non elite journals 3 PDF

Nom original: non elite journals 3.pdf

Titre: The weakening relationship between the Impact Factor and papers’ citations in the digital age

Auteur: Lozano, Lariviere, Gingras

Ce fichier au format PDF 1.5 a été généré par Microsoft® Word 2010, et a été envoyé sur Fichier-PDF le 25/10/2014 à 19:08, depuis l'adresse IP 212.195.x.x. La présente page de téléchargement du fichier a été vue 1119 fois.
Taille du document: 242 Ko (14 pages).
Confidentialité: fichier public

Aperçu du document

The weakening relationship between the Impact Factor
and papers’ citations in the digital age

George A. Lozano†, Vincent Larivière§ and Yves Gingras*

†

dr.george.lozano@gmail.com

Observatoire des Sciences et des Technologies (OST), Centre Interuniversitaire de Recherche
sur la Science et la Technologie (CIRST), Université du Québec à Montréal, CP 8888, Succ.
Centre-Ville, Montréal, QC. H3C 3P8 (Canada)

§

vincent.lariviere@umontreal.ca

École de bibliothéconomie et des sciences de l'information, Université de Montréal, C.P. 6128,
Succ. Centre-Ville, Montréal, QC. H3C 3J7 (Canada)
and
Observatoire des Sciences et des Technologies (OST), Centre Interuniversitaire de Recherche
sur la Science et la Technologie (CIRST), Université du Québec à Montréal, CP 8888, Succ.
Centre-Ville, Montréal, QC. H3C 3P8 (Canada)

*

gingras.yves@uqam.ca

Observatoire des Sciences et des Technologies (OST), Centre Interuniversitaire de Recherche
sur la Science et la Technologie (CIRST), Université du Québec à Montréal, CP 8888, Succ.
Centre-Ville, Montréal, QC. H3C 3P8 (Canada)

Abstract
Historically, papers have been physically bound to the journal in which they were
published but in the electronic age papers are available individually, no longer tied to
their respective journals. Hence, papers now can be read and cited based on their own
merits, independently of the journal’s physical availability, reputation, or Impact Factor.
We compare the strength of the relationship between journals’ Impact Factors and the
actual citations received by their respective papers from 1902 to 2009. Throughout most
of the 20th century, papers’ citation rates were increasingly linked to their respective
journals’ Impact Factors. However, since 1990, the advent of the digital age, the
strength of the relation between Impact Factors and paper citations has been
decreasing. This decrease began sooner in physics, a field that was quicker to make the
transition into the electronic domain. Furthermore, since 1990, the proportion of highly
cited papers coming from highly cited journals has been decreasing, and accordingly,
the proportion of highly cited papers not coming from highly cited journals has also been
increasing. Should this pattern continue, it might bring an end to the use of the Impact
Factor as a way to evaluate the quality of journals, papers and researchers.

2

Introduction
The Impact Factor (IF) was originally devised in the 1960s to guide academic libraries in their
journal purchases (Archambault & Larivière, 2009). Although several other types of citationbased measures of journal impact have been recently devised, such as the SCImago Journal Rank
(González-Pereira, Guerrero-Bote, & Moya-Anegón, 2009), the Eigenfactor (West & Bergstrom,
2010) and the source normalized impact per paper (Moed, 2010), the 2-year IF compiled by
Thomson Reuters is still the most widely used. The IF of a given journal for a given year is
defined as the mean citation rate, during that given year, of the papers published in that journal
during the previous 2 years. For example, a journal’s IF for 2011 considers citations received in
2011 by papers published in that journal during the years 2009 and 2010. Thus, the citation
window for individual papers ranges from 1 year to almost 3 years, with an average of 2 years.
Over the past few decades, IFs have slowly permeated into the collective consciousness of
scientists, and IFs have become self-reinforcing measures of journal quality, the papers therein,
and their authors. Researchers now consider IFs when choosing their publication outlets; journal
editors formulate policies explicitly designed to improve their IFs, and publishers advertise their
IFs on their web sites. IFs are often used as a surrogate for the actual number of citations a paper
recently published might eventually receive. Such a proxy might be partially justified given that,
independently of the quality of the paper, a journal’s IF is positively linked with the citations
received by its papers (Larivière & Gingras, 2010). Since the early 1990s, as citation data became
electronically available, interest and use of the IF has increased, and scholarly articles on the IF
have increased exponentially (Archambault & Larivière, 2009).
The digital age also brought forth another change. Since the creation in 1665 of the Journal
des Sçavans and the Philosophical Transactions of the Royal Society, which are considered to be

3

the first two scientific periodicals, researchers have mostly read actual printed journals, so papers
published in high profile journals with high circulation had a greater chance of being read and
cited than papers published in less widely available journals. Now that scientific information is
disseminated electronically, researchers are less likely to read entire journals; instead they
conduct electronic literature searches on particular topics and find specific articles from a wide
variety of journals. Hence, as long as the journal is listed in the main databases (e.g., Web of
Science, Scopus, or Google Scholar) and papers are available online, they can be read and cited
based on their own merits, unaffected by their journals’ physical availability, reputation, or IF.
Hence, before the electronic age, the citation rate of any given paper and its journal’s IF
mutually reinforced each other. A journal’s IF was (and still is) based on its individual papers’
citation rates, and the citation rate of any individual paper was affected by its journal’s circulation
and availability, which depended on its IF. Now the former is still true, but if new practices of
literature search and usage limit the effect of journal IF on paper citation rates, the correlation
between paper citation rate and IF should be decreasing over time. Additionally, the proportion of
highly cited papers coming from the highest IF journals should be diminishing over time, Here
we examine whether this is indeed the case, and consider the implications to the continued use of
the IF on the future of scientific publishing. Data for three groups of disciplines are presented:
natural and medical sciences taken altogether, physics, and social sciences, from 1900 to 2011.

Methods
We used Web of Science (WoS) data from Thomson Reuters from 1900 to 2011, covering all
areas of natural sciences, medical sciences and social sciences. The dataset covers the Century of
Science and Century of Social Sciences datasets from 1900-1944, and the Science Citation Index
(Expanded), the Social Sciences Citation Index and the Arts and Humanities Citation Index from
4

1945-2011 period. The disciplinary classification of journals used for natural and medical
sciences in general, and of physics and social sciences in particular, is an adaptation of the
classification used by the U.S. National Science Foundation (NSF), which categorizes each
journal in only one discipline and specialty.
The dataset included 25,569,603 natural and medical sciences papers 3,211,026 physics
papers and 879,494 social sciences papers. The total number of cited references analysed was
819,369,970. Humanities papers were excluded from the analysis because of their long citation
windows and high uncitedness rates (Larivière, Gingras, & Archambault, 2009), but citations in
humanities journals were included. To be included in the analysis, papers had to be published in a
journal for which an IF could be calculated. However, references made by excluded papers were
considered as citations for other papers included in the analysis. Some journals on some years did
not have an IF, either because their papers did not receive any citations during the 2-year citation
window, or because 2 years must elapse before new journals receive their first yearly IFs.
Given that Thomson Reuters does not compile IFs for the entire period studied, and that the
exact method by which it calculates IFs is not entirely clear, and hence irreproducible (Moed &
Van Leeuwen, 1995; Rossner, Van Epps, & Hill, 2007), the IF of each journal covered in the
database was recalculated. In the Thomsom Reuters IF, some types of publications are used to
count citations (the nominator), but do not themselves count as “papers” (the denominator). Here,
IF was calculated the same way as the Thomson Reuters IF, except that (1) this asymmetry
between the numerator and the denominator was corrected; except as noted above in regards to
papers in the humanities, if papers were counted, their citations were also counted, and (2)
citations to individual papers were counted during the entire 2 years following their respective
publication year. Hence, IF data was not available for the first 2 years, and full 2 year citation

5

windows were not possible for the last 2 years, leaving a complete dataset of both IFs and citation
rates from 1902 to 2009. A large proportion of papers remained completely uncited at the
beginning of the period (Larivière et al., 2009); to reduce their weight in the IF-citations
relationship, the analyses were also conducted excluding uncited papers, both in the calculation
of citation rates and IFs.
Two indicators were used to measure the strength of the relationship between IF of journals
and citations of papers. The first indicator was the coefficient of determination (r2). Each paper
was assigned the IF of the journal in which it was published and the citations it received during
two years following its publication year, and the r2 between the two series of variables was
calculated for each year. The second indicator is the yearly percentage of the most highly cited
papers published in the most highly cited journals, and the yearly percentage of the most highly
cited papers not published in the most highly cited journals.

Results
Figures 1, 2 and 3 present the r2 between IF and paper citations from 1902-2009, for all
disciplines of the natural sciences and medical sciences together (Fig. 1), physics (Fig. 2) and
social sciences (Fig. 3). For descriptive purposes, in cases where there was a clear break, these
data were split into 1902-1958, 1959-1990 and 1991-2009. For medical and natural sciences (Fig.
1), there was an increase of the correlation between IF and paper citation rates from 1902 until
the end of the 1990s. The strength of the relationship between IF and citations did not increase
steadily throughout the 20th century. Two dips occurred after the two World Wars, likely as a
result of changes in the research system. More interestingly and in contrast to the general pattern
throughout most of the 20th Century, since scientific information began to be disseminated

6

electronically, around 1990, the relationship between the IF and citation rates has been
weakening.
0,35
y = 0.003x - 5.7759
r2 = 0.579

Annual coefficient of determination (r2)

0,30

0,25
0,20
0,15
y = -0.0049x + 9.9863
r2 = 0.4788

0,10
0,05
0,00
1900

y = 0.0011x - 1.9418
r2 = 0.4666

1920

1940

1960

1980

2000

2020

FIG. 1. Coefficient of determination (r2) between the impact factor of journals and the 2-year
citation rate of their papers from 1902 to 2009, for all natural and medical sciences journals.

The same analyses were carried out with 2 disciplines thought to be at opposite ends of the
spectrum of how quickly they made the transition into the electronic domain: physics (Fig. 2) and
social sciences (Fig. 3). Given the smaller sample size, the variation of the r2 values between IF
and papers’ citation rates is larger, but in both cases there was a decrease during the last two
decades. Although the decrease is not significantly different in the two disciplines, in physics it
appears to start earlier, towards the end of the 1980s (Fig. 2).

7

Annual coefficient of determination (r2)

0,25

0,20

0,15

y = 0.0023x - 4.2978
r2 = 0.7291

0,10

y = -0.0026x + 5.4626
r2 = 0.5712

0,05

0,00
1900

1920

1940

1960

1980

2000

2020

FIG. 2. Coefficient of determination (r2) between the impact factor of physics journals and the 2year citation rate of their papers from 1902 to 2009.

Annual coefficient of determination (r2)

0,25

0,20
y = 0.0011x - 2.0527
r2 = 0.293
0,15

0,10

0,05
y = -0.0018x + 3.7039
r2 = 0.1824
0,00
1900

1920

1940

1960

1980

2000

2020

FIG. 3. Coefficient of determination (r2) between the impact factor of social sciences journals and
the 2-year citation rate of their papers from 1902 to 2009.

8

Although not shown, all results are similar and conclusions the same using Pearson’s r and
Spearman’s rank correlation as indicators. All analyses were also carried out excluding uncited
papers, both at the level of papers and in the calculation of the journal IFs. When uncited papers
are excluded, a clearer trend with fewer fluctuations emerges, but the strength of the relationship
between IF and citations remains within the same order of magnitude. So, removing uncited
papers does not result in a stronger relationship between the IF and citations.
Figures 4 and 5 present an additional indicator of the relationship between IF and paper
citations for all disciplines in the natural and medical sciences: the percentage of papers that are
both in the top 10% (and 5%) most cited and published in the top 10% (and 5%) highest IF
journals (Panels A of Figs. 4 and 5). By contrast, Panels B of these figures show the percentage
of the top 10 (and 5%) most papers that are not published in the top 10% (and 5%) highest IF
journals. Both figures show that the relationship between IF and citations has been weakening
steadily since 1990, as a larger proportion of top (5 and 10%) most cited papers are published
outside journals with top (5 and10%) IF.

9

6,5%

5,5%

75%

y = 0,0002x - 0,3969
R² = 0,3549

70%

Percentage of papers

Percentage of papers

6,0%

A

5,0%
4,5%
4,0%

y = -0,0003x + 0,6174
R² = 0,416

y = -0,0002x + 0,5033
R² = 0,293

3,5%

3,0%
1900

1920

1940

1960

1980

2000

2020

B

65%

y = 0.0017x - 2.8052
r2 = 0.2897

60%

55%

y = -0.0004x + 1.437
r2 = 0.0334

50%
1900

1920

y = -0.0028x + 6.0592
r2 = 0.7217

1940

1960

1980

2000

2020

FIG. 4. A) Percentage of the top 10% most cited papers published in the top 10% most cited
journals. B) Percentage of top 10% most cited papers that were not published in the top 10%
most cited journals.

More specifically, the percentage of the 10% most cited papers published in the 10% most
cited journals has been decreasing since 1990 (Fig. 4a), from about 5.25% to 4.50%.
Accordingly, the percentage of the 10% most cited papers not published in the 10% journals with
the highest IF has been increasing since 1990 (Fig. 4b); from 52% to about 56%. This pattern is
even more clearly evident when the same comparisons are made for the top 5% of papers and the
top 5% of journals (Fig. 5). In 1990 about 2.25% of the top 5% papers were published in top 5%
journals; but by 2009 this figure had fallen to 1.90 % (Fig. 5a). Similarly, in 1990 about 55% of
top 5% most cited papers were not published in the top 5% journals, but by 2009 the figure had
increased to 62% (Fig. 5b). Hence, the most important literature is increasingly coming from a
greater range of journals, not only the journals with the highest IF.

10

Percentage of papers

2,2%

A

75%

y = 0.0002x - 0.371
r2 = 0.8695

2,0%
1,8%

y = -0.0002x + 0.3732
r2 = 0.7103

1,6%
1,4%

70%
y = 0,0035x - 6,4769
R² = 0,7121

65%
60%

y = -2E-05x + 0,7562
R² = 7E-05

55%

1,2%
1,0%
1900

B

80%

y = 8E-07x + 0.0125
r2 = 5E-05

Percentage of papers

2,4%

1920

1940

1960

1980

2000

2020

50%
1900

1920

1940

y = -0,004x + 8,4464
R² = 0,8708

1960

1980

2000

2020

FIG. 5. A) Percentage of the top 5% most cited papers published in the top 5% most cited
journals. B) Percentage of top 5% most cited papers that were not published in the top 5% most
cited journals.

Discussion and conclusions
Impact factors were initially developed to assist libraries in their purchasing decisions, and hence
they have had a significant effect on journal circulation and availability. Here we show that
throughout most of the 20th century, the link between the IF and papers’ citations was getting
stronger, but, as predicted, this link has been weakening steadily since the beginning of the
electronic age. This change seems to have started slightly earlier in physics, a field that was
quicker to adopt electronic dissemination of information. Also during this time the percentage of
top papers coming from the top journals has been decreasing. Compounded with the fact that, in
general, citations have become more widely spread among journals (Larivière et al., 2009), the
electronic age and its new modes of disseminating and accessing scientific literature might bring
forth the end of the IF as a useful measure of the quality of journals, papers, and researchers and
have interesting implications for the future of scientific literature.

11

The IF has been repeatedly criticized as a suitable measure of journal quality (Aksnes,
2003; Morgan & Janca, 2000; Rossner et al., 2007; Rothenberg, 2008; Seglen, 1997; Vakil, 2005;
Whitehouse, 2001). The strongest arguments against its validity and use are: (1) some types of
publications within journals, such as letters and commentaries, are used to count citations (the
nominator), but do not themselves count as “papers” (the denominator), and hence inflate the
journal’s IF, (2) the IF depends on the number of references, which differs among disciplines and
journals, (3) the inclusion of journals in the database depends solely on Thomson Reuters, a
private company, and not on the fields’ practitioners, (4) the exact IF published by Thomson
Reuters cannot be replicated using publicly available data, (5) the distribution of citations/paper
is not normal, so at the very least the mode or median ought to be used instead of the mean, (6)
the 2-year span for papers followed by one year for citations is completely arbitrary and favours
high-turnover over long-lasting contributions, and (7) journal editors can manipulate and
artificially inflate their IFs. Our analysis identifies one more problem: the relationship between
paper quality and IF is weakening, so the IF is losing its significance as a measure of journal
quality.
Second, IFs are used as a proxy for paper quality. Except for the most recently published
papers that have not had a chance to be cited yet, there is no reason to use the IF as a proxy for a
paper’s quality. One can readily have access to any individual paper’s citation rate and determine
how the paper stands on its own, regardless of its journal’s IF. As the relationship between paper
citation rates and IF continues to weaken, and as more important papers increasingly appear in
more diverse venues, it will become even less justifiable to automatically transfer a journal’s
reputation and symbolic capital on to even its most recently published papers. This should force a
return to direct assessments of paper quality, by actually reading them.

12

Third, and even more troubling, is the 3-step approach of using the IF to infer journal
quality, extend it to the papers therein, and then use it to evaluate researchers. Our data shows
that the high IF journals are losing their stronghold as the sole repositories of high quality papers,
so there is no legitimate basis for extending the IF of a journal to its papers, and much less to
individual researchers. This is congruent with the finding that over the past decade in economics,
the proportion of papers in the top journals produced by people from the top departments has
been decreasing (Ellison, 2011). Moreover, given that researchers can be evaluated using a
variety of other criteria and bibliometric indicators (e.g., Averch, 1989; Leydesdorff &
Bornmann, 2011; Lozano, 2010; Lundberg, 2007; Põder, 2010), evaluating researchers by simply
looking at the IFs of the journals in which they publish is both naive and uninformative.
For the past few centuries journals were a convenient way to organize papers by subject,
but search engines now allow us to find individual papers on specific topics from across the
entire spectrum of journals, so highly subject-specific journals might become obsolete or begin to
amalgamate. Online, open-access journals, such as in the PLoS family of journals, and online
databases, such as the ArXiv system and its cognates, will continue to gain prominence. Using
these open-access repositories, experts could find publications in their respective fields and
decide which ones are worth reading and citing, regardless of the journal. If this pattern
continues, the IF will cease to be a meaningful measure of the quality of journals, papers and
researchers.

Acknowledgments
We thank Jean-Pierre Robitaille for comments on an earlier draft of this paper. G.A.L. thanks the
University of Tartu for allowing him free access to their online collections.

13

References
Aksnes, D.W. (2003). Characteristics of highly cited papers. Research Evaluation, 12(3), 159170.
Archambault, É., & Larivière, V. (2009). History of the journal impact factor: contingencies and
consequences. Scientometrics, 79(3), 635-649.
Averch, H.A. (1989). Exploring the cost-efficiency of basic research funding in chemistry.
Research Policy, 18(3), 165-172.
Ellison, G. (2011). Is peer review in decline? Econonic Enquiry, 49(3), 635-657.
González-Pereira, B., Guerrero-Bote, V.P., & Moya-Anegón, F. (2009). The SJR indicator: A
new indicator of journals' scientific prestige. Available at:
<http://arxiv.org/ftp/arxiv/papers/0912/0912.4141.pdf>
Larivière, V., Gingras, Y., & Archambault, É. (2009). The decline in the concentration of
citations, 1900-2007. Journal of the American Society for Information Science and
Technology, 60(4), 858-862.
Leydesdorff, L., & Bornmann, L. (2011). Integrated impact indicators compared with impact
factors: an alternative research design with policy implications. Journal of the American
Society for Information Science and Technology, 62(11), 2133-2146.
Lozano, G.A. (2010). A new criterion for allocating research funds: “impact per dollar”. Current
Science, 99(9), 1187-1188.
Lundberg, R. (2007). Lifting the crown-citation z-score. Journal of Infometircs, 1(2), 145-154.
Moed, H.F. (2010). Measuring contextual citation impact of scientific journals. Journal of
Informetrics, 4, 265-277.
Moed, H.F., & Van Leeuwen, T.N. (1995). Improving the accuracy of Institute for Scientific
Information journal impact factors. Journal of the American Society for Information
Science, 46(6), 461-467.
Morgan, V., & Janca, A. (2000). Revisiting the journal impact factor. Australasian Psychiatry,
8(3), 230-235.
Põder, E. (2010). Let's correct that small mistake. Journal of the American Society for
Information Science and Technology, 61(12), 2593-2594.
Rossner, M., Van Epps, H., & Hill, E. (2007). Show me the data. The Journal of Cell Biology,
179(6), 1091-1092.
Rothenberg, R. (2008). The impact factor follies. Epidemiology, 19(3), 372.
Seglen, P.O. (1997). Why the impact factor of journals should not be used for evaluating
research. British Medical Journal, 314(7079), 498-502.
Vakil, N. (2005). The journal impact factor: judging a book by its cover. American Journal of
Gastroenterology, 100(11), 2436-2437.
West, J.D., & Bergstrom, C.T. (2010). The EigenfactorTM Metrics: a network approach to
assessing scholarly journals. College and Research Libraries Journal, 71, 236-244.
Whitehouse, G.H. (2001). Citation rates and impact factors: should they matter? British Journal
of Radiology, 74(877), 1-3.

14