Journal of Informetrics 10 (2016) 685–692
Contents lists available at ScienceDirect
Journal of Informetrics
journal homepage: www.elsevier.com/locate/joi
A longitudinal analysis of link formation on collaboration
networks
Alireza Abbasi
School of Engineering and IT, The University of New South Wales (UNSW Australia), Canberra, Australia
a r t i c l e
i n f o
Article history:
Received 3 April 2016
Accepted 5 May 2016
Keywords:
Network dynamics
Link formation
Drivers of network change
Attachment behavior
Evolutionary analysis
Social network analysis
a b s t r a c t
Understanding the structural change and evolution of networks for predicting their dynamics is one of the fundamental problems in network related studies. In order to uncover the
dynamic structural patterns of a network over time, it is vital to investigate the ways nodes
behave at a local level. So, it is important to know the reasons why nodes stop a relationship
or select a new partner, compared to other alternatives, for establishing a link. This study
aims to understand the processes of network evolution by quantitatively examining the
attachment behaviors of nodes in a real collaboration network by identifying the characteristics of the existing nodes which can impact on their link formation process. To do so,
different link formation or attachment processes such as cohesiveness, cumulative advantage, assortative mixing, and structural position are examined. The results indicate that
structural position, the tendency to connect to the nodes in a strategic intermediating position in the network, is the most effective processes that expose the attachment behavior of
nodes during the evolution of a collaboration network. Understanding these effective processes can help to predict more precisely how the nodes’ local structure and consequently
the overall network structural change over time. This could support researchers, decision
makers or practitioners to manage the nodes (agents) in their social or technical networks
(systems) for reaching their organizational goals.
© 2016 Elsevier Ltd. All rights reserved.
1. Introduction
Our life is surrounded by a wide range of complex systems including natural networks (e.g., human brain, proteins,
and ant colonies), socio-economical networks (e.g., corporations’ organizational structure, world trade union) and sociotechnical networks (e.g., the World Wide Web, the Internet, and Facebook). The ubiquity of these networks has led to the
development of the interdisciplinary field of ‘network science’ covering topics such as ‘graph theory’ in mathematics and
computer science; ‘social network analysis’ in sociology, anthropology, business; and the analysis of complex networks in
statistical physics and biology. The concept of network is the same through all related fields: a network forms when at
least two nodes (also called actors in social networks in which they are often humans or organizations) often of a similar
type (e.g., humans, computers, organizations) connect, or can be considered virtually connected, to each other because of
commonalities (natural, social, and technical relationships) and/or shared goals.
Network structure, the way all nodes in a network are connected, is the main characteristics of networks and has been
studied vastly by researchers in different fields. Different forms of network structures and nodes’ positions may suggest either
E-mail address: a.abbasi@unsw.edu.au
http://dx.doi.org/10.1016/j.joi.2016.05.001
1751-1577/© 2016 Elsevier Ltd. All rights reserved.
686
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
benefits or restrictions for the nodes embedded in the network (Ahuja, Soda, & Zaheer, 2012; Burt, 1992). For instance, the
structural positions of nodes in their collaboration networks have shown to have positive effect on their productivity and
citation-based performance (Abbasi, Altmann, & Hossain, 2011; Abbasi, Wigand, & Hossain, 2014).
Understanding the structural positions of nodes in a network has been instrumental in driving a growing interest in the
study of networks especially from an evolutionary perspective, to consider not only the status of a network at a specific
point in time (static) but also during its evolution (dynamic) over time. Identifying how the networks change and evolve
and determining the drivers of these changes are fundamental issues which are not well understood yet (Ahuja et al., 2012).
Several studies such as (Glückler, 2007; Powell, White, Koput, & Owen-Smith, 2005) identified the following processes or
rules of attachment for nodes to explain the dynamics of networks during their evolution over time: Cumulative advantage,
the tendency of nodes to form links with highly connected nodes; Assortative mixing, the tendency of node to form links
to node with similar attributes such as connectivity rate; and Cohesivness, the tendency of node to form links with others
with past histories of connection. Another important link formation process which is often not emphasized properly in the
literature is the structural position of the nodes in their networks which is the main focus of this study. The structural position
of a node in a network reflects how it is connected to others. It is regarded as an attribute of the node reflecting its role and
influence on the overall network structure. The structural position of nodes in a network can be measured by how connected
and/or close on average to others they are and to what extend they intermediate by linking disconnected nodes (Freeman,
1979; Scott, 1991).
This research aims to determine the factors that affect the attachment behavior of nodes in a network. This will help to
identify nodes which make a network more resilient to the potential changes. To achieve this goal, it is pivotal to discover the
attachment or link formation processes of the nodes in a network over time. This can be accomplished through studying the
characteristics of the existing nodes that better attract new partners to investigate the potential effect of these characteristics
on the dynamics of the network. Nodes with strategic structural position, such as intermediating position measured by
high betweenness centrality, attract more partners for the future interactions, i.e., establishing new links (Abbasi, Hossain,
& Leydesdorff, 2012). Recognizing such strategic positions is beneficial for organizations to identify and invest on such
actors for information dissemination and viral marketing campaigns to help selling their products faster. Also, controlling
(removing or vaccinating) the actors in a disease outbreak network can improve the community’s resilience toward the
disease. Therefore, determining the strategic positions of the nodes has implication for decision makers and managers to
control and manage the evolution of networks.
The remainder of this paper reviews the literature on link formation processes during the evolution of networks in Section
2. The data sources and collection methods are described in Section 3. Section 4 provides the results of analysis discussing
the findings, and highlighting the implications, future directions and limitations of this study.
2. Link formation processes in social networks
Nodes in a network expend economic or human capital in order to build their social capital (i.e., their connections to
other nodes in the network) from which they hope to profit afterwards (Abbasi et al., 2014; Burt, 1992). However, in social
networks often the selection of a partner depends on both the mutual interests and decisions of both actors and also the
external environment (Glückler, 2007). For instance, managing interdependencies between firms (as actors) and gaining
access to resources are important external factors affecting firms’ alliance (link) formation. Or in a co-authorship network, a
link may form as a result of the deliberate or forced choice of authors or the external issues such as the power relationship (e.g.,
student-supervisor or researcher-sponsor) that requires including supervisors or sponsors as authors of a paper regardless
of their actual academic contribution. Understanding the link formation processes helps to uncover the structural changes
and dynamics of networks. The following processes have been recognized in different studies as the primary factors for node
to form links with other nodes:
2.1. Cohesiveness
Knowing and trusting a partner is an important motivation for establishing new relationships among people in their
social life. Repeated connections among people often build trust among them (Burt, 1992). The more interactions two
parties have, the better they will know each other and reinforce the trust among them. Trust is often one of the best reasons
for interaction among actors in social networks including financial interactions. For instance, Gulati (1995), in analyzing the
alliance formation among firms, found that the more frequent past alliances between two firms leads to more new alliances
between them. This is an important enabler not only for establishing new links but also facilitates existing connections.
Glückler (2007) explained this as an ‘embedding’ process and showed that “future ties form around strong ties by processes of
trust and indirect referrals”. Later Rosenkopf and Padula (2008), examining almost similar process, claimed that ‘cohesiveness’
(i.e., the history of connection between two actors) increases the likelihood of forming new ties with each other in the future.
Exploring the effect of existing connections among firms on their future alliances, it has been shown that actors’ positions
in the pre-existing network structure affect the formation of new ties (Gulati, 1995; Podolny, 1994; Walker, Kogut, & Shan,
1997). This process and the cumulative advantage process (see below) are based on a general mechanism through which a
relatively favorable position becomes a resource to generate additional gains (DiPrete & Eirich, 2006). Thus, it is hypothesized
that:
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
H1.
687
Scholars tend to form new links with their existing partners (with history of past connections).
2.2. Cumulative advantage
The cumulative advantage process, also known as ‘the Matthew effect’, is based on ‘the rich get richer’ principle originally
introduced by Merton (1968) and defined as: “the accruing of greater increments of recognition for particular scientific
contributions to scientists of considerable repute and the withholding of such recognition from scientists who have not yet
made their mark” (p. 57). Cumulative advantage process is regarded as a reason for inequality across any time-based process,
as the advantage gained by a person (or group of people) over another person (or group) accumulates over time (DiPrete &
Eirich, 2006). In other words, early success leads to greater gains through accumulating advantages by early starters (Powell
et al., 2005) and creates inequality compared to the late comers.
Since the prestigious or high status actors in a community are expected to gain more advantages over time, new actors
often tend to connect to such high status actors hoping to gain benefit through the association. For instance, new movie
actors accompany famous celebrities at the beginning of their career to gain benefits or new researchers often prefer to have
collaborations with well-established active scholars so as their joint publication receives more attention.
One of the main findings of the literature examining network embeddedness is that the high number of existing social
connections of the actors provide them an ability to gather information about a wider set of potential collaborations (Gulati,
1995). Thus, the actors with high status, resulting from their connectedness, are more likely to form additional connection
(Podolny, 1994). Therefore, it can be hypothesized that:
H2.
Scholars tend to establish new links with highly connected actors.
2.3. Assortative mixing
Newman (2002), analyzing a number of different networks, revealed that well-connected actors in social networks prefer
to attach to other well-connected actors. He referred to this phenomenon as assortative mixing. Assortative mixing process
can be considered as a specific form of homophily which is based on ‘birds of a feather flock together’ principle or ‘similarity
breeds connection’ (McPherson, Smith-Lovin, & Cook, 2001). Newman (2002) claimed that assortative networks, in which
nodes follow the assortative mixing phenomena by connecting to other nodes with similar connectivity, are more robust to
removal of their highest degree nodes as there are always other high degree nodes to replace them.
Assortative mixing or ‘structural homophily’ describes actors similarly embedded within the network (Ahuja, Polidoro,
& Mitchell, 2009). Although the early researchers suggested that other dimensions of homophily (e.g., similarity of resource
profiles) smooth collaboration, the network perspective highlights that nodes with similar network positions have a higher
chance of forming associations (Ahuja et al., 2009). So, assortative mixing is a form of homophily which considers only
the nodes’ similar number of connections in a network as an important reason for forming links among the nodes. Studies
empirically investigating this phenomenon found that highly connected organizations are likely to form alliances with other
highly connected organizations to ‘mitigate collaboration hazards’ (Gulati & Gargiulo, 1999). One can expect the same process
in collaboration networks. So, it is hypothesized that:
H3.
Scholars tend to form new links with other actors who have similar number of links.
2.4. Structural position
The way a node is connected, directly or indirectly, to other nodes in a network reflects the node’s structural position.
This manifests how the nodes can affect the flow of resources on the network. Historically, the global structural positions of a
node have been measured by Freeman (1979) based on short distance on average to other nodes (closeness) and the number
of times it is located on the shortest path along every pair of nodes (betweenness). The structural positions of nodes on the
collaboration networks have been shown to have positive effect on their performance (Abbasi et al., 2011) and, possibly
consequently, the intention of other nodes to collaborate with them (Abbasi et al., 2012).
Being close to other nodes in a network gives a node the benefit of being able to efficiently spread resources across the
network because of its proximity. This means that for example the actors in a social network can access faster to all other
members of the community if required. The intermediating or gatekeeping position of a node is also significant to manage
the flow of the resources through it especially if it is the only node bridging the lack of connection between other nodes
(structural holes). The nodes in this strategic position are playing a gatekeeping role to control the dissemination of resource
over the network which passes through them. Therefore, it is expected that nodes prefer to establish new links to nodes in
such strategic structural positions. So, it can be hypothesized that:
H4.
Scholars tend to form new links with actors who are close to others in the network.
H5.
Scholars tend to form new links with actors who are intermediators in the network.
688
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
Table 1
Authors and their co-authorship links statistics.
Cumulative numbers
Year
# of publications
# of authors
# of links
Avg. link/au.
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
367
338
380
437
358
615
568
583
428
481
965
798
854
1210
952
1554
1528
1618
1259
1340
1732
1329
978
2671
1980
2527
3540
3188
3300
2783
1.79
1.67
1.15
2.21
2.08
1.63
2.32
1.97
2.62
2.08
2001
2001–02
2001–03
2001–04
2001–05
2001–06
2001–07
2001–08
2001–09
2001–10
# of publications
# of authors
# of links
367
705
1085
1522
1880
2495
3063
3646
4074
4555
965
1732
2552
3714
4602
6072
7498
9003
10,130
11,343
1732
3061
4039
6710
8690
11,217
14,757
17,945
21,245
24,028
3. Methodology and data
A (social) network can be represented as a graph made of actors (e.g., individuals, organizations) or nodes (cells, web
pages) tied by one or more specific types of relations such as financial exchange, friendship, trade, and Web links (Wasserman
& Faust, 1994). Network information, nodes and links among them, should be transferred into a matrix format to be able to
quantify nodes and networks structure and apply mathematical and statistical analysis. Social network analysis (SNA) is a
set of very different methodologies for quantifying and visualizing the relationships in a network. In this study, SNA metrics
(e.g., centrality measures) are used in order to analyze the position of nodes (authors) in their collaboration networks.
In a co-authorship network, as a form of bibliometric network and academic collaboration network, the nodes are the
authors which a link between each pair represents the existence of at least one co-authored (or joint) publication. The
weights of links denote the number of publications that the two authors (co-authors) have jointly published.
A longitudinal network data set is created for empirically testing the hypotheses. Scopus, one of the main widely used and
authorized sources which provide bibliometric data, is used to retrieve meta-data of publications to form a co-authorship
network. The clause “information science” is searched on the titles, keywords, or abstracts of journal publications written in
English published between 2001 and 2010 inclusive. Indeed; the retrieved publications cannot be considered as representing
the world production in the field of information science but it illustrates a good portion of publications in this field which are
not limited to specific sub-fields; journals; institutions or countries.
After extracting the publications meta-data (e.g., title, publication date, journal/conference name, authors’ names and
affiliations including country and institution names) an application program (Abbasi & Altmann, 2011) is used to extract
the relationships between author (e.g., co-authorships) and to store the data in relational tables in its local database. It
also provides a function to find and merge authors with similar names and affiliations in order to avoid the authors’ name
disambiguation in addition to the existing cleaning techniques applied by Scopus. After cleansing of the publication data,
the resulting database contains 4555 publications reflecting the contributions of 11,343 authors with 24,028 co-authorship
links from 3196 different institutions (i.e., universities and private companies) across 99 countries.
4. Analysis and results
Table 1 illustrates the growth of the collaboration network indicating the numbers of authors and number of links among
them (considering the links weight for repeated links) during the temporal evolution of the network between 2001 and
2010 and also provides the cumulative numbers over time in the right hand side. As shown, the number of authors and
links increases rapidly but the average number of links per author rises smoothly, with fluctuations, from 1.92 (in 2001)
to 2.44 (in 2010). In order to study the evolution of the network, the cumulative numbers are used throughout the rest of
this research. For example, when referring to the statistics for 2005 academic collaboration network, it actually considers
cumulative numbers of the authors and their links (based on the papers published) between 2001 and 2005 inclusive.
During the evolution of a collaboration network, an attachment or a new link may happen: (1) between a new actor
(added in the following period) and an existing actor; (2) among two new actors; (3) among two existing actors who were
not connected before; and (4) among two existing actors who already had at least one connection (Abbasi et al., 2012). To
study the attachment processes, the mixture of the aforementioned options should be considered. For instance, to assess
the cumulative advantage process the new attachments among the existing actors and new actors, options 1 and 3, and for
the cohesiveness process, the new links among the existing connected actors, option 4, should be examined. The following
subsections show the result of testing our five hypotheses.
4.1. Cohesiveness
Assessing the cohesiveness process, the effect of the existing connections on the formation of new links is evaluated.
Table 2 shows the statistics about the number of existing links and the number of new links in the following year for the
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
689
Table 2
Authors’ link formation statistics over time.
Year
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
# of links
1732
3061
4039
6710
8690
11,217
14,757
17,945
21,245
24,028
# of new links
–
1329
978
2671
1980
2527
3540
3188
3300
2783
# of new attachments
Among NEW
authors
Between NEW &
existing authors
Among existing
disconnected
authors
Among existing
connected
authors
–
1281 (97%)
928 (95%)
2533 (95%)
1827 (92%)
2387 (94%)
3319 (94%)
2896 (91%)
2888 (88%)
2407 (86%)
–
19 (1%)
42 (4%)
109 (4%)
98 (5%)
124 (5%)
187 (5%)
231 (7%)
280 (8%)
299 (11%)
–
0 (0%)
0 (0%)
0 (0%)
3 (0%)
3 (0%)
1 (0%)
6 (0%)
4 (0%)
12 (0%)
–
29 (2%)
8 (1%)
29 (1%)
52 (3%)
13 (1%)
33 (1%)
55 (2%)
128 (4%)
65 (2%)
Table 3
Spearman’s rank correlations between authors’ degree and their number of new links in the next year.
Year
# of authors (n)
2001
2002
2003
2004
2005
2006
2007
2008
2009
847
1501
2219
3279
4079
5388
6699
8080
9126
a
Spearman’s rho (coefficient)
.038
− .040
.047
.020
− .008
.016
.022
.038a
.033a
Sig. (2-tailed)
.272
.118
.027
.245
.612
.237
.076
.001
.002
Correlation is significant at the 0.01 level (2-tailed).
period between 2001 and 2010. Furthermore, to assess the frequency of different types of links among the new and old
actors, the numbers of new attachments are classified as follows: (i) among new authors; (ii) between a new author and an
existing author; (iii) among existing authors who had no collaboration before; and (iv) among already connected authors (coauthors). As the results indicate, most of the new links form among the new actors and very few of the disconnected existing
actors attach to each other while repeated collaborations among the existing connected actors are relatively considerable.
Studying the attachment behavior of the existing actors from previous years shows that almost none of them forms a link
with other existing ones who are not connected to them. But the results indicate that very few of the existing actors make
new connections to their current co-authors and also new researchers (e.g., colleagues, research associates or students).
The moderately higher number of future attachments between the existing authors and the new authors compared to the
already connected authors reveals that the trust resulting from the history of connection among actors in the collaboration
network does not encourage (or affect) the existing scholars to form new links with the existing ones while otherwise is
expected. In other words, the trust built among scholars does not influence on their attachment behavior and therefore the
cohesive process is not a driver of actors’ attachment during the evolution of this collaboration networks over time.
4.2. Cumulative advantage
In order to assess the cumulative advantage process in the collaboration network, first the existing authors’ connectedness,
i.e., their number of links or degree centrality, at time t (e.g., 2002) and the number of new links to them at time t + 1
(e.g., 2003) are calculated. Then, the association between the connectedness of existing authors and their number of new
attachments in the following year is examined, using Spearman correlation rank test. As Table 3 shows only in 2008 and
2009 the correlation between the authors’ connectedness and their new attachments in the following years (2009 and 2010)
are statistically significant. Overall, there is no significant statistical association between the scholars’ current number of
connections and their future attachments in this dataset. This exposes that the authors with more extensive accounts of
links do not form more future links compared to others with low connectivity. In other words, our finding does not support
the cumulative advantage process that nodes with higher degree will attract more nodes in future (Barabási & Albert, 1999).
4.3. Assortative mixing
The assortativity coefficient is the Pearson correlation coefficient of degree between pairs of linked nodes (Newman,
2002). Positive values of the coefficients of the correlation test indicate a correlation between nodes of similar degree
(assortativity), while negative values indicate relationships between nodes of different degree (disassortativity). Therefore,
to examine the assortativity of the authors in their collaboration networks (authors with high (low) degree attempt to make
690
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
Table 4
Joint degree distribution matrix of academic collaboration network at 2003.
Degree
0
1
2
3
4
5
8
0
1
2
3
4
5
8
Sum of links
928
15
2
12
1
8
4
970
15
2
2
12
1
8
4
1
1
9
5
1
17
2
17
5
Table 5
Academic collaboration network’ Pearson degree correlation test results.
Pearson degree correlation
2002
2003
2004
2005
2006
2007
2008
2009
2010
r
p-value
n
.90a
.000
48
−.34
.017
50
−.24a
.004
138
.39a
.000
150
−.29a
.000
140
−.13
.041
221
−.04
.468
292
.50a
.000
412
−.11
.027
376
a
Correlation is significant at the 0.01 level (2-tailed).
Table 6
Correlation coefficients between authors’ centrality measures and their attachment frequency in each stage.
Structural positions factors
Closeness
Intermediating
*
**
n
P
P
Correlation coefficient with new links’ count in the next year
2001
2002
2003
2004
2005
2006
2007
2008
2009
847
0.032
.183*
1501
−0.029
.067*
2219
.043*
.099**
3279
0.018
.038*
4079
−0.008
.118*
5388
0.018
.131**
6699
.032**
.090**
8080
.035**
.057**
9126
.031**
.143**
Correlation is significant at the 0.05 level (2-tailed).
Correlation is significant at the 0.01 level (2-tailed).
links to other authors with high (low) connectivity), the degree in the previous timestamps has been considered. In other
words, for each link in t + 1 (e.g., 2002), the degree of both authors in t (e.g., 2001) have been used to calculate the Pearson
correlation coefficients at that year (e.g., 2002).
As an example, Table 4 demonstrates the joint degree distribution matrix for 2003. A value at cell (x, y) shows the number
of links between authors who have had degree centrality of x and y until 2002. As shown, the majority of links (928) are
among new authors (degree = 0 in 2002) and then between new authors and the existing authors (42) and there are only 8
links between authors with similar connectivity: five links between authors with degree 3 and two links between authors
with degree 1 and one link between an author with degree 5 and another with degree 8.
The Pearson degree correlations between pairs of co-authorships at each year are calculated and shown in Table 5. The
links which both authors have been new at that year (degree = 0 in the previous year) have been discarded in the analysis of
Pearson degree correlation. Although the results reveal significant relationships between pairs of links’ degrees for five years,
as shown only for three years (i.e., 2002, 2005 and 2009) the coefficients are positive (i.e., the collaboration networks are
assortative) reflecting the authors form links to the other authors with similar connectivity and other years show a negative
relationships indicating connectivity of authors with different degrees. Therefore, the contradictory results do not support
this process playing a role on the attachment behavior of the authors during the evolution of their collaboration networks
over time.
4.4. Structural positions
In order to investigate the effects of authors’ structural position on their attachment preferences, positional characteristics
(e.g., the closeness and intermediating roles) of the existing authors at time t (e.g., 2001) and the frequency of their new
links at time t + 1 (e.g., 2002) is measured. Using Spearman correlation rank test, the association between the structural
positions of the existing authors (enumerated through closeness and betweenness centrality measures) and their future
connectivity in the following year between 2001 and 2010 is examined. Table 6 demonstrates the correlation coefficients
and the significance of the association between the existing authors’ structural position factors and their future link formation
frequency. As an example, it is shown that the scholars’ closeness in 2001 is not significantly correlated to their number of
links in the following year while their intermediating structural position shows a positive and significant correlation with
the number of new links in 2002.
The results show that the correlation coefficients between intermediators, the actors with high betweenness centrality
values, and their future attachment frequency in the following year during the collaboration network evolution between
2001 and 2010 are always significant and much higher than the association of the authors’ closeness structural position. It
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
691
implies that the authors with intermediating position are more likely to have new links in the future due to the prestigious
role gained as a result of the strategic position in the network.
5. Conclusion and discussion
Complex networks surround our human life by a wide range of networks in nature and society. Many research problems
on different fields can be formulated and studied through a network approach (Wasserman and Faust, 1994; Watts, 2004).
One of the fundamental problems in network related studies is to understand the change and evolution of networks to
predict the dynamic of networks which requires accurate investigation of the dynamic structural patterns of networks over
time. However, considering the ways the actors behave at a local level helps to expose the global structural patterns of the
network (Panzarasa, Opsahl, & Carley, 2009) such as the reasons for and the ways to select a partner among other alternatives
to form a link. Therefore, one way to explain this research problem is to study the processes that affect the formation of
new links among the actors due to their specific characteristics or roles in a network. This research has been used a network
perspective to evaluate the characteristics of the actors (scholars) considering their positions and roles in their collaboration
(co-authorship) network expressed through their number of partners or the way the actors are connected to other actors
directly or indirectly.
Reviewing the literature on the network dynamics and link formation identifies the cumulative advantage, assortative
mixing and cohesiveness processes (Ahuja et al., 2009; Glückler, 2007; Gulati, 1995; Powell et al., 2005) as drivers of network
dynamics. There processes are used in this study to examine the especial characteristics of the authors which affect their link
formation over the evolution of a real collaboration network. Assessing the cohesiveness process, the results reject the first
hypothesis indicating that the history of old relationships do not encourage authors to form new links in the future as the
collaboration network evolves. As Table 2 indicates, considering very few new links by the existing authors in the following
year, more links form among the existing authors and the new authors over time. This suggests that the trust among already
connected scholars (co-authors in the previous years) is not a good motivation for more joint publications in the future.
The result of assessing the cumulative advantage process does not support the second hypothesis and reveals that the
new authors do not prefer to attach to the well-connected existing scholars in this collaboration network. Examining the
assortative mixing process, we found that the collaborations networks of only three years (i.e., 2002, 2005, and 2009) can be
considered assortative and for two year disassortative and the non-significant Pearson degree correlation was found for the
remaining periods. Therefore, the contradictory results indicate that the third hypothesis is not supported.
Furthermore, this study proposes another attachment process which considers the nodes’ structural position as a driver of
their link formation. The closeness and intermediating role of the authors in the global network structure are assessed since
these roles reflect the strategic positions of scholars in their collaboration network to access the other scholars or control
the flow of resources along the network. The result of the correlation test between the authors’ closeness position and their
future attachment frequency rejects the fourth hypothesis. But the last hypothesis is supported that the intermediating
position of authors in their network is a good driver of their future link formation.
In brief, this study shows that a strategic intermediating (structural) position in the network is the most effective processes
for scholars’ attachments during the evolution of their collaboration networks. The finding can explain the dynamics of the
networks over time and help to predict more precisely how the connectivity of authors and consequently the structure
of the network will change in the future. While studying the link formation or attachment processes among the scholars
in a collaboration network, we have to take into account the limited capability and resources of scholars as forming and
retaining connections needs variety of resources. Therefore, even if for instance a well-connected or well-positioned scholar
is expected to have more links in the future but the process cannot last forever due to the scarce capability and available
resources.
5.1. Implications and future directions
The findings of this study contribute to further development of a theoretical understanding of network dynamics and
evolution through the use of different multi-disciplinary theories. The findings may have implications in variety of fields such
as in: political science to invest on the actors who can influence other members of a community faster and more effectively
to spread the information that the political parties need so as to reach their goals; marketing research to invest on the actors
who can attract more customers; or disaster management to facilitate the dissemination of information and resources. On
the other hand, identifying the actors with a strategic position in a network can help to control and manage the networks in
different domains such as a in a disease outbreak network by removing or vaccinating the people who can easily and widely
disseminate the disease to others due to their strategic role such as intermediating; or the findings can help to control the
spreading of viruses attacking computers in computer networks.
This study also contributes to the network science field by proposing a methodology which can be extend the existing
models of network evolution such as Barabási and Albert (1999) to express actors’ preferential attachment. While the existing
model focuses on the connectivity of nodes as an effective predictor of their future attachments, rather this study shows
that nodes’ intermediating position, enumerated by betweenness centrality, better predict actors’ attachment behavior.
The metrics proposed and used for quantifying nodes’ attachment behavior can be extended in future research by using
more comprehensive metrics such as hybrid centrality measures which consider combining different structural attributes
692
A. Abbasi / Journal of Informetrics 10 (2016) 685–692
of nodes in a network (Abbasi, 2013) or extension of them to examine whether other characteristics of the actors can better
predict their attachment processes and the dynamics of network structure. For instance, understanding of the link formation
process can help to better realize the inter-organizational and inter-firm collaborations, or countries economic and political
cooperation. Besides, this line of research can further support and facilitate the ‘community detection’ research as the
attachment processes uncover which nodes and how they will form sub-groups or communities. In addition, future research
could benefit from the metrics and processes discussed for the link formation, especially the intermediating structural
position, to improve the algorithms and techniques for the ‘link prediction’ problem.
5.2. Limitations
It is inevitable to limit the focus of a research in order to achieve a depth understanding of the phenomena studied,
in almost all scientific fields. The most important limitation of the scientific research in general is the universality of the
findings based on an experiment or analysis of a single or several limited cases. The findings of this study have a similar
limitation. While academic collaboration networks are exemplars of social networks (Watts and Strogatz, 1998), but there
are limits to the generalizability of the social networks. For instance, the degree and frequency of the nodes’ attachment in
a collaboration network may vary depending upon the specific context of social networks. Therefore, there is a bias on the
generalizability of the findings of this research.
Author contribution
Alireza Abbasi: conceived and designed the analysis, collected the data, contributed data or analysis tools, performed the
analysis and wrote the paper.
References
Abbasi, A. (2013). h-Type hybrid centrality measures for weighted networks. Scientometrics, 96(2), 633–640.
Abbasi, A., & Altmann, J. (2011). A social network system for analyzing publication activities of researchers. In T. J. Bastiaens, U. Baumol, & B. J. Kramer
(Eds.), On collective intelligence (Vol. 76) (pp. 49–61). Berlin: Springer-Verlag Berlin.
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.
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(1), 403–412.
Abbasi, A., Wigand, R., & Hossain, L. (2014). Measuring social capital and its influence on individual performance. Library & Information Science Research,
36(1), 66–73.
Ahuja, G., Polidoro, F., Jr, & Mitchell, W. (2009). Structural homophily or social asymmetry? The formation of alliances by poorly embedded firms. Strategic
Management Journal, 30(9), 941–958.
Ahuja, G., Soda, G., & Zaheer, A. (2012). The genesis and dynamics of organizational networks. Organization Science, 23(2), 434–448.
Barabási, A. L., & Albert, R. (1999). Emergence of scaling in random networks. Science, 286(5439), 509.
Burt, R. S. (1992). Structural holes: the social structure of competition. Cambridge, MA: Harvard University Press.
DiPrete, T. A., & Eirich, G. M. (2006). Cumulative advantage as a mechanism for inequality: a review of theoretical and empirical developments. Annual
Review of Sociology, 32, 271–297.
Freeman, L. C. (1979). Centrality in social networks conceptual clarification. Social Networks, 1(3), 215–239.
Glückler, J. (2007). Economic geography and the evolution of networks. Journal of Economic Geography, 7(5), 619.
Gulati, R. (1995). Social structure and alliance formation patterns: a longitudinal analysis. Administrative Science Quarterly, 40, 619–652.
Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1398–1438.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: homophily in social networks. Annual Review of Sociology, 415–444.
Merton, R. K. (1968). The Matthew effect in science. Science, 159(3810), 56–63.
Newman, M. (2002). Assortative mixing in networks. Physical Review Letters, 89(20).
Panzarasa, P., Opsahl, T., & Carley, K. M. (2009). Patterns and dynamics of users’ behavior and interaction: network analysis of an online community.
Journal of the American Society for Information Science and Technology, 60(5), 911–932.
Podolny, J. M. (1994). Market uncertainty and the social character of economic exchange. Administrative Science Quarterly, 39(3), 458–483.
Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network dynamics and field evolution: the growth of interorganizational collaboration
in the life sciences. American Journal of Sociology, 110(4), 1132–1205.
Rosenkopf, L., & Padula, G. (2008). Investigating the microstructure of network evolution: alliance formation in the mobile communications industry.
Organization Science, 19(5), 669–687.
Scott, J. (1991). Social network analysis: a handbook. Sage.
Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 109–125.
Wasserman, S., & Faust, K. (1994). Social network analysis: methods and applications. Cambridge, New York: Cambridge Univ. Press.
Watts, D. J. (2004). The new science of networks. Annual Review of Sociology, 243–270.
Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of ‘small-world’networks. Nature, 393(6684), 440–442.