, V.
(2016). Age stratification and cohort effects in scholarly communica-
tion: A study of social sciences. Scientometrics, 109(2), 997–1016.
Thelwall, M. (2017). Three practical field normalised alternative in-
dicator formulae for research evaluation. Journal of Informetrics,
11(1), 128–151.

Thelwall, M. (2018a). Dimensions: A competitor to Scopus and the

Web of Science? Journal of Informetrics, 12(2), 430–435.

Thelwall, M. (2018b). Do females create higher impact research?
Scopus citations and Mendeley readers for articles from five
countries. Journal of Informetrics, 12(4), 1031–1041.

Thelwall, M., Bailey, C., Tobin, C., & Bradshaw, N. (2019). Gender
differences in research areas, methods and topics: Can people
and thing orientations explain the results? Journal of Informetrics,
13(1), 149–169.

Van den Besselaar, P., & Sandström, U. (2015). Early career grants,
performance, and careers: A study on predictive validity of grant
decisions. Journal of Informetrics, 9(4), 826–838.

Verleysen, F. T., & Ossenblok, T. L. (2017). Profiles of monograph
authors in the social sciences and humanities: An analysis of pro-
ductivity, career stage, coauthorship, disciplinary affiliation and
gender, based on a regional bibliographic database. Scientometrics,
111(3), 1673–1686.

Waltman, L., van Eck, N. J., van Leeuwen, T. N., Visser, M. S., &
van Raan, A. F. (2011). Towards a new crown indicator: An em-
pirical analysis. Scientometrics, 87(3), 467–481.

Wang, W., Yu, S., Bekele, T. M., Kong, X., & Xia, F. (2017).
Scientific collaboration patterns vary with scholars’ academic ages.
Scientometrics, 112(1), 329–343.

Yadav, M. S. (2010). The decline of conceptual articles and implications
for knowledge development. Journal of Marketing, 74(1), 1–19.
Yuret, T. (2018). Path to success: An analysis of U.S. educated elite
academics in the United States. Scientometrics, 117(1), 105–121.

Quantitative Science Studies

1348

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
q
s
s
/
a
r
t
i
c
e
–
p
d