Scientific systems in Latin America: performance, networks, and collaborations with industry
C. (2015). Natural resource industries as a platform for the
development of knowledge intensive industries. Tijdschrift voor Economische en Sociale Geografie
(Journal of Economic & Social Geography)
Scientific systems in Latin America: performance, networks, and collaborations with industry
Hugo Confraria 0 1
Fernando Vargas 0 1
0 UNU-MERIT and Maastricht University , Boschstraat 24, 6211 AX Maastricht , The Netherlands
1 JEL Classification O30 · O39 · O54
In this paper, we use a combination of bibliometric, social network and econometric approaches to increase our knowledge of how research institutions interact with the private sector in Latin America (LA). We first study recent trends in scientific output and specialization. On average, LA countries have been reducing the gap with the world leading regions. They have also tended to specialize in fields related to economic activities based on natural resources, such as Agricultural and Plant and Animal Sciences. However, collaborations with the private sector remain scarce. In this paper, we have built scientific networks composed by what we define as Research Departments (RD). These RDs belong to universities, research institutes and government agencies. We model the intensity of collaboration of a RD with industry as a function of its size, previous performance, and its position in the LA and national scientific networks. Our results show that the RDs which higher diversity of research partners in their national scientific network work more intensively with industry. Additionally, collaborations with industry are influenced by previous interactions with the private sector. Hugo Confraria and Fernando Vargas have contributed equally to this work.
University-industry collaborations; Co-publishing; Social networks; Bibliometrics; Technology transfer; Latin America
Understanding the main traits of scientific institutions that engage in collaborative work
with the industry is critical for improving policies aimed to increase science–industry
linkages. This is even more important in Latin American (LA) countries, where public
support for science, technology, and innovation (STI) has increased significantly since the
early 2000s, but the study of science–industry linkages has been relatively neglected. In
this study, we show that scientific systems in LA countries have improved in performance
while specializing in scientific fields that are related to the main economic activities of the
region. Furthermore, by analyzing scientific organizations characteristics and collaboration
networks we find that organizations that have, or that have access to, diversified sources of
knowledge work more intensively with the private sector.
One of the main motivations behind the increasing relevance of policies that promote
science–industry linkages in the LA region is the potential benefit in innovation and
technological capacities in the private sector. Indeed, Crespi (2012) summarizes main
results of impact evaluations conducted in LA finding that public policies that promote
collaboration between universities and industry increase the level of investments in
innovation and labor productivity in firms. Marotta et al. (2007) show that both in Chile
and Colombia firms that collaborate with universities are more likely to introduce product
innovations and to apply for patents. Despite these positive findings, the magnitude of STI
policies is still too modest to pull LA countries towards knowledge-based economies.
Evidence from innovation surveys shows that universities and research centers tend to be
less relevant partners for technological innovation in LA firms in comparison to the
nonLA OECD countries (OECD 2015).
Part of the reason behind the limited importance of scientific institutions in the typical
LA country has been attributed to the high relevance that the exploitation of natural
resources has in LA economies. Conventional views see natural resources-based industries
as activities with slow technological progress, where innovation is mostly driven by the
suppliers of machinery and equipment, and that has a reduced potential of producing
knowledge spillovers to other sectors (Lall 2000; Pavitt 1984). Nonetheless, some scholars
argue that there are certain specificities in the current context that creates a demand for
local knowledge for the exploitation of natural resources which open a ‘window of
opportunity’ for the development of local knowledge providers (Kaplan 2012; Marin et al.
2015; Urzu´ a 2011).
Marin et al. (2015), for example, argue that the intensification of the challenges in the
exploitation of natural resources together with changes in volume and requirements of the
global demand would favor the development of a domestic knowledge-intensive industry
built upon local scientific capacities. Following this rationale, economic growth would be
grounded in the capabilities acquired by each country in its specific area of resource
endowment. It would then advance along the new technological trajectories being opened
by research made in natural resources related fields. In this study, we provide insights
about scientific development around natural resources in LA.
After studying scientific systems at the country level, we focus our analysis on scientific
organizations and its patterns of collaboration with the private sector, measured by
copublications. There is an extensive body of literature on university–industry collaborations,
and some of these studies examine cross-country and disciplinary differences in the
patterns of co-authored scientific publications between university and industry (Godin 1996;
Hicks et al. 1996; Tijssen 2004). Nonetheless, not much is known about the characteristics
of scientific institutions which favor research collaborations with industry. The lack of
evidence is even more noticeable in the LA countries. The results of this study help to close
In what follows, we first update and discuss current literature related to our research
question; then in section three, we describe our data sources and the methodology.
Section four shows the evolution of scientific production process and major trends on the
specialization of LA countries. Section five describes LA scientific networks in five
selected disciplines: Agriculture, Engineering, Environmental, Geosciences, and Plant and
Animal Science. Section six presents and analyses main econometric results, and lastly, we
present conclusions in the final section.
2.1 Science–industry collaboration in Latin America
Science–Industry collaboration in LA has largely been built from a top–down perspective
as a result of S&T policies based on a supply-push focus (Crespi and Dutre´nit 2014;
Dutre´nit and Arza 2010). Although LA universities differ across countries about their
origins, a common feature is that they were initially oriented to undergraduate teaching. As
research activities became increasingly common, postgraduate programs were gradually
provided. Crespi and Dutre´nit (2014) describe how several of the most important public
research centers in LA were created during the period of supply-push science policies
(1950s–1980s). These centers have focused on supporting sectors considered relevant by
the policy makers (for example, coffee in Costa Rica, aeronautics and oil in Brazil, oil in
Mexico, nuclear technology in Argentina, and agriculture in most countries). During the
same period, private sector evolved in economic activities that remained fairly protected
from international pressures (either naturally or through intervention) (Crespi and Dutre´nit
2014). Consequently, the incentives to engage in technological updating and learning were
lessened. These singularities perhaps led LA firms to disregard local scientific institutions
when conducting their innovation activities (Crespi et al. 2010).
However, despite their relative scarcity, science–industry interactions have been key for
successful historical experiences in some industries in LA. For instance, Arza and Vazquez
(2012) highlight the importance of research by scientific institutions for agricultural
technological upgrading in Argentina, while Casas et al. (2000) discuss the role of research
institutions in successful experiences in biotechnology and other industries in Mexico. In a
similar line, Suzigan and Albuquerque (2011) argue for the importance of university
research for the development of the aircraft, steel and agricultural industry in Brazil.
Evidence from quantitative studies show that manufacturing firms that collaborate with
local universities increase their investments on R&D, are more likely to innovate and to
apply for patents, and reach higher levels of labor productivity (Crespi 2012; Marotta et al.
2.2 Measuring science–industry knowledge transfer
There are a wide variety of channels through which tacit and codified knowledge is being
transferred between universities and industry. Some mechanisms include the mobility of
students, personnel exchange, informal exchanges of information, public conferences,
consulting, collaborative and contract R&D projects, joint ventures, scientific publications
and patents (Cohen et al. 2002; Gray et al. 2013; Link et al. 2007; Meyer-Krahmer and
Schmoch 1998; Narin et al. 1997). According to Bekkers and Bodas Freitas (2008) the
relative importance of these different channels in different contexts is explained, to a large
degree, by the basic characteristics of the knowledge in question (tacitness, systemicness,
expected breakthroughs), the disciplinary origin of the knowledge involved, and to a lesser
degree the individual and organizational characteristics of those involved in the knowledge
Due to this variety of channels and mechanisms, there are methodological challenges in
measuring and assessing University–Science collaborative research. The impacts of
Science–Industry collaborations are usually spread in space and time, can be numerous, and
they are almost always difficult to separate from other parts of organizational life
(Bozeman 2000). This methodological challenge is compounded by problems of data
availability and measurability. For this reason, to focus on research collaboration between
research institutions and industry we adopt co-authored publications as a measure of
occurrence and intensity of collaboration (Godin 1996; Tijssen 2012).
2.3 Co-publications as a measure of science–industry collaboration
The analysis of co-authorship has become one of the standard ways of measuring research
collaborations between organizations (Lundberg et al. 2006). Co-authored publications
indicate the achievement of access to an often-informal network, and can be viewed as
successful scientific collaboration in themselves. They also suggest an indicating diffusion
of knowledge and skills. Moreover, co-authorship as an indicator is quantifiable and
invariant, while the measurement is not invasive (Abramo et al. 2009).
However, it should be emphasized that joint publications are just one type of the
different channels of knowledge transfer. Research financed by industry, co-patenting, or
even research collaborations that do not involve scientific publications are not captured by
academic databases. Hence, as the discussion of Grupp and Mogee (2004) reveals, relying
only on this single indicator to assess knowledge transfer activities of institutions and
countries may overestimate (underestimate) the performance of those which are naturally
more (less) inclined towards activities that lead to publishable outcomes. Furthermore,
some co-authored articles do not reflect real collaboration. A publication co-authored by
two institutions could suggest a collaboration that has not taken place, for example, if an
author has the two affiliations. Also, most scientific publications are about a specific topic
or research question, and interdisciplinary research may be left out of the publication
system (Porter and Rafols 2009). Therefore, co-authorship can never be more than a rather
imperfect or partial indicator of research collaboration (Katz and Martin 1997; Laudel
2002). In the LA context, it has been argued that co-authored publications are one of the
most important channels of knowledge transfer for researchers and firms (Dutre´nit and
Arza 2010). Therefore, for our study, we start from the assumption that companies need to
perform research to absorb and appropriate codified scientific and technical knowledge
(Aristei et al. 2016; Rosenberg 1990). Although “the traditional motivation of the
technologist is not to publish, but to produce his artifact or process without disclosing material
that may be helpful to his peers” (Price 1963), industrial researchers involved in scientific
production activities act strategically. They publish in order to build their reputations,
increase their visibility, reorient R&D agenda, establish intellectual claims and legal rights,
signal capabilities to attract potential partners, and remain effectively plugged in scientific
networks where new ideas are emerging (Godin 1996; Lee 2000; Li et al. 2015; Tijssen
Many of these papers are likely to be co-authored with researchers in the public sector.
These researchers, on the other hand, have a different set of motives to collaborate with
industrial researchers, namely to generate additional research funds, gain insights in the
area of research, look for business opportunities, increase the output of commercialization
activities and further the university’s outreach mission (Belkhodja and Landry 2007;
Bozeman and Gaughan 2007; D’Este and Patel 2007; Lee 2000; Wong and Singh 2013).
Consequently, science–industry co-authorships can constitute a strategic way of acting
that gives the researchers involved valuable insights in comparison with peers who are not
participating in such collaborations. In this study, we assume that scientific institutions are
always available for co-authorships with industry researchers, i.e., these institutions are not
actively selecting their potential private sector partners. On the other hand, we assume that
researchers in industries prefer to collaborate with institutions that exhibit certain structural
characteristics. These include a quality dimension and a measurement of the diversity of
knowledge. The latter can be studied by analyzing collaboration network structures.
2.4 Network position as correlate of performance
Scholars of social networks have consistently shown a significant association between
network position and performance. One line of research indicates that actors with a higher
number of direct ties will have access to additional sources of knowledge, ideas, and
resources, thereby enhancing performance (Ahuja 2000; Reagans and McEvily 2003).
Other research emphasizes the benefit of brokerage. Actors brokering between otherwise
disconnected actors are characterized by having a timing advantage, being in an
advantageous position for identifying arbitrage opportunities, having higher chances of creating
new knowledge or products, and being better able to capitalize on their existent capabilities
(Burt 2004, 2005; Zaheer and Bell 2005). The benefits of both types of network positions
have also been suggested (Reagans and McEvily 2003; Fleming et al. 2007). Despite these
different perspectives, the consensus has been that network positions correlate significantly
with actor performance in different areas.
If we consider scientific collaboration network studies, most of the previous work
focuses on the individual/researcher level. Some studies highlight the importance of
structural collaboration network positions as a driver of preferential attachments (Barabasi
et al. 2002; Moody 2004; Abbasi et al. 2012). Others try to understand if the location of a
researcher in a network can bring some advantages, for instance, a higher level of citations,
better access to knowledge sources, awareness of potential projects or access to more
funding (Abbasi et al. 2011; Ebadi and Schiffauerova 2015). However, to the best of our
knowledge, there is no evidence yet on the impact of structural collaboration network
positions on the level of collaboration with industry.
In this study, we use a mixed set of methodologies and metrics to analyze the LA
scientific system, its interactions, and its proximity to industry. Taking into consideration
the productive structure of the region, we focus on five natural resource related fields:
Agricultural, Engineering, Environmental, Geosciences, and Plant and Animals sciences.
Our contribution is to update trends and specializations patterns of scientific production in
LA countries using bibliometric analysis and descriptive statistics. Furthermore, we will
assess to what extent the specialized knowledge diversity of scientific institutions, proxy by
its position in the scientific networks, affect collaborations between science and industry.
Understanding the determinants of these collaborations will provide useful information for
the design of policies aimed at fostering science-industry linkages.
3.1 Data collection
We used the InCites™ (2017) tool proposed by Thomson Reuters, which is a web-based
research evaluation tool that facilitates national and institutional comparisons across long
time periods using indicators of publication output, productivity, specialization and
normalized citation impact. InCites™ provided output and citation metrics from the WoS™
(Web of Science™, Thomson Reuters), which in turn allowed us to access data and metrics
from a dataset of 22 million WoS™ papers from 1981 to 2013. All articles and reviews
from researchers with a LA affiliation, published between 2004 and 2013, were analyzed.
The metrics for comparisons between countries are created based on address criteria, using
the whole-counting method, that is, counts are not weighted by number of authors or
InCites™ classifies author addresses (affiliations) as “university,” “research institute,”
“government,” or “corporate.” In our work, an industry collaborative publication is one
that has at least one author with a “corporate” affiliation, and at least one author with an
affiliation with a LA “university” or “research institute.” It is important to keep in mind
that not all single affiliations of all publications in InCites™ are unified as “university,”
“research institute” or “corporate.”1 There are corporate affiliations that have not been
identified or unified yet; hence, they have not been classified as industrial publications.
Multinational enterprises (MNEs) are more likely to have been identified and unified as
“corporate.” Therefore, publications listed as industry (co)publications are a lower
boundary of the real private sector research output. We would expect that countries with a
lower presence of MNEs have larger differences between the number of publications
authored by industry captured by InCites™ and the real activity.
Another important caveat in our analysis is that LA’s research output may be
underestimated because its researchers often publish in journals that are not indexed in major
citation databases, such as WoS™ or Elsevier’s Scopus™.
3.2 Bibliometric analysis
In this section, we analyze the evolution of the science systems in LA, focusing mainly on
its output, productivity, specialization, quality, and linkages with industry. In addition to
publication output (number of articles and reviews) and research performance (publication
output relative to GDP and population), we calculate the percentage of publications of each
country that were co-authored with industry, and the share of total publication output
coauthored with international institutions. We also compute standard specialization indexes
to depict the relative specialization of each country in a given area (Balassa 1965). This
also serves to assess the overall level of specialization of each country (Laursen 2000).
Finally, to study the quality of the research output, we use two normalized measures of
1 There were few cases where the name of the institution was available but the classification type was
labeled as “unknown”. We assigned those cases to its correspondent type of affiliation after checking their
information on the web.
citation impact. These are values which evaluate the scientific influence or visibility of a
set of publications in a given period. For the Quality Citation Index (Bornmann and
Leydesdorff 2013), a country value of 1.2 indicates that the citation impacts of papers
published by researchers in this country are, on average, 20% points above the worldwide
average. For the Quality Top 10% Index a country value of “10” indicates that 10% of the
publications of that country are in the top 10% of the world, regardless of subject, year and
document type (Pudovkin and Garfield 2009). Therefore, that country can be considered as
performing at the same level as the world average. A value higher than “10”, indicates a
higher performance relative to the world average (see “Appendix” for more details).
3.3 Social network analysis
In this section, we describe the structure and patterns of collaborations of the LA scientific
network. We focus this part of our analysis on what we defined as the Research Department
level (RD).2 This unit of analysis is defined by an output measure. We assume that all
publications from one institution, in a determined scientific field, were produced by a
specific RD. For example, we treat all publications from one institution in two scientific
fields as research output from two different RDs that belong to the same institution. In
addition, we assume that the research performed in each area faces its specific conditions
and it is embedded in a particular scientific network, independent from other scientific
topics. Although these assumptions could be debatable, scientific research in each field
demands high levels of specialization and knowledge, which makes it very costly to get
involved in research in other disciplines (Jeffrey 2003). Hence, we expect that this
definition may include some errors but not a consistent bias.
As it was mentioned above, we define institutions conducting research in more than one
field as having different RDs operating separately in each one of them. To extract the
relevant scientific networks, we define a threshold to select the most prolific RDs in LA.
For each field studied, we select RDs with more than 50 publications in each of the two
5-year’s periods analyzed. Afterward, for each of these “elite” RD, we gather all partners
with five or more collaborations in the same field, in the same period. Thus, two RDs are
going to be linked if they have five or more co-authorships in the field and period. It is
worth mentioning that collaboration partners are not necessarily part of the “elite” RDs
group, given that they only need to satisfy the minimum of five co-publications with one
“elite” RD. This group of collaboration partners also includes RDs that are not from LA
institutions; however, we do not consider in our calculations those that are linked only with
one LA RD.
We perform this analysis in two periods of 5 years each (2004–2008, and 2009–2013).
Besides the graphical description of networks of both periods, we obtain information at the
RD (node) level, such as centrality indicators (degree, betweenness, and closeness), and
network features, namely the number of nodes, number of communities and average path
3.4 Econometric analysis
In this section, we set a model that allows for gathering new evidence of the characteristics
of the RDs working more closely with the industry. We define the percentage of
publications of RDs that are co-authored with the industry as the dependent variable, and we
2 RDs can belong to universities, research institutes (public or private) and governmental agencies.
relate it to a set of RD features that could influence such collaborations: (1) knowledge
production capacity; (2) research quality; (3) orientation towards industry; and (4)
Co-authorships with industry are far to be common in science. The occurrence of these
events can be represented as a case of corner outcomes with a corner at zero and a
continuous distribution for strictly positive values (upper-censored at 100). Wooldridge
(2002) suggests addressing these cases implementing “hurdle” or “two-tiered” models.
This allows explanatory variables to differently affect the participation decision, i.e., the
co-authorship of at least one publication, and the intensity of these collaborations,
measured as the percentage of the total publications of a RD that were produced jointly with
firms. Therefore, we firstly follow the specification of the two-tiered model developed by
Cragg (1971). In the “first-tier” of the model, we estimate the probability of participation in
co-publication with industry using a probit model. In the “second-tier” a truncated normal
model is used to estimate the intensity of the collaborations with industry, formally:
f ðw; yjx1x2Þ ¼ f1
Uðx1cÞg1ðw¼0Þ Uðx1cÞð2pÞ 21r 1 exp
where w is a dichotomous variable equal to 1 if the RD has at least one co-publication with
industry and 0 otherwise, and y is the percentage of publications of the RD co-authored
with the private sector. When w is equal to 0, then y also takes the value of 0. While w ¼ 1,
then y [ 0. Variables x1 and x2 are sets of characteristics of the RDs that affect the
likeliness to co-publish with industry and the intensity of these activities, respectively.
Hence, c captures the effects on the participation and b those associated with the intensity
of co-publication. This specification assumes conditional independence between the two
tiers of the model. In this case that means to assume that after controlling the observable
characteristics of the RDs, there is no correlation between the decision to participate and
the intensity of co-publications. We are aware that the latter assumption could be
debatable. Therefore, we also use the approach developed by Heckman (1979) as a consistency
check. Although this model is aimed to address the selectivity problem that arise when an
interval of the outcome variable is not observable, statistically is very similar to Cragg’s
model and its flexibility allows for correlation between the participation and intensity
equations. However, a variable that affects the participation but not the intensity of
collaborations with industry needs to be included to identify the model.
As we mentioned above, we model the participation of the RDs in collaboration with
industry and the intensity of co-publication as a direct function of the main RDs
characteristics. The first independent variable is the total number of scientific publications during
the period, which depicts the capacities of knowledge production of the RDs. This variable
is expected to have a positive effect on the relationships with the private sector since the
capability of a university to attract private enterprise collaboration is influenced by the size
of the group of academic researchers and their output (Abramo et al. 2010). Furthermore,
this variable is also a proxy for the size of the RD. Larger organizations may have more
resources available to assign for relationships with the private sector.
We also include a measurement of the scientific quality of the research output of the
RD, in the form of a citation impact index. In principle, we expect an ambiguous effect of
quality in co-publications between science and industry. On the one hand, highly cited
institutions enjoy reputational benefits that make them perceived as more desirable partners
for research by the private sector, increasing the likeliness of this type of collaboration. On
the other hand, institutions that produced highly cited publications may be mainly focused
on academic research, leaving few resources available to create linkages with industry. The
empirical evidence is also mixed. Some studies have shown a correlation between
universities’ citation impact and their intensity of collaboration with industry (Abramo et al.
2010; Balconi and Laboranti 2006; Giunta et al. 2016). However, further analysis,
examining specifically the Italian situation, showed that enterprises do not necessarily
choose partners with higher scientific influence (Abramo et al. 2009).
The orientation of a RD towards working closely with the private sector certainly will
affect the share of co-publications (Bozeman and Gaughan 2007; Giunta et al. 2016). We
proxy this factor, using a variable that calculates the previous record of science–industry
collaborations of the RD. By measuring prior partnership, we are also able to control for
the pre-existent linkages with industry that could have been developed at the institutional
or personal3 (researchers) level.
Finally, we consider that the diversification of knowledge within each RD positively
affects its closeness to the private sector. In particular, we assume that industrial research
projects in which companies involve RDs are significantly more complex and uncertain
than the common ones (Hall et al. 2003). Hence, RDs that possess or have access to diverse
but complementary expertise, even within the same scientific field, are going to be working
more intensively with the industry. Unfortunately, the level of disaggregation of the
publication data by scientific field does not allow us to test this directly. Nevertheless, we
make use of the social network features of each RD to proxy their internal knowledge
diversification. Specifically, we assume that RDs that have more diversified internal
knowledge sources are more likely to have a more varied set of research partners.4
Accordingly, we include variables that provide information about the linkages of the RDs
and its relevance in the scientific network. We rely on three commonly used measures of
network centrality (Freeman 1978): degree, betweenness, and closeness (see “Appendix”
for more details).
By including this type of variables in our estimation, we are also controlling for other
mechanisms that are taking place in parallel. Namely, RDs that are relatively better
connected in their scientific network could be given preference in work collaborations, since
they have earlier access to sources of knowledge and ideas (Burt 2005). Higher centrality
can also lower the cost of screening other RDs for future partnerships, help to diffuse the
scientific challenges in which companies are interested and increase its scientific reputation
thereby attracting top researchers (Godin 1996; Lee 2000; Li et al. 2015; Tijssen 2004). On
the other hand, working with highly connected RDs can be risky for companies because it
increases the potential damages of leakages of relevant information of the firms. Finally,
we can also expect that geographical proximity plays a role in shaping these collaborations
(Bozeman and Corley 2004; Giunta et al. 2016; Pinch et al. 2003). Hence, we include RDs
information regarding their relevance in both LA and their national scientific networks.
3 The choice of a university partner by an enterprise often develops not only based on objective information
but also through personal contacts. Selection and maintenance of relationships is strongly conditioned by
social proximity (Granovetter 1985).
4 Unlike the alliances between firms, driven by complementary knowledge for learning purposes (Baum
et al. 2010), we assume that co-publications between researchers from different scientific institutions can
also be driven by the opportunity to increase their publication productivity levels, therefore joint research
could be conduct between researchers and institutions with fairly similar specialization patterns and levels of
Usually, quantitative studies assessing causality based on statistics and data from
networks are subject to endogeneity biases. In our case, it would be in the causal direction of
the relationship between the linkages of a RD within their scientific network and the
intensity of collaborations with industry. We try to address this potential problem by using
information from two separate periods of time. This enables us to analyze RDs
characteristics and position in the network in one period and the collaborations with the private
sector in the following period. Furthermore, from the management literature, we know that
previous alliances tend to remain or to be repeated because routines decrease asymmetries
of information among partners and facilitate the estimation of future returns of joint
activities (Gulati 1995). At the same time, processes of path dependence induced by the
influence of initial conditions on future developments may also occur here (Thune and
Gulbrandsen 2014). The choice of the 5-year time span is a compromise between
robustness of results and timeliness.
We control for differences in the intrinsic degree of proximity to industry of different
scientific fields. We also include a set of country dummies to control for idiosyncratic
characteristics and specific science–industry policies. Also, we include a dummy variable
that controls characteristics of RDs that are part of universities, relative to other types of
institutions. Finally, we allow errors to be correlated among RDs that belong to the same
4 Science in LA: trends and specialization
LA’s long-term world percentage of publication output in WoS™ has increased from
1.32% in 1981 to 5.03% in 2013. In 2013, all LA countries accounted for 71,391
publications in WoS™. Brazil´s share of publication output is particularly high when compared
with other countries of the region (around 55% of LA output in 2013), reflecting
differences in the size of the economies. According to our analysis, the share of world scientific
output from Brazil increased at a constant rate from 1993 to 2006, when publications
skyrocketed to the levels seen in Brazil in 2013.5 Other countries that show higher average
shares of scientific output than LA in the last decade are Mexico, Argentina, and Chile.
Table 1 provides data adjusting scientific output by other characteristics of the
countries. This allows for an assessment of the scientific “productivity” per billion of USD and
per million of inhabitants.
LA countries are ranked in Table 2 by aggregate scientific production from 2004 to
2013. Although Brazil has the highest number of publications, it has the lower scientific
impact. This may happen due to a significant percentage of articles being published in
national journals that had recently been included in the databases (Collazo-Reyes 2013).
Countries with smaller scientific systems tend to rely more intensively on international
collaborations. The average LA country has 75% of its scientific outputs co-published with
a foreign institution, while that figure goes down to 42% when considering the top 4 largest
science systems in the region (Brazil, Mexico, Argentina, and Chile).
In general, although LA’s scientific impact is growing, it remains relatively low when
compared to the world average. Despite their low productivity and scientific output, Peru
and Panama perform best in these terms, probably because more than 85% of their
5 Vargas et al. (2014) argue that, in areas such as Agricultural Sciences the increase of output since 2006
was due to the expansion of Brazilian journals in WoS and an increase in the number of issues published by
lirza iecox tireanng lieh ilaoobm lezaeeun abu rayugu reu tisaacoR rcaodu aaanm iilavo tleaaaum iraacagu raaaugy iicanonm rsaundo tiia llaavdoS tiaenAm
B M A C C V C U P C E P B G N P D H H E L
n ) ra
io % e
t ( v
r 0 e
te – th
I 0 – ts
/scPO 08 tiaon reaeg
4 153 rabo a
o – v
l ly )
l r e
31 co eay rag
– y ( e
ity aP 90 42 trs P av
tiv D du D y
cdu /scG 08 Idn foG lrea
o o – y
r 4 21 an $ (
P D 0 S ts
Plant and Animal Science
Plant and Animal Science 2.50
Table 2 Top five subject areas, in the nine LA with higher scientific output (2009–2013). Source: Own
Country Relative specialization intensity (rank)
Mexico Space Science
Chile Space Science
b RSI = Share of a country’s papers in a given field, relative to the share of world papers in that field
c SII = Specialization Intensity Index. This measure provides a ratio to assess whether a country is
“specialized” or “not specialized.” It grows with the specialization intensity of a country
publications are co-authored with researchers outside their country (Van Raan 1998).
Chile, by far the most productive country in the region, has also increased its research
output and maintained a medium level of scientific impact. As regard to the levels of
collaboration with industry, we can appreciate a relatively low percentage mainly in the
countries that are not so dependent on international collaboration or have larger science
systems—compared to countries like the United States or Germany (higher than 2%).
These results are in line with Tijssen (2012), who showed that LA and North Africa are the
regions in the world with the lowest intensity of science–industry co-authorship.
Countries often try to invest strategically in research areas critical to their economic
development. Creation of specific local knowledge may increase innovation capacities of
incumbents, but also promote the birth of start-ups or spin-offs. These trends run in parallel
with others that do not necessarily operate in the same direction. Historical and cultural
influences, strengths of scientific establishments, as well as incentives and government
funding for scientific research play a relevant role in defining the revealed scientific
specialization of a country. The size of the scientific system also matters, since larger science
systems have the capacity for more diversity and greater coverage of the full scope of
sciences. In contrast, smaller systems may be limited in their ability to invest in specific
domains. We explore the outcome of these trends through a specialization analysis based
on the 22 Essential Science Indicators (ESI) areas.6 Table 2 contains the five subject areas
of higher specialization for the nine countries in LA with more than 1% of LA total
scientific output over the 2009–2013 period. Table 2 also provides information on
aggregate specialization level (given by the SII index) for each of these nine countries.
Research specialization is quite similar across these LA countries. In aggregate terms,
the top 5 areas with the largest output from LA, relative to the world are Agricultural
Sciences (15.7%), Plant and Animal Science (12.3%), Space Science (9.3%), Environment/
Ecology (7.7%) and Microbiology (7.3%). The higher LA specializations are in
Agricultural Sciences and Plant and Animal Sciences, which is in line with the high importance of
agricultural, livestock and agro-industrial activities in the region.
The cases of Peru and Chile are interesting because they revealed high specialization7 in
subject areas different from the other countries of the sample. The specialization of Peru is
related to issues in public health (prevention of HIV, tuberculosis, and lupus) in which they
also have a high scientific impact (Van Noorden 2014). Chile’s high specialization in
Space Science is related to its excellent infrastructure of giant telescopes housed in the
Atacama Desert. According to Catanzaro et al. (2014), funding for astrophysics has
increased from $2 million in 2006 to $6.8 million in 2010. Over the same period, the
number of faculty positions has almost doubled. This has led not only to an increase in the
number of publications in this field but also to an increase in quality. In contrast,
Economics and Business, Materials Science, Computer Science, Psychiatry/Psychology and at
a certain level Engineering seem to be neglected research disciplines across LA countries.
In summary, scientific activity has been growing in LA countries during the last decade
but not at a pace that allowed it to catch-up with the rest of the world. Only four countries
show productivity levels closer to the world averages. Co-publications with international
6 The Essential Science Indicators schema (Thomson Reuters) comprises 22 subject areas in science and
social sciences and is based on journal assignments. Arts and Humanities journals are not included. Each
indexed journal (11,000+) is found in only one of the 22 subject areas and there is no overlap between
7 If a country has a scientific output structure equal to the world, the value of the indicator will be zero. The
size of SII is an indication of how strongly each country is specialized.
institutions are frequent and highly relevant for the scientific impact (quality) of smaller
scientific systems. On the other hand, collaborations with industry are scarce even when
research specialization seems to be influenced by economic specialization.
In what follows, we will focus on the study of five main scientific fields: Agricultural,
Engineering, Environmental, Geosciences, and Plant and Animals sciences. We choose
Agricultural, Geosciences and Plant and Animals as they are closely related to the natural
resources-based economic activities in which LA countries are more intensive. We also
include engineering and environmental sciences because we assume that this type of
knowledge needs to be consistently applied across the main economic activities of the
5 Network analysis
The data requirements for extracting the scientific networks explained in Sect. 3.3 only
allow us to include in our analysis RDs from the following LA countries: Argentina,
Brazil, Chile, Colombia, Costa Rica, Cuba, Mexico, Panama, Peru, Uruguay, and
Venezuela. Table 3 gives some network summary statistics from the five scientific fields that we
are analyzing. Network graphs are available in the “Appendix” (“Network graphs of all
scientific areas in 2004–2008 and 2009–2013” section).
The 55% growth of LA scientific production between 2004–2008 and 2009–2013 is
roughly proportional to the increase in the number of RDs (nodes) in all subject areas
(networks). In both periods, the average path length is less than 4, implying that knowledge
that is created in one node has the potential to be diffused in few steps to the rest of the
Interestingly, the change in the number of communities8 does not follow a common
trend. Engineering, Plant and Animal, and to some extent, Environmental Sciences shows a
remarkable increment in the number of RDs in the LA network. However, there are limited
changes in the number of communities,9 suggesting that newcomers were rapidly attached
to well-established groups of collaborators. On the other hand, Agricultural and
Geosciences at least double the number of knowledge communities. It can be interpreted that
evolving networks are creating new niches of knowledge, either with new local actors or
increasing diversification of knowledge sources through new international collaborations.
Nevertheless, it should be borne in mind that geographical proximity may also be playing a
role in the creation and evolution of these research communities.
Shortest average paths together with an increasing number of communities are signals
that a network structure is evolving towards a structure that facilitates both knowledge
creation and knowledge diffusion. However, attention needs to be paid to the fact that in
almost all scientific fields studied it is common to observe that two neighboring countries
(in geographic terms) are only connected to each other through a RD that is based in a third
country. Even when this situation gives potential brokerage power to the external RD, it is
not clear what the impact is for the performance of LA scientific networks. Clearly, this is a
topic that requires further research.
8 Communities were detected applying the ‘Leading eigenvector method’ available in the R software
9 Communities are group of nodes that are densely connected between them and more sparsely connected
with nodes from other communities. In our case, communities are interpreted as groups of RDs that tend to
collaborate more intensively in research with each other.
% of RD
Average % of
We also found that the increases in the number of RDs, between both periods, are not
reflected in significant changes in the percentage of RDs collaborating with the industry.
Therefore, we can assume that the share of RDs collaborating with the industry among the
new incumbents is the same as the proportion of RDs connected to the industry in the
previous period. On the other hand, the average percentage RDs of co-publications with
industry fell in all scientific fields, except for engineering. The latter suggests that the new
RDs that co-publish with the industry are doing it less intensively than the average RD of
the previous period.
As we mentioned before, for the econometric implementation we also estimate
centrality measures for local/national networks. For each country, these networks are formed
by all elite national RD (same threshold defined before) and its research partners. Foreign
institutions are also included in the network. However, those that have collaborations with
only one local RD are considered peripheral and are subsequently dropped from the
network. After application of these filters, we are left with data only from Argentina,
Brazil, Chile, and Mexico.10
6 Econometric analysis
In this section, we present the results of the estimation of Cragg (1971) model described in
Sect. 3.4, run using the user-developed craggit routine in the Stata software.11 We pooled
data from the LA scientific networks presented in Sect. 5. After the application of data
requirements to the nodes gathered from these networks and dropping outliers on the
outcome variable,12 we end up with a database of 324 observations from four LA countries
(Argentina, Brazil, Chile, and Mexico) in the five selected scientific topics.
10 We could get the national network of Venezuela, but data requirements for the econometric estimations
left these observations out of the final dataset.
11 As a consistency check, we also estimated a two-step Heckman selection model, using the same software.
Those results are available in the appendix.
12 We define as an outlier a RD for which the outcome variable is more than three standards deviations
above/below the mean.
)(3 ire1T .54601** .)(01199 .58770** .)(01811 .00750 .)(00237 .04140 .)(05257 −.13703 .)(01305 .10410 .)(05368 .5378*0 .)(01339 .3313*1** .)(02323 .49240 .)(01983 .70260 .)(07474 .543*6*0 .)(08272
reeegD ()2 ire1T .04536** .()01200 .0587*4* .()01831 .00076 .()00263 .04063 .()02553 −.01319 .()03101 .10092 .()03522 .58004* .()03154 .3203***1 .)(03249 .45360 .)(03911 .62180 .)(47055 .5649**0 .)(22880
ire2T .35079 .()02484 −.1388*2 .()05444 −.01093 .()10186 .07422** .()81118 −.10071 .()74156 .56720 .()62168 .40092 .)(27337 .6822**9 .)(59463 .52605 .)(09338 .889*6*8 .)(54445 −.30411 .)(46612
in ra * *
tihw tceonN )(1 ire1T .44058** .)(01165 .*71033* .()10122 .00089 .()00231 .43051* .()24096 −.01010 .()31098 .53109 .()34084 .*43608* .()30053 .78313*** .()13038 .*30507 .()94200 .9170**0 .)(97307 .5258**0 .)(33202
o io )
itra itca 008
o l 2
b b –
lla pu 04
o f 0
c o 2
a g n s
r n e e
trIsydun rebuNm tilayuQ lirzaB liehC iceoxM tlircuugA iireenngE irnonvEm isceecnoG tiirseynvU
98 ) * )
8 * 3 0 )5
ire2 .1173 .158 7692 .259 .272 595
7 . 1 7 .6
T − ( 2 ( − (
) )7 *0 ) * )
7 0 * 8
)(3 ire1T .16642 .(2277 .11550 .(4590 −.4938 .(1605 .88681 .(0672 423
1* ) * ) s
−.5448 .(08585 .18431* .(03091 324 teeeennBw ()4 ire1T .45470*** .()01195 .62130*** .()01436 .00680 .()00242 .42610* .()02538 −.01557 .()03134
7 )49 *9 )2 ** )2 ) )4 ** )8 )
teBw )(4 ire1T .0130 .(350 .9500 .(3090 .92161 .(3320 .14380 .(23100 .36670 .(5440 .55890 .(3220 .61542 .(92062
iceoxM tlilrrcauugA iireenngngE itrenonnvEm issceeecnoG tiirseynvU ilrzeadoNm lirzeadoNm ilreazdoNm ilreazdoNm ttsannoC
percentage of co-publications with industry are more prone to keep engaging in these
collaborations. Across the different specifications of the model, the sign and statistical
significance of this effect remain. The positive impact of past collaborations in the intensity
equation is not statistically significant in the craggit estimation. However, the
Heckmanmodel not only confirms the positive relation but in this specification the coefficients are
Interestingly, there is no clear relationship between academic quality and engagement in
research with the private sector. The signs of the coefficients (positive for participation and
negative for intensity) may suggest that higher academic quality favors participation in
research with the industry, but for RDs with higher levels of citation impact, the
collaboration intensity with industry is relatively smaller. One possible explanation for the latter
is that RDs that produce highly cited research are mainly focused on academic research and
not so much on generating linkages with industry.
Country dummies show that there are no significant differences at this level on the
likeliness of RDs to engage in research collaboration with the private sector. However, the
Brazilian RDs that do participate in industry collaborations are doing it more intensively
than their counterparts in Argentina, Chile, and Mexico. Finally, engineering sciences are
consistently the research field with most collaboration with industry in both, participation
and intensity, a result that we expected given the applied orientation of the engineering
activities. The result in the intensity equation also holds for Geosciences, probably due to
the importance of mining operations in the sample of countries included in our analysis.
The position of the RDs in the LA scientific network does not seem to be related to its
relationship with the private sector. Indeed, none of the network centrality measurements
tested (degree, estimations 2 and 3; betweenness, estimations 4 and 5; and closeness,
estimations 6 and 7) at the LA level show statistically significant coefficients. The
unimportance of these RDs features, in the global LA context, contrast with the results
observed when we consider the node characteristics of the RDs in the national/local
network. Indeed, our most important finding is that two of the centrality measurements of
the national/local scientific networks (estimations 3 and 5) show positive and significant
effects on the intensity of the collaboration with industry.
Although not relevant in the participation equation, RDs with higher values of local
degree and betweenness engage more intensively in research activities with the private
sector. These results suggest that the RDs that have or have access to a more diversified set of
knowledge sources in their countries are more prone to engage intensively in research with
industry. The mechanism behind this finding may be related to the fact that RDs which are
better connected can provide different strands of specialized knowledge that allow them to
tackle the type of challenges proposed by the industry adequately. At the same time, these
RDs can provide benefits to firms by lowering the costs of screening other RDs for future
partnerships, decreasing the risk of knowledge lock-in, attracting high-qualified researchers,
and providing a more effective diffusion of the scientific challenges of the company.
Furthermore, nodes in brokerage positions (higher betweenness) are characterized by
having a timing advantage. They are not only more likely to be first recipients of
information from diverse groups but also occupy a privileged position from which they can
assess the relevance of new information (Burt 2005). Therefore, in a competitive process in
which timing is rewarded, a brokerage position of RDs in national borders may be
providing a crucial advantage for collaboration with industry. However, as previous research
has also suggested (Liao and Phan 2016) since the participation equation is not significant
in the local network, we should be careful when arguing that these two types of network
positions will lead deterministically to more science–industry collaborations.
Larger RDs can cover a wider spectrum of scientific topics, and they also have more
resources that could be used to establish relations with the private sector (e.g., TTOs),
making them more prone to engage in collaboration with industry. Our results confirm this
showing that even after controlling for networks centrality features, the size of the RD is
revealed as a strong predictor of the likeliness of performing research with the private
sector. On the other hand, the intensity of these collaborations decreases with the size of
the RD. We are aware that the relation between size and centrality may raise a
multicollinearity problem. The availability of more resources in larger RDs can also increase its
centrality. However, we are confident that theoretically, both variables are not measuring
the same characteristics of the RDs, that is, more co-publications do not necessarily imply
more diversity in co-publications partners. Therefore, both size and centrality must be
included in the estimations. Excluding these aspects will give rise to a problem of omitted
variables. Nevertheless, this potential problem needs to be considered when interpreting
In Table 6 (in the “Appendix”) we present the results of the estimations of a two-step
Heckman model for specifications 3, 5 and 7 of the model. Based on the results of the
Craggit estimations we use the size of the RD as the exclusion variable, i.e., affecting the
participation decision but not the intensity equation. Most of the results of the previous
estimations hold. However, in this set of estimations, the previous collaboration with
industry increases not only the likeliness of participation in co-publication but also the
intensity of these collaborations. An extra 1% of co-publications with industry in one
period increase these activities in 0.35–0.45% in the next period. Despite this change, the
degree and betweenness values of the RDs in their national scientific networks have a
positive effect on the intensity of collaborations with industry.
In this paper, we use a combination of bibliometric, social network and econometric
techniques to increase the understanding of LA scientific systems and its relationship with
the private sector. We studied recent trends in the scientific outcome, the linkages that exist
between RDs within and between LA countries, and RDs collaboration activities with
We found that the LA share of global scientific publications started to increase at a
higher rate since 1993, thus revealing a trend for convergence with the world leading
regions. This increase has been mainly driven by Brazil and most notably in subject areas
such as Agricultural, and Plant and Animal Sciences. Moreover, when analyzing the
relative scientific output normalized by GDP (Docs/GDP) and population (Docs/Pop), the
results show that in the most recent years Chile, Uruguay, Argentina, and Brazil have
levels of scientific productivity higher than the world average. Furthermore, specialization
of scientific systems in LA tends to follow economic specialization, focusing on scientific
fields related to natural resources. However, in the last decade, most LA countries have an
average industry collaboration percentage below 1%. This is a low number when compared
to the rest of the world. There are differences between fields (Engineering and Geosciences
show higher levels than other sciences) but in general, collaborations between science and
industry, measured as co-publications, are scarce.
The growth of scientific production can also be appreciated by the increasing number of
RDs embedded in the LA scientific networks. However, the structures of these networks
are not evolving in the same way. We find preliminary evidence that suggests that LA
Geosciences and Agricultural Sciences networks are evolving towards structures that
facilitate both knowledge creation and diffusion. It is worth noting that collaborations
between RDs of different LA countries remain low. In most of the fields studied, linkages
between LA countries are scarce even when these countries tend to specialize in similar
scientific fields. Understanding if this lack of integration between LA scientific institutions
is harming potential gains of complementary knowledge is a matter of further research.
The main finding is that the RDs that have a more diverse set of knowledge sources, within
their scientific discipline, are the ones that are working more closely with industry. Besides
possessing different sources of complementary knowledge within the same discipline, that
can tackle more effectively private sector challenges; firms may perceive these RDs as
having a higher reputation and stronger research capabilities. Furthermore, by being in
brokerage positions, RDs are not only more likely to be early recipients of information from
diverse groups but also occupy a privileged position from which they can assess the
relevance of new information. This timing advantage may be a crucial element for collaboration
with industry. Although interesting, we cannot determine which of these is dominating.
Complementing this analysis with qualitative approaches and primary data which
consider other types of technology transfer activities and sources of funding for research would
certainly improve the understanding of LA knowledge production, transfer and diffusion
systems. Furthermore, focusing on the publication analysis of science–industry linkages at
the level of technologies, rather than scientific fields, is a matter of further research.
Acknowledgements The authors would like to thank Robin Cowan for his comments and insights on an
earlier version of this paper. We thank the National Commission for Scientific and Technological Research
of Chile (CONICYT) for providing access to InCites™ data and Oscar Ravanal Echeverr´ıa for his technical
support in data collection and facilitating our work in CONICYT facilities. This article also benefited from
comments made by the participants of the Globelics 2015 Conference (Havana, 2015) and the EMAEE 2015
Conference (Maastricht, 2015). This research received financial support from the Competitiveness and
Innovation Division of the Inter-American Development Bank.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution,
and reproduction in any medium, provided you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons license, and indicate if changes were made.
Specialization and citation impact measures
Revealed specialization intensity This index assesses the relative specialization of each
country in a given area. It is an adaptation of the Revealed Comparative Advantage Index,
proposed by Balassa (1965), which compares the specialization intensity of a subject area s
in country i with the equivalent relative specialization intensity of that subject area for all
RSI ¼ Ps=P
where Pis accounts for the number of publications in subject area s in country i, Pi accounts
for the total number of publications in that same country i, Ps accounts for the total number
of publications in subject area s worldwide, and finally P accounts for the total number of
publications in the world.
Specialization Intensity Index This measure provides a ratio which in the numerator
displays the square of the difference between specialization intensity of class s in country i
and specialization intensity of that class in the world, while the same denominator shows
the sum of the weighting of all subject areas in country i, with this ratio summed up across
all s subject areas. This Chi square of sectoral specialization is adapted from Laursen
(2000) and provides a concentration measure that grows with the specialization intensity of
SII ¼ X
Xsi= Ps Xsi
Pi Xsi= Ps Pi Xsi
Pi Xsi= Ps Pi Xsi
Quality Citation Index This score calculates the mean citation rate of a country’s set of
publications in a specific subject area, period, and document type, divided by the mean
citation rate of all publications in that subject area/period/document type:
PiP¼1 lf i
Quality Top 10% Index This index shows the proportion of publications belonging to the
top 10% most cited documents in a given subject category, year and publication type:
Degree This measure of centrality accounts for the total number of links that a node has
in a network. In the case of the networks that we are studying it will account for the total
number of different research partners with whom each RD collaborates. RDs with higher
degree number could be considered popular among their peers, enjoying benefits from
reputation. Furthermore, they also hold what could be regarded as a more diversified set of
research partners. However, regularly, maintaining links is a costly endeavor, and then we
would expect to find limits on the utility of getting new linkages. We use the normalized
version of the indicator implemented by the igraph package of the R software. Formally:
where lði; jÞ ¼
CDðiÞ ¼ ðn 1 1Þ Xj¼n1 lði; jÞ ð2Þ
if there is an edge between i and j , and n is the number of nodes of the
1 X gjkðiÞ
3Þðn þ 2ÞÞ j6¼k gjk
where n is the number of nodes of the network, gjkðiÞ is the number of shortest paths that
pass through node i, and gjk is the total number of shortest paths.
Closeness This index is defined by the inverse of the average shortest path to all other
nodes in the network. An RD with higher values of closeness would require less effort to
reach any other source of information. At the same time, at least theoretically, it could
access new knowledge more quickly than others. We use the normalized version of the
indicator implemented by the igraph package of the R software. Formally:
CCðiÞ ¼ ðn
X dði; jÞ
where n is the number of nodes of the network, dði; jÞ is the length of the shortest path
between nodes i and j.
Network graphs of all scientific areas in 2004–2008 and 2009–201314,15
See Figs. 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10.
Betweenness This index accounts for the total number of shortest paths13 in which a node
is involved. Under the assumption that shortest paths are preferred in the diffusion of
knowledge in a network, RT with higher betweenness values may be connecting
knowledge from two very distant RD, broadening the scope of potential sources of information
and allowing them to play a role of broker of knowledge. We use the normalized version of
the indicator implemented by the igraph package of the R software. Formally:
13 The shortest path is the minimum distance, accounted by links, between two nodes of a network.
14 Networks are visualized using the Fruchterman–Reingold algorithm. Circle nodes are for LA RD. Square
nodes are for non-LA RD. Edge thickness represent the normalized number of co-publications.
15 AR = Argentina, BR = Brazil, CL = Chile, CO = Colombia, CR = Costa Rica, CU = Cuba,
MX = Mexico, PA = Panama, PE = Peru, UR = Uruguay, VE = Venezuela. AT = Austria, AU =
Australia, BE = Belgium, CA = Canada, CH = Switzerland, CN = China, CZ = Czech Republic,
DE = Germany, DK = Denmark, EG = Egypt, ES = Spain, FI = Finland, FR = France, IL = Israel,
IT = Italy, JP = Japan, NG = Nigeria, NL = Netherlands, NO = Norway, NZ = New Zealand,
PL = Poland, PT = Portugal, RU = Russia, SE = Sweden, UK = United Kingdom, US = United States,
ZA = South Africa.
Fig. 6 Network structure of Environmental Sciences (2009–2013)
titIsenn .34520* .(20305 – – −.00277 .()04300 .37130** .()30047 −.20942 .()53401 .80510 .()47250 .01900 .()49310 .34491 .()59410 .63650 .()27970 .811*31* .()76760 −.15990 .)(06500
lssseeonC )(11 ititircaapP .56140*** .()63100 .51260*** .()62100 .16100 .()28003 .28720 .()81309 −.59203 .()36400 .85900 .()90406 .14670** .()81301 .80*061** .()82740 .29940 .()31030 .38760 .()04940 .64*520 .()19820
teeeennBw ()10 tiitiracapoP .04543*** .()01664 .06213*** .()01600 .00067 .()00279 .04242 .()02990 −.01540 .()04596 .00976 .()04310 .05875 .()03590 .12859*** .()03678 .04085 .()03745 .06302 .()05233 .05536* .()02936
ttiIsenn .3350*8 .(17088 – – −.00293 .()04041 .2760**0 .()34067 −.14054 .()42069 −.00557 .()59020 .20702 .()26047 .83416** .()41804 .37200 .()25901 .08410** .()36302 −.19501 .()06905
reeegD )(9 iititrcaapP .615*40** .()44910 .778*50** .()70320 .50700 .()95200 .41040 .)(00930 −.73310 .)(36730 .41010 .)(57640 .8*3750 .)(31030 .3*3*131* .)(33030 .29440 .)(95230 .07260 .)(70640 .3*4650 .)(97030
tiitraonobhw tlitrceaonN )(8 tiitiracaponP .4458*0** .()10696 .7133*0** .()10440 .00890 .)(00286 .43510 .)(30095 −.10100 .)(36029 .15390 .)(45290 .6438**0 .)(32410 .3783**1* .)(36060 .30570 .)(33670 .91700 .)(50710 .25580 .)(32060
no io )
tira itac 008
o l 2
b b –
lla up 40
o f 0
c o 2
tw ) t
e 0 r
B ( P
ty 9 )0 ** )5
tsen .3079 .533 4234 .647
In − (2 .1 (0
a ) )
e ip 2 2
reegD ()9 tiracP .21646 .(2886 .01155 .(0566
2 )4 ** )2
.408 199 189 784
0 . 8 .
2 . 0
− ( 1 (
7 6 4 6
9 8 0 4
5 8 6 7
4 . 1 .
. 2 . 0
2 ( 0 (
0 45 )5 1 )5
.258 .973 0175 .116 .098 889
2 3 . 8 1 .2
− ( 9 ( − (
69 892 8 )71 8*6 )15 7 )37 9 61 0
.980 .092 .169 .507 .246 .121 .088 .505 24 .381 .678 .000
2 ( 6 ( − ( 0 ( 3 1 6 0
2* ) * )
6 * 0 7
.014 053 539 379 4 195 230 000
5 .1 .8 .0 2 .1 .7 .0
− ( 0 ( 3 1 6 0
0* ) )
2 * 1 5
.839 381 436 394 4 703 043 000
4 .1 .8 .0 2 .1 .0 .0
− ( 0 ( 3 1 7 0
1* ) * )
0 * 4 2
.484 411 963 13 895 958 000
5 .1 .8 .02 24 .1 .1 .0
− ( 0 ( 3 1 6 0
Abbasi, A., Altmann, J., & Hossain, L. (2011). Identifying the effects of co-authorship networks on the
performance of scholars: A correlation and regression analysis of performance measures and social
network analysis measures. Journal of Informetrics, 5(4), 594–607. doi:10.1016/j.joi.2011.05.007.
Abbasi, A., Hossain, L., & Leydesdorff, L. (2012). Betweenness centrality as a driver of preferential
attachment in the evolution of research collaboration networks. Journal of Informetrics, 6(3), 403–412.
Abramo, G., D’Angelo, C. A., & Costa, F. Di. (2010). University–industry research collaboration: A model
to assess university capability. Higher Education, 62(2), 163–181. doi:10.1007/s10734-010-9372-0.
Abramo, G., D’Angelo, C. A., Di Costa, F., & Solazzi, M. (2009). University–industry collaboration in Italy:
A bibliometric examination. Technovation, 29(6–7), 498–507. doi:10.1016/j.technovation.2008.11.003.
Ahuja, G. (2000). The duality of collaboration: Inducements and opportunities in the formation of interfirm
linkages. Strategic Management Journal, 21(3), 317–343. doi:10.2307/3094190.
Aristei, D., Vecchi, M., & Venturini, F. (2016). University and inter-firm R&D collaborations: Propensity
and intensity of cooperation in Europe. The Journal of Technology Transfer, 41(4), 841–871. doi:10.
Arza, V., & Vazquez, C. (2012). Firms’ linkages with universities and public research institutes in
Argentina: Factors driving the selection of different channels. Prometheus, 30(1), 47–72. doi:10.1080/
Balassa, B. (1965). Trade liberalisation and “revealed” comparative advantage. The Manchester School, 33
(2), 99–123. doi:10.1111/j.1467-9957.1965.tb00050.x.
Balconi, M., & Laboranti, A. (2006). University–industry interactions in applied research: The case of
microelectronics. Research Policy, 35(10), 1616–1630. doi:10.1016/j.respol.2006.09.018.
Barabasi, A. L., Jeong, H., Neda, Z., Ravasz, E., Schubert, A., & Vicsek, T. (2002). Evolution of the social
network of scientific collaborations. Physica A: Statistical Mechanics and its Applications, 311(3–4),
Baum, J. A. C., Cowan, R., & Jonard, N. (2010). Network-independent partner selection and the evolution of
innovation networks. Management Science, 56(11), 2094–2110. doi:10.1287/mnsc.1100.1229.
Bekkers, R., & Bodas Freitas, I. M. (2008). Analysing knowledge transfer channels between universities and
industry: To what degree do sectors also matter? Research Policy, 37(10), 1837–1853. doi:10.1016/j.
Belkhodja, O., & Landry, R. (2007). The Triple–Helix collaboration: Why do researchers collaborate with
industry and the government? What are the factors that influence the perceived barriers?
Scientometrics, 70(2), 301–332. doi:10.1007/s11192-007-0205-6.
Bornmann, L., & Leydesdorff, L. (2013). Macro-indicators of citation impacts of six prolific countries:
InCites data and the statistical significance of trends. PLoS ONE, 8(2), e56768. doi:10.1371/journal.
Bozeman, B. (2000). Technology transfer and public policy: A review of research and theory. Research
Policy, 29(4–5), 627–655. doi:10.1016/S0048-7333(99)00093-1.
Bozeman, B., & Corley, E. (2004). Scientists’ collaboration strategies: Implications for scientific and
technical human capital. Research Policy, 33(4), 599–616. doi:10.1016/j.respol.2004.01.008.
Bozeman, B., & Gaughan, M. (2007). Impacts of grants and contracts on academic researchers’ interactions
with industry. Research Policy, 36(5), 694–707. doi:10.1016/j.respol.2007.01.007.
Burt, R. S. (2004). Structural holes and good ideas. American Journal of Sociology, 110(2), 349–399.
Burt, R. S. (2005). Brokerage and closure: An introduction to social capital. New York: Oxford University
Casas, R., de Gortari, R., & Santos, M. J. (2000). The building of knowledge spaces in Mexico: A regional
approach to networking. Research Policy, 29(2), 225–241. doi:10.1016/S0048-7333(99)00062-1.
Catanzaro, M., Miranda, G., Palmer, L., & Bajak, A. (2014). South American science: Big players. Nature,
510(7504), 204–206. doi:10.1038/510204a.
Cohen, W. M., Nelson, R. R., & Walsh, J. P. (2002). Links and impacts: The influence of public research on
industrial R&D. Management Science, 48(1), 1–23. doi:10.1287/mnsc.18.104.22.16873.
Collazo-Reyes, F. (2013). Growth of the number of indexed journals of Latin America and the Caribbean:
The effect on the impact of each country. Scientometrics, 98(1), 197–209.
Cragg, J. G. (1971). Some statistical models for limited dependent variables with application to the demand
for durable goods. Econometrica, 39(5), 829–844. doi:10.2307/1909582.
Crespi, G. A. (2012). Incentivos fiscales a la innovacio´ n empresarial. Las instituciones fiscales del mañana
(pp. 137–175). Washington DC: Inter-American Development Bank.