Available via license: CC BY 4.0
Content may be subject to copyright.
Contents lists available at ScienceDirect
Research Policy
journal homepage: www.elsevier.com/locate/respol
The gender gap in early career transitions in the life sciences
☆
Marc J. Lerchenmueller
⁎
, Olav Sorenson
Yale School of Management, Yale University, 165 Whitney Avenue, New Haven, CT, 06511, United States
ARTICLE INFO
JEL classification:
O3 Research and Development
J44 Professional Labor Markets
J71 Labor Discrimination
Keywords:
Science careers
Gender gap
Productivity paradox
Differential returns
National Institutes of Health
ABSTRACT
We examined the extent to which and why early career transitions have led to women being underrepresented
among faculty in the life sciences. We followed the careers of 6,336 scientists from the post-doctoral fellowship
stage to becoming a principal investigator (PI) –a critical transition in the academic life sciences. Using a unique
dataset that connects individuals’National Institutes of Health funding histories to their publication records, we
found that a large portion of the overall gender gap in the life sciences emerges at this transition. Women become
PIs at a 20% lower rate than men. Differences in “productivity”(publication records) can explain about 60% of
this differential. The remaining portion appears to stem from gender differences in the returns to similar pub-
lication records, with women receiving less credit for their citations.
1. Introduction
Despite a narrowing of the gender gap, women remain under-
represented in the science, technology, engineering, and mathematics
(STEM) academic labor force. According to the National Science
Foundation, women earn about half of the doctoral degrees in science,
yet represent a mere 22% of the faculty at the full professor level at
Research I institutions in the United States (NSF, 2015). This continuing
gap, in part, reflects the fact that many of today's senior faculty received
their degrees thirty or more years ago. But that fact alone cannot ac-
count for this gap. Thirty years ago, women already accounted for more
than 30% of doctoral degrees earned in the life sciences (Hill et al.,
2010).
In attempting to explain this gap, a large body of research has
documented that women produce less measurable output than men.
Women, for example, publish fewer papers (Cole and Zuckerman, 1984;
Long, 1992; Xie and Shauman, 1998), the papers that they publish
appear in less prominent journals (Brooks et al., 2014;Lerchenmüller
et al., 2018) and receive fewer citations (Larivière et al., 2013; King
et al., 2016), and women receive the prestigious first and last author-
ships on co-authored articles less often (West et al., 2013; Filardo et al.,
2016). Although these differences in publication records may them-
selves stem from factors such as discrimination, disparity in the time
spent on childcare, or insufficient mentoring, to the extent that these
elements of the research record factor into hiring, promotion, and
funding decisions, one would expect fewer women to attain and retain
faculty positions. But, even when men and women have equivalent
research records, a parallel literature, based primarily on audit studies,
suggests that hiring and promotion committees still prefer men over
women (Steinpreis et al., 1999; Moss-Racusin et al., 2012).
We extend this literature on the gender gap in STEM faculty by
examining the extent to which disparate publication records versus
differential returns to similar records account for a critical early career
transition in the life sciences, from being a lab member to being a
principal investigator (PI). Because researchers in the academic life
sciences require substantial resources –equipment and personnel –for
their research, acquiring these grants has effectively become a pre-
cursor to being viable for tenure at a research-oriented university (Jena
et al., 2015).
This shift to analyzing the correlates of a critical career transition –
as opposed to identifying cross-sectional differences between men and
women in their publication records –forwards our theoretical under-
standing of the underrepresentation of women in STEM in at least two
respects. Most importantly, it examines whether differential publication
records could actually account for the gender gap. Most prior studies
have not been capable of disentangling cause from effect. The gender
gap at the faculty level might arise from women publishing fewer or less
prominent papers (Xie and Shauman, 1998). But the direction of
https://doi.org/10.1016/j.respol.2018.02.009
Received 10 February 2017; Received in revised form 6 February 2018; Accepted 13 February 2018
☆
The Initiative on Leadership and Organization at the Yale School of Management provided generous financial support. Marc Lerchenmueller also received funding through a grant
from the German Research Foundation (DFG; LE 3426/1-1). We thank Meshna Koren, from Elsevier, for facilitating our access to the Scopus API for extracting comprehensive citation
information, and Amandine Ody-Brasier and seminar participants at Copenhagen Business School, the European School of Management and Technology (ESMT), and the Yale School of
Management for helpful comments on earlier versions of this paper. The usual disclaimer applies.
⁎
Corresponding author.
E-mail addresses: marc.lerchenmueller@yale.edu (M.J. Lerchenmueller), olav.sorenson@yale.edu (O. Sorenson).
Research Policy 47 (2018) 1007–1017
Available online 16 April 2018
0048-7333/ © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
T
causality could run in the reverse direction: Women might have less
impressive publication records because they have not had the time and
resources for research that come from being senior faculty at research-
oriented institutions (Merton, 1968).
Second, our approach allows us to isolate whether –and if so, where
–men and women receive differential returns to their publication re-
cords. Although audit studies suggest that these differential returns
exist (e.g., Moss-Racusin et al., 2012), because those studies, by design,
hold constant all elements of the publication record, they cannot de-
termine whether women receive less credit for some specific element of
their research portfolios or whether the individuals evaluating appli-
cants simply have a preference for candidates of a particular gender
among those with equal qualifications.
Our analysis focuses on a set of similar men and women –those who
had received a postdoctoral (F32) training grant from the National
Institutes of Health (NIH). We examined the rates at which men and
women funded by those grants transitioned to being independent re-
searchers, becoming a PI on an NIH R01 grant, and the extent to which
their publishing records could account for those transitions. We first
document that the transition to being a PI on an R01 grant can explain a
substantial share of the gender gap in the life sciences. Women ex-
perienced 20% lower rates of transition than men. We then explored
what factors might account for this disparity. Adjusting flexibly for
differences in publication records could explain about 60% of this
gender gap. But even women with similar publication records received
R01 grants at lower rates than men. We then examined the extent to
which women might receive less credit for their publication records
(differential returns). These differential returns, particularly in the ex-
tent to which women benefited from citations, could account for the
remainder of the gap.
In addition to the theoretical implications of the results, our study
also contributes empirically to the literature on the gender gap in STEM
in at least two additional respects. First, most of the prior studies on
gender differences in productivity have analyzed samples of scientists
who received their doctoral degrees in the 1970s or earlier. We update
these findings by studying a sample of scientists who received their
degrees in the 1980s, 1990s, and 2000s, a period during which the gap
between the numbers of men and women enrolled in doctoral programs
in the life sciences closed (Hill et al., 2010).
Second, prior research has focused on the average differences in
publication records and on the linear effects of those differences on pay
or promotion. But many of the returns in science come from being in
the right-hand tail, to being unusually productive or producing research
of particular importance, to being perceived as a star (Merton, 1968).
We therefore introduce an empirical approach that allows us to capture
heterogeneity in the returns to the research record across the dis-
tribution of the various dimensions of that record. Doing so can explain
a substantial amount of additional variance. But the gender gap in the
transition to being a PI remains even allowing for these non-linearities.
2. Career transitions
In trying to understand why women remain underrepresented in
STEM fields, researchers have commonly characterized the process as
being similar to a pipeline with an almost continuous series of leaks
(e.g., Berryman, 1983; Etzkowitz et al., 2000; Lautenberger et al.,
2014). Although this view has been criticized as being overly linear and
insufficiently sensitive to the importance of social context outside of
school or the workplace (Xie and Shauman, 2003), research in this vein
has usefully documented the fact that the proportion of women in STEM
fields declines through the college years, during graduate school, and as
one considers ever more senior positions in these fields (Berryman,
1983; Shen, 2013; Lautenberger et al., 2014). Recent research suggests
that the gender gap in the pipeline emerges even before college, as high
school students begin to form their career ambitions and expectations
(Morgan et al., 2013; Legewie and DePrete, 2014).
However, this pipeline view obscures the fact that most of the loss of
women appears to occur within a short segment of the career, and one
relatively far down the line. Consider the academic life sciences, the
largest among the STEM fields: Women have reached near parity in
both of the primary paths for entry, having a medical degree or a
doctorate in a life sciences field (Lautenberger et al., 2014; Shen, 2013).
They still appear almost equally represented in residency and post-
doctoral training positions in research laboratories (Lautenberger et al.,
2014; NPA, 2011). Yet, women hold only 40% of assistant professor-
ships and no more than 30% of associate professorships in the life
sciences (Jena et al., 2015). Their underrepresentation in the field
emerges in the space of only two to ten years out of a career of forty or
more. Returning to the pipeline analogy, it is less that the pipe drips
continuously along the way and more that it is gushing at one or two of
the joints between segments.
Given this fact, we see value in shifting the focus of analysis to
understanding these critical career transitions where the gap widens
most rapidly –in this case, on the transition to becoming an in-
dependent researcher in the life sciences. Individuals who complete a
relevant graduate degree –a medical degree (MD) or a doctorate (PhD)
–first move into a junior faculty position, either directly or following
post-doctoral training. Because of the increasingly expensive nature of
research in the life sciences, junior faculty must then find a means of
funding their research. That usually means winning a major grant.
Those who fail to do so have low odds of securing long-term (tenured)
academic positions.
One can readily see from the much lower proportion of women at
the associate professor level relative to the assistant professor level that
women clear these hurdles at lower rates. What might account for
differences in the transition rates experienced by men versus by
women? We focused on two potential disparities: differences in pub-
lication records and differences in the returns to those publication re-
cords.
2.1. The productivity paradox
In academia as in many other settings, productivity represents an
important determinant not only of who gets hired but also of who gets
promoted. Given the up-or-out nature of the tenure-track job ladder,
moreover, it also determines who remains in academia.
Productivity in academia, particularly in the sciences, means pub-
lications. Much attention therefore has been given to gender differences
in publication records, the so-called “productivity paradox”(Cole and
Zuckerman, 1984). Women publish fewer articles than men (Cole and
Zuckerman, 1984; Long, 1992; Stack, 2002), and place them in less
prominent outlets (Brooks et al., 2014;Lerchenmüller et al., 2018).
Articles written by women, moreover, receive fewer citations, an im-
portant metric used to assess the influence of scientific research
(Larivière et al., 2013).
On the articles they do publish, women appear in less prestigious
authorship positions (Jagsi et al., 2006; Filardo et al., 2016). In the life
sciences, the first and last authorships carry particular prestige. By
convention, the individual who led the research and who analyzed and
wrote up the results receives the first authorship. Last authorship goes
to the head of the laboratory, who often receives credit not just for
funding the research but also for conceiving of it. Interior authorships,
meanwhile, go to those who assisted with data collection or analysis.
Although women have reached parity in their probability of appearing
in the first author position (West et al., 2013), this average belies the
fact that women remain less likely to receive this prime position on
articles published in the most prestigious journals (Lerchenmüller et al.,
2018).
Overall, the reasons for these “productivity”differences remain a
puzzle. Women may suffer discrimination both in the research lab and
in the publication process, with consequences for their publication re-
cords. They may also find themselves with less time for research, either
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1008
because they engage in more non-research activities at work or because
they must shoulder a disproportionate share of the responsibilities at
home (e.g., Craig and Mullan, 2011). Women may also choose different
research paths. Leahey (2007), for example, has argued and provided
evidence that women specialize less than men. Since specialization can
allow researchers to produce more articles and increases the odds that
they receive attention from others active in the field, it could account
for multiple aspects of the productivity paradox.
But the productivity paradox may also come from comparing apples
to oranges, or at least shoots to plants. Most of the studies on the gender
gap in productivity have examined cross sections of authors or articles,
pooling individuals across all career stages. If publication and citation
rates rise over time and if fewer women transition to senior positions
(perhaps due to bias in the evaluation process), then the average
woman would occupy an earlier career stage than the average man in
the population and one would observe these gender gaps in the cross-
section even if men and women at the same career stages had equiva-
lent publication records.
However, to the extent that productivity differences between men
and women do appear early in their careers, one could see how such
easily quantifiable differences in publication records could lead to dif-
ferential rates of hiring, grant awarding, and promotion for men and
women –regardless of whether these differences emerge from dis-
crimination, from disparities in the allocation of parenting and other
responsibilities, or from differential choices in their research agendas.
We nevertheless have little direct evidence regarding the extent to
which differences in publication records might account for critical ca-
reer transitions.
2.2. Undervalued research records
Although academia ostensibly operates as a meritocracy, at least
two lines of research suggest that differences in productivity might not
account for the paucity of senior women on science faculties. First, in a
series of audit studies, researchers have sent out equivalent resumes or
curriculum vita, altering only the names of the candidates to signal the
gender of the individual. Steinpreis et al. (1999), for example, ma-
nipulated the names of applicants for an assistant professor position in
psychology and found that psychologists preferred candidates with
stereotypically male names over those with female names. Moss-
Racusin et al. (2012) repeated this design more recently for candidates
for a lab manager position and again found a preference for applicants
with male names.
Although these audit studies suggest that women receive lower re-
turns to the same research records, this evidence remains inconclusive.
On the one hand, the design of these studies holds constant every ele-
ment of the research record. The same pattern of results would emerge
even if the individuals screening the applications had only a slight
preference for candidates of a particular gender among those equally
qualified. On the other hand, the results of these studies have also not
been consistent. Williams and Ceci (2015), for example, using the same
study design, found that faculty preferred assistant professor candidates
with female names over those with male names in every field studied,
except for economics.
In a second line of research, a small number of studies have ex-
amined promotion rates and found residual effects for gender even after
controlling for the number of publications. Long et al. (1993),Leahey
et al. (2010), and Lutter and Schröder (2016) for example, have re-
ported gender differences in the rates of promotion to tenure among
biochemists, American sociologists, and German sociologists, respec-
tively. After adjusting for the number of publications (and sometimes
other dimensions of the publication record), these studies find lower
promotion rates for women relative to men. These studies, however, do
not provide direct evidence for the proportion of the overall gender gap
that might stem from the productivity paradox because they have either
entered gender in their regressions after or simultaneous to their
measures of publication records, meaning that one cannot assess the
extent to which adjusting for publication records might have narrowed
the gender gap.
These literatures nonetheless suggest the possibility of differential
returns –that women receive less credit for equivalent publication re-
cords. These differential returns could emerge in at least a couple of
ways. Evaluators may simply place less value on the articles written by
women or on the citations received by them. Such a pure form of dis-
crimination would obviously place women at a disadvantage in selec-
tion and promotion processes. Or, it may reflect a preference for similar
others. Research in social psychology has found that both men and
women tend to evaluate same-sex individuals more favorably for si-
milar levels of performance than individuals of the opposite sex
(Greenberg, 1978). Given that men still account for the majority of
evaluators –such as editors and grant application reviewers –in the life
sciences and elsewhere, both forms of discrimination seem plausible.
But one could also imagine a more subtle dynamic. Perhaps the
articles and citations themselves receive the same weight regardless of
the gender of the authors but the allocation of credit for those articles
differs systematically across men and women. Modern science has be-
come a team sport, with ever larger groups of scientists involved in
research projects (Wuchty et al., 2007). In mixed gender research
groups, readers may perceive the men on the team as having con-
tributed more to the research than the women. Consistent with this
idea, Sarsons (2017) found that men in economics benefited much more
from coauthored articles than women did, in terms of their odds of
being promoted to tenure, and that this disparity appeared largest for
mixed-gender coauthorships.
3. NIH funding and life science careers
The NIH is the largest funder of life science research in the United
States, with an annual budget of roughly $30 billion (NIH, 2016a)–
more than four times that of the National Science Foundation. The NIH
supports intramural (NIH executed) and extramural research, with
more than 80% of its budget going to the latter through competitive
grants awarded to individuals and institutions, primarily in the United
States.
A useful feature of the life sciences for our research is the fact that
academic research in this field operates as a “soft money”environment
in the United States. Rather than being guaranteed salaries and research
funds by their universities, life scientists must compete for grants to
fund their own positions and to finance their resource-intensive re-
search projects. The funding trail therefore provides a good means for
assessing the relationship between publication records and career ad-
vancement in the life sciences.
We focused on the receipt of the first R01 grant. The R01, a project-
based renewable research grant awarded to scientists who have de-
monstrated research competence in a specific area (Azoulay et al.,
2011), serves as the primary funding mechanism for the NIH. Although
some other programs support the research of independent investigators,
no other program comes close in importance to the R01.
1
These grants
account for almost half of all NIH grant dollars and they represent the
primary funding source for most academic biomedical research groups
in the United States. There are about 27,000 outstanding awards, with
roughly 4,000 new ones approved each year. Each award provides an
average of $1.7 million in support spread over three to five years (Li,
2017).
Researchers typically receive their first R01 around the age of 42. It
represents an important milestone in their careers, as both a financial
enabler and as an indicator of their ability to conduct research in-
dependent from a more senior scientist (Garrison and Deschamps,
1
Since these other programs may have somewhat different selection criteria, we re-
stricted our analysis to the receipt of an R01.
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1009
2014). As part of its charge to develop the biomedical research work-
force, the NIH has a longstanding commitment to identifying and
supporting promising young scientists on the way to independence.
Since the 1970s, the NIH has sought to identify New Investigators,
applicants who have not previously received an R01. The NIH segre-
gates their applications into a separate pool, reviewing them relative to
other early career scientists. NIH policy, moreover, requires the agency
to award grants to new and experienced principal investigators at
comparable rates (NIH, 2011).
One of the difficulties in almost any research on transitions involves
the definition of the set at risk.
2
Defining the population at risk too
narrowly precludes the researcher from gaining insight into crucial
intermediate stages in the process. Defining it too broadly increases the
odds that individuals differ in meaningful ways not captured in the
covariates. Consider, for example, the transition to tenured faculty. At
one extreme, one might want to follow all who completed a relevant
doctoral program as being at risk. But doing so would then include
many who had no interest in an academic career. At the other extreme,
one might limit the sample to assistant professors. That restriction,
however, would exclude all of those who had pursued academic posi-
tions but failed to obtain one.
To ensure that our sample included a relatively homogeneous group
of individuals while still encompassing the crucial period when much of
the gender gap emerges, we focused our analysis on individuals selected
by the NIH to receive an F32 postdoctoral fellowship award.
Established in 1974, the Kirchstein National Research Service Award
(NRSA) Fellowship program (or F grant mechanism) represents the only
means by which the NIH directly supports the basic preparation of in-
dividuals for careers in biomedical research (Mantovani et al., 2006).
The F32 grant, by far the most common of these grants, targets scien-
tists in their early postdoctoral years, with the average recipient being
about 32 years old. The fellowship offers up to three consecutive years
of mentored research support, with an average annual grant size of
about $50,000 (Jacob and Lefgren, 2011).
3
The support packages in-
clude a stipend, tuition support, and an allowance to defray other
miscellaneous costs related to research training (Mantovani et al.,
2006). Fig. 1 situates the F32 postdoctoral fellowship award and the
R01 mechanism within a typical life science career and reports the
approximate proportion of women at each career stage (NPA, 2011
Lautenberger et al., 2014; Jena et al., 2015).
The F32 fellowship identifies individuals likely to pursue scientific
careers. F32 grant recipients have a demonstrated interest in and
commitment to pursuing an academic career. About two-thirds of F32
fellows remain employed in academia eight years after the completion
of the fellowship. To ensure further that our analysis does not include
individuals who have no or limited interest in pursuing an academic
career, we exclude from the risk set F32 recipients who did not produce
a single publication during their F32 fellowship period (roughly four
out of ten F32 recipients). A gender gap in this cohort of committed and
nationally-competitive individuals should therefore not reflect differ-
ences in the careers that men and women would prefer to pursue.
4. Gender gap in funding
We used a January 2016 download from the NIH ExPORTER data-
base to track scientists. The database includes the names of funded
scientists and a unique identifier (ID) assigned by the NIH to each
scientist. Because grant applicants must use their assigned ID in all
subsequent NIH grant applications, with failure to do so punishable by
disqualification and potentially by federal law, these identifiers have
extremely high fidelity across grants. The database, which covers the
period running from 1985 to 2015, records grant budget periods, areas
of research inquiry, and publications citing the grant, as well as other
information.
We used the forenames of the funded scientists to infer their gender,
using the Genderize database (Lerchenmueller, 2016). Genderize as-
sociates a name with the probability of being a man or a woman based
on the occurrence of that name in a number of official sources, such as
the Social Security Administration records, and in social media sources
that verify the gender of the users (Wais, 2015).
4
For example, the
database designates the forename “Chris”as male with 93% confidence,
based on 8,631 verified records. For our analyses, we only included
cases where the algorithm assigned a 90% or greater probability to the
individual being of a specific gender. Using this confidence threshold,
we could assign a gender to 88% of the F32 recipients in our download
of the NIH ExPORTER database.
At the postdoctoral level, men and women differ somewhat in the
financial support that they receive from the NIH, with men accounting
for roughly 60% of these awards. Fig. 2 depicts the percentage shares of
women receiving F32 awards and R01 grants since 1985. Although
women still receive fewer F32 awards than men, their proportion has
climbed in lockstep with the proportion of women obtaining terminal
Fig. 1. Career stages, gender representation, and typical timing of first F32 and R01 award.
2
The NIH does not release data on unsuccessful grant applications and, even if it did,
defining the risk set as those who had applied might prove too narrow.
3
F32 fellows have a strong incentive to complete at least two years of supported
training, as the NIH Revitalization Act of 1993 specifies that recipients must reimburse
any support if they leave the fellowship prior to the completion of two years.
4
Genderize has an advantage over other databases primarily in its scope, incorporating
name data on 216,286 unique names from 79 countries and 89 languages. Comparisons of
Genderize to other automated algorithms have found that it provides the most accurate
gender assignments (Coding News, 2015).
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1010
degrees in the life sciences. The remaining differential has largely been
a function of the fact that somewhat fewer women apply for these
grants. Conditional on application, the success rates do not differ by
gender (Pohlhaus et al., 2011). Men and women also receive roughly
the same levels of funding, with average award amounts differing by a
mere $353 (not statistically significant).
But men and women receiving F32 awards have much different
trajectories following this post-doctoral training. Fig. 2, which depicts
the proportions of F32 and R01 awards going to women over time,
reveals a substantial gender gap in R01 awards, which lags that in post-
doctoral awards by at least 20 years. Table 1 reports the grant transition
matrix for F32 recipients by gender, focusing on those who received
their awards before 2006 (to allow for at least ten years to observe
transitions). The R01 award represents the most common source for
future funding, irrespective of gender. Note also that the other Research
Program Grants, which account for many of the other transitions, often
serve as intermediate awards on the path toward receiving an R01.
Overall, men secure follow-on funding at an eight percentage point
higher rate than women. The gender gap in funding levels stems almost
entirely from this disparity in transition rates, as men and women re-
ceive awards of roughly equivalent sizes (with R01s for men having
annual budgets of only $2,412 more than for women, a difference of
less than half of one percent).
Fig. 3 provides a sense of when these differences emerge, depicting
Kaplan–Meier estimates of the cumulative transition probabilities for
men (blue) and women (red).
5
Over 60% of the men and 70% of the
women did not transition to an R01 grant during our observation
window. Women who received grants also received them later in their
careers, on average. At every point in time after the fifth year from the
F32 receipt, a smaller proportion of women than of men have received
their first R01. For example, 10 years after the receipt, 25% of the men
had received an R01 grant compared to only 17% of the women. These
unconditional transition rates, however, do not say anything about why
the gap widens.
5. Correlates of the gender gap
To assess the extent to which publication records versus differential
returns to those records might account for the gender gap, we con-
nected the grant data to article-level data from the PubMed database
(for further details of the sample construction, see Appendix A). In total,
our sample for estimation comprised 6,336 F32 recipients (60% male
and 40% female).
Our data set consists of one observation per publication per person.
The F32 budget start date served as the beginning of the time at risk and
we considered the R01 budget start date the time of the transition
event. If a scientist did not receive an R01 grant by the end of 2009, we
considered the person right-censored. Our data set includes 74,188
publication-person observations (11.7 publications per person on
average), covering 68,834 person-years (10.9 years on average).
We created a number of variables to capture different dimensions of
publication records. We included a (logged) count of the number of ln
(articles) on which individuals had been listed as an author. Because the
norms in the life sciences assign the first author position to the person
responsible for leading the execution and reporting of the research, one
might expect that first authors would receive more credit for any
publications. We therefore calculated a Percent (first author) variable to
capture this effect.
6
We also incorporated measures of the importance of these pub-
lications. Having been a coauthor on a publication in a leading journal
may count more than having an article in a less prominent outlet. We
therefore computed the proportions of publications appearing in jour-
nals with 5-year journal impact factors (JIF) of over five and up to ten
(Percent (JIF 5-10)) and exceeding ten Percent (JIF >10). The first
category includes a number of important field journals, while the latter
encompasses the most prominent journals in science and medicine. The
proportion of publications in journals with JIFs of five or less served as
Fig. 2. Women's representation across NIH grant programs 1985–2009.
Table 1
Transition matrix for F32s (1985–2005).
Men(%) Women(%)
Prior grant
No 98 96
Yes 2 4
Prior type
F30-31 90 92
Other 10 8
Post grant
No 60 68
Yes 40 32
Post type
R01 65 61
R
a
26 28
Other 9 11
N8,140 5,487
a
One of 23 other R-mechanism grants (excluding R01 mechanism).
Post grants considered up to fiscal year 2015.
Fig. 3. Kaplan–Meier survival rate estimates.
5
Since we have a continuous clock, we used the asymptotic variance estimate to derive
confidence intervals for
St
ˆ()
(Kalbfleisch and Prentice, 2002).
6
We used proportions for these variables rather than raw counts to reduce collinearity
with the number of publications.
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1011
the baseline category.
Our models also included the average number of citations, ac-
cording to Scopus (a database maintained by Elsevier), received by all
articles published by the individual up to that point in time. Citation
counts have often been used as an article-level metric of research
quality but even if other factors influence them they clearly capture the
attention received by the research. Our models included the (logged)
average number of citations received by all articles published by the
individual (ln (avg. citations)).
7
To ensure that our data reflect the
number of citations received up to a particular point in time, this
variable has been calculated at the time of each publication.
These dimensions of the research record align well with the criteria
by which the NIH claims to evaluate proposals. The agency suggests
that evaluators should use five criteria: (1) Does the applicant have a
record of accomplishments in the field? (2) Will the scientist convert
funding into research output? (3) How significant is the proposed re-
search? (4) How innovative is it? And, (5) does the researcher operate
in an environment that will support the research? The final two criteria
seem least connected to these measures of publication records, though
the prominence of the journal and the number of citations received
often reflect some combination of quality, significance, and innova-
tiveness. Note also that our models control for the institution's ag-
gregate success in securing R01 grants, adjusting explicitly for the in-
stitutional environment (the fifth criterion).
5.1. Estimation
To understand better the factors underlying gender differences in
the transition to principal investigator, we turned to parametric sur-
vival analysis. These models have the advantage of exploiting in-
formation on both the occurrence and the timing of events.
We began by estimating a non-parametric baseline hazard rate,
without any covariates:
=+>> >
→
ht t tTtTt
t
() limPr ( Δ | )
Δ
,
tΔ0 (1)
where Trepresents a random variable for the time of R01 receipt
and tdenotes the amount of time that has passed since individual ihas
received the F32 award. We estimated this function using kernal
smoothing, averaging values of the function over a moving window.
8
Fig. 4 displays these unconditional hazard rates for men and for
women. The rate at which these scientists received their first R01
awards peaked at about eight years after their F32 awards. Interest-
ingly, the hazard rate of receiving an R01 grant rose more steeply for
men than for women. Irrespective of gender, however, the hazard rate
had a non-monotonic relationship with t,reflecting the fact that the
transition to R01 becomes increasingly less likely if it does not happen
within eight years of the F32. Beyond 15 years, the hazard rate falls
almost to zero. We therefore limited our observation window to the 15
years following the receipt of an F32.
Given the non-monotonic shape of the hazard function, we chose a
log-logistic form of time dependence.
9
We estimated time to first R01
grant with an accelerated failure-time (AFT) model. One can interpret
the exponentiated coefficients of these models as time ratios, with va-
lues in excess of one indicating a delay in the arrival of the event and
values below one reflecting an acceleration of the arrival rate.
The models also include a number of control variables. We account
for other sources of funding with a count of the number of non-R01
grants received from the NIH since the receipt of the F32 (interim
grants). We also included an indicator variable for whether the in-
dividual had received an NIH grant prior to the F32 (prior grants) and a
count of the number of grant extensions.
10
We also accounted for the
potential effect of the number of articles published prior to F32 receipt
(ln (prior articles)).
Because the composition of the research teams of which individuals
have been a member may influence the allocation of credit, we calcu-
lated a covariate for the average number of coauthors on an article (avg.
team size). Each coauthor probably receives less credit for research
produced by a larger team. We also computed the percent female authors
across all publications as an additional control variable.
Past research outside the life sciences has found that academics with
more specialized research agendas appear to enjoy greater success (e.g.,
Leahey, 2006). We therefore included a measure of specialization,
based on medical subject header (MeSH) terms (keywords). We calcu-
lated a Herfindahl-index measure of specialization, summing the
squared proportions of MeSH terms associated with an individual's ar-
ticles. Because this measure correlates strongly with the number of
articles, we regressed this measure on the (logged) number of articles
and used the residuals from that regression as our measure of speciali-
zation.
We also accounted for the strength of the institutional environment
by including the institution's percentile rank in terms of its number of
NIH R01 awards. Finally, we included fixed effects for area of research
inquiry (using the two-digit letter code for the supporting NIH
Institute/Center embedded in the F32 grant number) and for grant
vintage (dividing F32 grants into five year cycles, 1985–2009) to con-
trol for field and period effects, such as changes in the NIH budget.
Table 2 provides descriptive statistics for the variables used in our
models. All independent variables and the control variables for interim
grants, grant extensions, specialization, average team size, and percent
female authors update at the time of publishing an article. All other
control variables remain constant for an individual over time. Inter-
estingly, although men and women differ in terms of their numbers of
publications, they appear quite similar on almost every other dimension
of their publication records.
6. Results
Table 3 reports the results of the log-logistic regressions of time to
first R01 grant. The first model, including only a covariate for gender,
Fig. 4. Baseline hazard rate estimates.
7
Using the average instead of the sum helps us to distinguish publication quantity from
article-level attention.
8
In principle, one could estimate h(t)bydifferentiating the cumulative hazard function
with respect to t, using the Kaplan–Meier estimate of S(t). But the Kaplan–Meier estimator
creates a step function for S(t). One therefore cannot differentiate it directly.
9
Comparisons of log-logistic models to estimates using other forms of time dependence
revealed that the log-logistic models fit the data better based on the Akaike Information
Criterion (AIC).
10
The NIH data do not record family events, such as child birth; however, NIH policies
allow the extension of fellowship and career development grants when the grantee has
family responsibilities that delay the research (NIH, 2016b).
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1012
indicates that women, on average, have a 20% slower rate of transi-
tioning to the receipt of an R01. Adjusting for the various control
variables reduces this gap by roughly 25% (to 15%).
Models 3 through 6 then examine the extent to which differences in
publication records might account for the remaining gap. Model 3 ac-
counts only for the number of articles. Not surprisingly, publications
have a large effect on the expected time to receiving a first R01. A
doubling in the number of articles reduces the expected time to R01 by
roughly 20%. Accounting for this effect, moreover, reduced the un-
explained gender gap to 12%. Model 4 then introduces the various
other dimensions of the publication record. Interestingly, the propor-
tion first authorships and the proportion of publications in prestigious
journals have little influence on the expected time to first R01. Citations
do, however, have a large effect: a doubling in the number of citations
per article also reduces the expected time to receiving an R01 by about
22%. The residual gender gap, however, remained fairly stable at about
12%, because men and women did not differ meaningfully on these
dimensions.
6.1. Functional form
An unstated assumption in much of the past literature not just on
academic productivity but also on the effects of productivity on success
Table 2
Descriptive statistics for F32-transition models.
Men Women
Mean SD Mean SD
Publication record
Ln (articles) 2.05 0.88 1.84 0.85
Pct (first author) 0.32 0.23 0.34 0.25
Pct (JIF [0-5]) 0.66 0.29 0.65 0.30
Pct (JIF (5-10]) 0.18 0.21 0.20 0.23
Pct (JIF> 10) 0.16 0.22 0.15 0.22
Ln (avg. citations)
a
4.25 0.79 4.26 0.77
Control variables
Interim grants 0.54 0.82 0.46 0.81
Grant extensions 0.14 0.35 0.16 0.37
Prior grants 0.03 0.17 0.04 0.20
Specialization (resid.) 0.00 0.01 0.00 0.01
Avg. team size 5.11 2.62 5.22 3.91
Pct female authors 0.22 0.23 0.53 0.28
Status host institution 96.73 8.97 95.99 10.88
Ln (prior articles) 1.25 0.86 1.16 0.80
N3,822 2,514
a
Based on 3,817 male and 2,512 female F32 grant holders with citations to
their work.
Table 3
Log-logistic regression of time to first R01 –publication records.
(1) (2) (3) (4) (5) (6)
Sex only model Add controls Add productivity Add quality metrics Add funct. form Add complements
Sex 1.20
**
1.15
**
1.12
*
1.12
*
1.08
*
1.08
*
(0.04) (0.04) (0.06) (0.05) (0.04) (0.04)
Publication record
Ln (articles) 0.71
**
0.70
**
0.98 1.00
(0.03) (0.03) (0.05) (0.05)
Pct (first author) 0.87
†
0.64
**
0.52
**
(0.07) (0.09) (0.08)
Pct (JIF 5-10) 1.05 1.04 1.00
(0.10) (0.17) (0.17)
Pct (JIF> 10) 0.84
†
0.83 0.85
(0.08) (0.11) (0.11)
Ln (avg. citations) 0.69
**
0.77
**
0.77
**
(0.02) (0.03) (0.03)
Control variables
Interim grants 0.62
**
0.56
**
0.60
**
0.68
**
0.69
**
(0.01) (0.02) (0.02) (0.02) (0.01)
Grant extensions 0.97 0.95 0.98 0.98 0.98
(0.04) (0.05) (0.05) (0.04) (0.04)
Prior grants 0.75
**
0.70
**
0.72
**
0.77
**
0.75
**
(0.06) (0.07) (0.07) (0.06) (0.06)
Specialization (resid.) 0.09
**
0.04
**
0.03
**
0.16
*
0.19
†
(0.05) (0.03) (0.02) (0.14) (0.16)
Avg. team size 1.00 1.02
†
1.05
**
1.04
**
1.04
**
(0.01) (0.01) (0.01) (0.01) (0.01)
Pct female authors 1.02 1.02 1.00 0.99 0.99
(0.06) (0.08) (0.08) (0.06) (0.06)
Status host institution 1.00 1.00 1.00 1.00 1.00
(0.00) (0.00) (0.00) (0.00) (0.00)
Ln (prior articles) 0.84
**
0.85
**
0.85
**
0.89
**
0.89
**
(0.01) (0.02) (0.02) (0.02) (0.02)
Research field fixed effects (19) NO YES YES YES YES YES
Grant vintage fixed effects (4) NO YES YES YES YES YES
Functional form fixed effects (15) NO NO NO NO YES YES
Complements fixed effects (45) NO NO NO NO NO YES
Log-likelihood −3,170 −2,530 −2,469 −2,332 −2,294 −2,244
Observations 68,776 68,776 68,776 68,614 68,614 68,614
†
Significant level: 10%.
* Significant level: 5%.
** Significant level: 1%.
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1013
in other contexts has been that output has a linear or log-linear re-
lationship to outcomes. For example, researchers will include the count
of articles or the logged count as an independent variable as a means of
adjusting for research output. But the relationship may prove more
complex for a variety of reasons. On the one hand, the first publication
might have inordinate importance, as a sort of proof of concept. Or,
publications might have increasing marginal returns to the extent that
large numbers of them lead to the individual being perceived as a star
(Merton, 1968). On the other hand, many evaluations in academia
compare individuals to “peers”–this implicit competition might mean
that where one falls on the distribution (relative output) matters more
than absolute output.
To relax these functional form assumptions, we calculated time-
varying distributions of our five measures of the publication record and
created vectors of indicator variables to reflect the quartile of the dis-
tribution into which the scientist fell at any particular point in time. In
total, we included 15 variables to capture this distributional informa-
tion, three quartile indicators for each of the five dimensions of the
publication record.
Model 5 includes these quartile fixed effects in the models. Their
inclusion improved the model fit(p< 0.01). As one might expect,
these quartile indicators largely absorbed the effect of the logged
number of articles. Surprisingly, however, their inclusion increased the
predictive power associated with the linear term for the proportion of
first authorships. This effect emerged because all of the action on that
dimension comes from the top quartile. Fig. 5a–c displays the predicted
values including the time-varying quartiles relative to the predicted
values using only the continuous measures for three of the variables
(article count, proportion first author, and citations).
11
As one can see,
the deviation for first authorships occurs for those who have been first
authors on the majority of their publications. Without the quartile fixed
effects, the effect in the first three quartiles of the distribution drives the
average estimate. Fig. 5d–f depicts the number of men and women in
each of these quartiles. Overall, adjusting flexibly for functional form
reduced the residual gender gap by another third to 8%.
6.2. Complementarity
Regression estimates of promotion also typically assume that the
various components of productivity have additive effects on the out-
come. For example, being a first author on an article in Science should
have the same effect as being a first author in a less prominent outlet
and being an interior coauthor on an article in Science. But even casual
observation of how decisions get made in academia suggests that these
various measures may interact in important ways. First authorships, for
example, may prove particularly valuable if they occur on publications
in highly visible outlets (JIF > 10). To allow for these interactions, we
created a set of indicator variables for every possible combination of the
quartile variables, a total of 45 additional terms. Although these terms
proved jointly significant (p< 0.01), suggesting that the various di-
mensions of publication records do interact in important ways in de-
termining grant receipt, their inclusion neither narrowed nor widened
the unexplained gender gap.
6.3. Differential returns
All of the models thus far have assumed that men and women
benefit equally from their publication records. But, as noted above,
women may receive less credit for the same output –in essence, they
may receive differential returns to their publication records.
To assess this possibility of differential returns, we first estimated
gender-specific models of the time to first R01. Doing so effectively
interacts gender with every other variable in the model. Table 4 reports
the results of these models in the first two columns. Note that we ex-
cluded all fixed effects from these models so that any differences in the
returns would only appear in reported coefficients.
12
Although all of the
point estimates appear somewhat different for men and women, the
90% confidence intervals around these estimates substantively overlap
in all cases except for one, citations. While a doubling in the number of
Fig. 5. Non-linear vs. linear effects on probability of transition to R01 at year eight of follow-up.
11
To allow for a visual comparison of these time-varying effects, we computed the
predicted probabilities of survival (i.e., not receiving a R01 grant) at eight years after the
receipt of the F32 (the peak transition year). Kinks in the lines illustrate the effect of
shifting into the next quartile of the respective distributions.
12
The functional forms of the relationships of the publication measures to time of
grant receipt did not vary across men and women (prob >χ
2
= 0.44).
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1014
citations per paper reduced the time to the first R01 grant by about 23%
for men, the same increase in citations per paper only reduced the time
to R01 by roughly 16% for women.
The final model then pools the estimates again, including interac-
tion terms for gender and the various dimensions of the publication
record. These models include the field, period, functional form, and
complements fixed effects. Only one interaction has a significant effect:
citations. A woman with the same number of average citations per
publication appears to benefit about 12% less from them in terms of
time to receiving her first R01. Note that women received a slightly
greater proportion of their citations from first authorships (23.1%) re-
lative to men (22.5%) in our data, so this disparity does not come from
women receiving less credit for articles on which they had been interior
authors.
Although the main effect of gender in this model would appear to
suggest that women actually transition more rapidly to R01 than men,
all else equal, note that one cannot interpret the gender coefficient in
the same manner when estimated with an interaction term. The coef-
ficient represents the time to R01 for women relative to men for those
with zero citations. But scientists without any citations have almost no
chance of being granted an R01. A more useful way of calibrating this
coefficient calculates at what number of citations men appear ad-
vantaged relative to women. That occurs when the average citations per
article exceeds roughly 2.1 (= e
(1/.88)/1.5
), a level less than the fifth
percentile of the citations distribution.
7. Discussion
The gender gap in academic STEM employment has attracted much
attention. Research and policy agendas have been focused on two types
of effects. As in other settings, there has been concern that women face
a“glass ceiling”–a level beyond which they simply cannot advance.
Research has also called attention to the idea of a “leaky pipeline”–that
the number of women active in STEM professions declines from early
education to college to post-doctoral training and at every subsequent
career stage (Etzkowitz et al., 2000).
We document that a large share of this gap emerges in a relatively
short period of time, as men and women move from being a member of
another researcher's lab to leading their own lab. Rather than women
dripping out of the STEM career pipe every centimeter along the way,
they appear to pour out at one of the critical junctures. We therefore
shifted the lens to focus on this period where the gap widens most ra-
pidly.
In particular, we analyzed the rates at which men and women re-
ceived their first R01 grant from the NIH. Among those who had al-
ready held post-doctoral grants and who had published, women had
20% lower transition rates to an R01. Although not a measure of pro-
motion per se, the importance of funding in the life sciences means that
an R01 has effectively become a precursor to receiving tenure at a re-
search university (Jena et al., 2015).
Why does this gap in funding emerge? In trying to understand the
underrepresentation of women in STEM and recognizing that the pub-
lication record plays a prime role in determining who gets hired,
funded, and promoted in academia, past research has documented a
number of dimensions on which women experience worse outcomes
than men, from fewer publications overall to publishing in less pro-
minent outlets (e.g., Cole and Zuckerman, 1984; Long et al., 1993;
Stack, 2002; Lerchenmüller et al., 2018).
But these studies have not been able to connect these gender gaps in
publication records to the paucity of women at the senior levels of the
professorate for two reasons. First, many of them have been cross-sec-
tional. One cannot even say then in which direction causality might
run. Less impressive publication records might contribute to the un-
derrepresentation of women in science but gender gaps in the pub-
lication record might also emerge as an artifact of comparing the re-
cords of more senior men to more junior women. Second, even those
studies that have been longitudinal in their design have not structured
and reported their analyses in a manner that allows one to determine
what portion of the gender gap in STEM might stem from differences in
the publication records of men and women.
We therefore estimated the extent to which publication records
could account for the lower rates at which women received their first
R01 grants. Various dimensions of the publication record, most notably
the number of publications and the average number of citations re-
ceived per article, can account for roughly 60% of the gender gap in the
receipt of these grants. We adopted extremely flexible functional forms,
allowing publication records to have non-linear and even non-mono-
tonic effects and to have complementarities between aspects of the
publication record –for example, allowing first authorship on a paper
in a top-tier journal to count more than first authorship on a paper in a
less prominent one. These flexible functional forms substantially
Table 4
Log-logistic regression of time to first R01 - differential returns.
(7) (8) (9)
Men Women Pooled
Sex 0.67
†
(0.15)
Publication record
Ln (articles) 0.74
**
0.67
**
1.01
(0.03) (0.05) (0.05)
Pct (first author) 0.87 1.01 0.51
**
(0.09) (0.16) (0.08)
Pct (JIF 5-10) 0.98 1.13 0.95
(0.11) (0.19) (0.17)
Pct (JIF> 10) 0.81
*
0.70
*
0.86
(0.08) (0.11) (0.11)
Ln (avg. citations) 0.67
**
0.76
**
0.74
**
(0.02) (0.04) (0.03)
Differential returns
Sex × ln (articles) 0.96
(0.04)
Sex × pct (first author) 1.05
(0.14)
Sex × pct (JIF 5-10) 1.11
(0.17)
Sex × pct (JIF > 10) 0.93
(0.13)
Sex × ln (avg. citations) 1.12
**
(0.05)
Control variables
Interim grants 0.60
**
0.63
**
0.69
**
(0.02) (0.03) (0.01)
Grant extensions 1.00 1.02 0.98
(0.06) (0.08) (0.04)
Prior grants 0.82
†
0.63
**
0.75
**
(0.09) (0.09) (0.06)
Specialization (resid.) 0.03
**
0.03
†
0.21
†
(0.04) (0.05) (0.18)
Avg. team size 1.05
**
1.06
**
1.04
**
(0.01) (0.02) (0.01)
Pct female authors 1.08 0.91 1.00
(0.11) (0.11) (0.06)
Status host institution 1.00 1.00 1.00
(0.00) (0.00) (0.00)
Ln (prior articles) 0.85
**
0.85
**
0.89
**
(0.02) (0.03) (0.02)
Research field fixed effects (19) NO NO YES
Grant vintage fixed effects (4) NO NO YES
Functional form fixed effects (15) NO NO YES
Complements fixed effects (45) NO NO YES
Log-likelihood −1,540 −817 −2,238
Observations 44,712 23,902 68,614
†
Significant level: 10%.
* Significant level: 5%.
** Significant level: 1%.
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1015
improve the explanatory power of publication records. In contrast, the
typical approach –assuming a linear or log-linear relationship between
the number of publications and grant receipt –underestimates the
importance of being near the top of the distribution, of being highly
prolific. But even allowing for extremely flexible functional forms and
for complementarity between dimensions of the publication record
could not fully explain the gender gap in the receipt of R01 grants.
A subsequent exploration of potential differences in the returns to
the same features of the publication record suggested that women
benefit less from the same number of average citations per article (but
that they experience similar returns to every other dimension of the
research record). These differential returns can account for the re-
maining gender gap in funding. Although our results would appear
consistent with audit studies, some of which have suggested that
women with equivalent records receive less favorable outcomes than
men (e.g., Moss-Racusin et al., 2012), people rarely put their average
number or even their total number of citations on their resumes. Our
results therefore appear to point to a novel specific dimension on which
women receive lower returns than men.
Our research design does not, however, allow us to say precisely
why these differential returns occur. Note that the models do control for
the proportion of women coauthors (and that men and women did not
differ significantly on the effects of that variable). The differential re-
turns to citations therefore would not seem to stem from women re-
ceiving less of the credit when coauthoring with men.
One possibility is that evaluators err in their estimates of the in-
fluence of research. They may effectively overestimate the importance
of prior research done by men relative to that done by women. In
contrast to other elements of the publication record, NIH applications
do not typically include information on the citations that applicants
have received. This absence of explicit information may allow more
latitude for cognitive biases –even implicit ones –to creep into eva-
luations. Consistent with that idea, men and women did not differ in
their apparent returns to any of the dimensions of the publication re-
cord that appear on the grant applications. One potential remedy worth
considering therefore would involve including explicit information on
citations on grant applications.
Another possibility is that evaluators perceive the research done by
women as less valuable than that done by men and that this bias applies
most strongly to the most novel and most influential research. That idea
seems consistent with some of the prior research on stereotyping which
suggests that many scientists and engineers perceive science as a male
occupation (Joshi, 2014). Unfortunately, however, if that explains the
effect, then it becomes hard to imagine any simple policy intervention
that could rectify the situation.
But differences in publication records –the number of articles and
the average number of citations per article –appear even more im-
portant than differential returns in explaining the gender gap in
funding. Why men and women differ on these dimensions, however,
also remains an open question. In our definition of the sample, we tried
to rule out gender differences in career preferences by focusing our
analysis on cohorts of F32 post-doctoral grant recipients –scientists
with an interest in and commitment to pursuing an academic career
(Mantovani et al., 2006).
Biases may, of course, directly influence the ability of women to
publish and the number of citations that they receive. If reviewers or
editors perceive publications written by women as less important or of
lower quality than those written by men, women might receive more
rejections before finally placing a paper or go through more rounds of
review. Either could slow down their rate of publication. Published
research by women, moreover, may receive less recognition by others,
in the form of citations.
Blind review, particularly of a form where the editors did not have
information on the identities of the authors before coming to a decision,
could help to limit gender differences in the journal evaluation process.
When orchestras began to have musicians audition from behind screens
so that the judges could not guess the gender of the musician, the
gender balance of orchestras rapidly shifted from being mostly men to
the majority being women (Goldin and Rouse, 2000). But if the same
biases exist among readers, the consumers of research, blind review
would not necessarily eliminate disparities in citation rates.
Men and women may also differ in their output because of differ-
ences in the time that they have available for research. Some of these
differences likely stem from the home. Even among dual-career couples,
women typically shoulder most of the burden in childcare and in
maintenance of the household (Craig and Mullan, 2011). But a large
share of these differences may also emerge from the workplace. Women
often do more than their fair share of administration and service in
academic settings.
These differences in productivity might also stem from differential
access to mentoring and role models. One of the difficulties in ex-
panding the representation of women in the life sciences and elsewhere
has been the very paucity of senior women. Not only does this absence
mean that junior women have fewer role models who they may con-
sider relevant but also it means that they may not have access to senior
women who can act as mentors. Mentors can play a number of im-
portant roles, from providing their junior colleagues with a better un-
derstanding of how the publication and grant application processes
work to introducing them to potential collaborators and to gatekeepers
in the field (Preston, 2004; Etzkowitz et al., 2000). If mentors favor
those of the same gender in these processes, women may find them-
selves disadvantaged in this early access to tacit knowledge and social
capital. Understanding better the ultimate source of these differences in
publication records therefore represents an important question for fu-
ture research.
Although our focus has been on early career transitions, our results
may also have relevance for gender stratification at later career stages
as well. Funding continues to matter at more senior levels, determining
who can pursue their research agendas and therefore who can publish
and receive accolades for their contributions. These factors therefore
may continue to disadvantage women even if they receive an R01 and
earn tenure.
Appendix A
We began the construction of our dataset by connecting F32 grant
recipients to their publication records by first associating the grant
holders with articles that had acknowledged these F32 grants (ac-
cording to the NIH ExPORTER database). Those articles then served as a
means of connecting the NIH data to the Authority author codes in
PubMed, which allowed us to identify their entire publication records
both prior to and after the period covered by the F32 grants. The
Authority disambiguation algorithm, which has been assigned to all
authors of PubMed articles written prior to 2009, identifies which cases
of authors of different articles with the same name have a high prob-
ability of referring to the same individual. It has been found to have a
greater than 99% accuracy (Lerchenmueller and Sorenson, 2016).
Our initial dataset included 7,623 F32 grants that had been ac-
knowledged in at least one article published by the respective grantees
prior to 2009 (the end of the period covered by the Authority algo-
rithm). From this sampling frame, we could match the unique author
identifiers from NIH ExPORTER to the unique author identifiers avail-
able in PubMed for 7,169 F32 recipients (94%).
We excluded scientists who had published ten years or more prior to
the receipt of their F32. Given that the F32 should almost immediately
follow a doctoral degree, either these records include an error or these
individuals would have had to have published as a high school student
or as an undergraduate.
From this set, we excluded scientists for whom we could not de-
termine the dates of their publications for a large share of their pub-
lication records. Our longitudinal analysis required us to assign pub-
lications to specific dates so publications without these dates effectively
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1016
add noise to the analysis. Specifically, we dropped from the analysis
scientists whose publication portfolios included ten articles or fewer
with publication dates missing for two or more of these articles (i.e.,
more than 20% of their articles). We also excluded scientists, with more
than ten publications, who had missing dates for more than 10% of
their articles. These restrictions reduced the sample to 6,549 F32 sci-
entists (but it did not alter the overall gender distribution of the
sample).
In cases where PubMed recorded only a month and year of pub-
lication (as opposed to a specific date within the month), we assigned a
random publication day (1–30) from a uniform distribution to avoid
tied spells in the time-to-event analysis. In cases where PubMed only
recorded publication years (11% of records), we assigned a random
publication day from the full year (1–365), from a uniform distribution.
Although the NIH represents the major source of third-party funding
in the life sciences, some individuals may have benefited from other
sources of funding that we could not observe and may therefore have
entered the F32 program at a later stage in their careers or may have
received a major award prior to receiving their first R01. We therefore
dropped from the analysis individuals in the long tail of the distribu-
tion, with more than 50 articles (97th percentile of the distribution).
Note, however, that including these cases did not substantively change
the results.
References
Azoulay, Pierre, GraffZivin, Joshua S., Manso, Gustavo, 2011. Incentives and creativity:
evidence from the academic life sciences. RAND J. Econ. 42 (3), 527–554.
Berryman, Sue E., 1983. Who Will Do Science? Minority and Female Attainment of
Science and Mathematics Degrees. Rockefeller Foundation, New York.
Brooks, Chris, Fenton, Evelyn M., Walker, James T., 2014. Gender and the evaluation of
research. Res. Policy 43 (6), 990–1001.
Coding News, 2015. Gender Detection. Tech. Rep. Coding News (accessed 11.07.17).
http://codingnews.info/post/gender-detection.html.
Cole, Jonathan R., Zuckerman, Harriet, 1984. The productivity puzzle: persistence and
change in patterns of publication of men and women scientists. Adv. Motiv. Achiev.
2, 217–258.
Craig, Lyn, Mullan, Killian, 2011. How mothers and fathers share childcare. Am. Sociol.
Rev. 76 (6), 834–861.
Etzkowitz, Henry, Kemelgor, Carol, Uzzi, Brian, 2000. Athena Unbound: The
Advancement of Women in Science and Technology. Cambridge University Press,
New York.
Filardo, Giovanni, da Graca, Briget, Sass, Danielle M., Pollock, Benjamin D., Smith, Emma
B., Ashley-Marie Martinez, Melissa, 2016. Trends and comparison of female first
authorship in high impact medical journals: observational study (1994–2014). BMJ
352. http://dx.doi.org/10.1136/bmj.i847.
Garrison, Howard H., Deschamps, Anne M., 2014. NIH research funding and early career
physician scientists: continuing challenges in the 21st century. FASEB J. 28 (3),
1049–1058.
Goldin, Claudia, Rouse, Cecilia, 2000. Orchestrating inpartiality: the impact of ‘blind’
auditions on female musicians. Am. Econ. Rev. 90, 715–741.
Greenberg, Jerald, 1978. Allocator-recipient similarity and the equitable division of re-
wards. Soc. Psychol. 41 (4), 337–341.
Hill, Catherine, Corbett, Christianne, St. Rose, Andresse, 2010. Why so few? Women in
Science, Technology, Engineering, and Mathematics. Tech. Rep. American
Association of University Women, Washington, DC (accessed 11.07.17). https://
www.aauw.org/files/2013/02/Why-So-Few-Women-in-Science-Technology-
Engineering-and-Mathematics.pdf.
Jacob, Brian A., Lefgren, Lars, 2011. The impact of NIH postdoctoral training grants on
scientific productivity. Res. Policy 40 (6), 864–874.
Jagsi, Reshma, Guancial, Elizabeth A., Worobey, Cynthia Cooper, Henault, Lori E., Chang,
Yuchiao, Starr, Rebecca, Tarbell, Nancy J., Hylek, Elaine M., 2006. The “gender gap”
in authorship of academic medical literature –a 35-year perspective. N. Engl. J. Med.
355 (3), 281–287.
Jena, Anupam B., Khullar, Dhruv, Ho, Oliver, Olenski, Andrew R., Blumenthal, Daniel M.,
2015. Sex differences in academic rank in US medical schools in 2014. JAMA 314
(11), 1149–1158.
Joshi, Aparna, 2014. By whom and when is women's expertise recognized? The inter-
active effects of gender and education in science and engineering teams. Adm. Sci. Q.
59 (2), 202–239.
Kalbfleisch, John D., Prentice, Ross L., 2002. The Statistical Analysis of Failure Time Data,
2nd edition. Wiley, New York.
King, Molly M., Bergstrom, Carl T., Correll, Shelley J., Jacquet, Jennifer, West, Jevin D.,
2016. Men Set Their Own Cites High: Gender and Self-Citation Across Fields and Over
Time. (accessed 11.07.17. arXiv:1607.00376.
Larivière, Vincent, Ni, Chaoqun, Gingras, Yves, Cronin, Blaise, Sugimoto, Cassidy R.,
2013. Global gender disparities in science. Nature 504, 211–213.
Lautenberger, Diana M., Dandar, Valerie M., Raezer, Claudia L., Sloane, Rae Anne, 2014.
The State of Women in Academic Medicine: The Pipeline and Pathways to
Leadership. Tech. Rep. Association of American Medical Colleges, Washington, DC
(accessed 11.07.17). https://members.aamc.org/eweb/upload/The%20State%20of
%20Women%20in%20Academic%20Medicine%202013-2014%20FINAL.pdf.
Leahey, Erin, 2006. Gender differences in productivity –research specialization as a
missing link. Gender Soc. 20 (6), 754–780.
Leahey, Erin, 2007. Not by productivity alone: how visibility and specialization con-
tribute to academic earnings. Am. Sociol. Rev. 72 (4), 533–561.
Leahey, Erin, Keith, Bruce, Crockett, Jason, 2010. Specialization and promotion in an
academic discipline. Res. Soc. Stratif. Mob. 28 (2), 135–155.
Legewie, Joscha, DePrete, Thomas A., 2014. Pathways to science and engineering ba-
chelor's degrees for men and women. Sociol. Sci. 1, 41–48.
Lerchenmüller, C., Lerchenmueller, M.J., Sorenson, O., 2018. Long-term analysis of sex
differences in prestigious authorships in cardiovascular research supported by the
National Institutes of Health. Circulation 137, 880–882. http://dx.doi.org/10.1161/
circulationaha.117.032325.
Lerchenmueller, Marc J., 2016. Gender Designation via Genderize API With Unlimited
Request. https://figshare.com/articles/Genderize_unlimited_API_request/4563814.
Lerchenmueller, Marc J., Sorenson, Olav, 2016. Author disambiguation in PubMed: evi-
dence on the precision and recall of Authority among NIH-funded scientists. PLOS
ONE 11 (7), e0158731.
Li, Danielle, 2017. Expertise versus bias in evaluation: evidence from the NIH. Am. Econ.
J.: Appl. Econ. 2 (9), 60–92.
Long, J. Scott, 1992. Measures of sex-differences in scientific productivity. Soc. Forces 71
(1), 159–178.
Long, J. Scott, Allison, Paul D., McGinnis, Robert, 1993. Rank advancement in academic
careers: sex differences and the effects of productivity. Am. Sociol. Rev. 58 (5),
703–722.
Lutter, Mark, Schröder, Martin, 2016. Who becomes a tenured professor, and why? Panel
data evidence from German sociology, 1980–2013. Res. Policy 45, 999–1013.
Mantovani, Richard, Look, Mary V., Wuerker, Emily, 2006. The Career Achievements of
National Research Service Award Postdoctoral Trainees and Fellows: 1975–2004.
Tech. Rep. National Institutes of Health, Bethesda, MD.
Merton, Robert K., 1968. The Matthew effect in science. Science 159 (3810), 56–63.
Morgan, Stephen L., Gelbgiser, Dafna, Weeden, Kim A., 2013. Feeding the pipeline:
gender, occupational plans, and college major selection. Soc. Sci. Res. 42 (4),
989–1005.
Moss-Racusin, Corinne A., Dovidio, John F., Brescoll, Victoria L., Graham, Mark J.,
Handelsman, Jo, 2012. Science faculty's subtle gender biases favor male students.
Proc. Natl. Acad. Sci. 109 (41), 16474–16479.
National Postdoctoral Association, (NPA), 2011. Postdoctoral Scholars, Gender, and the
Academic Career Pipeline. Tech. Rep. National Postdoctoral Association, Rockville,
MD (accessed 11.07.17). http://c.ymcdn.com/sites/www.nationalpostdoc.org/
resource/resmgr/Docs/postdoc-gender-fact-sheet-20.pdf.
NIH, 2011. A History of New and Early Stage Investigator Policies and Data. Tech. Rep.
National Institutes of Health (accessed 11.07.17). https://grants.nih.gov/policy/
new_investigators/history.htm.
NIH, 2016a. National Institutes of Health Budget. Tech. Rep. National Institutes of Health
(accessed 11.07.17). http://www.hhs.gov/about/budget/budget-in-brief/nih/index.
html.
NIH, 2016b. NIH Family-Friendly Initiatives. Tech. Rep. National Institutes of Health
(accessed 11.07.17). https://grants.nih.gov/grants/family_friendly.htm.
NSF, 2015. Women, Minorities, and Persons with Disabilities in Science and Engineering:
2015. Special Report NSF 15-311. Tech. Rep. National Center for Science and
Engineering Statistics, Arlington, VA (accessed 11.07.17). https://www.nsf.gov/
statistics/2017/nsf17310/.
Pohlhaus, Jennifer Reineke, Jiang, Hong, Wagner, Robin M., Schaffer, Walter T., Pinn,
Vivian W., 2011. Sex differences in application, success, and funding rates for NIH
extramural programs. Acad. Med. 86 (6), 759–767.
Preston, Anne E., 2004. Leaving Science: Occupational Exit from Scientific Careers.
Russell Sage Foundation, New York, NY.
Sarsons, Heather, 2017. Recognition for group work: gender differences in academia. Am.
Econ. Rev.: Papers Proc. 107 (5), 141–145.
Shen, Helen, 2013. Inequality quantified: mind the gender gap. Nature 495 (7439),
22–24.
Stack, Steven, 2002. Gender and scholarly productivity: 1970–2000. Sociol. Forum 35
(3), 285–296.
Steinpreis, Rhea E., Anders, Katie A., Ritzke, Dawn, 1999. The impact of gender on the
review of the curricula vitae of job applicants and tenure candidates: a national
empirical study. Sex Roles 41 (7), 509–528.
Wais, K., 2015. genderizeR: Gender Prediction Based on First Names. Tech. Rep. Github
(accessed 11.07.17). https://github.com/kalimu/genderizeR.
West, Jevin D., Jacquet, Jennifer, King, Molly M., Correll, Shelley J., Bergstrom, Carl T.,
2013. The role of gender in scholarly authorship. PLOS ONE 8 (7), e66212.
Williams, Wendy M., Ceci, Stephen J., 2015. National hiring experiments reveal 2:1 fa-
culty preference for women on STEM tenure track. Proc. Natl. Acad. Sci. 112 (17),
5360–5365.
Wuchty, Stefan, Jones, Benjamin F., Uzzi, Brian, 2007. The increasing dominance of
teams in production of knowledge. Science 316 (5827), 1036–1039.
Xie, Yu, Shauman, Kimberlee, 1998. Sex differences in research productivity: New evi-
dence about an old puzzle. Am. Sociol. Rev. 63 (6), 847–870.
Xie, Yu, Shauman, Kimberlee, 2003. Women in Science: Career Processes and Outcomes.
Harvard University Press, Cambridge.
M.J. Lerchenmueller, O. Sorenson Research Policy 47 (2018) 1007–1017
1017
