Research Policy 44 (2015) 545–559

Contents lists available at ScienceDirect

Research Policy
journal homepage: www.elsevier.com/locate/respol

The impact of multilevel networks on innovation

Jiancheng Guan a,b,∗ , Jingjing Zhang a , Yan Yan a
a
School of Management, University of Chinese Academy of Sciences, 100190 Beijing, PR China
b
School of Management, Fudan University, 200433 Shanghai, China

a r t i c l e i n f o a b s t r a c t

Article history: Different from previous studies analyzing the influence of network structure on actors’ performance by
Received 24 April 2014 using a single network perspective, this article explores the impact of multilevel networks on innovation.
Received in revised form Using a sample of 41,007 patents in the field of alternative energy of the USPTO database, we construct
14 December 2014
inventor collaboration networks at city and as well as country levels. In the empirical analysis, we use
Accepted 14 December 2014
panel data and negative binomial regression models with fixed effects. To keep our results reliable, we
use an instrumental variables approach to solve potential endogeneity problems and perform a series of
Keywords:
robustness tests. The results show that inter-country collaboration network structure moderates the rela-
Multilevel collaboration networks
Innovation performance
tionships between inter-city collaboration network structure and innovation performance. Our findings
Alternative energy show that when country’s centrality and structural holes are high, the positive effects of city’s centrality
and structural holes on innovation performance are enhanced, and the negative effects of city’s cluster-
ing coefficient are weakened. Implications of the findings for complex innovation network theory and
innovation policies are discussed.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction studies have highlighted, for instance, the importance of strength-

We live in an increasingly globalized world leading to exten-
sive interconnection between countries (Scellato et al., 2014). For
instance, with rising awareness of environmental protection, rapid
development in green technologies has substantially stimulated
collaborations at the international and national level (Melville,
2010). Meanwhile, the popularization of email and inexpensive
international communication enable cross-country collaboration
in researches (Ding et al., 2010). Cooperation and research net-
works are the key to enhance the competitiveness of regions (Cooke
et al., 1997). In recent years, the need to know the consequences of
climate change has increased greatly (Paschen and Ison, 2014) as
climate change has impact on economic development, population
growth and poverty (The United Nations Framework Convention
on Climate Change). Technology has an important role in enabling
industry worldwide to make the transition toward more environ-
mentally sustainable modes of operation (United States Patent and
Trademark Office) and such sustainable technologies ensure that
resources are used more efficiently and harmful emissions are
reduced (Bohnsack et al., 2014). Various forms of collaborations
between countries or cities in the area of technologies relevant
to address climate change each play an essential role. Recent
∗ Corresponding author. Tel.: +86 10 82680807.
E-mail address: guanjianch@ucas.ac.cn (J. Guan).
ening cross-national cooperation to develop alternative energy
technology (Philibert, 2004). The researches of these technolo-
gies must be deployed and adopted on a global scale which will
involve producers, regional and national innovation systems in
nations around the world (Mowery et al., 2010). Participants form
these collaborations for spreading the costs of huge investments,
sharing and pooling risks, and to gain access to complementary
abilities and resources (Marquardt, 2013). International collabora-
tive linkages have the capacity to transfer disembodied knowledge
over long distances (Herstad et al., 2014). It is clear that, from
a long-term development perspective, a collaborative network
encourages close relationships and creates more value among part-
ners (Adegbesan and Higgins, 2011). Not surprisingly research on
collaboration networks has increasingly attracted the attention of
scholars (Chen and Guan, 2010; Wagner and Leydesdorff, 2005).
Although previous studies have explored collaboration net-
works at the country level by analyzing several international
scientific data sets (Cantner and Rake, 2014), there is still relatively
little empirical evidence on the global technological collaboration
network (De Prato and Nepelski, 2012). Some scholars suggest that
it is worth considering sub-national or regional systems of innova-
tion to fully understand innovation process (Feldman and Florida,
1994). However, many existing collaboration network studies focus
on country, regional, organizational and industrial innovation net-
works, but they rarely include city collaboration networks (Fleming
et al., 2007; Gay and Dousset, 2005; Phelps, 2010). Cities are not

http://dx.doi.org/10.1016/j.respol.2014.12.007
0048-7333/© 2014 Elsevier B.V. All rights reserved.
546 J. Guan et al. / Research Policy 44 (2015) 545–559

only basic units in economic competition but are also important innovation at multiple levels. A sample of 41,007 patents in the field
targets for government policies. Cities are centers for the integra- of alternative energy taken from the USPTO is collected to test the
tion of human capital and incubators of invention (Bettencourt hypotheses. Afterwards, characteristics of the collaboration net-
et al., 2007). Economic and political regimes are not only operative works of inter-city and inter-country are obtained with the help of
at national and global scales, but also at local, regional and city’s methodologies derived from network analysis. Finally, conclusions
scales. Cities utilize local tax to develop, create and concentrate are given and specific literature contributions are discussed.
knowledge which increases their attractive pull for highly skilled
and creative individuals who, by locating in urban centers, con-
2. Theory and hypotheses
tribute in turn to the generation of further knowledge spillovers
(Florida, 2005). Certainly considerable progress would be made if
To understand the process of innovation, we explain the nature
data sources identifying innovation activity at the city or county
of innovation in views of the Schumpeterian re-combinatory per-
level were made available (Audretsch and Feldman, 1996). For this
spective, which view the innovation as novel recombination of
reason innovation collaboration at a city level should receive more
existing knowledge (Cantner et al., 2011). Previous work suggested
theoretical and empirical attention (Neal, 2011). In the past few
that innovation is essentially liable to be embedded in a way that
decades, research on networks at a city level has mainly focused
the innovativeness will be affected by external and extra factors
on the transportation network and most of them approach this
(Freel, 2003) and explores innovation using this notion by recom-
problem from a geographical perspective (Singh and Marx, 2013).
bining resource within and outside boundaries (Belussi et al., 2010).
Recent studies have utilized social network to understand the
Collaboration and the establishment of a composite knowledge
influence of an innovator’s network structure on its innovation
base are very important for innovation processes (Hölzl and Janger,
performance (Rost, 2011; Bercovitz and Feldman, 2011). However,
2014). As cities form and maintain collaborations with each other,
recent work has exclusively focused on the benefits that innova-
they weave a network consisting of extensive communication ties
tors obtained by occupying advantageous positions (Gilsing et al.,
(Ynalvez and Shrum, 2011). Central cities have more opportunities
2008). There is still a gap which neglects to test the impact of
for knowledge transfer and learning. However, the benefits from
the higher level network structure on the lower level innovation
the central position in inter-city collaboration network would be
processes. Attempts to integrate these levels, either empirically or
influenced by the national-level characters. Prior research did not
theoretically (Pan et al., 2012) or study interactions of two levels
explore the effect of a nation’s position on its internal innovative
(Schwab, 2007) hardly exist. A great deal of research regarding mul-
performance.
tiple levels is concerned with the effect of individual-organizational
This study explains the theory about how cities and nations
interactions on individual outcomes (Hrebiniak and Alutto, 1972;
are connected at the network level from two perspectives. On the
Paruchuri, 2010). Some scholars argued for multilevel models pro-
one hand, we explain these from internal and external innova-
vide insights into a richer perspective innovation helping us to
tion theory. Prior studies demonstrated that internal and external
understand innovation phenomena at and across different levels
knowledge acquisition activities are complementary for innova-
(Gupta et al., 2007). Combining this issue we observe that, in spite
tion (Cassiman and Veugelers, 2006). Today, we need to expand
of their evident importance, surprisingly little attention has been
the external use of innovation to conduct purposive inflows and
devoted to city–country interactions. Our research is designed to fill
outflows of information and knowledge to accelerate internal inno-
this gap by putting forward new proposals that concern both city
vation (Poot et al., 2009). A nation’s internal innovation activities
and country levels. The analyses at one level are linked to those
are carried out by innovation cities connected in an inter-city
at the other, and the innovation mechanisms may be different at
collaboration network. It is noted that international collabora-
each level. Relationships among cities bring collaborative trust and
tion network brings the country external information flows which
information on a small scope. Collaboration among countries can
would influence its internal innovation network. To be specific,
bring some fresh and heterogeneous knowledge in a broader scope.
inter-country ties act as channels of communication that provides
Thus, we will study the interaction of the two levels of collaboration
knowledge spillovers from other countries. Furthermore, the influ-
networks.
ence of external information flows on its internal network relies
Acquisition of knowledge is crucial for the economic develop-
on the amount of spillovers coming from the inter-country net-
ment of cities and from there for the growth and innovation of
work, which in turn depends on the country’s position within the
regions (C. Phelps et al., 2012). The structure of networks in which
network. These flows within the nations in turn differentially influ-
actors are embedded influences their potential for knowledge
ence the ability of cities and regions to assimilate and process
acquisition (Gonzalez-Brambila et al., 2013; Gulati, 2007; Schilling
information, thereby changing their innovation activities. Thus, we
and Phelps, 2007). A good position occupied by a city presents
combine inter-country collaboration and the inter-city collabora-
advantages which let a central city enjoy profits in terms of infor-
tion at the network level. On the other hand, we explain these
mation collection and processing (Zaheer and Bell, 2005). Given
from interaction theory. Environmental influences all operate as
the above-described problem, the effects of domestic and of for-
interacting determinants that influence each other bidirectionally
eign knowledge on technology innovation are different (Almeida,
(Bandura, 2001). Thus, innovation activities would be influenced by
1996), and prior study argued that geographical distance influence
the interaction between innovators and their environmental con-
knowledge flows (Criscuolo and Verspagen, 2008). Yet, a holistic
texts. Given the above-described reality, the innovation cities are
view is needed taking domestic as well as global networks into
involved as actors and at the same time embedded in the broader
account. This raises the following questions: Does the structure
environment (i.e. national context). Thus, from these views, we con-
of inter-country network influence collaborations among cities in
nect cities and nations at the network level, and investigate their
inter-city network? If so, what network structure of inter-country
interaction effects.
will enhance a city’s innovation? We aim to examine how the effect
of a city’ position on its innovation activities varies with the coun-
try’s position in the inter-country collaboration. 2.1. City’s network structure and innovation output
This article, studying to above mentioned problems, is organized
as follows. First, we offer a review of the major literature on net- In an inter-city innovation collaboration network, actors with
work structure and innovation. Second, a model and hypotheses are different egocentric network properties have different opportuni-
proposed to account for the mechanisms and processes underlying ties to acquire new knowledge, essential to innovation activities
 J. Guan et al. / Research Policy 44 (2015) 545–559
(Acs et al., 2002; Tsai, 2001). Previous empirical studies have shown
a close relation between the network structure and actors’ innova-
tion performance (Grewal et al., 2006; Phelps, 2010). Most prior
studies emphasized positive effects of networks on the inter-group
level (Schilling and Phelps, 2007; Tsai and Ghoshal, 1998). In this
study, we aim to contribute to the inter-country and intro-country
network theory. Centrality is a structural characteristic capturing
patterns of knowledge exchange in an organizational unit (Breschi
and Catalini, 2010). Centrality can reflect the extent to which a city
acts as an intermediary, the degree of access to resources or the way
it controls them (Brass, 1984; Freeman, 1979). Partners occupy-
ing central structural positions are in a better situation for sharing,
integrating and utilizing complementary, heterogeneous resources
such as information, technology and knowledge (Borgatti, 2005).
Moreover, central nodes occupy critical gate-keeping positions
(Valente, 2012). Cities with high centrality enjoy more advan-
tages in relation to patent output. First, existing evidence supports
that cities with high centrality have an information advantage
(Becker, 1970; Gilsing et al., 2008). Such cities have access to unique
information and become rich information resources (Perry-Smith,
2006). A central city becomes a hub of information and knowledge.
Previous work illustrated that actors with high centrality exercise
more control in the network to which they belong (Rowley, 1997).
Nodes or clusters in the network can be connected via a key node
which has the unique advantage to determine the flow of resources.
A central city also provides opportunity to easily bridge the infor-
mation gap between tripartite relationships. Finally, cities with
high centrality have a greater capacity to bear risk and to reach a
position of high status and high prestige (Ahuja et al., 2012). Hence,
such a city has more power and can attract new cities gravitating
around it (Sytch et al., 2012). The costs associated with coopera-
tion as well as communication decrease and these cities end up
with more social capital (Li et al., 2013).
Based on the above arguments and discussion, we propose the
following hypothesis:
Hypothesis 1. A city’s centrality positively affects its patent out-
put.
An actor filling structural holes in a collaboration network is tied
to disconnected clusters of other actors (Burt, 2005, 2007) and as
such may be able to acquire fresh information (Granovetter, 1995).
Structural holes refer to the absence of links that would otherwise
connect actors. The existence of structural holes indicates that a
third node can act as a broker or intermediary (Scott and Carrington,
2011). Burt’s classic work argued and demonstrated that struc-
tural holes are opportunities to broker the flow of knowledge and
information between actors, and control the projects that bridging
actors from opposite sides of the hole (Burt, 2004). Such actors have
information benefits and enjoy more autonomy in decision mak-
ing (Shipilov and Li, 2008; Wang et al., 2014). The actors, which
bridge these holes, attain an advantageous position yielding ben-
efits in terms of information and control (Burt, 2002; Zaheer and
Soda, 2009). Meanwhile, innovative cities may have more innova-
tive firms, university and individuals (Etzkowitz and Leydesdorff,
2000; Rothaermel and Hess, 2007). Thus, cities can possess the
firms’ and other’s advantage of information and control through
spanning structural holes. In addition, moreover, such a city may be
free of constraints thereby enjoying autonomy in inventive activi-
ties (Shipilov, 2009). The benefits of structural holes, situating cities
at the confluence of different social domains, are non-redundant
and heterogeneous resources and greater freedom or autonomy
to exploit them. According to resource-based view (RBV), actors
endowed with the heterogeneous resources are able to produce
more economically and/or satisfy the need (Peteraf, 1993). Thus
the cities spanning more structural holes have obvious competitive
advantages. There are also debates that structural holes may pose
547
an action problem because the dispersed actors may be difficult
to coordinate and mobilize in sparse networks (Obstfeld, 2005). In
general, we assert that information and autonomy advantage struc-
tural holes afforded can overwhelm the disadvantage the action
problem produced.
Based on the above discussion, we propose the following second
hypothesis:
Hypothesis 2. A city able to fill structural holes has a position
which positively affects its patent output.
The clustering coefficient is a measure of the degree of inter-
connectivity in the neighborhood of a node (Watts and Strogatz,
1998). The traditional definition of clustering coefficient for node
A is “the probability that two randomly selected friends of A are
friends with each other” (Easley and Kleinberg, 2010). For example,
a node whose neighbors are all connected to each other has C = 1. On
the contrary, a node with no links between its neighbors has C = 0
(Ravasz et al., 2002). The clustering coefficient indicates whether
one actor’s direct contacts typically also know each other (Gulati
et al., 2012). Clustering enhance the information transmission
capacity of the network (Guan and Zhao, 2013). A high clustering
coefficient leads, on the one hand, to easier diffusion of resources,
more third party and repeated links providing trust in a group and
greater risk sharing (Granovetter, 1985). On the other hand, high
clustering may lead to a chaotic view (too many perspectives) and
repeated information, which may harm innovation performance.
Excessive structural connectivity would decrease the creative dis-
tinctiveness of clusters. Such clusters are filled by traditional ideas
rather than fresh opinions. A high clustering coefficient means that
the city’s partners can collaborate with each other freely, result-
ing in a loss of its brokerage role. In addition, closure’s associations
with distrust, character assassination and maintenance costs are as
strong as its association with trust (Burt, 2004). Based on the above
discussion, we propose the following third hypothesis:
Hypothesis 3. A city’s clustering coefficient negatively affects its
patent output.
2.2. The moderating role derived from characteristics of a country
network
The relationship between a city’s characteristics and its patent
output is influenced by other factors too (Tsai, 2001). It also depends
on the country’s position in the inter-country collaboration net-
work. Thus, we need to analyze the interaction between the city
and the country level. Interaction theory holds that innovation
behavior occurs in a social environment. Environmental factors
will not only have a direct impact on individual behavior but also
strengthen or weaken the relations between individual indepen-
dent variables and dependent variables (Eisingerich et al., 2010).
Gupta et al. (2007) suggested that all innovation is at least a two-
level phenomenon, which involves the actor and the environment
in which the actor is embedded. Given the above-described reality,
we should take into account the broader environment – country –
within in which the city is embedded. International collaborations
form very influential, rapidly changing, and highly dynamic net-
works which feed back into the local and national levels, influencing
the innovation process (Wagner and Leydesdorff, 2005).
When a country has a high centrality, the country will get
more opportunities of knowledge learning, access and utilization
improving the efficiency of innovation (Tsai and Ghoshal, 1998).
Countries occupying a central position in the network always per-
form better than other nodes without these characteristics (Mason
and Watts, 2012). The benefits from a country’s high centrality may
enhance positive effects on innovation performance of its cities.
Core nodes with a higher reputation than peripheral nodes will
548 J. Guan et al. / Research Policy 44 (2015) 545–559

have easier access to critical knowledge and build up their own

innovation ability faster. As such, we conjecture that cities situated
in countries with a high centrality have more advantages than cities
in peripheral countries.
As for cities, also countries that fill structural holes obtain an
advantageous position leading to benefits in terms of information
and control. Structural holes theory proposes that structural holes
exist between two acquaintances if they connect with the same
individual but are not linked with each other (Burt, 1993). Such a
country spanning structural holes can be more flexible in terms of
national policy and international collaboration. Domestic cities can
enjoy the benefits of national resources and policy. Countries span-
ning more structural holes are confronted with fewer constraints
from other partners. Its cities can also avoid external restrictions,
and thus be more productive in the inter-city network.
Networks which present several triads will result in high clus-
tering coefficient (Scellato et al., 2011). A triad is a network
structure in which two actors are connected with each other and
to at least one common third party (Lee et al., 2014). The coun-
tries with high clustering coefficient are short of structural holes,
featuring the plenty of connections among their partners in the net-
work. Norms and homogeneous information may diffuse in a dense
and highly clustering ego network, so deviant behavior is curtailed
and performance will be suppressed (Gargiulo and Benassi, 2000).
Due to the redundancy of knowledge and information transmit- Fig. 1. Knowledge network in the alternative energy field.
ting within the network, the generation of new ideas is hindered
(Granovetter, 1973). The diversity of knowledge and capacity of
transferring in inter-city network will be limited. The influence of areas. In effect, alternative energy requires concurrent integration
inter-city network on cities will decrease. of mixed technologies. The USPTO categorizes technologies into
Similarly, we can propose the following hypotheses. more than 400 classes and 100,000 subclasses which can be thought
of as valid proxies for knowledge elements (Yayavaram and Ahuja,
Hypothesis 4. The structure of the inter-country collaboration
2008). Fig. 1 shows the knowledge network of alternative energy in
network moderates the relationship between the inter-city collab-
the period 2008–2012. Each node represents a knowledge element,
oration network structure and innovation performance.
and edges represent co-occurrence relations. If two knowledge ele-
Hypothesis 4(a). A country’s centrality moderates the relation- ments, indicated by industrial subclasses, are co-present in a patent,
ship between the characteristics of its cities’ network and the cities’ there is a tie between them. Yellow corresponds to wind power and
patent output. blue represents knowledge in the field of solar energy. We clearly
find that these technologies are completely intertwined. Moreover,
Hypothesis 4(b). A country’s structural holes moderate the rela-
testing our hypotheses in such a large field enhances the reliability
tionship between the characteristics of its cities’ network and the
of the results. Thus, we claim that using alternative energy patents
cities’ patent output.
as the focus of our research is an appropriate choice. It is, more-
Hypothesis 4(c). A country’s clustering coefficient moderates the over, effective to choose the USPTO database. Patents applied in
relationship between the characteristics of its cities’ network and the USPTO are generally considered to have high value in economic
the cities’ patent output. and technological properties embodying high level of technological
innovation. Patents contribute more innovation and have become
3. Methods common measure of innovation output (Giuri et al., 2007). In all,
our data pool contains 41,007 patent documents over the period
3.1. Research background, sample, and procedures 1976–2012.
Each patent in USPTO contains the names and addresses of the
We tested hypotheses using a longitudinal dataset including inventors (cities, states and countries of residence, while states are
patenting activities that were active in the field of alternative available for the patents originating in the USA). Such rich data
energy. The reasons to choose data on alternative energy will information enables us to construct inter-city and inter-country
be presented below. Firstly, with increased understanding of the collaboration networks (Blomkvist et al., 2014) for overlapping 5-
effects of greenhouse gases on climate change, research and devel- year windows from 2000 to 2011. We use a variety of matching
opment of alternative energy sources continue to rise (Moss et al., algorithms to extract the cities and countries information and uti-
2010). Thus, alternative energy technology patents have risen enor- lize software Sci2 (Team, 2009) to construct networks. Specifically,
mously in recent years. It really has become a hot spot (Ferdous there is an edge between two cities if the inventors in them have
et al., 2013). co-patented at least once. If researchers from two nations have
Secondly, environmental protection transcends all boundaries co-authored a patent, there will be a tie between two nations. Col-
of time and space (Wapner, 1997). Hence it cannot be managed laboration networks can be roughly divided into 8 periods. This
through distant and indirect participation. The energy problem is study also collects data before 2000, but this part of the data is
by its very nature a global problem requiring global collaboration. not included in the regression analysis as these early networks are
Over 62% of alternative energy patents belong to more than one too sparse. In addition, our collected patent data for 2013 is incom-
international class which means that new trends for the integration plete since it is too recent. Thus, the total sample of our longitudinal
of alternative energy are expected. Each specific sub-technology in observations ends in 2012. By the end of 2012, 72 countries applied
the field of alternative energy is connected to innovation in other alternative energy patents in the USPTO. We analyze 8 countries
 J. Guan et al. / Research Policy 44 (2015) 545–559
350
300
250
200
150
100
50
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Fig. 2. Alternative energy patents by countries by periods (two y-axes).
thoroughly, namely the G7 countries (the United States, Britain,
France, Germany, Japan, Italy, and Canada) and China. The reason
we chose the G7 countries is that they are the most developed coun-
tries in the world leading in science and technology. Moreover,
these countries’ innovation systems and R&D resources are (rel-
atively) stable (King, 2004). As the present second largest economy
in the world, scientific and technological innovation is recognized
as an important strategy in China (Huang, 2010). These eight coun-
tries’ alternative energy patents cover 80.93% of the total in the
world. Fig. 2 with two y-axes depicts the country distribution of
alternative energy patents for each period. Because of the huge gap
between the USA and other nations, we put the patent number
of the U.S on the right y-axis and these of other countries on the
left y-axis. It is noteworthy that, although China has high num-
bers of patents in some areas (Hu, 2010); the number of patents
in alternative energy grows much slower than for other countries.
The USA’s patent output is far ahead in the field of alternative
energy and maintains a steady growth trend over the entire period.
Japan and Germany display a similar pattern of development. How-
ever, their patent output is only about one-seventh of the United
States’. A report of the Royal Society states that the current govern-
ment policy does not provide enough funding for green technology
(Shepherd, 2009), leading to a decline in the British number of
patents on alternative energy. In contrast, France, favored by ample
natural resources, is the European Union’s second largest producer
and consumer of renewable energies. Its output in patents on alter-
native energy increases every year. Nevertheless, China’s output
catches up with Italy and gradually gets close to the level of Canada.
We also study the patent collaboration networks at the coun-
try level. The inter-country collaboration network in 2007–2011
is shown in Fig. 3. A darker color is used to indicate the quantity
of each region, and the width of the lines across two countries
corresponds to the strength of the collaborative relation, i.e. num-
ber of collaborations. On the one hand, we analyze similarities and
Fig. 3. Inter-country network.
549
2500
JP
2000 DE
FR
1500
GB
1000 CA
500 CN
IT
US
differences among countries using color codes. As can be seen from
the graph, the USA, Japan, and the Western European countries are
the first echelon members representing high-producing countries.
Australia, China, and India belong to the second echelon producing
a moderate level output. South American countries and Africa are
the third echelon members whose patent output is relatively rare.
When analyzing connected lines between countries we find that
countries with high patent output always have collaborative rela-
tionships with other countries with high output, whereas countries
with low patent output tend to collaborate with countries with high
output rather than geographical neighboring countries.
Patent output shows significant differences across regions.
Patent output of the USA states is demonstrated in Fig. 4. The color
reflects the alternative energy patent output. The darker the color,
the more patents. Through enquiring the GDP (Gross Domestic
Product) of American States in 2011, we find that states’ GDP has a
striking consistency with its patent output. Three of the top states in
GDP are Texas, New York, and California. Moreover, from Fig. 4, we
see that alternative energy patent output of these three states is rel-
atively high. Moreover, the last three states in GDP are Wyoming,
Montana, and Vermont. Correspondingly, a relatively low patent
output is found in these states. We consider that a city or region
with relatively rapid economic development pays more attention
to environmental protection and the development of alternative
energy, and they have more resources and the ability to develop
alternative energy.
In order to see inter-city collaboration network more clearly,
we take patent collaboration network of the USA in 2007–2011
as an example, and draw the largest connected component. Fig. 5
displays patent collaboration network between cities in the USA.
Nodes represent cities and lines represent collaborations. Edges are
presented in this figure if and only if their weights exceed 2 and
cities whose degrees exceed 15 are showed in Fig. 5. The width of
edges reflects the number of collaborations. The actual network is
a dense network with more than 1500 nodes and 13,000 edges.
3.2. Measures
3.2.1. Dependent variable
The patent output. We use the number of alternative energy
patents granted in the year following the collaboration network
period as the measure of innovation output. The choice of this
measure is based on existing studies showing that using patents
number can be considered a measure of innovation activity (Lahiri
and Narayanan, 2013). For example, if we study the collaboration
network at a city level in 2000–2004, the dependent variable is the
number of city’s alternative energy patents granted in 2005.
3.2.2. Independent variables
3.2.2.1. City and country centrality. After constructing collabora-
tions networks at city and county levels, we calculate the value
550 J. Guan et al. / Research Policy 44 (2015) 545–559

Fig. 4. Patent output of states in the USA.

of the centrality for each city and country using Pajek (Batagelj
and Mrvar, 1998). Previous research found that different centrality
measures may be highly collinear which could result in multi-
collinearity among these variables (Salman and Saives, 2005). Thus,
we choose only one centrality measure. We use betweenness cen-
trality as a measure of structural centrality (Cantner and Rake,
2014; Peterman et al., 2014). Betweenness centrality measures
the ability of the node occupying a critical gate-keeping position
to act as an intermediary (a broker) (Valente, 2012). Nodes with
high betweenness centrality control the information flow, grasp
business opportunities, and have access to intermediary benefits.
Betweenness centrality is calculated as shown in Eq. (1). The sym-
bol gjk denotes the total number of shortest paths from node j to
node k, gjk (ni ) refers to the number of paths that go through node
ni and g is the number of nodes in the network.
 that are directly or indirectly invested in the connection with j. The
gjk (ni )/gjk
j<k
CB (ni ) = (1)
(g − 1)(g − 2)
Both city and country betweenness centrality are calculated in
each period using Pajek. These indicators are scaled and reflect the
actor’s ability as an intermediary.
3.2.3. Structural holes in the cities and in the countries networks
Burt uses the term structural holes to indicate the absence of
connections among two actors (Burt, 1992/2005). We adopt Burt’s
(1992) measure of constraint to compute structural holes using
Pajek. A high constraint means that the actor’s ego network is
densely connected, whereas a low constraint measure represents
that actor’s partners are sparsely connected to one another (Koka
and Prescott, 2002). The CO index of node i is calculated as in Eq.
(2):
  2
COi = (pij + piq pqj ) (2)
j q,q =
/ i,q =
/ j
In detail, pij is the proportion of i’s relations invested in contact-
ing j, and the total in parentheses is the proportion of i’s relations
value of constraint measure COi is sometimes bigger than 1 (Lee,
2010). To arrive at a measure of structural holes Si , we subtract the
aggregate constraint measure COi from 2 (Wang et al., 2014) as is
shown in Eq. (3). The higher the index, the richer a city or country
is in structural holes.
Si = 2 − COi (3)

Fig. 5. Inter-city network in the USA.

 J. Guan et al. / Research Policy 44 (2015) 545–559
3.2.4. Clustering coefficients for cities and countries
To calculate the actual clustering coefficient, we determine how
many pairs of collaborators have a shared association, or how many
triads are “closed”. Three different links between three actors can
yield a triad: suppose actor X is connected to actor Y who is con-
nected to person Z, if actor X is connected to actor Z, then the three
links among actors X, Y, and Z form a closed triad (i.e., a triangle).
The clustering coefficient shown by Eq. (4) is a standard method to
explore how clustered these networks are.
2E(i)
CC(i) =
k(i)(k(i) − 1)
where E(i) is the number of edges between node i’s neighbors, and
k(i) is the degree of node i. Hence CCi = 0 if none of the neighbors of
a city or country are connected, and CCi = 0 if all of the neighbors
are connected.
3.2.5. Control variables
Pre-sample: To control for unobserved heterogeneity in mem-
ber’s patenting activities, we choose a pre-sample as the control
variable (Blundell et al., 1995). The pre-sample is the number of
patents acquired by a city in the 5 years prior to its entry into the
sample.
Firms: To control for the scale influence, we select firms as a size
variable, which is measured by the number of firms which are active
in the focal period in the city.
Inventors: We also include inventors in the focal period as the
control variable, measured by the number of inventors in the focal
period in the city.
Knowledge: This variable indicates the knowledge breadth of the
city, which is calculated by the number of knowledge elements (i.e.
industrial subclasses) that the city’s patents involved.
Network characteristics: There are other factors that affect the
dependent variable. Some other network characteristics in city col-
laboration network in each period are calculated using Pajek. We
include: (1) Number of collaborations. It reflects the amount of
collaborations in the inter-city network. (2) Density of the net-
work. This variable means the density of the inter-city network.
It is calculated by the ratio of the number of actual connections and
their possible maximum connection number. (3) Average path. It
measures the average length of a path in the inter-city network,
quantifying the information transmission speed in the network.
3.3. Statistical models
The data used in this study have two main features. On the one
hand, the dependent variable, that is, the number of patents, is a
non-negative integer which can be dealt with in a negative binomial
model (Hausman et al., 1984). The negative binomial model is an
extension of the standard Poisson model (Baba et al., 2009) which
can account for over-dispersion. The basic Poisson model that was
used by Miaou et al. (1991) to estimate the conditional probability
is as follows:
y Subramanian, 2013). After tests, the biggest VIF (variance inflation
i i e−i
p(Yi = yi ) = p(yi ) = i = 1, 2, 3, . . ., n.
yi !
where
k
 x ˇ
i = E(Yi ) = i [exi ˇ ] = i [e j=1 ij j ] i = 1, 2, 3, . . ., n.
Utilizing the Poisson regression model requires that the mean
of the data is restrained to be equal to the variance (Demidenko,
2013), which means Var(Yi ) = E(Yi ) = i . However, countable data
frequently suffer from over-dispersion. We found that the mean of
dependent variable is 1.24 and the SD is 3.39. The evidence of over-
dispersion in the data suggests that negative binomial models are
551
more appropriate (Cameron and Trivedi, 2013). The negative bino-
mial regression model used in our study has the following form:
 (yi + (1/˛))
 1
1/˛  ˛ yi
i
p(Yi = yi ) =
 (1/˛) (1 + yi ) 1 + ˛i 1 + ˛i
yi = 0, 1, 2, 3, . . .
where
k
 x ˇ
(4) i = E(Yi ) = i [exi ˇ ] = i [e j=1 ij j ] i = 1, 2, 3, . . ., n. (7)
and we obtain the variance of Yi :
Var(Yi ) = i + ˛2i (8)
In above formulas, ˛ refers to dispersion parameter, and we
can see that the variance exceeds the mean in the negative bino-
mial regression model. On the other hand, since we use 2000–2012
panel data, we need to determine if we model the data using a fixed
effect or a random effect. Due to that cities of a nation are related
to each other, the observations across cities are not independent.
Further, the choice of country samples is not random; besides, we
consider that the difference in patent output in countries should be
controlled. Consequently, we utilize the models with country-level
fixed effect and city-level random effect in the analysis. To be spe-
cific, the functional form of our models which are considered for i
is as follows:

(˛j Xj +ˇk Zk +i +εit ))
i = e( (9)
where ˛j displays the coefficient of the “j” independent variable of
an city “i”, and ˇk means the fixed effect for country k. In addition,
i represents the random effects for city i, and εij is the error term
in our models. As a consequence of the above, this issue reduces
the bias because now the results illustrate the change in the inno-
vation output between cities in the same country. Therefore, the
comparison is not across countries but for different independent
variable values within the same country. In order to control for
possible heteroscedasticity, we conduct a White test (rejecting het-
eroscedasticity at the 0.01 level).
4. Results
Table 1 displays the means, standard deviations, and correla-
tions of the variables included in our study. Supporting the prior
literature on innovation performance, a city’s centrality is posi-
tively related to its patent output (r = 0.096, p < 0.01), and structural
holes are positively related to its patent output (r = 0.254, p < 0.01).
However, a negative correlation for the relationship between a
city’s clustering coefficient and patent output is found (r = −0.186,
p < 0.01).
Table 2 summarizes the results from negative binomial regres-
sions. To prevent collinearity when interactions are entered, we
centralize the interaction terms (Aiken and West, 1991; Faems and
factor) in Table 2 is 4.85, indicating multicollinearity among vari-
(5)
ables is not a big concern (Chatterjee and Price, 1991). A Durbin test
(Durbin and Watson, 1950) is also done to verify the autocorrela-
tion in models, and we infer that models have no serial correlation
(H0 : no serial correlation, p value is 0.96).
(6)
To test Hypothesis 1–3, we regress a city’s patent output on
its centrality (BC), structural holes (SH), and clustering coeffi-
cient (CC) together with control variables. Model 1 in Table 2
presents the results. As expected, the baseline model indicated
that pre-sample which can reflect city-level innovation capabili-
ties is positively associated with innovation output. We also note
the non-significant effect of density on innovation. High density on
552 J. Guan et al. / Research Policy 44 (2015) 545–559

the one hand may bring dissemination of trust and make coordi-

.094**
nation process fluid, on the other hand may reduce the diversity

13

1
of competence and experiences (Soda, 2011). In the light of above-
mentioned reasoning, the effect of density on innovation could be

−.585**
.097**
non-significant. As predicted by Hypothesis 1, a city’s centrality
positively affects its patent output (r = 1.082, p < 0.01). A city’s struc-
12

1
tural holes have a positive effect on its patent output (r = 0.486,
p < 0.01), and hence Hypothesis 2 is supported. A city’s clustering

−.652**
.820**
−.091**
coefficient has a negative effect on its patent output (r = −0.206,
11

p < 0.01), and Hypothesis 3 is also verified.

1
To test Hypotheses 4(a), after entering control variables, we
regress a city’s patent output on its network characteristic vari-

.254**
−.057**
.085**
.081**
ables, its country’s clustering coefficient, together with interaction
10

1
terms of the network characteristics at the country level and the city
level. We put CC country * CC city , CC country * BC city , and CC country * SH
−.152**

.096** city into model 2–4 in turn. As shown, we obtain significant coeffi-
.018*
.013
−.011

cients of CC country * CC city and CC country * BC city (r = −8.368, p < 0.01;

9

r = −3.387, p < 0.01, respectively). Nonetheless, the coefficient of CC

country * SH city (r = −0.245, p > 0.1) is not significant. In conclusion
−.376**

−.186**
.029**

−.020*
.017

.014

we may say that Hypothesis 4(a) is partially corroborated.

We then test Hypotheses 4(b), after entering control variables,
8

we regress a city’s patent output on its network characteristic vari-

ables, and its country’s centrality, together with interaction terms
−.328**
.226**
−.043**
.047**
−.011
−.015
.009

of the network characteristic at the country level and the city level.
7

We add BC country * CC city , BC country * BC city , and BC country * SH city

into model 5–7 in turn. As expected we obtain a non-significant
−.128**
.677**
−.376**
−.062**

.706**
−.053**

BC country * CC city (r = 1.478, p > 0.1), a significant BC country * BC city

.010
.02*

(r = 3.695, p < 0.01) and a significant BC country * SH city (r = 0.513,

6

p < 0.05), thereby partially supporting Hypothesis 4(b).

Finally, after entering control variables, we regress a city’s
−.374**

.135**
−.627**
.628**
−.097**
−.027**

−.710**
.104**
−.002

patent output on its network characteristic variables, its country’s

structural holes, together with interaction terms of the network
5

characteristic at the country level and the city level. We put SH

country * CC city , SH country * BC city , and SH country * SH city into model
.162**
.416**
−.122**
.138**
.125**
.395**
.106**
−.034**
.049**
−.300**

8–10 in sequence. We acquire a significant SH country * CC city

4

(r = 13.55, p < 0.01) and a significant SH country * BC city (r = 9.465,

p < 0.01). Nonetheless, the coefficient of SH country * SH city (r = 0.466,
.697**

−.287**

.434**

.379**
.078**
−.035**
−.048**

.160**

.052**
.038**

p > 0.1) is not significant which means that hypothesis 4(c) is only
−.019*

partially accepted. The Likelihood ratio test of improvement of fit

3

compares a model with the baseline model (i.e. model 1) shows

that all the models improve significantly on baseline model.
.539**
.357**
.425**

−.188**
.158**
.199**

.255**
−.505**
.057**
.050**
.020*
−.002

To clearly understand the meaning of the interaction effect,

the implication of the coefficients for explanatory variables and
2

interaction terms in Table 2 must be understood. For instance, the

coefficient estimates for BC city are positive in Models 3,6 and 9
.291**
.454**
.461**
.148**
−.138**

−.284**
.644**
.289**

.126**
−.129**
.772**
.032**

−.120**

if we assume that the values of moderating variables are equal to

1

zero, which means when we remove the impact of the interaction

with BC city , we obtain the positive effect of BC city on patent output.
Similarly, the effect of SH city is positive, while that of CC city is neg-
11.37

0.13
0.16

0.61

0.14

3.39
0.03

0.05

0.40
0.06
0.30

0.20
0.05

ative when moderating variables are equal to zero. The significant

4285
SD

interaction terms are taken into account to understand the effect

of explanatory variables. For example, the coefficient estimates for
interaction terms between explanatory variables and BC country are
Means, standard deviations, and correlations.

5.33
0.94
5.42
14.13

3.82
0.53

1.45
0.19
0.34

1.24
0.02

0.01

1.90

positive and significant. These results indicate that BC country posi-

6723

tively moderates the relationships between explanatory variables

M

and innovation output. Similarly, SH country positively moderates

above relationships, while CC country negatively moderates above-
mentioned relationships.
5 Number of collaborations

We conducted some additional analyses to confirm the robust-

ness of our empirical results. First, we used the ordinary least
squares (OLS) analysis to re-estimate the model. Second, we put
6 Density of net
7 Average path

the logarithm of the patent number plus 1 as the dependent vari-

4 Knowledge
1 Presample

13 SH country
12 BC country
11 CC country
3 Inventors

14 Patents

able to test the model. Third, we use a hierarchical linear modeling

10 SH city

p < .05.
p < .01.
9 BC city
8 CC city
2 Firms

(HLM) framework to test the effects of network structure of cities

Table 1

and national levels moderating hypotheses. Our results are consis-

*

**

tent with and similar to previous results. Finally, even though we

Table 2
Results of regression models.

Model 1 2 3 4 5 6 7 8 9 10

Pre-sample 0.018** 0.018** 0.018** 0.018** 0.018** 0.018** 0.018** 0.018** 0.018** 0.018**
Firms 0.051** 0.059** 0.055** 0.055** 0.055** 0.054** 0.055** 0.054** 0.053** 0.053**
Inventors 0.126** 0.123** 0.123** 0.124** 0.123** 0.122** 0.123** 0.125** 0.124** 0.125**
Knowledge 0.038** 0.040** 0.039** 0.039** 0.039** 0.040** 0.039** 0.039** 0.040** 0.039**
Number of collaborations 0.160 0.402** 0.299** 0.250** 0.313** 0.328** 0.314** 0.432** 0.333** 0.249**

J. Guan et al. / Research Policy 44 (2015) 545–559

Density of net 0.489** 0.608 0.517 −0.027 0.460 0.335 0.153 0.807 0.165 −0.260
Average path 0.070+ 0.170** 0.145** 0.124** 0.113** 0.126** 0.127** 0.133** 0.123** 0.098*
CC city −0.206** −0.204** −0.194** −0.203** −0.207** −0.196** −0.201** −0.205** −0.192** −0.204**
BC city 1.082** 1.093** 1.788** 1.074** 1.056** 2.073** 1.185** 1.091** 1.823** 1.043**
SH city 0.486** 0.455** 0.452** 0.470** 0.477** 0.455** 0.453** 0.470** 0.456** 0.477**
CC country 3.259** 1.101** 0.862**
BC country −0.961** −0.835** −0.878**
SH country −6.241** −3.306** −2.031**
CC country * CCcity −8.368**
CC country * BCcity −3.387**
CC country * SHcity −0.245
BC country * CCcity 1.478
BC country * BCcity 3.695*
BC country * SHcity 0.513*
SH country * CCcity 13.55**
SH country * BCcity 9.465**
SH country * SHcity 0.466
Constant −2.812** −3.748** −3.346** −3.201** −2.690** −2.800** −2.776** 8.769** 3.220* 0.919**
Log likelihood −15,397 −15,366 −15,385 −15,391 −15,386 −15,385 −15,385 −15,379 −15,383 −15,393
Likelihood ratio test 62.06** 24.17** 10.82** 20.64** 22.93** 22.34** 34.35** 27.71** 7.38*

Notes: N = 13,339. Country fixed effects included. The coefficients of Number of collaborations have multiplied by 10,000.
+
p < .1.
*
p < .05.
**
p < .01.

553
554 J. Guan et al. / Research Policy 44 (2015) 545–559
derived our within-city hypotheses from theoretical models sug-
gesting a causal effect of centrality on innovation output (Li et al.,
2013; Sytch et al., 2012), the issue of the direction of causality needs
to be carefully considered. We conducted two additional analyses
to investigate causality. The first set estimates the effect of cen-
trality in the period t − 1 on the innovation output in the period
t, and the second set estimates the effect of innovation output in
the period t − 1 on centrality in the period t. The comparison of the
effects does suggest directionality mostly from the structural prop-
erty to the innovation: the standardized coefficient for predicting
innovation output with a previous period’s centrality is 0.10. On
the contrary, the previous period’s innovation output has no effect
(with a standardized coefficient 0.00) on centrality.
In order to verify the existence of endogenous variables in
regressions, we use a Hausman test to test and verify the endogene-
ity of the explanatory variables (Hill et al., 2008). The small p-value
of Hausman chi-square test indicates that there is a significant dif-
ference between the instrumental-variable (IV) and ordinary least
squares (OLS) coefficients, and instrumental variables approach is
appropriate (Song and Knaap, 2004). After testing we find that a
city’ clustering coefficient and structural holes don’t suffer from an
endogeneity problem (2 = 8.33, p > 0.1; 2 = 9.75, p > 0.1, respec-
tively), but that the variable “city centrality’ has this problem
(2 = 30.92, p < 0.05). In this article, a city’s centrality predicts its
patent output, and the city’s patent output predicts the city’s
centrality, which results in an endogeneity problem (Wooldridge,
2012). This means that we have to deal with this endogeneity. To
eliminate endogenous impact of the independent variables, we use
two-stage least squares (2SLS) regression. The difficulty in applying
a two-stage least squares method lies in the choice of instrumental
variables. The variables used must be correlated with the inde-
pendent variables that they are instrumenting for, and have to be
uncorrelated with the primary regression’s error term (Gould and
Gruben, 1996). The instruments are required to only affect depen-
dence variable through the independent variables that they are
instrumenting for (Acemoglu et al., 2000). Based on the purpose
of this study, we select the number of cities in the prior inter-city
network and the density of the inter-city network. The number of
cities in the prior inter-city network is related with a city’s network
characteristics but weakly related with its patent output. The den-
sity of the inter-city network in our study is a property of the whole
network, rather than the ego density of a specific node which only
considers direct ties (Gilsing et al., 2008). We find the density of
the inter-city network cannot significantly predict a city’s patent
output under controlling BC city (see Table 2).
Table 3 displays results of the fixed effects generalized least
squares regression model with instrumental variables, which are
consistent with previous findings which are represented by Table 2.
Although two models in Table 3 (models 7 and 10) fail to improve
model fit on baseline model, each model in Table 3 displaying dif-
ferent interaction terms helps us to thoroughly understand the
relationships assumed before. The significant interaction terms
confirm the results without the instrumental variables and provide
strong evidence to support our hypotheses.
To show the interaction effect in the form of a diagram Fig. 6
plots interactions graphs. We divide the city’ network characteris-
tics according to their median into high and low groups. We can
see from Fig. 6 that the main effects of a city’s centrality and struc-
tural holes on patent output are positive, while the main effect of
the clustering coefficient on patent output is negative. Upward in
Fig. 6, we draw the interactions between a country’s centrality and
inter-city network characteristics. As shown, when a country’s cen-
trality is high, the positive effects of a city’s centrality and structural
holes on innovation performance are enhanced, and the negative
effect of a city’s clustering coefficient is weakened. In the middle of
Fig. 6, we can see that when a country’s structural holes are high, the
positive effects of a city’s centrality and structural holes on innova-
tion performance are enhanced, and the negative effect of a city’s
clustering coefficient is weakened. At the bottom of this picture, it
is easy to see that when a country’s clustering coefficient is high, the
positive effects of a city’s centrality and structural holes on innova-
tion performance are weakened, and the negative effect of a city’s
clustering coefficient is enhanced.
5. Discussion
Earlier empirical research, focused on individual and organi-
zational levels, and typically ignored the interaction of structural
properties of networks at multiple levels. Our research, however,
goes beyond prior findings. The present study suggests that there
are complex relationships between a city’s network characteristics
and a city’s patent output, the forms of which vary with the corre-
sponding inter-country network. Our findings are consistent with
the interaction theory which states that a city’s innovation output is
produced by the joint effect of the city itself and contextual factors
related to the country to which it belongs (Paruchuri, 2010). Based
on a longitudinal and multilevel networks study, we find that the
network characteristics of cities where inventors live have signifi-
cant effects on their technological innovation, which are moderated
by the characteristics of countries in the international network.
Specifically, it is not recorded in the existing literature, and the
results of this study show that the relationships between the city’s
network characteristics and innovation performance depend on
three factors: the country’ clustering coefficient, centrality, and
structural holes. When the centrality and structural holes of the
country are high, the positive effects of a city’s centrality and struc-
tural holes on innovation performance are enhanced, and negative
effects of the city’s clustering coefficient are weakened. The coun-
try’s clustering coefficient plays an opposite moderating role. These
results are robust to plenty of additional tests and model specifica-
tions.
5.1. Theoretical implications
Our findings, apart from having managerial implications, have
implications for research on innovation processes, network stud-
ies and policy research. Firstly, our study has implications for
regional innovation research (Tsai, 2001). Previous studies have
neglected the effect of external links on the internal innovation pro-
cess (Nooteboom, 1999). Analyses on multilevel innovation were
ignored for a long time. Studies have illustrated that in five main-
stream management journals during the period 1990–2006, only
10% of innovation articles involved multilevel empirical analysis
(Gupta et al., 2007). Even so, the scarce studies that conducted mul-
tilevel empirical analysis on innovation networks were focused on
the individual and organization level (Rothaermel and Hess, 2007),
which are not even about the levels of nations and regions. Even
if some scholar considers these two levels, no work has been done
about the innovation mechanism from the perspective of two levels
networks (Sotarauta and Kautonen, 2007). In this paper, we build
multilevel collaboration networks to verify that the network posi-
tion of city does influence its innovation output. According to our
findings, the city’s network characteristics play an important role
in the process of innovation production. Network characteristics
on the country level may amplify or weaken the effects of a city’s
characteristics on its innovation output. Local governments should
encourage inventors to strengthen collaboration and information
exchanges in the region. Making a firm the center of a collabora-
tion network and using its own competitive advantage to occupy
structural holes are actions advantageous to a firm (Zaheer and
Bell, 2005). According to our theoretical and empirical analysis,
Table 3
Regressions with instrumental variables.

Model 1 2 3 4 5 6 7 8 9 10
** ** ** ** ** ** ** ** **
Pre-sample 0.221 0.221 0.221 0.221 0.221 0.219 0.221 0.221 0.221 0.221**
Firms 0.111** 0.110** 0.110** 0.111** 0.110** 0.108** 0.112** 0.110** 0.109** 0.111**
Inventors −0.012 −0.011 −0.011 −0.011 −0.012 −0.011 −0.012 −0.010 −0.010 −0.011
Knowledge 0.015** 0.015** 0.015** 0.015** 0.015** 0.016** 0.015** 0.015** 0.015** 0.015**
Number of collaborations 0.444** 0.468** 0.449** 0.424** 0.441** 0.464** 0.407** 0.583** 0.506** 0.442**
Average path 0.037 0.037 0.041 0.040 0.046 0.048 0.047 0.021 0.029 0.035
CCcity −0.038** −0.040** −0.033** −0.019+ −0.040** −0.001 −0.019+ −0.040** −0.030* −0.020+
BCcity 0.059 0.061 1.060+ 0.650+ 0.066 8.638** 0.697+ 0.024 1.116* 0.617+
0.231** 0.228** 0.220** 0.202** 0.234** 0.155* 0.194** 0.232** 0.217** 0.201**

J. Guan et al. / Research Policy 44 (2015) 545–559

SHcity
CCcountry 2.084+ −0.724 −1.055+
BCcountry 0.858 0.754 0.558
SHcountry −6.168** −2.833* −0.811
CCcountry * CCcity −5.551*
CCcountry * BCcity −4.441*
CCcountry * SHcity −2.433**
BCcountry * CCcity 5.119**
BCcountry * BCcity 28.890**
BCcountry * SHcity 1.780**
SHcountry * CCcity 10.332**
SHcountry * BCcity 12.216**
SHcountry * SHcity 6.114**
Constant −0.866** −1.164** −0.734* −0.639* −1.056** −1.100** −1.033** 10.913** 4.528+ 0.712+
Wald chi2 24,187 24,209 24,201 24,261 24,205 24,240 22,970 24,224 24,207 24,132

Notes: N = 13,339. Country fixed effects included. The coefficients of number of collaborations have multiplied by 10,000.
First stage regression:

BCcity = .29967 − .09975 ∗ Density of network − .00014 ∗ Number of cities in the prior network
(−3.89∗∗ ) (−5.71∗∗ )
+.00012 ∗ Pre-sample + .00060 ∗ Firms − .00030 ∗ Inventors+.00014 ∗ Knowledge + .00001 ∗ Number of collaborations
(2.8∗∗ ) (8.77∗∗ ) (−1.16) (1.98∗ ) (5.50∗∗ )

+.00222 ∗ Average path − .01870 ∗ CCcity + .03858 ∗ SHcity + .00609 ∗ CCcountry − .15669 ∗ SHcountry + .04086 ∗ BCcountry
(1.36) (−16.31∗∗ ) (22.75∗∗) (0.40) (−4.39∗∗ ) (2.86∗∗ )

R2 = 0.10, observations = 13,339. T-statistics are in parentheses.

+
p < .1.
*
p < .05.
**
p < .01.

555
556 J. Guan et al. / Research Policy 44 (2015) 545–559

Fig. 6. Interaction of city’s characteristics and country’s characteristics.

this conclusion can also be applied to the city level. At the same
time, the national government should encourage international
collaboration (Freeman, 1991), rather than local protection. The
relative importance of these cities varies with the countries’ loca-
tions in the higher-level network. Thus, these findings give us new
insights that a country’s collaboration innovation can influence the
optimal ways of its cities innovation activities. Specifically, we can
identify what structure or position of a city can help the city to
achieve benefits from the country network.
Secondly, this study has implications for network studies. Net-
work studies have traditionally assumed that similar profits accrue
to those partners who occupy similar positions and examined
actors at only one level (Ahuja, 2000). However, this study shows
that the benefits accruing to cities occupying strategic locations
depend on the country’s position in the network. Even when
cities or regions occupy a similar structural position in the net-
work of countries, their abilities to obtain benefits from that
position varies with the organization of the international collab-
oration network. This may explain the inconsistent results found
in innovation research on network characteristics (Ahuja, 2000;
Burt, 1997; Wang et al., 2014). Furthermore, this study shows a
way of combining hitherto unconnected processes at two levels,
 J. Guan et al. / Research Policy 44 (2015) 545–559
inter-city and inter-country, to generate a much needed macro-
level study. In addition, this study has complemented the research
on relationships between network characteristics and innovation
performance. This study is based on the early research that an
actor’s position in the network will influence its innovativeness
(Gilsing et al., 2008) which has provided crucial and important
insights for our research. We are inspired to consider a more
comprehensive understanding of various network characteristics.
We include three network characteristics, i.e. centrality, structural
holes and clustering coefficient, to show their influence on an
actor’s performance. We further discuss different effects of a city’s
location with its country’s location in the international network.
“Small world effects” as the important feature of complex network
is an important research aspect. Future work could examine the
relationship between the small world effects and innovation, and
its dynamic mechanism involved in the innovation activities in our
future study. What’ more, bidirectional effects between national
and city networks on innovation should be considered in the future
study. The conclusions of this paper also suggest that internal inno-
vation processes such as inter-city collaborations are modified by
external innovation processes. Future research could test how other
factors such as psychological elements (Reinholt et al., 2011), cap-
ital, capability (Almeida et al., 2011), and dynamics play a role in
multilevel networks (Moliterno and Mahony, 2011).
Finally, we utilize a longitudinal study to explore the effects of
multilevel collaboration networks on innovation performance. It is
imperative to discuss the dynamic social structures among differ-
ent levels. One should pay more attention to the macro-level and
the micro-level in future studies. Besides this, network structures
should be considered to be constantly evolving and this evolution-
ary dynamics can influence an actors’ innovation output over time.
More research work on dynamic social structures should be carried
out in the future. Although environmental problems and alterna-
tive energy receive increasing attentions, there are few empirical
researches on innovation networks for alternative energy. We
established collaboration networks in an emerging field, that is,
the alternative energy field, so we are also expanding research into
new frontier areas. The findings of present study should definitely
contribute to the development of alternative energy.
5.2. Practical implications
What we find is interesting and valuable for technology pol-
icy. The conclusion in this paper suggests that the benefits and
advantages generated by the collaboration networks do not only
necessarily depend on the own structural characteristics of the
innovator alone. Rather, the structural characteristics of both the
inter-city and inter-nation networks jointly decide the benefits
brought by collaborations. Government policy makers could use
these insights in collaborating with new partners by not only
concentrating on the collaborations with other cities, but also con-
cerning the collaborative ties of nations.
Due to the inter-city and inter-country collaboration relation-
ships, this research also gives us the important implication that
policy makers should have multiple approaches. First, policies
should help cities, especially the firms located in them, to occupy
a central position in relation to information transmission in an
intercity network and span large structure holes. To obtain better
innovation output, the firms of a city should strengthen its tech-
nical cooperation, and play an intermediary role in information
transmission. Second, a country in a central position promotes the
innovation performance of its cities. National governments should
stimulate the firms in their countries to control the information
flow in international technical collaboration, and strengthen inter-
actions with abroad sources of innovation and their openness to
others (Lee et al., 2010). Through multinational negotiations and
557
support on the key technology, national governments can make
countries in important positions. Meanwhile, local government
policy makers should encourage local firms in their cities to raise
their ability to absorb and transfer imported technology, as well
as take the proper collaborative strategy to span structural holes
and occupy central places. Moreover, energy policy makers should
be interested in the conclusion of the study. Alternative energy
technology policies upon excessive regional protection and even
national protection will block the knowledge flow, and finally
affects the regional innovation performance.
5.3. Limitations
This study has a number of limitations. The sample of this study
comes from the high technology industry; the generalizability of
its conclusions to other types of industry is not shown. To some
extent, the stability of the results may be limited by the self scale
of the city and country. Although we include several size variables
in this article, scaling issues should be taken into account in the
future innovation research (Gao and Guan, 2009). In this article, we
use network characteristics at the country level as moderating vari-
ables, but we cannot completely include all potential moderating
factors. It is necessary to explore other moderating factors, such as
national policy, patent protection, and openness to enrich future
research. Specifically, the relationship between the small world
effects and innovation, and its dynamic mechanism involved in the
innovation activities should be extensively explored in our future
study.
Acknowledgments
This study is supported by a Grant from National Natural Science
Foundation of China (No. 71373254). The authors are very grateful
for the very valuable comments and suggestions from Editor Prof.
Martin Kenney and two anonymous reviewers, which significantly
improved the quality of the paper. The authors wish to thank Pro-
fessor Ronald Rousseau for his very careful English corrections and
suggestions to the paper, which significantly improved the read-
ability of the article.
References
Acemoglu, D., Johnson, S., Robinson, J.A., No. w7771 2000. The Colonial Origins
of Comparative Development: An Empirical Investigation. National Bureau of
Economic Research.
Acs, Z.J., Anselin, L., Varga, A., 2002. Patents and innovation counts as measures of
regional production of new knowledge. Res. Policy 31 (7), 1069–1085.
Adegbesan, J.A., Higgins, M.J., 2011. The intra-alliance division of value created
through collaboration. Strateg. Manag. J. 32 (2), 187–211.
Ahuja, G., 2000. Collaboration networks, structural holes, and innovation: a longi-
tudinal study. Adm. Sci. Q. 45 (3), 425–455.
Ahuja, G., Soda, G., Zaheer, A., 2012. The genesis and dynamics of organizational
networks. Org. Sci. 23 (2), 434–448.
Aiken, L.S., West, S.G., 1991. Multiple Regression: Testing and Interpreting Interac-
tions. Sage, CA, USA.
Almeida, P., 1996. Knowledge sourcing by foreign multinationals: patent citation
analysis in the US semiconductor industry. Strateg. Manag. J. 17 (S2), 155–165.
Almeida, P., Hohberger, J., Parada, P., 2011. Individual scientific collaborations and
firm-level innovation. Ind. Corp. Change 20 (6), 1571–1599.
Audretsch, D.B., Feldman, M.P., 1996. R&D spillovers and the geography of innovation
and production. Am. Econ. Rev. 86 (3), 630–640.
Baba, Y., Shichijo, N., Sedita, S.R., 2009. How do collaborations with universities affect
firms’ innovative performance? The role of Pasteur scientists in the advanced
materials field. Res. Policy 38 (5), 756–764.
Bandura, A., 2001. Social cognitive theory: an agentic perspective. Annu. Rev. Psy-
chol. 52 (1), 1–26.
Batagelj, V., Mrvar, A., 1998. Pajek – program for large network analysis. Connections
21 (2), 47–57.
Becker, M.H., 1970. Sociometric location and innovativeness: reformulation and
extension of the diffusion model. Am. Sociol. Rev. 35 (2), 267–282.
Belussi, F., Sammarra, A., Sedita, S.R., 2010. Learning at the boundaries in an Open
Regional Innovation System: a focus on firms’ innovation strategies in the Emilia
Romagna life science industry. Res. Policy 39 (6), 710–721.
558 J. Guan et al. / Research Policy 44 (2015) 545–559
Bercovitz, J., Feldman, M., 2011. The mechanisms of collaboration in inventive teams:
composition, social networks, and geography. Res. Policy 40 (10), 81–93.
Bettencourt, L., Lobo, J., Strumsky, D., 2007. Invention in the city: increasing returns
to patenting as a scaling function of metropolitan size. Res. Policy 36 (1),
107–120.
Blomkvist, K., Kappen, P., Zander, I., 2014. Superstar inventors—towards a
people-centric perspective on the geography of technological renewal in the
multinational corporation. Res. Policy 43 (4), 669–682.
Blundell, R., Griffith, R., Van Reenen, J., 1995. Dynamic count data models of techno-
logical innovation. Econ. J. 105 (429), 333–344.
Bohnsack, R., Pinkse, J., Kolk, A., 2014. Business models for sustainable technologies:
exploring business model evolution in the case of electric vehicles. Res. Policy
43 (2), 284–300.
Borgatti, S.P., 2005. Centrality and network flow. Soc. Netw. 27 (1), 55–71.
Brass, D.J., 1984. Being in the right place: a structural analysis of individual influence
in an organization. Adm. Sci. Q. 29 (4), 518–539.
Breschi, S., Catalini, C., 2010. Tracing the links between science and technology: an
exploratory analysis of scientists’ and inventors’ networks. Res. Policy 39 (1),
14–26.
Burt, R.S., 1997. The contingent value of social capital. Adm. Sci. Q. 42 (2), 339–365.
Burt, R.S., 1993. The Social Structure of Competition. Explorations in Economic Soci-
ology. Russell Sage Foundation, New York, pp. 65–103.
Burt, R.S., 1992. The Social Structure of Competition. Networks and Organizations:
Structure, Form, and Action. Harvard Business School Press, Boston, MA, pp.
57–91.
Burt, R.S., 2002. The Social Capital of Structural Holes. The New Economic Sociology:
Developments in an Emerging Field. Harvard University Press, New York, pp.
148–190.
Burt, R.S., 2005. Brokerage and Closure: An Introduction to Social Capital: An Intro-
duction to Social Capital. Oxford University Press, New York.
Burt, R.S., 2007. Secondhand brokerage: evidence on the importance of local struc-
ture for managers, bankers, and analysts. Acad. Manag. J. 50 (1), 119–148.
Burt, R.S., 2004. Structural holes and good ideas. Am. J. Sociol. 110 (2), 349–399.
Cameron, A.C., Trivedi, P.K., 2013. Regression Analysis of Count Data. Cambridge
University Press, Cambridge, UK.
Cantner, U., Joel, K., Schmidt, T., 2011. The effects of knowledge management on
innovative success – an empirical analysis of German firms. Res. Policy 40 (10),
1453–1462.
Cantner, U., Rake, B., 2014. International research networks in pharmaceuticals:
structure and dynamics. Res. Policy 43 (2), 333–348.
Cassiman, B., Veugelers, R., 2006. In search of complementarity in innovation strat-
egy: internal R&D and external knowledge acquisition. Manag. Sci. 52 (1), 68–82.
Chatterjee, S., Price, B., 1991. Regression Diagnostics. Sage, New York.
Chen, Z., Guan, J., 2010. The impact of small world on innovation: an empirical study
of 16 countries. J. Informetr. 4 (1), 97–106.
Cooke, P., Gomez Uranga, M., Etxebarria, G., 1997. Regional innovation systems:
institutional and organisational dimensions. Res. Policy 26 (4/5), 475–491.
Criscuolo, P., Verspagen, B., 2008. Does it matter where patent citations come
from? Inventor vs. examiner citations in European patents. Res. Policy 37 (10),
1892–1908.
De Prato, G., Nepelski, D., 2012. Global technological collaboration network:
network analysis of international co-inventions. J. Technol. Transf., 1–18,
http://dx.doi.org/10.1007/s10961-012-9285-4.
Demidenko, E., 2013. Mixed Models: Theory and Applications with R. John Wiley &
Sons, Hoboken, NJ.
Ding, W.W., Levin, S.G., Stephan, P.E., Winkler, A.E., 2010. The impact of informa-
tion technology on academic scientists’ productivity and collaboration patterns.
Manag. Sci. 56 (9), 1439–1461.
Durbin, J., Watson, G.S., 1950. Testing for serial correlation in least squares regres-
sion: I. Biometrika 37 (3/4), 409–428.
Easley, D., Kleinberg, J., 2010. Networks, Crowds and Markets. Cambridge University
Press, New York.
Eisingerich, A.B., Bell, S.J., Tracey, P., 2010. How can clusters sustain performance?
The role of network strength, network openness, and environmental uncer-
tainty. Res. Policy 39 (2), 239–253.
Etzkowitz, H., Leydesdorff, L., 2000. The dynamics of innovation: from National Sys-
tems and Mode 2 to a Triple Helix of university–industry–government relations.
Res. Policy 29 (2), 109–123.
Faems, D., Subramanian, A.M., 2013. R&D manpower and technological perfor-
mance: the impact of demographic and task-related diversity. Res. Policy 42
(9), 1624–1633.
Feldman, M.P., Florida, R., 1994. The geographic sources of innovation: technologi-
cal infrastructure and product innovation in the United States. Ann. Assoc. Am.
Geogr. 84 (2), 210–229.
Ferdous, K., Uddin, M.R., Uddin, M.R., Khan, M.R., Islam, M., 2013. Preparation and
optimization of biodiesel production from mixed feedstock oil. Nature 1 (4),
62–66.
Fleming, L., King, C., Juda, A.I., 2007. Small worlds and regional innovation. Org. Sci.
18 (6), 938–954.
Florida, R.L., 2005. Cities and the Creative Class. Psychology Press.
Freel, M.S., 2003. Sectoral patterns of small firm innovation, networking and prox-
imity. Res. Policy 32 (5), 751–770.
Freeman, C., 1991. Networks of innovators: a synthesis of research issues. Res. Policy
20 (5), 499–514.
Freeman, L.C., 1979. Centrality in social networks conceptual clarification. Soc. Netw.
1 (3), 215–239.
Gao, X., Guan, J., 2009. A scale-independent analysis of the performance of the
Chinese innovation system. J. Informetr. 3 (4), 321–331.
Gargiulo, M., Benassi, M., 2000. Trapped in your own net? Network cohesion, struc-
tural holes, and the adaptation of social capital. Org. Sci. 11 (2), 183–196.
Gay, B., Dousset, B., 2005. Innovation and network structural dynamics: study of the
alliance network of a major sector of the biotechnology industry. Res. Policy 34
(10), 1457–1475.
Gilsing, V., Nooteboom, B., Vanhaverbeke, W., Duysters, G., van den Oord, A., 2008.
Network embeddedness and the exploration of novel technologies: technologi-
cal distance, betweenness centrality and density. Res. Policy 37 (10), 1717–1731.
Giuri, P., Mariani, M., Brusoni, S., Crespi, G., Francoz, D., Gambardella, A., Garcia-
Fontes, W., Geuna, A., Gonzales, R., Harhoff, D., 2007. Inventors and invention
processes in Europe: results from the PatVal-EU survey. Res. Policy 36 (8),
1107–1127.
Gonzalez-Brambila, C.N., Veloso, F.M., Krackhardt, D., 2013. The impact of network
embeddedness on research output. Res. Policy 42 (9), 1555–1567.
Gould, D.M., Gruben, W.C., 1996. The role of intellectual property rights in economic
growth. J. Dev. Econ. 48 (2), 323–350.
Granovetter, M., 1985. Economic action and social structure: the problem of embed-
dedness. Am. J. Sociol. 91 (3), 481–510.
Granovetter, M., 1995. Getting a Job: A Study of Contacts and Careers. University of
Chicago Press, Chicago, IL.
Granovetter, M.S., 1973. The strength of weak ties. Am. J. Sociol. 78 (6), 1360–1380.
Grewal, R., Lilien, G.L., Mallapragada, G., 2006. Location, location, location: how net-
work embeddedness affects project success in open source systems. Manag. Sci.
52 (7), 1043–1056.
Guan, J., Zhao, Q., 2013. The impact of university–industry collaboration networks
on innovation in nanobiopharmaceuticals. Technol. Forecast. Soc. Change 80 (7),
1271–1286.
Gulati, R., 2007. Managing Network Resources: Alliances, Affiliations and other Rela-
tional Assets. Oxford University Press Oxford, New York.
Gulati, R., Sytch, M., Tatarynowicz, A., 2012. The rise and fall of small worlds: explor-
ing the dynamics of social structure. Org. Sci. 23 (2), 449–471.
Gupta, A.K., Tesluk, P.E., Taylor, M.S., 2007. Innovation at and across multiple levels
of analysis. Org. Sci. 18 (6), 885–897.
Hausman, J.A., Hall, B.H., Griliches, Z., 1984. Econometric Models for Count Data with
an Application to the Patents-R&D Relationship. National Bureau of Economic
Research Cambridge, MA, USA.
Herstad, S.J., Aslesen, H.W., Ebersberger, B., 2014. On industrial knowledge bases,
commercial opportunities and global innovation network linkages. Res. Policy
43 (3), 495–504.
Hill, R.C., Griffiths, W.E., Lim, G.C., 2008. Principles of Econometrics. Wiley, Hoboken,
NJ/Chicago, IL.
Hölzl, W., Janger, J., 2014. Distance to the frontier and the perception of innovation
barriers across European countries. Res. Policy 43 (4), 707–725.
Hrebiniak, L.G., Alutto, J.A., 1972. Personal and role-related factors in the develop-
ment of organizational commitment. Adm. Sci. Q. 17 (4), 555–573.
Hu, A.G., 2010. Propensity to patent, competition and China’s foreign patenting
surge. Res. Policy 39 (7), 985–993.
Huang, K., 2010. China’s innovation landscape. Science 329 (5992), 632–633.
King, D.A., 2004. The scientific impact of nations. Nature 430 (6997), 311–316.
Koka, Balaji R., Prescott, John E., 2002. Strategic alliances as social capital: a multi-
dimensional view. Strateg. Manag. J. 23 (9), 795–816.
Lahiri, N., Narayanan, S., 2013. Vertical integration, innovation, and alliance portfolio
size: implications for firm performance. Strateg. Manag. J. 34 (9), 1042–1064.
Lee, J., 2010. Heterogeneity, brokerage, and innovative performance: endogenous
formation of collaborative inventor networks. Org. Sci. 21 (4), 804.
Lee, J.-Y., Bachrach, D.G., Lewis, K., 2014. Social network ties, transactive memory,
and performance in groups. Org. Sci. 25 (3), 951–967.
Lee, S., Park, G., Yoon, B., Park, J., 2010. Open innovation in SMEs—an intermediated
network model. Res. Policy 39 (2), 290–300.
Li, E.Y., Liao, C.H., Yen, H.R., 2013. Co-authorship networks and research impact: a
social capital perspective. Res. Policy 42 (9), 1515–1530.
Marquardt, M.J., 2013. The Global Advantage. Routledge, TX, USA.
Mason, W., Watts, D.J., 2012. Collaborative learning in networks. Proc. Natl. Acad.
Sci. 109 (3), 764–769.
Melville, N.P., 2010. Information systems innovation for environmental sustainabil-
ity. MIS Q. 34 (1), 1–21.
Miaou, S., Hu, P., Wright, T., Davis, S., Rathi, A., 1991. Development of relation-
ships between truck accidents and highway geometric design. Phase I. Technical
memorandum prepared by the Oak Ridge National Laboratory. Federal Highway
Administration, Washington, DC.
Moliterno, T.P., Mahony, D.M., 2011. Network theory of organization: a multilevel
approach. J. Manag. 37 (2), 443–467.
Moss, R.H., Edmonds, J.A., Hibbard, K.A., Manning, M.R., Rose, S.K., van Vuuren, D.P.,
Carter, T.R., Emori, S., Kainuma, M., Kram, T., 2010. The next generation of sce-
narios for climate change research and assessment. Nature 463 (7282), 747–756.
Mowery, D.C., Nelson, R.R., Martin, B.R., 2010. Technology policy and global warm-
ing: why new policy models are needed (or why putting new wine in old bottles
won’t work). Res. Policy 39 (8), 1011–1023.
Neal, Z., 2011. Differentiating centrality and power in the world city network. Urban
Stud. 48 (13), 2733–2748.
Nooteboom, B., 1999. Innovation and inter-firm linkages: new implications for pol-
icy. Res. Policy 28 (8), 793–805.
Obstfeld, D., 2005. Social networks, the tertius iungens orientation, and involvement
in innovation. Adm. Sci. Q. 50 (1), 100–130.
 J. Guan et al. / Research Policy 44 (2015) 545–559
Pan, R.K., Kaski, K., Fortunato, S., 2012. World citation and collaboration networks:
uncovering the role of geography in science. Sci. Rep. 2 (902.), 2045–2322.
Paruchuri, S., 2010. Intraorganizational networks, interorganizational networks, and
the impact of central inventors: a longitudinal study of pharmaceutical firms.
Org. Sci. 21 (1), 63–80.
Paschen, J.-A., Ison, R., 2014. Narrative research in climate change
adaptation—exploring a complementary paradigm for research and governance.
Res. Policy 43 (6), 1083–1092.
Perry-Smith, J.E., 2006. Social yet creative: the role of social relationships in facili-
tating individual creativity. Acad. Manag. J. 49 (1), 85–101.
Peteraf, M.A., 1993. The cornerstones of competitive advantage: a resource-based
view. Strateg. Manag. J. 14 (3), 179–191.
Peterman, A., Kourula, A., Levitt, R., 2014. Balancing act: government roles in an
energy conservation network. Res. Policy 43 (6), 1067–1082.
Phelps, C., Heidl, R., Wadhwa, A., 2012. Knowledge, networks, and knowledge net-
works: a review and research agenda. J. Manag. 38 (4), 1115–1166.
Phelps, C.C., 2010. A longitudinal study of the influence of alliance network struc-
ture and composition on firm exploratory innovation. Acad. Manag. J. 53 (4),
890–913.
Philibert, C., 2004. International energy technology collaboration and climate change
mitigation. OECD/IEA Information Paper, Paris.
Poot, T., Faems, D., Vanhaverbeke, W., 2009. Toward a dynamic perspective on
open innovation: a longitudinal assessment of the adoption of internal and
external innovation strategies in the Netherlands. Int. J. Innov. Manag. 13 (2),
177–200.
Ravasz, E., Somera, A.L., Mongru, D.A., Oltvai, Z.N., Barabási, A.-L., 2002. Hierar-
chical organization of modularity in metabolic networks. Science 297 (5586),
1551–1555.
Reinholt, M., Pedersen, T., Foss, N.J., 2011. Why a central network position isn’t
enough: the role of motivation and ability for knowledge sharing in employee
networks. Acad. Manag. J. 54 (6), 1277–1297.
Rost, K., 2011. The strength of strong ties in the creation of innovation. Res. Policy
40 (4), 588–604.
Rothaermel, F.T., Hess, A.M., 2007. Building dynamic capabilities: innovation driven
by individual-, firm-, and network-level effects. Org. Sci. 18 (6), 898–921.
Rowley, T.J., 1997. Moving beyond dyadic ties: a network theory of stakeholder
influences. Acad. Manag. Rev. 22 (4), 887–910.
Salman, N., Saives, A.L., 2005. Indirect networks: an intangible resource for biotech-
nology innovation. R&D Manag. 35 (2), 203–215.
Scellato, G., Franzoni, C., Stephan, P., 2014. Migrant scientists and international net-
works. Res. Policy 43 (1), 1–20.
Scellato, S., Noulas, A., Lambiotte, R., Mascolo, C., 2011. Socio-spatial properties of
online location-based social networks. ICWSM 11, 329–336.
Schilling, M.A., Phelps, C.C., 2007. Interfirm collaboration networks: the impact
of large-scale network structure on firm innovation. Manag. Sci. 53 (7),
1113–1126.
Schwab, A., 2007. Incremental organizational learning from multilevel information
sources: evidence for cross-level interactions. Org. Sci. 18 (2), 233–251.
Scott, J., Carrington, P.J., 2011. The SAGE Handbook of Social Network Analysis. SAGE
Publications, London.
559
Shepherd, J.G., 2009. Geoengineering the Climate: Science, Governance and Uncer-
tainty. Royal Society, London.
Shipilov, A.V., 2009. Firm scope experience, historic multimarket contact with part-
ners, centrality, and the relationship between structural holes and performance.
Org. Sci. 20 (1), 85–106.
Shipilov, A.V., Li, S.X., 2008. Can you have your cake and eat it too? Structural holes’
influence on status accumulation and market performance in collaborative net-
works. Adm. Sci. Q. 53 (1), 73–108.
Singh, J., Marx, M., 2013. Geographic constraints on knowledge spillovers: political
borders vs. spatial proximity. Manag. Sci. 59 (9), 2056–2078.
Soda, G., 2011. The management of firms’ alliance network positioning: implications
for innovation. Eur. Manag. J. 29 (5), 377–388.
Song, Y., Knaap, G.-J., 2004. Measuring the effects of mixed land uses on housing
values. Reg. Sci. Urban Econ. 34 (6), 663–680.
Sotarauta, M., Kautonen, M., 2007. Co-evolution of the Finnish national and local
innovation and science arenas: towards a dynamic understanding of multi-level
governance. Reg. Stud. 41 (8), 1085–1098.
Sytch, M., Tatarynowicz, A., Gulati, R., 2012. Toward a theory of extended contact:
the incentives and opportunities for bridging across network communities. Org.
Sci. 23 (6), 1658–1681.
Team, S., 2009. Science of Science (Sci2) Tool. Indiana University and SciTech Strate-
gies.
Tsai, W., 2001. Knowledge transfer in intraorganizational networks: effects of
network position and absorptive capacity on business unit innovation and per-
formance. Acad. Manag. J. 44 (5), 996–1004.
Tsai, W., Ghoshal, S., 1998. Social capital and value creation: the role of intrafirm
networks. Acad. Manag. J. 41 (4), 464–476.
Valente, T.W., 2012. Network interventions. Science 337 (6090), 49–53.
Wagner, C.S., Leydesdorff, L., 2005. Network structure, self-organization, and the
growth of international collaboration in science. Res. Policy 34 (10), 1608–1618.
Wang, C., Rodan, S., Fruin, M., Xu, X., 2014. Knowledge networks, collaboration
networks, and exploratory innovation. Acad. Manag. J. 57 (2), 484–514.
Wapner, P., 1997. Governance in Global Civil Society. Global Governance: Drawing
Insights from the Environmental Experience. MIT Press, Cambridge, pp. 65–84.
Watts, D.J., Strogatz, S.H., 1998. Collective dynamics of ‘small-world’networks.
Nature 393 (6684), 440–442.
Wooldridge, J.M., 2012. Introductory Econometrics: A Modern Approach. Cengage
Learning, Mason, OH.
Yayavaram, S., Ahuja, G., 2008. Decomposability in knowledge structures and its
impact on the usefulness of inventions and knowledge-base malleability. Adm.
Sci. Q. 53 (2), 333–362.
Ynalvez, M.A., Shrum, W.M., 2011. Professional networks, scientific collaboration,
and publication productivity in resource-constrained research institutions in a
developing country. Res. Policy 40 (2), 204–216.
Zaheer, A., Bell, G.G., 2005. Benefiting from network position: firm capabilities, struc-
tural holes, and performance. Strateg. Manag. J. 26 (9), 809–825.
Zaheer, A., Soda, G., 2009. Network evolution: the origins of structural holes. Adm.
Sci. Q. 54 (1), 1–31.