You are on page 1of 14

A Macro Analysis of Productivity Differences Across

Fields: Challenges in the Measurement of


Scientific Publishing

Fredrik Niclas Piro, Dag W. Aksnes, and Kristoffer Rrstad


Nordic Institute for Studies in Innovation, Research and Education (NIFU), N-0302 Oslo, Norway.
E-mail: {fredrik.piro,dag.w.aksnes,kristoffer.rorstad}@nifu.no

While many studies have compared research productiv- disciplines, in particular for the social sciences, arts, and
ity across scientific fields, they have mostly focused on humanities, where a substantial share of relevant books or
the hard sciences, in many cases due to limited pub-
journals is not indexed in the Web of Science (Hicks,
lication data for the softer disciplines; these studies
have also typically been based on a small sample of 2004). The novelty of this study is the completeness and
researchers. In this study we use complete publication volume of publication data it draws on, and the level of
data for all researchers employed at Norwegian univer- detail included about academic employees in every disci-
sities over a 4-year period, linked to biographic data for pline. Although our focus is on differences in average pro-
each researcher. Using this detailed and complete data
ductivity between major areas of research at the macro level,
set, we compare research productivity between five
main scientific domains (and subfields within them), researchers academic position, age, and gender are also
across academic positions, and in terms of age and considered, as they have been shown to be important vari-
gender. The studys key finding is that researchers from ables associated with productivity differences (Kyvik, 1991,
medicine, natural sciences, and technology are most 1993). We also investigate publishing practices (books, book
productive when whole counts of publications are used,
chapters, and journal articles), coauthorship practices,
while researchers from the humanities and social sci-
ences are most productive when article counts are frac- and publication frequencies across scientific domains and
tionalized according to the total number of authors. The subfields.
strong differences between these fields in publishing These questions have been analyzed in several previous
forms and patterns of coauthorship raise questions as studies, but usually on the basis of more limited data sets;
to whether publication indicators can justifiably be
some studies cover a broad range of fields, but based on a
used for comparison of productivity across scientific
disciplines. much smaller sample (e.g., Dundar & Lewis, 1998; Shin &
Cummings, 2010), some studies only cover selected disci-
plines (e.g., Bourke & Butler, 1996; Costas, van Leeuwen, &
Introduction Bordons, 2010; Seglen & Aksnes, 2000), while the study
The purpose of this study is to further the knowledge base most comparable to ours in terms of sample size and cover-
about differences between scientific domains in levels of age of disciplines does not include books and book chapters
scientific/scholarly publishing productivity. The study is (Archambault & Larivire, 2010). In Norway, scientific
based on the total scientific publication output of 11,571 productivity has been investigated in several previous
researchers, at the main Norwegian universities, over a studies by Kyvik (e.g., Kyvik, 1991, 1993, 2003), but in
4-year period. contrast to our study, these studies used self-reported,
Most previous productivity studies are restricted to the nonverified data sets.
data available from larger bibliometric databases, such as the Productivity studies at the individual level have shown
ISI Web of Science (Thomson Reuters). Such analyses often that the number of publications per person depends on a
produce a distorted picture of the research output in some wide range of factors, such as: age, gender, academic posi-
tion and rank, the availability of research funds, teaching
loads, equipment, research assistants, workload policies,
Received April 11, 2012; revised May 25, 2012; accepted May 26, 2012
departmental culture and working conditions, departmental
2013 ASIS&T Published online in Wiley Online Library size, and organizational context (Bonaccorsi & Daraio,
(wileyonlinelibrary.com). DOI: 10.1002/asi.22746 2005; Dundar & Lewis, 1998; Kyvik, 1993; Ramesh &

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY, ():, 2013
Singh, 1998; Smeby & Try, 2005). Nevertheless, few studies we have linked persons with identical names manually using
consider disciplinary differences as a key factor for produc- available data and information.
tivity in their analytical models of research performance; in The publications of staff at individual departments were
effect, they treat faculty demographics as control variables assigned to five broad fields, the humanities, social sciences,
(Shin & Cummings, 2010, pp. 582583). The completeness natural sciences, medicine, and technology; this assignment
of our data set makes it possible to investigate productivity was based on the departments reporting of their scientific
from the opposite point of view: research productivity is profile in national R&D statistics (departments select from a
analyzed across scientific domains, using individual list of 58 scientific subfields, within the five main domains).
researcher characteristics as mediating variables. The sixth domain used in national statistics (agriculture,
fishery, and veterinary sciences) was not used because there
were too few researchers in these subfields. Researchers
Methods and Data
from veterinary medicine were included in the medicine
In this study we use publication data from the national category, while researchers from agriculture and fishery
Norwegian database, FRIDA. In 2004, Norway imple- sciences were included in the natural sciences.
mented a bibliometric model for performance-based funding The analyses cover the 4-year period 20052008 and the
of research institutions. These institutions budget alloca- study is limited to the four major research universities in
tions are now partially based on their scientific and scholarly Norway (Oslo, Bergen, Trondheim, and Troms), which
publishing, as documented in FRIDA. This database account for 72% of national research output in the higher
includes all types of scientific publications, in all fields of education sector. Only research personnel with at least one
research in the higher education sector. Bibliographic data publication during the time period were included.
are collected through a common documentation system used Scientific productivity can, in principle, be measured
by all institutions, resulting in complete, verifiable, and relatively easily, by quantifying research output in terms of
structured data for use in bibliometric analysis. publications. In practice a number of issues make this more
In the FRIDA database, publication activity is reported difficult, and must be addressed (Seglen, 2001). In particu-
by the institutions as standard bibliographic references, lar, the choice and weighting of publication types and the
which are analyzable by publication channel and type of attribution of author credit are important. Many publications
publication. A dynamic authority record, covering 19,000 are the result of collaborative efforts involving more than
controlled scientific and scholarly publication channels one researcher, and are thus multiauthored. Two different
ensures that references from nonscientific publications are principles and counting methods are typically applied for
not entered into the system. Publication data from profes- multiauthor papers, in bibliometric studies. The most
sional bibliographic data sources (e.g., the Web of Science) common is whole counting, where all contributing authors
are imported to the FRIDA system, to facilitate the registra- receive credit for a whole publication. A second approach is
tion of publications by the employees. The database is there- fractionalized counting where credit is divided equally
fore well suited to productivity analyses across subject between all authors; if an article has five authors, each
fields, as a large-scale database, with complete data covering author is credited with 1/5 (0.2) of the article. It can be
not only journal articles, but also monographs and book argued that these counting methods are complementary: the
chapters. whole or integer count gives the number of papers in which
To provide information about authors individual charac- each person participated; a fractionalized count gives the
teristics (institutional affiliation, position, age, and gender), number of papers creditable to each person, assuming all
the bibliographic database was coupled to another database, authors made equal contributions to the paper, and that all
the Norwegian Research Personnel Register (providing the contributions add up to one (Moed, 2005). While these are
official Norwegian R&D statistics, compiled by NIFU), to the two main counting methods applied in bibliometric
introduce individual-level data. This database has biographi- studies, many other variants have been used, and a study
cal information for all researchers in the higher education by Gauffriau, Larsen, Maye, Roulin-Perriard, and von Ins
sector and institute sector in Norway (with a few exceptions, (2007) identified 19 different methods of counting publica-
such as students and foreign guest researchers). The tions. As Larsen (2008, p. 235) observes, 40 years of pub-
researchers have unique IDs in both the Research Personnel lication counting have not resulted in general agreement on
Register and the FRIDA publication database. However, the definitions of methods and terminology nor in any kind of
IDs are not identical and the linking is based on information standardization.
on the full name of the persons as well as their institutional In this study, three simple and commonly applied count-
affiliations. For a large number of persons, there is a one- ing methods are employed, allowing the following publica-
to-one correspondence, and homonyms (different persons tion indicators to be calculated and compared:
with identical names) do not represent a problem. In a
previous study, based the Norwegian Research Personnel 1. Average number of publications (whole counts) per
Register, we found that the frequency of homonyms using person.
the full name of the persons was only 1.4% of the Norwe- 2. Average number of fractionalized publications per person
gian population of researchers (Aksnes, 2008). In our study (each publication divided by the number of authors).

2 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi
TABLE 1. Average and median number of authors per publication by field, and publication types (article equivalents) across disciplines (n = 41,066).

Avg. authors per publication


Median authors per Avg. authors per (excl. 100+ authors Percentage journal Percentage book Percentage
Subject field publication publication publications) articles articles monographs

Humanities 1 1.4 1.4 39.0 37.5 23.5


Social Sciences 2 2.6 2.6 49.2 31.2 19.6
Technology 3 4.3 4.3 69.2 28.1 2.7
Natural Sciences 4 19.6 5.6 88.8 9.2 2.0
Medicine 6 8.4 8.3 93.9 4.6 1.5

3. Average number of article equivalents per person (frac- which often have several hundred authors; if the average
tionalized publication counts combined with a weighting number of publications is calculated for each field excluding
of monographs as equal to 5 articlesboth in journals all publications with more than 100 authors there is a sub-
and books). stantial drop in the average number of authors in the natural
sciences, and medicine emerges as the scientific domain
The latter weighting system of publication types (also with the highest average number of authors. The results in
applied in the Norwegian funding model) attempts to make Table 1 illustrate that whole counts of publications may
the research efforts behind articles and books comparable to skew the data, and should therefore be fractionalized by the
article publications. Kyviks (1991) review of such weighting number of contributing authors.
procedures from several other studies shows that it is standard In Figures 15 and Table 2 we compare research produc-
to consider a full monograph as equivalent to 46 articles. tivity across scientific domains and by academic position.
All figures can be found the Appendix. Large variations in
scientific publishing productivity are apparent between
Results
fields (Figure 1). As expected, the number of publications
Between 2005 and 2008 almost 60,000 publications (whole counts) per researcher is much higher in the natural
(n = 59,861) were produced by the researchers in this study sciences, medicine, and technology than in the humanities
(n = 11,571). About 60% of the researchers are based in and social sciences.
the two largest scientific fields, medicine (33%) and the However, due to the high number of authors per publica-
natural sciences (28%). Another 12% of the researchers tion in technical/medical/natural science publications, pro-
were from technology. A total of 27% of the researchers ductivity drops significantly when we use fractionalized
come from the softer disciplines: 16% from social sciences counts. In contrast, fractionalized productivity measures
and 11% from the humanities. result in substantially higher averages in both the humanities
The use of different publication forms varies substan- and social sciences than in the other fields. This impact of
tially between scientific areas (Table 1). In the humanities, fractionalization on productivity measures for the natural
more than 60% of publications are books and book chapters. sciences is particularly significant in the subfield of physics,
In the social sciences the figure is ~50%. In contrast, books due to the effect of the CERN-papers (Table 1). At the other
and book chapters play a minor role in the natural sciences end we find mathematics and informatics with the lowest
and medicine. In technology, book chapters are more differences between these two measures. A second observa-
common due to many proceedings papers in conference tion that can be made from Figure 1 is that differences
series produced in this field. The overall conclusion that is between disciplines are much greater when looking at frac-
apparent in Table 1 is that, while the natural sciences and tionalized publications per person compared to whole counts.
medicine may be reasonably compared based on their pro- The average number of publications based on fractionalized
duction of journal articles only, any comparison with the counts varies from 1.03 in medicine to 3.18 in humanities (see
humanities and social sciences will be inadequate and Appendix), that is a ratio of 3.08, while the ratio for the
partial, unless book chapters and monographs are included lowest and highest whole counts at the field level was 1.69
in measures of publication activity. (3.64 in humanities and 6.16 in natural sciences).
The five scientific fields also differ substantially in their Previous studies have also shown large differences in
coauthorship practices. Research in humanities subjects is publication patterns according to researchers academic
typically an individual game, while social science publica- position. In order to compare like with like we added data
tions are typically coauthored by two or three researchers. on a range of common academic positions, and a category
Publishing in other disciplines is generally the result of Other personnel which includes adjunct professors and
teamwork, with the highest median number of authors found other scientific positions (e.g., researchers), administrative
in medicine (six authors). Looking at the average number of and technical personnel as well as retired persons. Scientific
authors, the natural sciences have by far the highest number productivity varies greatly between the various academic
(19.6), although this is largely due to the CERN-papers positions (Figure 2).

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013 3
DOI: 10.1002/asi
FIG. 1. Mean number of publications per person: total, all research personnel.

FIG. 2. Mean number of publications per person by academic position.

Regardless of the publication measure used, professors out to be less productive than PhD students based on
constitute the most productive academic group; this is fractionalized counts. Figure 3 shows how the research
perhaps to be expected as full professors at Norwegian productivity varies for professors only across scientific
research universities are expected to devote ~50% of their domains.
time to research. Postdoctorates exceed associate pro- The pattern for professors is very similar to that observed
fessors publishing based on whole counts, but associate for the whole sample: the professors in the humanities and
professors are more productive after fractionalization. The social sciences are less productive than their colleagues in
largest decrease in productivity after fractionalization is the hard sciences based on whole counts, but this picture
found in the case of medical doctors/physicians, who turn is reversed after fractionalization, where productivity is once

4 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi
FIG. 3. Mean number of publications per person, full professors (n = 2,405).

FIG. 4. Mean number of article equivalents per year by scientific field and academic position.

again highest in the humanities and lowest in medicine. One When we compare between academic positions using
noticeable exception in this professors-only analysis is article equivalents only (Figure 4), another confirmation of
found in the results for technology, where the gap in- previous findings emerges. Across every academic position,
publishing output compared to the humanities and social personnel from the humanities and social sciences are most
sciences is much smaller than that found in the analysis productive. In every academic position (except for medical
of all researchers (Figure 1). doctors/physicians, where there are no personnel from the

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013 5
DOI: 10.1002/asi
Mean number of publications per person by subfield (n = 37).
FIG. 5.

6 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi
TABLE 2. Average number of publications per person by subfield.

Fractionalized article
Subfield Scientific domain No. of researchers Whole counts Fractionalized counts equivalents

Archaeology Humanities 109 3.09 2.60 3.21


History 205 3.72 3.15 3.87
Languages 356 3.50 2.93 3.40
Literary sciences 164 4.55 4.17 5.43
Musicology 50 2.60 2.27 2.85
Philosophy sciences 146 3.55 3.16 3.99
Theology/history of religion 94 4.24 3.93 4.91
Economics Social Sciences 142 4.39 2.32 2.49
Educational sciences 286 4.28 2.84 3.49
Human geography 71 3.75 2.16 2.41
Jurisprudence/criminology 279 3.61 3.25 4.40
Media/journalism sciences 73 3.96 3.00 3.93
Political sciences 194 5.04 3.29 3.86
Psychology 412 5.94 2.10 2.22
Social anthropology 126 3.11 2.54 3.16
Sociology 206 4.61 3.00 3.48
Biosciences Natural Sciences 1,048 5.21 1.34 1.35
Chemistry 463 6.56 1.68 1.70
Fisheries 97 3.37 0.83 0.83
Geosciences 419 5.08 1.23 1.25
Informatics 301 6.65 2.27 2.31
Mathematics 305 5.29 2.15 2.21
Physics 515 9.48 1.72 1.75
Basic medicine/odontology Medicine 1,543 5.04 0.93 0.94
Health sciences 488 7.08 1.80 1.83
Clinical medicine 1,560 5.30 0.87 0.88
Clinical odontology 132 4.50 1.24 1.24
Biotechnology Technology 113 4.80 1.02 1.06
Building 111 4.40 1.50 1.64
technology/architecture
Chemical technology 126 6.57 1.80 1.88
Electro/technical sciences 93 4.14 1.31 1.37
Engineering 103 4.57 1.75 1.82
Environment technology 69 4.65 1.78 1.80
ICT 428 5.96 2.28 2.32
Marine technology 78 6.14 2.27 2.32
Material technology 139 5.26 1.32 1.32
Mineral resources/petroleum 72 4.31 1.60 1.60

humanities) we also find the same pattern of productivity 9.48), but this is also the scientific domain with the largest
rankings by disciplinary area: humanities are first, followed spread in subfield scores. Medicine has the second highest
by social sciences, technology, natural sciences, and lastly median score (5.17, ranging from 4.507.08), followed by
medicine. technology (4.72, ranging from 4.146.57), social sciences
The results so far have been based on five broad scientific (4.28, ranging from 3.115.94), and lastly humanities (3.55,
domains. Further analyses were performed for all subfields ranging from 2.604.55). The whole counts for subfields are
containing 50 researchers or more (accounting for a total of thus in line with the overall trends found for broad fields
11,116 out of the full data set of 11,571 researchers). Table (Figure 1).
2 helps to illustrate (1) whether the overall ranking described The fractionalized scores show some interesting results.
is also valid at the subfield level, and (2) whether there are Within the humanities and social sciences, psychology gets
any substantial variations in subfield productivity within the lowest score (2.21), indicating that this subfield,
each scientific domain. although classified as part of the social sciences, has a pub-
When looking at publication whole counts, physics lication pattern more in line with medicine. The highest
(9.48) and health sciences (7.08) have the highest scores. fractionalized scores at the subfield level are found in the
Psychology has the highest score in the social sciences subjects literary sciences (4.17), and theology/history of
(5.94), while literary science (4.55) is the humanities sub- religion (3.93) from the humanities. Outside of the humani-
field with the highest score. The highest median score is ties and social sciences, Information and Communication
found in the natural sciences (5.29, ranging from 3.37 Technologies (ICT) has the highest fractionalized score

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013 7
DOI: 10.1002/asi
(2.28). The subfields from the humanities also have the Bozemans (2005) study where the strong relationship
highest median fractionalized score (3.15, ranging from between research collaboration and research productivity
2.274.17), followed by social sciences (2.84, ranging from (measured by the number of publications) was no longer
2.103.39). Natural sciences (1.68, ranging from 0.83 significant when the publications were fractionalized.
2.27), and technology (1.67, ranging from 1.022.82) are However, we do not know of any other studies that show that
both above medicine (1.08, ranging from 0.871.80). This is the relationship between scientific domains and productivity
also in accordance with the overall trends (Figure 1). is completely reversed after fractionalization, as is found in
The fractionalized article equivalents for subfields show our study. One possible reason for this is the fact that we
a near identical pattern to those of fractionalized scores: included data on monographs and book chapters.
the median score in humanities subfields is highest (3.87, In Archambault and Larivires (2010) study of 13,479
ranging from 2.855.43), followed by social sciences (3.48, researchers in Quebec, it was shown that productivity by
ranging from 2.224.40). The median value for technology domain changed after fractionalization in the same direction
subfields (1.72, ranging from 1.062.32) is slightly above as in our study. That is, the superiority in productivity
that of natural sciences (1.70, ranging from 0.832.31), the observed in medicine, natural sciences, and engineering
opposite relationship than for the fractionalized scores. compared to social sciences and humanities was drastically
Again, the lowest median value is found in medicine (1.09, reduced. In the Canadian study, based on journal articles
ranging from 0.881.83). The highest average score is found only, the productivity ratio between mean number of whole
for literary sciences (5.43), and again, ICT (2.32) is the counts per researcher in social sciences (6.9) and natural
subfield outside humanities and social sciences with the sciences (17.4) was 2.52, but was reduced to 1.95 after
highest score. The three lowest scores are found in fisheries fractionalization. In the humanities, the ratio compared to
(0.83), clinical medicine (0.88), and basic medicine/ natural sciences was 6.96 when using whole counts and 1.95
odontology (0.94). after fractionalization. Because our study also includes
The main picture that emerges from these results is that monographs and book chapters, which together account for
there is a reverse association for subfield scores based on a very large share of the publications in the social sciences
whether publications have been fractionalized or not and humanities, the initial productivity differences when
(Figure 5). That is, the higher the whole counts scores are looking at whole counts were smaller than found in the
for a subfield, the lower its fractionalized scores will be. In Canadian study. The whole count ratio between the natural
Figure 5 the subfields are listed by the number of whole sciences and social sciences was 1.38, but after fractional-
counts (regardless of scientific domain). ization the ratio was 1.67 in favor of the social sciences, and
increasing to 1.96 after weighting the publications by type.
Discussion The same pattern was seen for the humanities. At first, using
whole counts, the productivity ratio between the humanities
We have identified large productivity differences between and natural sciences was 1.69 in favor natural sciences,
five main scientific domains and their subfield areas. In which changed to 2.01 in favor of the humanities after frac-
line with previous studies, we find that whole counts of tionalization, and 2.44 after weighting of publication type
publications are higher in the technical, natural, and medical (see Appendix). Comparing results like this to the Canadian
fields (compared to humanities and social sciences) (e.g., study is interesting, because our study takes it further by
Archambault & Larivire, 2010; Dundar & Lewis, 1998; stating that the remaining productivity differences are com-
Shin & Cummings, 2010). The analysis goes on to indicate, pletely eliminated as books and book chapters are also
however, that simple adjustments to the methods used in included; they even contribute to a reversed productivity
counting publications (i.e., fractionalization based on landscape.
number of authors and weighting of publication types) lead Our results clearly demonstrate that the choice between
to very substantial changes in the comparative performance whole counts and fractionalized counts (with or without a
across disciplinary areas. A noticeable example is that of full weighting to take account of different publication types)
professors in medicine and the humanities: determines the outcome of the productivity analysis. This
suggests that none of the three indicators (whole counts,
1. Whole counts: a full professor in medicine publishes on
fractionalized counts, and article equivalents) used here
average 138% more publications than a professor in
humanities. provides good indicators by themselves for comparing pro-
2. Article equivalents: a full professor in the humanities ductivity across scientific domains. However, it is possible
publishes on average 75% more than a professor in medi- that other weighting principles might work to make such
cine once the publications are fractionalized and weighted comparisons more valid and meaningful. In citation analy-
by publication type. ses, the average citation rate varies considerably between the
different scientific disciplines. As a response, various refer-
Our results imply that it is more demanding in terms of ence standards and normalization procedures have been
research effort to produce e.g., 10 articles each with 10 developed. The most common is the average citation rate of
authors than a single-authored article, which both would the journal or field in which the particular papers have been
result in one article equivalent. This is in line with Lee and published (Moed, 2005). It may be that similar reference

8 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi
standards can be developed for scientific productivity. The natural sciences); technology was third (6.44, which is 1.5
main problems that need to be addressed relate to how much times as high as originally, ratio: 0.90); the social sciences
credit each author should be given in coauthored publica- were fourth (6.15, compared to 5.33 originally, ratio: 0.87);
tions, and the relative weight that should be given to the and the humanities came out at the bottom (6.10, compared
differing publication formats of journals, book chapters, and to 5.99 originally, ratio: 0.86). These findings are interesting
monographs. As seen in our analyses, if scientific domains because they show that when it comes to the most research-
are to be comparable, one needs to attribute to each author a intensive and productive academic group, field differences
higher score for each paper than the fraction according to the can be adequately neutralized by using the 0.5 point method.
number of authors. In order to adjust for the identified field When all academic groups and not just full professors were
differences, the extra weighting would have to be substan- included, large field differences still remained, which in the
tial. To test this approach, we performed an analysis where case of medicine may be exemplified by many physicians in
coauthored publications gave 0.5 points to each author, hospitals who do publish occasionally, but at a rather low
regardless of the total number of authors, to investigate frequency compared to other scientific personnel at univer-
whether this would neutralize the field differences. Analyz- sities who are not professors.
ing fractionalized article equivalents we found that the mean Another option to adjust measures of publication rates
score increased in all scientific domains, but the increase would be to introduce standardized scores, e.g., Hays (1970)
was lowest in domains with a low share of coauthored pub- suggested the equation Z = (x M)/s, where Z is a
lications; the humanities (which was up 0.09 points: from researchers standardized score; x is the persons raw
3.93 to 4.02) and social sciences (up 0.46 points: from 3.16 score; M is the average score for the discipline to which
to 3.62). The largest increase was found in medicine, where s/he belongs; and s is the standard deviation for that
the mean number went up by 164% (from 1.04 to 2.75), discipline. Based on such standardization, almost all
followed by natural sciences (from 1.61 to 3.27) and tech- differences between scientific fields would disappear
nology (from 1.81 to 2.82). The main focus of this exercise, (see Figure 6; real numbers have been standardized around
however, is to compare the distances between the fields 100 for visual purposes).
when all coauthored publications are boosted to 0.5 points Even though field differences in productivity have been
per author. To investigate the gap between fields, their rela- more or less eliminated, two exceptions stand out: the
tive publication output can be expressed as ratios between first is that the humanities remain above other scientific
the most productive field based on the original analysis, the domains in fractionalized publications. The second, and
humanities. When humanities output was taken as equal to more substantial exception, is the low values seen for tech-
1.00, the ratios for the other disciplines were: 0.80 for social nology compared to other fields after publications have
sciences, 0.46 for technology, 0.40 for natural sciences, and been fractionalized.
0.26 for medicine. The new analysis drastically reduces the Nevertheless, it seems likely that it will be more difficult
gaps between subject ratios, to: 0.90 for social sciences (up to provide acceptable reference standards for productivity
0.10), 0.70 for technology (up 0.24), 0.81 for natural sci- than for citation rates. This is because scientific productivity
ences (up 0.41), and 0.68 for medicine (up 0.42). Although has been shown to depend on numerous factors, such as
it does not eliminate the differences between fields, the new academic position, gender, age, time for research/teaching
weights given to coauthored papers reduced much of the loads, availability of research funds (and equipment,
field differences, although the ranking remained fairly research assistants, etc.), department culture and working
stable: humanities and social sciences at the top, medicine at conditions, and departmental size and organizational
the bottom, and natural sciences changing position to over- context. These findings suggest that productivity differences
take technology. across fields are not a matter of weighting procedures and
When we performed the same analysis for professors fractionalization schemes, but may be the result of differ-
only, the field differences almost completely disappeared. ences in other meaningful variablesat both the individual
Professors from the humanities and social sciences were no and contextual levels. In the next sections we discuss our
longer rated as the most productive groups, as they were results in light of former studies that have looked at produc-
overtaken by the professors in natural sciences and medi- tivity differences, based on some of the issues discussed
cine. Originally, professors in humanities had the highest here. In general, our results differ from past studies based on
average rate of publications (5.99), followed by social the same Norwegian universities, covering the same scien-
sciences (5.33, ratio: 0.89), technology (4.25, ratio: 0.70), tific domains (Kyvik, 1991, 1993, 2003; Kyvik & Olsen,
natural sciences (3.61, ratio: 0.60), and medicine (2.75, 2008). This is likely due to important differences between
ratio: 0.46). The distance between the highest and lowest these studies and ours: this study is based on a much larger
mean number was thus more than two. After providing all data set (11,500 compared to 1,2001,500 persons) and also
authors with 0.5 points for coauthored publications, the uses registration databases for data on individual research-
results changed markedly: the highest value was found in ers, while the other studies used questionnaire data and
natural sciences (7.10, more than twice as high compared to nonverified self-reported numbers of publications. Finally, it
the original analysis); medicine was second (6.64, which is may be that productivity patterns may have changed over
2.4 times as high as originally, ratio: 0.90, compared to time, and these studies cover differing time periods. Thus,

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013 9
DOI: 10.1002/asi
FIG. 6: Standardized number of publications by scientific domains.

the results from former Norwegian studies are not directly production of publications increases with age and reaches a
comparable to the current study. peak at some point during an academic career, from which it
then declines (see, e.g., Barjak, 2006; Gingras, Larivire,
Macaluso, & Robitaille, 2008; Gonzalez-Brambila &
Productivity and Gender
Veloso, 2007; Aksnes, Rrstad, Piro, & Sivertsen, 2011b).
Previous studies have shown large gender differences in Based on the same data used in this study, Aksnes et al.
scientific productivity (Aksnes, Rrstad, Piro, & Sivertsen, (2011b) found an identical pattern of age-related rise
2011a; Kyvik & Teigen, 1996) with female researchers and decline across scientific domains; but despite similari-
being shown to be less productive. In Aksnes et al. (2011a), ties in overall age patterns, there were also noticeable dif-
this pattern was found for almost all age groups and scien- ferences. For example, scholars in the humanities and social
tific domains. Male researchers in our study appear to be sciences remained productive for a longer period. In these
more productive when using all three productivity measures fields, researchers aged between 65 and 69 remained very
(see Appendix); this holds for all scientific domains (not productive.
shown in the tables), thus making it important to look at how In this study, looking at 10-year age intervals, productiv-
the gender balance differs between fields (gender is also ity (fractionalized counts) increases with age in all scientific
highly correlated with academic position, where women are domains, up to 5059 years (humanities and natural sci-
less represented in the most productive group, the profes- ences) or 6069 (all other fields). Therefore, the presence of
sors), because a high share of women is associated with low many young researchers appears to dilute rates of publica-
productivity. A total of 35.6% of the sample are women, and tion productivity. Technology is a particularly interesting
the highest shares of women are found in medicine (44.5%), case herethe most productive age group (6069) consti-
the humanities (41.3%), and the social sciences (40.3%). tutes only 7.5% of researchers in this field (the mean share of
The share of female researchers is much lower in natural 6069-year-olds in all fields is 13.9%). In technology we
sciences (27%) and technology (19.5%). These results do also find the highest share of researchers in both the 2029
not support gender as an explanatory factor for subject field and the 3039 age groups. Both in that study and the present
differences in productivity, since women are better repre- one, we may question whether the decision to exclude
sented in the two disciplines where productivity is highest, researchers without a paper has affected the age-specific
based on fractionalized counts. results. Arguably, there are scientific staff in the four main
universities in Norway in the older age groups who did not
publish any papers throughout the years 20052008. A study
Productivity and Age
by Gringas et al. (2008) found that productivity for research-
Although the results of previous studies have not always ers who kept publishing increased in line with our results,
been entirely consistent, a curvilinear relationship between but once researchers without a paper were included in the
age and productivity is quite firmly established. The average denominator, the productivity began to decrease from age

10 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi
50 years. While the lack of nonproductive researchers in consuming work on behalf of their supervisors, which would
our data in particular may affect the results for the older age otherwise reduce these senior researchers own publication
groups, we do not believe that it impacts the interpretation of output. Within a group of researchers led by a professor, the
the key result, which is the divide between young and more professor will typically get her/his name on all publications,
established researchers. while the PhD students will be authors only in the publica-
The two most productive scientific domains measured by tions they have been directly involved in (Kyvik, 1991). This
fractionalized counts (humanities and social sciences) have means that, instead of high numbers of PhDs being seen as
a much lower share of researchers under 40 years (28.6 and an impediment to the overall productivity of a group (i.e.,
30.2%, respectively) than the natural sciences and technol- PhD students and postdocs) they may enhance productivity
ogy (51.6 and 62%, respectively), which supports the idea in some segments of that group. This was also found (with
that some of the productivity differences between these reversed causation) in Smeby and Try (2005), where pro-
fields may be related to a different age composition, with the ductivity decreased with age, but where older researchers
humanities and social sciences benefiting from a more had a positive impact on the overall group level, Further-
mature pool of researchers. However, medicine also has a more, Gingras et al. (2008) showed that the decline in
low share of young researchers (29.9% under 40 years) and researchers productivity about age 50 was subsequently suc-
still has a low productivity level (possibly because many ceeded by an increase in the average number of papers in
physicians are in dual positions in universities/hospitals and highly cited journals and among highly cited papers (which
publish at a lower frequency than full-time university rises continuously until retirement), meaning that productiv-
employees). ity and quality/impact are two features that should both
be considered before describing a development as being
negative due to reduced productivity.
Productivity and Academic Position
Norwegian R&D statistics also include data on time
Many studies have shown that individual publication pro- available for research, according to academic positions,
ductivity tends to increase moving up the hierarchy of aca- institutions, and domains (R&D equivalents). These data
demic positions (Kyvik, 1991; Aksnes et al. 2011a), where show that the tenured personnel tend to have ~40% of their
professors are the most prolific personnel. It is also well time to use on research, while the figures for PhD students
established that PhD students are much less productive in and postdocs are around 7080%. These differences are
terms of the number of publications they produce than more mainly due to the fact that the tenured personnel have much
established researchers. This is not surprising, as the PhD larger teaching obligations. If we had applied the R&D
students are still in training as researchers. In our study, we equivalents on the publication data in our study the differ-
find that technology has a completely different composition ences between the natural sciences and technology, on the
of research roles when compared to the other scientific one hand, and the humanities and the social sciences on the
fields. In technology, 19.5% are professors and 39.9% PhD other would be even larger (due to the different composi-
students, while in the humanities the distribution is 31.3% tion of the academic personnel). This reinforces the obser-
and 13.9%, respectively. The share of professors is also low vation that the work effort needed to produce one article
in medicine, as many of the more experienced researchers equivalent in the first domain is significantly higher than in
are employed at university hospitals as medical doctors/ the latter.
physicians. The composition in the natural sciences and
technology includes postdoctorates and PhD students as the
Different Publishing Practices
largest proportion of personnel, while in the other domains,
professors/associate professors are in the majority. Hence, in Using the Norwegian Research Personnel Register, we
the humanities and social sciences the majority of research- were able to interpret our results in light of different
ers are found to be in the most productive academic compositions of researchers in the different disciplinary
positions, whereas in natural sciences and technology, the fields, based on researchers age, gender, and academic
majority of the personnel are found in the least productive position. In the next sections we briefly discuss two other
academic positions. explanatory factors behind productivity differences across
This discussion indicates that some of the excess produc- fields: the share of unproductive researchers and patterns of
tivity in humanities and social sciences may be explained by collaboration.
different compositions of research personnel based on Our study included only those researchers who had pub-
gender, age, and academic position. The differences also lished at least one or more publications during the 4-year
indicate that the reduced productivity in technology may period, which represents a limitation in our data set, as only
be due to the large number of younger and less advanced researchers who have published are registered in the data-
researchers (PhD students) and the low share of professors base. Arguably, many of the researchers who are not found
and associate professors. However, this should not automati- in the database (and have not published) are newly
cally be interpreted as a management argument for hiring appointed, and thus unproductive. Nevertheless, many
more senior personnel. PhD students represent an important researchers are not in the database simply because they have
work force in the science system and do a lot of time- not published in the time period. This is only a problem for

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013 11
DOI: 10.1002/asi
our conclusions if there are systematic differences between teams. These again were more productive than individuals
the scientific domains in their share of unproductive without a team.
researchers. We do not have data to verify/falsify this, but in We have shown in our paper that productivity results are
Kyviks (1991) 20-year old study at the same university highly dependent on whether publication whole counts or
departments there were certainly field differences in the fractionalized publication counts are used. This is due to the
share of unproductive researchers. The highest share of large differences in the average number of authors behind
unproductive researchers was found in the humanities (19%, each article, across the different scientific fields. However,
ranging from 4% in theology to 28% in languages), followed designing weighting procedures that are able to resolve this
by the social sciences (14% unproductive researchers, coauthorship dilemma (or field differences in book vs.
ranging from 8% in sociology to 20% in psychology), journal article publishing) may still not be enough to unify
natural sciences and technology (13%, ranging from 8% in the two indicators, as several factors at the individual level
geosciences to 20% in physics) and medicine (10%, ranging (age, gender, academic position) are both strongly related to
from 5% in biomedical sciences to 12% in odontology). We productivity and are unevenly distributed across scientific
therefore cannot conclude that the share of unproductive fields.
researchers is similar in all fields. If this pattern described Our study is based on data on Norwegian scientists.
above is also valid for our study sample, the superiority of Although it is a large-scale study, the question remains
the hard sciences for the whole counts analysis would be whether these findings have general validity in other set-
even greater, while the soft sciences superiority in frac- tings. As shown above, several previous studies have pro-
tionalized counts analysis would be reduced. Such patterns vided consistent results concerning the publication patterns
of unproductiveness may thus have skewed our data. across fields. The higher productivity in some fields based
Another aspect that must be considered in trying to on whole counts has been documented in numerous other
explain productivity differences between scientific fields is studies, given different coauthorship practices across fields.
the great difference in the average number of authors con- However, our inclusion of monographs and book chapters
tributing to a paper (as seen in Table 1), which obviously has not only added further evidence to this, but also showed that
a great impact on the number of whole count publications what may be an unfavorable gain of much coauthorship,
(Lee & Bozeman, 2005). There are various reasons for these may be an unreasonable disadvantage after fractionalization
field differences in publishing practices. Research in the once the publication numbers are based on books as well.
natural sciences and medicine is often large scale: it may The findings in this article of various author characteristics
require contributions from different research laboratories being associated with productivity, such as academic posi-
with different experimental equipment, or field research tion, gender, and age, are also much in line with previous
based on observations in different parts of the world carried studies, although the limitation in our study of not including
out by several research teams. This results in a large number the unproductive researchers may have led to an overesti-
of authors in many publications. Kyvik (1991, p.6970) mation of the productivity especially in the older age groups,
argues that scientists in the hard sciences have become where productivity numbers are lower in studies that have
more dependent on each other to make progress than have included the unproductive researchers as well.
researchers in the soft sciences. In the latter fields, inde- Although empirically robust, it should be acknowledged
pendence may actually be regarded as fruitful, in order to that the exact figures in terms of average productivity are
develop scientific paradigms. This idea was supported likely to vary among countries. Cultural factors related to
in Kyviks study, in that several of the few humanities gender equality, academic opportunities, professional inte-
researchers who had tried to coauthor a publication reported gration, availability of resources, time for research, etc.
poor experiences. One possible explanation for this pattern differ at national levels and may affect scientific perfor-
is that the final products in the humanities often have an mance. Moreover, it is evident from our findings that differ-
essayistic and individual form, making it difficult to agree on ent compositions within research populations (e.g., in terms
both content and style for the collaborating persons. In other of academic position and age) would have given different
fields, where content is more codified and standardized, it overall results.
may be easier to coauthor publications. Still, the expectation
that collaboration automatically generates more publications
Conclusion
may be questioned, as it has been by Lee and Bozeman
(2005, p. 694), who point out that: The number of collabo- In our study, scientists from medicine, the natural sci-
rators is not the same as the number of collaborations. If one ences, and technology are found to be more productive than
collaborates ten times with one person, how does that differ, their colleagues from the humanities and social sciences
in productivity effects, from collaborating ten times with ten when productivity is measured using whole counts of pub-
different persons? Permanent collaboration/collaborators lications. This is arguably due to the large share of multiau-
seems to be the key factor in publication patterns, as in thored papers in these disciplines. After fractionalization of
Rey-Rocha, Martin-Sempere, and Garzn (2002), where publications, the picture is completely reversed, with schol-
researchers belonging to consolidated teams were shown to ars from the humanities and social sciences being rated as
be more productive than their colleagues in nonconsolidated the most productive. We have therefore argued that these

12 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi
simple productivity measures cannot be used in comparative Bourke, P., & Butler, L. (1996). Publication types, citation rates and
assessments of productivity levels across domains, as they evaluation. Scientometrics, 37(3), 437494.
Costas, R., van Leeuwen, T.N., & Bordons, M. (2010). A bibliometric
are not neutral in terms of the varying research efforts classificatory approach for the study and assessment of research perfor-
involved in single-authored and coauthored publications. In mance at the individual level: The effects of age on productivity and
contrast to citation indicators, where advanced and valid impact. Journal of the American Society for Information Science and
methods for adjusting for field differences exist, it is much Technology, 61(8), 15641581.
more complex to produce reference standards for scientific Dundar, H., & Lewis, D.R. (1998). Determinants of research producti-
vity in higher education. Research in Higher Education, 39(6),
productivity. Not only are there large differences in publish- 607631.
ing practices (both in types of publication and the extent of Gauffriau, M., Larsen, P.O., Maye, O., Roulin-Perriard, A., & von Ins, M.
collaboration) across domains, but variables such as position (2007). Publication cooperation and productivity measures in scientific
and age of researchers also play a role. In evaluations of research. Scientometrics, 73(2), 175214.
research performance, and increasingly in funding alloca- Gonzalez-Brambila, C., & Veloso, F.M. (2007). The determinants of
research output and impact: A study of Mexican researchers. Research
tions, the productivity of scientists is usually one of the Policy, 36(7), 10351051.
issues reviewed. In this respect, studies like ours may be Gringas, Y., Larivire, V., Macaluso, B., & Robitaille, J-P. (2008). The
useful in assessing the productivity levels of groups or effects of aging on researchers publication and citation patterns. PLoS
personnel. ONE, 3(12), e4048.
Hays, W.L. (1970). Statistics. London: Holt, Rinehart and Winston.
Hicks, D. (2004). The four literatures of social science. In H.F. Moed, W.
Glnzel, & U. Schmoch (Eds.), Handbook of quantitative science and
Acknowledgment technology research (pp. 473495). Dordrecht, Netherlands: Kluwer
Preliminary results of this study were presented at the Academic Publishers.
Kyvik, S. (1991). Productivity in academia. Scientific publishing at
11th International Conference on Science and Technology Norwegian universities. Oslo, Norway: Universitetsforlaget.
Indicators, Leiden, September 10th, 2010: Dag W. Aksnes, Kyvik, S. (1993). Academic staff and scientific productivity. Higher
Kristoffer Rrstad, Gunnar Sivertsen, Fredrik Niclas Piro Education Management, 5(2), 191202.
(2010): Productivity differences across fields a macro Kyvik, S. (2003). Changing trends in publishing behaviour among univer-
analysis. Book of abstracts 11th International Conference on sity faculty, 1980-2000. Scientometrics, 58(1), 3548.
Kyvik, S., & Bruen Olsen, T. (2008). Does the aging of tenured academic
Science and Technology Indicators, Leiden (pp. 1416). staff affect the research performance of universities? Scientometrics,
76(3), 439455.
Kyvik, S., & Teigen, M. (1996). Child care, research collaboration and
gender differences in scientific productivity. Science, Technology &
References
Human Values, 21(1), 5471.
Aksnes, D.W. (2008). When different persons have an identical author Larsen, P.O. (2008). The state of art in the publication counting. Sciento-
name. How frequent are homonyms? Journal of the American Society for metrics, 77(2), 235251.
Information Science and Technology, 59(5), 838841. Lee, S., & Bozeman, B. (2005). The impact of research collaboration on
Aksnes, D.W., Rrstad, K., Piro, F.N., & Sivertsen, G. (2011a). Are female scientific productivity. Social Studies of Science, 35(5), 673702.
researchers less cited? A large-scale study of Norwegian scientists. Moed, H.F. (2005). Citation analysis in research evaluation. Dordrecht,
Journal of the American Society for Information Science and Technol- Netherlands: Springer.
ogy, 62(4), 628636. Ramesh, B.A., & Singh, Y.P. (1998). Determinants of research productivity.
Aksnes, D.W., Rrstad, K., Piro, F.N., & Sivertsen, G. (2011b). Age and Scientometrics, 43(3), 309329.
scientific performance. A large-scale study of Norwegian scientists. In Rey-Rocha, J., Martin-Sempere, M.J., & Garzn, B. (2002). Research
E. Noyons, P. Ngulube, & J. Leta (Eds.), Proceedings of the 13th Inter- productivity of scientists in consolidated vs. non-consolidated teams:
national Conference of the International Society for Scientometrics and The case of Spanish university geologists. Scientometrics, 55(1), 137
Infometrics (ISSI 2011) (pp. 3445). Durban, South Africa: ISSI. 156.
Archambault, E., & Larivire, V. (2010). Individual researchers research Seglen, P.O. (2001). Evaluating biology. A scientometric study of a univer-
productivity: A comparative analysis for counting methods. In Book of sity department. NIFU Report 6/2001. Oslo, Norway: NIFU.
Abstracts 11th International Conference on Science and Technology Seglen, P.O., & Aksnes, D.W. (2000). Scientific productivity and group
Indicators (STI), Leiden 2010 (pp. 2224). Leiden, Netherlands: Univer- size: A bibliometric analysis of Norwegian microbiological research.
siteit Leiden, CWTS. Scientometrics, 49(1), 125143.
Barjak, F. (2006). Research productivity in the internet era. Scientometrics, Shin, J.C., & Cummings, W.K. (2010). Multilevel analysis of academic
68(3), 343360. publishing across disciplines: Research preference, collaboration, and
Bonaccorsi, A., & Daraio, C. (2005). Exploring size and agglomeration time on research. Scientometrics, 85(2), 581594.
effects on public research productivity. Scientometrics, 63(1), 87 Smeby, J.C., & Try, S. (2005). Departmental context and faculty research
120. activity in Norway. Research in Higher Education, 46(6), 593619.

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013 13
DOI: 10.1002/asi
Appendix. Average number of publications per person 20052008, by field and academic position.

Fractionalized Fractionalized article


Field Positiona Number of persons Whole counts counts equivalents

Humanities Professors 407 5.40 4.82 5.99


Social sciences 542 7.58 4.41 5.33
Natural sciences 651 13.43 3.53 3.61
Medicine 537 12.88 2.74 2.75
Technology 268 11.97 4.09 4.25
Humanities Associate professors 282 3.51 3.12 3.94
Social sciences 366 3.99 2.61 3.10
Natural sciences 318 5.25 1.65 1.66
Medicine 226 5.85 1.53 1.60
Technology 86 5.27 1.87 1.91
Humanities Post docs 115 3.40 2.86 3.49
Social sciences 143 4.53 2.65 3.32
Natural sciences 484 5.58 1.50 1.51
Medicine 376 4.71 0.88 0.88
Technology 173 5.39 1.83 1.85
Humanities PhD-students 180 1.76 1.57 1.80
Social sciences 340 2.06 1.27 1.46
Natural sciences 782 3.20 0.74 0.74
Medicine 809 2.60 0.46 0.46
Technology 548 2.89 0.97 0.97
Humanities Medical doctor/ Physician 0 N/A N/A N/A
Social sciences 12 1.75 0.38 0.38
Natural sciences 3 1.00 0.19 0.19
Medicine 886 4.63 0.69 0.70
Technology 0 N/A N/A N/A
Humanities Othersb 315 2.64 2.16 2.62
Social sciences 500 3.13 1.75 2.03
Natural sciences 956 4.23 0.97 0.99
Medicine 966 4.34 0.81 0.82
Technology 300 3.56 1.08 1.13
Humanities Total 1,299 3.64 3.18 3.93
Social sciences 1,903 4.47 2.65 3.16
Natural sciences 3,194 6.16 1.58 1.61
Medicine 3,800 5.37 1.03 1.04
Technology 1,375 5.27 1.77 1.81
Total All fields 11,571 5.23 1.78 1.96
Professors 2,405 10.47 3.83 4.28
Associate professors 1,278 4.61 2.24 2.58
Post docs 1,291 4.99 1.61 1.75
PhD students 2,659 2.71 0.82 0.87
Medical doctor/physician 901 4.58 0.68 0.69
Othersb 3,037 3.85 1.18 1.29
Men All fields 7,449 6.00 1.99 2.19
Women 4,122 3.85 1.39 1.54
a
Persons who have changed academic position during the time period are classified and included in more than one category.
b
Includes a variety of other positions: adjunct professors and other scientific positions (e.g., researchers), administrative, and technical personnel, as well as
retired persons.

14 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY 2013
DOI: 10.1002/asi

You might also like