RESEARCH ARTICLE - Ricerca sull'intelligenza artificiale specializzata al MIT

RESEARCH ARTICLE

Measuring and interpreting the differences of the
nations’ scientific specialization indexes by
output and by input

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

j o u r n a l

Giovanni Abramo1

, Ciriaco Andrea D’Angelo1,2

, and Flavia Di Costa2

1Laboratory for Studies in Research Evaluation, Institute for System Analysis and Computer Science (IASI-CNR),
National Research Council of Italy, Rome, Italy
2Department of Engineering and Management, University of Rome “Tor Vergata,” Rome, Italy

Citation: Abramo, G., D’Angelo, C. A., &
Di Costa, F. (2022). Measuring and
interpreting the differences of the
nations’ scientific specialization
indexes by output and by input.
Quantitative Science Studies, 3(3),
755–775. https://doi.org/10.1162/qss_a
_00206

DOI:
https://doi.org/10.1162/qss_a_00206

Peer Review:
https://publons.com/publon/10.1162
/qss_a_00206

Received: 20 April 2022
Accepted: 20 Luglio 2022

Corresponding Author:
Giovanni Abramo
giovanni.abramo@iasi.cnr.it

Handling Editor:
Ludo Waltman

Copyright: © 2022 Giovanni Abramo,
Ciriaco Andrea D’Angelo, and Flavia Di
Costa. Published under a Creative
Commons Attribution 4.0 Internazionale
(CC BY 4.0) licenza.

The MIT Press

Keywords: allocation efficiency, bibliometrics, disciplinary profiles, research efficiency, scientific
specialization index

ABSTRACT

This paper compares the national scientific profiles of 199 countries in 254 fields, tracked by
two indices of scientific specialization based respectively on indicators of input and output.
For each country, the indicator of inputs considers the number of researchers in each field. IL
output indicator, named Total Fractional Impact, based on the citations of publications indexed
in the Web of Science, measures the scholarly impact of knowledge produced in each field.
For each country, the approach allows us to measure the deviations between the two profiles,
thereby revealing potential differences in research efficiency and/or capital allocation across
fields, compared to benchmark countries.

INTRODUCTION

Policy-makers who have knowledge of the scientific specializations of their country can better
formulate research policies and funding priorities, including by specific field, and can better
assess the effectiveness of their initiatives in relation to strategic priorities. Whether public or
private, Tuttavia, stakeholders face major challenges in identifying scientific priorities and
then parceling their investments (King, 2004; May, 1997). What is necessary is not only knowl-
edge of the home nation scientific profile but also its relation to those of other countries, at
regional and global levels.

The measurement of research activity and the construction of a national scientific profile
can be carried out by considering either the input employed (resources and capital investment,
research personnel, eccetera.) or the output produced (competenza, scientific publications, patents,
eccetera.); questo è, the knowledge developed and its scholarly impact (Sugimoto & Larivière, 2018).

In a previous work, for purposes of tracing the scientific profiles of countries, we proposed
an index of scientific specialization based on scholarly impact of 2010–2019 Web of Science
( WoS) publications in each subject category (SC) (Abramo, D’Angelo, & Di Costa, 2022UN). By
producing a specialization profile for each country in relation to all SCs (254), we were able to
identify the distinctive characteristics of individual countries and country clusters.

Tuttavia, if we consider the whole process of scientific research production as a black box,
the calculation of specialization indices can also be carried out by considering input indicators

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

alongside the output indicators. The former approach traces the profile of a country through
the sectoral distribution of research investments; the latter through the relative distribution of
its scientific production.

From an operational point of view, tracing the research profile of a country on the basis of
input indicators is a challenging task, because at the global level, gathering input data disag-
gregated by field is formidable, even more so by univocal classification of those fields. Input
dati, or production factors according to the microeconomic theory of production, are labor (l)
and capital (K ); questo è, all resources other than labor used to conduct research activities. While
K data are not available, in this paper we go some way to overcoming the obstacle concerning
L data. Infatti, the bibliometric approach allows not only measurement and classification of
produzione, through observation of scientific output, but indirectly also the input, limited to the
research staff. Infatti, having understood how to disambiguate authors’ identities and their
country affiliations, this makes it possible to measure the size of the research staff of a country
and to classify it per SC based on the prevalent SC in which each author’s publications fall. È
then possible to measure the scientific specialization of countries with input data (limited to L),
in a similar way as with output data.

It is then interesting to check whether and to what extent the resulting scientific profiles are
different. The share of research fields showing deviations between the two indices would
reveal differences in research efficiency and/or allocation of K across fields, compared to
benchmark countries. Infatti, because research output is a function of L and K, if a field spe-
cialization index is higher by input than by output, a possible explanation is that the country
has historically invested less K in that field than in others and/or that the productivity of the
researchers, compared to other countries, is lower in that field. When the share of such fields
surpasses one half, the inference would be that the country is entering the area of imbalance
across fields, in the efficiency of their research and/or capital allocations. Were K data avail-
able and accounted for, those differences would reveal directly field-level comparative advan-
tages across countries.

Essentially, to move the national research profile towards alignment with strategic objec-
tives, governments can act on two levers: differentiated allocation of public funds across fields,
and/or differentiation of productivity incentives by scientific fields, although the latter would
not be easy in practice. In any case, the effects of these interventions on field outputs of
research, and on shifting the scientific profile, is in part dependent on the status of productivity
across these very fields.

The objectives of the present work are therefore, for each country

(cid:129) produce two specialization profiles, respectively based on input and output indicators,

corresponding to each of the 254 SCs of the WoS classification scheme;

(cid:129) analyze the two specialization profiles of countries by input and output indicators; E
(cid:129) assess the deviations between the two profiles for individual countries and country

clusters;

all this in a manner supportive of policy-makers intending to formulate research policies and
priorities for funding by field.

The next section of this paper reviews the relevant literature. Sezione 3 describes the data
and indicators used for analysis, and the methodology adopted for construction of the special-
ization profiles. Sezione 4 presents the results of the analysis and Section 5 comments the main
findings and discusses the policy implications.

Quantitative Science Studies

756

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

2. LITERATURE REVIEW

Scholars have generally applied frameworks from business or economics in studying special-
ization levels in scientific research. The most common approach is by “revealed comparative
advantage” (Aksnes, Sivertsen et al., 2017; Allik, Realo, & Lauk, 2020; Bongioanni, Daraio
et al., 2015; Cimini, Zaccaria, & Gabrielli, 2016; Horta & Veloso, 2007; Leydesdorff & Wagner,
2009; Li, 2017; Patelli, Cimini et al., 2017; Sandström & Van den Besselaar, 2018). Examining
a field at international level, this approach “reveals” the comparative advantage of a country in
proportions of labor factor, or output produced, compared globally or to a selection of coun-
tries. All comparative advantage indices used in international economics originate from the
Balassa or “RCA” index (Balassa, 1965). The first to transfer RCA to investigation of speciali-
zation in scientific research was Frame (1977), who introduced the so-called “activity index.1”
This indicator is typically based on one of the easily measured macroscopic bibliometric var-
iables: total publications from a country; total citations received by the country’s publications
(Aksnes, van Leeuwen, & Sivertsen, 2014; Harzing & Giroud, 2014); and in some case more
sophisticated combinations of output and impact (Abramo, D’Angelo, & Di Costa, 2014;
Abramo et al., 2022UN).

The value of the activity index is given by the ratio of two ratios. The first one measures the
share of research effort (or output) of a country in a given field with respect to the national
total, and the second one measures the same share but at a global level. The indicator is
expressed as an absolute value or transformed on a scale [−100; +100] for easier understand-
ing and comparison.

Subsequent to detailed analysis of its technicalities, Glänzel (2000), and Schubert and
Braun (1986) have provided interpretations of this indicator. Other authors have explored the-
oretical problems in the construction of the activity index and related indicators (Aksnes et al.,
2014; Rousseau, 2018, 2019; Rousseau & Yang, 2012).

The bibliometric indicators generally used are based on output data extracted from biblio-
graphic repertories ( WoS, Scopus) Quale, despite coverage problems (by discipline, lingua,
country, eccetera.), have become the de facto standard for measuring research, and more generally,
for studies in the field of the so-called “science of science” (Archambault, Vignola-Gagné
et al., 2006; Hicks, 1999; Waltman, 2016). Compared to other approaches of measuring
research, bibliometrics clearly has the advantage of access to data, gathered by repository pub-
lishers according to globally standardized procedures.

In contrasto, input data are generally collected through local and international surveys,
under the auspices of national research councils or international organizations, such as OECD
and UNESCO. Although such entities collect and regularly update their data, none have the
mandate or capacities to apply standard classification systems, so none can provide data suf-
ficient for reliable study of specialization. Given the inaccessibility of data on inputs, scholars
interested in the investigation of specialization at macro (cioè., country) level have thus far
engaged solely with data on outputs.

D'altra parte, there is no shortage of analyses on input and output data at meso level
(cioè., surveys of data on a small set of local institutions, enabling evaluation of their speciali-
zation). Heinze, Tunger et al. (2019), Per esempio, described research and teaching profiles for
68 public universities in Germany (from 1992 A 2015) and produced specialization maps for

1 Activity index (AI) was originally defined as the ratio between the country’s share in the world’s publication
output in the given field and the country’s share in the world’s publication output in all fields (Frame, 1977).

Quantitative Science Studies

757

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

each of them. Fuchs and Heinze (2021) then revised the analysis on an updated data set (1992
A 2018). Teixeira, Rocha et al. (2012) adapted one output and three input measures from the
RCA index of Balassa (1965) in the study of field-by-field diversity (specialization and/or diver-
sificazione) of Portuguese higher education institutions.

Thus far however, in measurement of specialization at macro/country level, for the reasons
explained above, there remain no works using input data. In this paper we try to fill this gap,
using the bibliometric approach.

3. DATA AND METHODS

Observing the authorship of scientific publications, then taking on the task of disambiguating
the author identities, and tagging by country affiliation and field of specialization, we are ulti-
mately able to measure the size of a country’s research staff in a given field. This input measure
can then be used to construct the country’s sectoral specialization profile in terms of inputs, In
the manner of traditional approaches dealing only with outputs. In the following, we explain
the methodological details.

The data set for the analysis is the same as previously used by Abramo et al. (2022UN), Quale
applied the rule-based scoring and clustering algorithm of Caron and van Eck (2014) to data
extracted from the in-house WoS database of the Centre for Science and Technology Studies
(CWTS) at Leiden University (updated to the 13th week of 2021). For this algorithm, biblio-
metric metadata on authors and their publications are taken as input, and clusters of publica-
tions likely to be written by the same author are taken as output. The algorithm considers four
categories of bibliographic elements:

(cid:129) author name (first and last name, affiliation, email);
(cid:129) article (shared coauthors, grant numbers, address not linked to authors);
(cid:129) source (SC, journal); E
(cid:129) citation (self-citations, cocitations, bibliographic coupling).

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The higher the number of shared bibliographic elements (source, topic, coauthors, emails,
affiliations, references, eccetera.) between two publications, the stronger is the evidence that these
are written by the same author.

Based on scoring values and thresholds, defined on a verified seed set, the algorithm

develops clusters of publications and assigns them to an individual.

Ovviamente, the algorithm is far from being error free, especially for authors with popular
names, or production of highly diversified and heterogeneous bibliographic elements, UN
circumstance that could lead to splitting their portfolio in two or more clusters.

Tuttavia, at the aggregate country level, this latter error, as extensively explained in the
theory and methodology of the previous work, will have only marginal effects on analytical
risultati. Referring to Abramo et al. (2022UN), an important note is that to increase robustness of
the analysis, the data set excludes those clusters that fail to comply with one or more of the
following conditions:

(cid:129) contain at least 10 publications (excludes “occasional” researchers, for whom clustering

has lower confidence levels);

(cid:129) of which at least one publication is after 2018 (designed to exclude researchers no

longer active); E

Quantitative Science Studies

758

The nations’ scientific specialization indexes by output and by input

(cid:129) with a “research age”2 of minimum 5 years (designed to include only “established”

researchers).

Through such “cuts” we effectively exclude small clusters, related to very young or occa-
sional researchers but also those related to researchers no longer active (per esempio., who are now
retired). We also exclude part of those clusters deriving from the splitting of authors with
popular names and/or with highly diversified scientific production, caused by the Caron
and van Eck algorithm. All this allows us to have a higher confidence that the resulting data
set actually represents the research staff of a given country, at present.

The final data set consists of over 2 million clusters, accounting for over 120 million author-
ships, related to almost 17 million unique publications. On average each cluster contains 58
publications, and each unique publication is coauthored by eight distinct clusters.

For field classification purposes, we use the WoS scheme, including 254 SCs3. Each cluster
in the data set is provided with the 2010–2019 related WoS indexed publications4 and is
associated with a field, given by the “prevalent” SC of its publications (cioè., the one hosting
most of his or her scientific production)5. In the input-based approach, the specialization index
(IB)SIjk of country k, in the SC j is

IBð

ÞSIjk ¼ RSjkP
jRSjk

k RSjk
P

k RSjk

;

(1)

where RSjk = research staff, operationalized as number of clusters of the country k in the SC j.

The higher the value of SIjk compared to 1, the more specialized the country k is in SC j, COME
the share of its research staff is higher than the expected value observed at world level. If SIjk is
less than 1 it means that no specialization is involved in SC j for country k.

In the output-based approach, instead, we use the composite indicator proposed in Abramo
et al. (2022UN), and called Total Fractional Impact (TFI ), which is a combination of publication
volume and field normalized citation impact. The TFI of a country k in SC j, is defined as
X

(2)

TFIjk ¼

Njk
i¼1 fik

;

ci
C(cid:1)j

Dove

Njk = number of publications of country k, in SC j
fik = fractional contribution of coauthors of country k to publication i. For a publication with

n coauthors, m of which are affiliated to country k, fik is equal to m/n6

ci = citations received by publication i (counted at the 13th week of 2021)
C(cid:1)j = average citations received by all cited publications of the same year and SC j of

publication i 7

2 Given by the difference between the first and the last publication year assigned to the cluster.
3 In WoS each publication inherits the SC of the hosting journal.
4 Only articles, recensioni, letters, and proceedings papers.
5 Clusters with more than one prevalent SC are around 2% and are counted multiple times.
6 Note that according to the CvE algorithm, each cluster (and thus each author) is associated with one and

only one country.

7 Abramo, Cicero, and D’Angelo (2012) demonstrated that the average of the distribution of citations received

for all cited publications of the same year and SC is the best-performing scaling factor.

Quantitative Science Studies

759

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Applying Total Fractional Index, we can measure the output-based index of specialization

(OB)SIjk of country k in SC j as

OBð

ÞSIjk ¼ TFIjkP
jTFIjk

k TFIjk
P

k TFIjk

(3)

In this case a value higher than 1 implies that country k is specialized in SC j, as the share of

TFI in such SC is higher than the expected value observed at world level, and vice versa.

Countries can be more or less concentrated (diversified) in terms of scope (number of SCs)
of research. We will assess that by the Gini index, or Gini coefficient, which measures variable
distribution across a population (Gini, 1921). A higher Gini coefficient indicates greater
inequality in the distribution of input (produzione) across SCs, with high-input SCs receiving much
larger shares of the total input for research. The Gini coefficient ranges from 0 A 1, con 1
representing perfect inequality (concentration) E 0 representing perfect equality
(diversification).

4. RESULTS

The analyses of the current paper, come segue, are aimed at comparing the distributions of
SIjk calculated from input and output data. For this, we construct 199 × 254 matrices con-
taining the SI values, by input and output, for a set of 199 countries in each of the 254 WoS
SCs. For reasons of space, we present only a few examples of possible data elaborations. IL
complete data on all 199 countries in 254 SCs are found in Abramo, D’Angelo, and Di Costa
(2022B).

As a first example, Figura 1 shows, for China, the distribution of SIs detected for the SCs of
Biomedical Research (14 in all). The SI values measured through output are never greater than
unity; instead, when measured through input, five fields out of the 14 reach levels greater than
unity. IL (OB)SI values are higher than the (IB)SI values in only four cases: Among these, IL
highest absolute values are in Toxicology (0.759 by output data, 0.639 by input data). In abso-
lute value, the greatest gap is in Medical Laboratory Technology (0.882 vs. 2.859), followed by
Virology (1.136 vs. 0.592) and Oncology (1.175 vs. 0.678). It therefore emerges that for China,
in general, there is a significant lack of specialization in this set of SCs, and above all a gap in
capital investment and/or productivity, compared to other countries.

Figura 2 shows the comparison for the United States, looking at the SI values for input and
output in the 20 SCs that are greatest by world output. In 15 out of 20, IL (OB)SI value is
higher than the (IB)SI value based on input, with a maximum deviation in Medicine, General
& Internal; in this field, for the United States, IL (OB)SI is 1.368, compared to an SI by input of
0.831. At the opposite extreme for these 20 SCs is Chemistry, Multidisciplinary which shows
an (IB)SI of 1.267 versus an (OB)SI of 0.743 by output, or in other words, 41% less. Also for the
stati Uniti, whether for specialization index by input or output, there are nine SCs with
values greater than unity, and of these, eight SCs represent the particular case where both
SI values are greater than unity (Astronomy & Astrophysics; Biochemistry & Molecular Biology;
Cardiac & Cardiovascular Systems; Clinical Neurology; Neurosciences; Oncology; Public,
Environmental & Occupational Health; Surgery). For these eight SCs, the percentage variation
between the two SI values was within the ±10% in 10 out of 20 cases.

Figura 3 instead examines Biochemistry & Molecular Biology, looking at the 20 largest
countries by overall world share in the SC. For these, the radar graph shows a mismatch in
the values of the specialization indices for some countries: especially for Russia (1.865 by

Quantitative Science Studies

760

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Figura 1. China: specialization indices for the subject categories in “Biomedical research”.

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

Figura 2. stati Uniti: specialization indices for the 20 subject categories that are largest by world output.

Quantitative Science Studies

761

The nations’ scientific specialization indexes by output and by input

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

Figura 3. Biochemistry and Molecular Biology: specialization indices of the 20 largest countries
by world share of output.

input vs. 1.040 by output), followed by Poland (1.572 vs. 1.191) and South Korea (1.316 vs.
0.911). Eight other countries on the list have SI values by input that are higher than those cal-
culated by output; the opposite relation is seen in nine countries. The difference between
values of the indicator falls in ±10% for eight countries out of the total 20 (Australia, Iran, Italy,
Japan, Spain, Svizzera, United Kingdom, stati Uniti).

Tavolo 1 provides an examination of the specialization profiles for the major European coun-
tries in terms of research output, specifically their top five SCs by specialization index based on
input ((IB)SI ) and output data ((OB)SI ). All five of these European countries show a strong pres-
ence of “top” SCs (Di 1/3 of the total, for both input and output) in the humanities and
social sciences. Also interesting is that the intersection between the two sets of categories is
rather limited: For France, Germany and Netherlands, two SCs appear in both columns; Italy
and Spain have only one with a double appearance, and the United Kingdom has none.
Finalmente, in this table, the top values of (IB)SI are greater than the corresponding top values
Di (OB)SI in 24 del 30 total cases.

In Table 2, for the top seven countries by share of output, we look into the two SCs char-
acterized by maximal difference between (IB)SI and (OB)SI, both negative and positive. In
other words, for each country, columns 2–3 report the SCs with evident gaps in either or both
of capital investment and productivity, given that the specialization indexes by output data do
not align with what emerges concerning inputs. For China, Per esempio, the maximal nega-
tive case ((OB)SI − (IB)SI ) is found in Medicine, Research & Experimental, and in Mathemat-
ics, Interdisciplinary Applications; for Russia, this is found in Chemistry, Applied and Mining
& Mineral Processing.

Columns 4–5 report the opposite situation (cioè., SCs with maximal difference of SI by output
data over input data), evidently due to higher capital allocation and/or productive efficiency

Quantitative Science Studies

762

The nations’ scientific specialization indexes by output and by input

Tavolo 1. Major European countries: top five SCs by specialization indices

Country
France

Acoustics

Input data

SIjk
2.997

Literary Reviews

Output data
SC

Imaging Science & Photographic Technology

2.700

Critical Care Medicine

2.369

Logic

Mechanics

2.299

Geochemistry & Geophysics

2.031

Physics, Fluids & Plasmas

SIjk
2.584

2.315

2.271

2.230

2.079

Germany

Letteratura, German, Dutch, Scandinavian

8.230

Letteratura, German, Dutch, Scandinavian

8.116

Medical Ethics

7.091

Psychology, Psychoanalysis

3.190

Microscopy

Social Sciences, Mathematical Methods

3.124

Radiology, Nuclear Medicine &

Medical Imaging

Psychology, Educational

2.793

Dermatology

Italy

Instruments & Instrumentation

3.124

Arte

Geography, Physical

Architecture

Mineralogy

Limnology

3.035

Architecture

2.810

Andrology

2.790

Medical Laboratory Technology

2.598

Engineering, Geological

Netherlands

Development Studies

6.191

Psychology, Mathematical

6.170

Public Administration

Ethnic Studies

5.060

Regional & Urban Planning

Social Sciences, Mathematical Methods

4.793

Primary Health Care

Public Administration

4.198

Social Issues

3.331

2.441

1.982

1.968

3.180

3.170

2.716

2.239

2.212

4.365

4.123

3.573

3.523

3.189

Spain

Literary Theory & Criticism

7.705

Letteratura, Romance

10.502

Psychology, Biological

Letteratura, Romance

4.220

Food Science & Tecnologia

3.566

Horticulture

Psychology, Multidisciplinary

3.554

Agriculture, Multidisciplinary

Education & Educational Research

3.412

Ornithology

United Kingdom

Ethnic Studies

7.587

Dance

Development Studies

6.295

Letteratura, British Isles

History of Social Sciences

5.966

Theater

Social Sciences, Biomedical

5.861

Cultural Studies

Classics

5.646

Medieval & Renaissance Studies

Quantitative Science Studies

3.095

2.501

2.436

2.269

7.250

6.601

6.184

5.531

5.068

763

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Tavolo 2.

Subject categories with min(max) (OB)SI − (IB)SI difference for the top seven countries by share of output

Country
China

Medicine, Research & Experimental

(OB)SI − (IB)SI
−1.977

SC
Computer Science, Cybernetics

(OB)SI − (IB)SI
+2.182

Mathematics, Interdisciplinary Applications

France

Letteratura, British Isles

Imaging Science & Photographic Technology

Germany

Medical Ethics

Social Sciences, Mathematical Methods

Japan

Engineering, Ocean

Limnology

−1.860

−1.378

−1.255

−6.423

−2.131

−1.605

−1.524

Physics, Condensed Matter

Logic

Letteratura

Psychology, Biological

+1.186

+2.271

+1.025

+1.004

Quantum Science & Tecnologia

+0.977

Cell & Tissue Engineering

+1.253

Quantum Science & Tecnologia

+1.240

Russia

Chemistry, Applied

−13.760

Letteratura, Slavic

Mining & Mineral Processing

United Kingdon

Ethnic Studies

Social Sciences, Biomedical

stati Uniti

Education, Special

Poetry

−4.691

−4.696

−3.561

−1.844

−1.518

Paleontology

Poetry

Medieval & Renaissance Studies

+2.649

Limnology

Anatomy & Morphology

+0.809

+0.620

+11.708

+2.255

+3.173

compared to other SCs. For China, Per esempio, such virtuous cases occur in Computer Sci-
ence, Cybernetics and in Physics, Condensed Matter, while for the United States in Limnology
and in Anatomy & Morphology.

Tavolo 3 reports, for each of the top 20 countries by share of output, the shares of SCs
con (IB)SI greater than unity; (OB)SI greater than unity; E (OB)SI greater than (IB)SI. Within
this group of 20 we quickly note some G7 countries, such as the United States, United
Kingdom, Germany, and Canada, at the bottom of the table, but also another G7 country—
Italy—near the top of the list. The first four countries in the list have about 70% of SCs with
(OB)SI greater than (IB)SI, the last four about 50%. It should be noted, Tuttavia, that the
latter case describes capital allocation and efficiency of research that are more balanced
across fields.

4.1. Concentration/Diversification in Country Disciplinary Profiles

The disciplinary profile of a country can be more or less specialized in a few SIs or distributed
in many (diversified or “balanced”). In this regard, there are interesting differences between
countries when considering SIs based on input or output data. Tavolo 4 shows, for the top
20 countries by share of output, the value of the GINI coefficient (output data) and the relative
coefficients of variation of the distributions of SI values for the 254 SCs (input and output data).
For all 20 countries except Iran, the GINI value for their (IB)SI distribution is greater than the
value for (OB)SI. Russia, Iran, and India, in view of the high values of GINI coefficients calcu-
lated in both modes, are the countries with highest level of concentration of sectoral

Quantitative Science Studies

764

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Tavolo 3.

Share of subject categories with (IB)SI and (OB)SI above one, E (OB)SI higher than (IB)SI for top 20 countries by share of output

Country
Turkey

Italy

Brasile

Poland

Russia

India

Japan

Svizzera

France

South Korea

Netherlands

Sweden

Iran

Spain

Germany

Australia

United Kingdom

stati Uniti

China

Canada

* With at least one researcher.

No of SCs*
219

Of which with
(IB)SI > 1 (%)
33.8

Of which with
(OB)SI > 1 (%)
42.5

Of which with
(OB)SI > (IB)SI (%)
83.1

238

222

207

210

224

234

236

219

246

234

202

242

248

250

254

232

250

35.7

29.3

35.1

30.9

30.5

25.9

37.2

36.0

32.0

46.7

47.0

41.1

41.3

37.1

55.2

54.4

54.3

34.5

57.2

42.9

33.3

41.0

28.0

34.8

31.3

39.7

34.3

33.3

55.3

52.1

38.1

45.5

38.3

58.9

57.6

50.4

30.6

56.8

76.5

72.5

71.6

69.6

67.1

66.1

62.0

61.4

61.2

61.0

60.7

59.9

58.7

57.7

55.2

50.8

50.4

50.0

49.2

specializations. By contrast, the lowest values are recorded for the United States and Canada.
Examining still further, Russia not only has the highest values of both GINI indicators (cioè., IL
profile strongest in specialization) Ma, along with China, India, and Iran, also has the lowest
differences between the two values (ΔGINI 0.444). Basically, in all four of these countries,
input and output are concentrated in certain fields functional to a specific industrialization
modello, most probably of historic character. The contrary situation of great difference between
input and output distribution is observed for Switzerland (0.457 vs. 0.313), Sweden (0.412 vs.
0.279), and Turkey (0.579 vs. 0.466). On observing the variation coefficient, instead of GINI,
similar trends in disciplinary profiles emerge: The largest differences between coefficients for
distribution of (IB)SI and (OB)SI are for Switzerland and Poland; the smallest for Russia and
Iran. Overall (Tavolo 4) the values of variation coefficient fall in the intervals 0.604–2.025 for
(IB)SI and 0.439–2.220 for (OB)SI.

Quantitative Science Studies

765

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Tavolo 4. Dispersion of national disciplinary profiles and GINI concentration indexes for top 20 countries by share of output

Country
Russia

Iran

India

Brasile

China

Poland

Japan

Turkey

South Korea

Netherlands

United Kingdom

France

Svizzera

Italy

Australia

Spain

Germany

Sweden

Canada

stati Uniti

Input data

Output data

GINI
coefficient
0.750

Variation
coefficient
2.025

GINI
coefficient
0.706

Variation
coefficient
2.220

0.599

0.595

0.580

0.540

0.576

0.513

0.579

0.533

0.440

0.416

0.395

0.457

0.408

0.394

0.366

0.372

0.412

0.356

0.327

1.248

1.182

1.434

1.020

1.515

0.958

1.197

1.111

0.873

0.865

0.709

1.552

0.756

0.821

0.791

0.872

0.812

0.922

0.604

0.607

0.576

0.519

0.517

0.471

0.466

0.460

0.363

0.351

0.324

0.313

0.311

0.300

0.291

0.289

0.279

0.244

0.243

1.262

1.137

1.186

0.962

1.000

0.850

0.971

0.878

0.661

0.749

0.583

0.591

0.569

0.564

0.751

0.684

0.544

0.460

0.439

ΔGINI
0.044

–0.008

0.019

0.061

0.023

0.105

0.047

0.113

0.073

0.077

0.065

0.071

0.144

0.097

0.094

0.075

0.083

0.133

0.112

0.084

Δ Variation
coefficient
–0.195

–0.014

0.045

0.248

0.058

0.515

0.108

0.226

0.233

0.212

0.116

0.126

0.961

0.187

0.257

0.040

0.188

0.268

0.462

0.165

Figures 4 E 5 compare the national disciplinary profiles of the United States and Russia,
the two countries already noted at the antipodes in specialization/differentiation of scientific
profiles in terms of (IB)SI and (OB)SI. A first observation is that for both indices, the values for
the United States never exceed 4.5. On the contrary, the trends for Russia show pronounced
oscillations: (IB)SI, while in the range 0–4 for 237 del 254 SCs, presents a number of sharp
peaks, two of which are close to the value 16; for (OB)SI the trend is to even more oscillations,
although with peaks not surpassing 8.

Finalmente, we investigated the relationship between the dispersion of the national profiles of
the top 20 countries by share of output and the balance of efficiency of research and/or capital
allocation across fields. The correlation analyses showed that countries with high dispersion
are those more balanced (for (IB)SI, Pearson correlation coefficient: 0.543; Spearman correla-
tion coefficient: 0.583; for (OB)SI, 0.420 E 0.514, rispettivamente).

Quantitative Science Studies

766

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Figura 4. United States and Russia: dispersion of national disciplinary profiles, SI based on input data.

For all 199 countries examined, Figures 6 E 7 show, on input and output sides, the world
quantile maps of the GINI coefficient of the SI specialization index. Both maps show the pres-
ence of balanced vs. unbalanced research profiles, the former being typical of developed
countries, the latter of developing countries. Tuttavia, not only the “top” countries seen ear-
lier, but almost all (189/199) nations show a higher value of input-based than output-based
GINI coefficient (cioè., profiles that are more distributed on the input side). The largest differ-
ences are found for Latvia (0.879 vs. 0.650), Luxembourg (0.844 vs. 0.630), and Croatia (0.741
vs. 0.530).

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

Figura 5. United States and Russia: dispersion of national disciplinary profiles, SI based on output data.

Quantitative Science Studies

767

The nations’ scientific specialization indexes by output and by input

Figura 6. GINI coefficient of specialization index (SI)—world map based on input.

4.2. Clusters of Countries by Research-System Disciplinary Profile

In the previous sections we used specialization indices based on input and output data to
reveal the scientific profile of countries, and especially to compare their disciplinary charac-
terization with respect to all other countries. Such indices can also be used to group countries
by similarity of respective profiles. We do this by grouping according to Ward’s dissimilarity
(Ward, 1963), after principal component analysis (PCA) for reduction of the 254 SC speciali-
zation indices to seven principal components8, beginning from both input and output data.
The results are shown Tables 5 E 6, for input and output. There is partial overlapping in
the composition of the identified groups but also an evident partial reconfiguration of the
clusters when considering one or the other sides of data.

Taking either approach, the first cluster lacks the top countries by share of output seen

earlier, including only East African countries, with Ghana also in the output approach.

China, India, and Iran gather in a cluster in both approaches, but the other associated
countries change: Taking the input approach, the cluster includes a concentration of Middle
Eastern, Asian, and North African countries, united (apart from a few) by linguistic-cultural
factors, among which are some “tigers of the East” (Indonesia, Malaysia, Thailand).

Russia occupies a cluster as the sole top country, along with three post-Soviet countries also
(Belarus, Kazakhstan, Ukraine). Note that many of the other post-Soviet countries appear in
cluster 7 in the input approach, without any top country by share of output; and in cluster 3
in the output approach (along with Poland as a top country).

8 “Principal components” are new variables constructed as linear combinations of initial variables. The initial
variables are the SIs on 254 SCs, combined so that the new variables are uncorrelated and most information
within the initial variables is stored in the first components. Here, 254-dimensional data yields 254 principal
components, but PCA maximizes information in the first ones, achieving a reduced data set focused on the
first few components but without important loss of information. Specifically, the first seven components
explain about 50% of the variability of the original information, both with input and with output data.
Hence, we limit our analysis to these seven components and to as many clusters of countries.

Quantitative Science Studies

768

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Figura 7. GINI coefficient of specialization index (SI)—world map based on output.

Clusters 5 (input data) E 6 (output data) are quite similar, with the top countries all
English-speaking plus the Netherlands in the input approach, and Netherlands plus Sweden
in the output approach.

France, Germany, Italy, and Switzerland are all present in clusters 6 (input data) E 7 (fuori-
put data). Spain groups with these only for the input approach, while considering the output
side, it appears as the sole top country of a cluster together with a number of Latin American
countries. The situation of Japan is also singular, being associated with Brazil and Poland in
the input approach and with France, Germany, Italy, and Switzerland in the output approach.

Tavolo 5. Clustering of countries (based on Ward’s dissimilarity), after principal component analysis related to input data, reducing the 254
subject categories specialization indexes to seven principal components

Cluster
1

Top countries
–

Ethiopia; Kenya; Tanzania; Uganda

Other countries

Brasile; Japan; Poland

Argentina; Bulgaria; Cameroon; Ecuador; Mexico; Nigeria; Peru; Uruguay;

Venezuela

China; India; Iran

Algeria; Bangladesh; Colombia; Egypt; Iceland; Indonesia; Iraq; Jordan;
Kuwait; Malaysia; Morocco; Oman; Pakistan; Qatar; Romania; Saudi
Arabia; Serbia; Sri Lanka; Thailand; Tunisia; United Arab Emirates;
Vietnam

Russia

Belarus; Kazakhstan; Ukraine

Australia; Canada; Netherlands;

United Kingdom; stati Uniti

Belgium; Ireland; Israel; New Zealand; Norway

France; Germany; Italy; South Korea;

Austria; Chile; Denmark; Finland; Greece; Hungary; Lebanon; Portugal;

Spain; Sweden; Svizzera;
Turkey

–

Singapore; Taiwan

Croatia; Cyprus; Czech Republic; Estonia; Ghana; Latvia; Lithuania;

Luxembourg; Philippines; Slovakia; Slovenia; South Africa

Quantitative Science Studies

769

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Tavolo 6. Clustering of countries (based on Ward’s dissimilarity), after principal component analysis related to output data reducing the 254
subject categories specialization indexes to seven principal components

Cluster
1

Top countries by share of output
–

Ethiopia; Ghana; Kenya; Tanzania; Uganda

Other countries

China; India; Iran

Brasile; Poland; South Korea; Turkey

Russia

Spain

Australia; Canada; Netherlands;
Sweden; United Kingdom;
stati Uniti

Algeria; Egypt; Iraq; Jordan; Luxembourg; Morocco; Pakistan; Qatar;
Saudi Arabia; Singapore; Tunisia; United Arab Emirates; Vietnam

Bangladesh; Bulgaria; Cameroon; Croatia; Czech Republic; Greece;
Indonesia; Kuwait; Latvia; Lebanon; Lithuania; Malaysia; Nigeria;
Oman; Portugal; Romania; Serbia; Slovakia; Slovenia; Sri Lanka;
Taiwan; Thailand

Belarus; Kazakhstan; Ukraine

Argentina; Chile; Colombia; Cyprus; Ecuador; Estonia; Iceland;
Mexico; Peru; Philippines; South Africa; Uruguay; Venezuela

Belgium; Denmark; Finland; Ireland; Israel; New Zealand; Norway

France; Germany; Italy; Japan;

Austria; Hungary

Svizzera

Allo stesso tempo, with the input data, these four countries correspond to a profile that assim-
ilates that of South Korea and Turkey, countries that instead associate with Brazil and Poland in
an output cluster.

Figures 8 E 9 show the ranking of the countries determined by input and output data
rispettivamente, but now limiting the analysis solely to principal components 1 E 2: UN

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

Figura 8. Dispersion of national disciplinary profiles for top 20 countries by share of output, based on the first two principal components
related to the input data. AU: Australia; BR: Brasile; CA: Canada; CH: Svizzera; CN: China; DE: Germany; FR: France; IN: India; IR: Iran; IT:
Italy; JP: Japan; KR: South Korea; NL: Netherlands; PL: Poland; RU: Russia; SE: Sweden; SP: Spain; TR: Turkey; UK: United Kingdom; US:
stati Uniti.

Quantitative Science Studies

770

The nations’ scientific specialization indexes by output and by input

Figura 9. Dispersion of national disciplinary profiles for top 20 countries by share of output, based
on the first two principal components related to the output data. AU: Australia; BR: Brasile; CA: Can-
ada; CH: Svizzera; CN: China; DE: Germany; FR: France; IN: India; IR: Iran; IT: Italy; JP: Japan;
KR: South Korea; NL: Netherlands; PL: Poland; RU: Russia; SE: Sweden; SP: Spain; TR: Turkey; UK:
United Kingdom; US: stati Uniti.

representation still more partial on an even greater restriction of the overall information con-
tained in the data9. Comparing the two graphs, we see that the rightmost cluster, containing
technically and scientifically advanced countries (Australia, Canada, Netherlands, United
Kingdom, stati Uniti) remains substantially unchanged in composition (with the exception
of Sweden, present only for output data), while the other clusters present different recombina-
tions of countries, the only other being the outlier character of Russia, isolated in both graphs.

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

5. DISCUSSION AND CONCLUSIONS

National research systems can be analyzed in terms of their scientific profiles, and their capital
allocation and productive efficiency, through the application of scientific specialization indi-
ces (SIs), in this way supporting policy-makers as they work to define and pursue the research
priorities of their countries. in questo documento, we have constructed indices of scientific specializa-
zione, calculated from both input and output data, for a set of 199 countries, operating in 254
WoS SCs. One of the aims was to conduct a comparative analysis drawing on the results of the
different SIs, more specifically: to produce, for each country, a dual specialization profile for
each SC; for each country and field, to measure the deviations between the values of the two
indices; and to observe how distinctive or common features of individual countries or clusters
of countries, in terms of their SIs for different fields, may vary depending on the point of view of
the index used.

For the calculation of the output-based specialization indices, we used the Total Fractional
Impact (TFI) (cioè., the sum of the impact of the individual publications produced by the country

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

9 Note that in Figure 8, PC1 is not centered on zero. The distribution of PC1 is indeed centered on zero for the
total 198 countries, but for the 20 largest in our analysis, in the input approach the values are all positive
with an average of 6.7.

Quantitative Science Studies

771

The nations’ scientific specialization indexes by output and by input

in each SC). Given that the rate of international collaboration (and therefore coauthorship) In
research varies from country to country, we adopted fractional counting to take into account
the contribution to each publication by researchers from each country. For calculation of the
input-based indices, we used the number of authors from the country in the SC, accepting that,
due to lack of information, we could not account for invested capital.

A value above one for SI in a given SC indicates a specialization of the country in that SC,
evidently because it presents some particular interest. Tuttavia, based on the construction of
the SI as a ratio of ratios, values higher than one are also naturally observed in all those SCs
where the share of either TFI or of researchers, although low in value at national level, È
nevertheless higher than the corresponding value at world level. This phenomenon is observed
for some nationally specific SCs of Art & Humanities, such as “Literature, German, Dutch,
Scandinavian” and “Literature, British Isles,” for example, where Germany and the United
Kingdom are at the top for the relative specialization indices.

Looking at the top 20 countries by share of output, the analysis of their share of SCs pre-
senting differences in indices on output and input sides revealed that most of the G7 countries
are characterized by very balanced capital allocation and efficiency of research across fields.
Exceptions would be Japan and especially Italy, which falls in a group of opposite character,
along with Turkey, Brasile, Poland, and Russia.

D'altra parte, the presence of SCs with large shares of the country’s total fractional
impact or researchers, and with SIs much higher than one, is clearly informative of the
research system structure, and reflects policy choices that have enhanced the concentration
on certain SCs over others.

Depending on the distribution of SI values among SCs, a country can therefore have a more
or less specialized or diversified disciplinary profile. In this regard, we observed that for all
countries but one (Iran), the GINI coefficient for distribution of (IB)SI is higher than for (OB)SI.
Russia, with the highest values of GINI coefficient on both input and output sides (0.750,
0.706), is the country with the strongest profile of specialization. Russia, along with Iran and
India, is also one of the countries with the smallest difference between the two concentration
indices: countries that have concentrated most of their resources on only a few sectors, follow-
ing a historic industrialization model that has accumulated expertise in specific sectors. IL
contrary profiles of the greatest differences between the (OB)SI and (IB)SI are instead seen in
Sweden, Svizzera, Canada, and the United States: countries that have diversified their
researchers across fields, and which have even more nuanced profiles of specialization when
measured through their output.

After PCA, reducing the 254 SC specialization indices to seven principal components, we
were able to identify seven clusters of countries by similarities in their profiles. There is partial
overlapping in the composition of the identified groups, but also an evident partial reconfig-
uration of the clusters when considering one or the other sides of data. China, India, and Iran,
and four of the English-speaking countries (Australia, Canada, United Kingdom, stati Uniti)
on the other, compose the nuclei of two groups that maintain similar specialization profiles
regardless of the approach.

In concluding, we note that the proposed analysis is not free of the intrinsic limits of the
bibliometric approach, inevitably with effects on analytical results. In particular, scientific pub-
lications in international scientific journals indexed in WoS represent only part of the total
output from research activity. This emerges as a criticality especially where the repertoires pro-
vide very low coverage, for example in the fields of Art & Humanities (Aksnes & Sivertsen,
2019), which are fields also suffering from uneven coverage. The choice of field classification

Quantitative Science Studies

772

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

scheme also remains critical. In this work, we implemented the one available in WoS, Quale
covers 254 SCs. The repertoire choice of a high number of fields allows good detail in profiling
the specializations of countries, but on the other hand reduces confidence in the analyses,
especially for smaller countries.

Other limitations concern citations as a proxy of scholarly impact, as not all citations are
positive or indicate real use by citing authors; and citations are not representative of all uses
(Abramo, 2018; Bornmann & Daniel, 2008; Tahamtan & Bornmann, 2018; Tahamtan, Safipour
Afshar, & Ahamdzadeh, 2016).

Finalmente, on the input side, the author name disambiguation algorithm is not free of errors,
which have an effect also on the accuracy of the output produced by each country. Most
importantly, when extracting research staff from publications’ metadata, we are not able to
account for unproductive researchers or researchers who do not publish in journals indexed
in WoS. Inoltre, due to a lack of data on capital investment by country (and even more so
by relative fields), the methodological approach to measurement of inputs considers only the
numbers of researchers. But research obviously depends on instrumental resources, not only
human, and ignoring investment differentials between countries certainly leads to analytical
bias. The difference in specialization of a country across fields, from the input and output
sides, can in fact have two explanations: higher/lower productivity of the country’s researchers
but also their higher/lower access to instrumental resources, compared to their colleagues in
other countries. For now, the distinction between the two determinants remains difficult to
investigate given the lack of data and of a collection framework that is both comprehensive
and detailed. D'altra parte, Tuttavia, we are addressing the question of higher/lower
differentials in the productivity of researchers by at least examining the feasibility of measure-
ment with respect to an international benchmark, country by country.

ACKNOWLEDGMENT

We are indebted to the Centre for Science and Technology Studies (CWTS) at Leiden Univer-
sity for providing us with access to the in-house WoS database from which we extracted data
as the basis of our elaborations.

AUTHOR CONTRIBUTIONS

Giovanni Abramo: Conceptualization, Investigation, Methodology, Supervision, Validation,
Writing—Original draft, Writing—Review & editing. Flavia Di Costa: Data curation, Investiga-
zione, Writing—Original draft. Ciriaco Andrea D’Angelo: Data curation, Formal analysis, Inves-
tigation, Methodology, Validation, Writing—Original draft.

COMPETING INTERESTS

The authors have no competing interests.

FUNDING INFORMATION

The research project received no funding.

DATA AVAILABILITY

Being subject to Clarivate-WoS license restrictions, the raw data cannot be made publicly
available. The complete results of our elaborations for all 199 countries in 254 SCs can be
found in Abramo et al. (2022B).

Quantitative Science Studies

773

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

REFERENCES

Abramo, G. (2018). Revisiting the scientometric conceptualization
of impact and its measurement. Journal of Informetrics, 12(3),
590–597. https://doi.org/10.1016/j.joi.2018.05.001

Abramo, G., Cicero, T., & D’Angelo, C. UN. (2012). How important is
choice of the scaling factor in standardizing citations? Journal of
Informetrics, 6(4), 645–654. https://doi.org/10.1016/j.joi.2012.07
.002

Abramo, G., D’Angelo, C. A., & Di Costa, F. (2014). A new biblio-
metric approach to assess the scientific specialization of regions.
Research Evaluation, 23(2), 183–194. https://doi.org/10.1093
/reseval/rvu005

Abramo, G., D’Angelo, C. A., & Di Costa, F. (2022UN). Revealing the
scientific comparative advantage of nations: Common and dis-
tinctive features. Journal of Informetrics, 16(1), 101244. https://
doi.org/10.1016/j.joi.2021.101244

Abramo, G., D’Angelo, C. A., & Di Costa, F. (2022B). Specialization
indexes of countries for 254 subject categories, by input and by
produzione [Data set]. Zenodo. https://doi.org/10.5281/zenodo
.6881520

Aksnes, D. W., & Sivertsen, G. (2019). A criteria-based assessment
of the coverage of Scopus and Web of Science. Journal of Data
and Information Science, 4(1), 1–21. https://doi.org/10.2478/jdis
-2019-0001

Aksnes, D. W., Sivertsen, G., van Leeuwen, T. N., & Wendt, K. K.
(2017). Measuring the productivity of national R&D systems:
Challenges in cross-national comparisons of R&D input and
publication output indicators. Science and Public Policy, 44(2),
246–258. https://doi.org/10.1093/scipol/scw058

Aksnes, D. W., van Leeuwen, T. N., & Sivertsen, G. (2014). IL
effect of booming countries on changes in the relative speciali-
zation index (RSI) on country level. Scientometrics, 101(2),
1391–1401. https://doi.org/10.1007/s11192-014-1245-3

Allik, J., Realo, A., & Lauk, K. (2020). The scientific impact derived
from the disciplinary profiles. Frontiers in Research Metrics and
Analytics, 5, 569268. https://doi.org/10.3389/frma.2020
.569268, PubMed: 33870047

Archambault, É., Vignola-Gagné, É., Côté, G., Larivière, V., &
Gingras, Y. (2006). Benchmarking scientific output in the social
sciences and humanities: The limits of existing databases. Scien-
tometrics, 68(3), 329–342. https://doi.org/10.1007/s11192-006
-0115-z

Balassa, B. (1965). Trade liberalisation and ‘revealed’ comparative
advantage. Manchester School of Economic and Social Studies,
33(2), 99–123. https://doi.org/10.1111/j.1467-9957.1965
.tb00050.x

Bongioanni, I., Daraio, C., Moed, H. F., & Ruocco, G. (2015). Com-
paring the disciplinary profiles of national and regional research
systems by extensive and intensive measures. Proceedings of ISSI
2015-Istanbul: 15th International Society of Scientometrics and
Informetrics Conference (pag. 684–696).

Bornmann, L., & Daniel, H.-D. (2008). What do citation counts
measure? A review of studies on citing behavior. Journal of
Documentation, 64(1), 45–80. https://doi.org/10.1108
/00220410810844150

Caron, E., & van Eck, N.-J. (2014). Large scale author name disam-
biguation using rule-based scoring and clustering. In E. Noyons
(Ed.), Proceedings of the Science and Technology Indicators
Conferenza 2014 (pag. 79–86). Universiteit Leiden.

Cimini, G., Zaccaria, A., & Gabrielli, UN. (2016). Investigating the
interplay between fundamentals of national research systems:
Performance, investments and international collaborations.

Journal of Informetrics, 10(1), 200–211. https://doi.org/10.1016
/j.joi.2016.01.002

Frame, J. D. (1977). Mainstream research in Latin America and the

Caribbean. Interciencia, 2(3), 143–148.

Fuchs, J. E., & Heinze, T. (2021). Two-dimensional mapping of
university profiles in research. ISSI2021: 18th International
Conference on Scientometrics & Informetrics (pag. 425–434).
KU Leuven, Belgium.

Gini, C. (1921). Measurement of inequality of incomes. IL
Economic Journal, 31(121), 124–126. https://doi.org/10.2307
/2223319

Glänzel, W. (2000). Science in Scandinavia: A bibliometric
approach. Scientometrics, 48(2), 121–150. https://doi.org/10
.1023/UN:1005640604267

Harzing, A.-W., & Giroud, UN. (2014). The competitive advantage of
nazioni: An application to academia. Journal of Informetrics, 8(1),
29–42. https://doi.org/10.1016/j.joi.2013.10.007

Heinze, T., Tunger, D., Fuchs, J. E., Jappe, A., & Eberhardt, P.
(2019). Research and teaching profiles of public universities in
Germany. A mapping of selected fields. Wuppertal: BUW.

Hicks, D. (1999). The difficulty of achieving full coverage of inter-
national social science literature and the bibliometric conse-
quences. Scientometrics, 44(2), 193–215. https://doi.org/10
.1007/BF02457380

Horta, H., & Veloso, F. M. (2007). Opening the box: Comparing
EU and US scientific output by scientific field. Technological
Forecasting and Social Change, 74(8), 1334–1356. https://doi
.org/10.1016/j.techfore.2007.02.013

King, D. UN. (2004). The scientific impact of nations. Nature,
430(6997), 311–316. https://doi.org/10.1038/430311a,
PubMed: 15254529

Leydesdorff, L., & Wagner, C. (2009). Macro-level indicators of the
relations between research funding and research output. Journal
of Informetrics, 3(4), 353–362. https://doi.org/10.1016/j.joi.2009
.05.005

Li, N. (2017). Evolutionary patterns of national disciplinary profiles
in research: 1996–2015. Scientometrics, 111(1), 493–520. https://
doi.org/10.1007/s11192-017-2259-4

May, R. M. (1997). The scientific wealth of nations. Scienza, 275,

793–796. https://doi.org/10.1126/science.275.5301.793

Patelli, A., Cimini, G., Pugliese, E., & Gabrielli, UN. (2017). The sci-
entific influence of nations on global scientific and technological
development. Journal of Informetrics, 11(4), 1229–1237. https://
doi.org/10.1016/j.joi.2017.10.005

Rousseau, R. (2018). The F-measure for research priority. Journal of
Data and Information Science, 3(1), 1–18. https://doi.org/10
.2478/jdis-2018-0001

Rousseau, R. (2019). Balassa = revealed competitive advantage =
activity. Scientometrics, 121(3), 1835–1836. https://doi.org/10
.1007/s11192-019-03273-y

Rousseau, R., & Yang, l. (2012). Reflections on the activity index
and related indicators. Journal of Informetrics, 6, 413–421.
https://doi.org/10.1016/j.joi.2012.01.004

Sandström, U., & Van den Besselaar, P. (2018). Funding, evaluation,
and the performance of national research systems. Journal of
Informetrics, 12(1), 365–384. https://doi.org/10.1016/j.joi.2018
.01.007

Schubert, A., & Braun, T. (1986). Relative indicators and relational
charts for comparative assessment of publication output and cita-
tion impact. Scientometrics, 9(5–6), 281–291. https://doi.org/10
.1007/BF02017249

Quantitative Science Studies

774

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D

F
/

3
3
7
5
5
2
0
5
7
7
2
3
q
S
S
_
UN
_
0
0
2
0
6
P
D

B
sì
G
tu
e
S
T

o
N
0
9
S
e
P
e
M
B
e
R
2
0
2
3

The nations’ scientific specialization indexes by output and by input

Sugimoto, C. R., & Larivière, V. (2018). Measuring research: Che cosa
everyone needs to know. Oxford: Oxford University Press. https://
doi.org/10.1093/wentk/9780190640118.001.0001

Tahamtan, I., & Bornmann, l. (2018). Core elements in the process
of citing publications: Conceptual overview of the literature.
Journal of Informetrics, 12(1), 203–216. https://doi.org/10.1016
/j.joi.2018.01.002

Tahamtan, I., Safipour Afshar, A., & Ahamdzadeh, K. (2016).
Factors affecting number of citations: A comprehensive review
of the literature. Scientometrics, 107(3), 1195–1225. https://doi
.org/10.1007/s11192-016-1889-2

Teixeira, P. N., Rocha, V., Biscaia, R., & Cardoso, M. F. (2012).
Competition and diversity in higher education: An empirical
approach to specialization patterns of Portuguese institutions.
Higher Education, 63(3), 337–352. https://doi.org/10.1007
/s10734-011-9444-9

Waltman, l. (2016). A review of the literature on citation impact
indicators. Journal of Informetrics, 10(2), 365–391. https://doi
.org/10.1016/j.joi.2016.02.007

Ward, J. H. (1963). Hierarchical grouping to optimize an objective
function. Journal of the American Statistical Association, 58,
236–244. https://doi.org/10.1080/01621459.1963.10500845

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

:
/
/

D
io
R
e
C
T
.

io
T
.

e
D
tu
q
S
S
/
UN
R
T
io
C
e
–
P
D