1
The effect of powerful alters: investigating the moderating effect
of power between centrality and performance
Joshua de Groot – 10034323
University of Amsterdam – Master Thesis Msc Business Administration Track: Strategy
Academic year: 2016-2017
Supervisor: Nathan Betancourt Submission data: 23-06-2017
2 Statement of Originality
This document is written by Joshua de Groot who declares to take full responsibility for the
contents of this document. I declare that the text and the work presented in this document is
original and that no sources other than those mentioned in the text and its references have
been used in creating it. The Faculty of Economics and Business is responsible solely for the
3 Contents
Abstract 4
1. Introduction 5
2. Literature review 10
2.1 Basic network definitions and Social capital 10
2.2 Centrality and Performance 15
2.3 Power and Powerful connections 17
3. Methodology 20
3.1 The Nobel Prize network 21
3.2 Independent variable: centrality 24
3.3 Dependent variable: performance 26
3.4 Moderator variable: power (authority) 27
3.5 Analysis 27
4. Results 29
4.1 Significant correlations 29
4.2 Direct effects of winning 32
4.3 Moderation effect of winning 32
4.4 Direct effects and moderation effect on productivity 33
5. Discussion 35
5.1 Managerial and practical implications 35
5.2 Limitations and further research 37
6. Conclusion 40
References 41
4 Abstract
The present paper investigates the moderating effect of power on the positive relation between
degree centrality and performance. Previous research within the broad field of social networks
have established that network advantage, which stems from ego’s favorable position within
the network, is generally positively related to performance. However, there are also studies
that have found variances in performance from actors in identical network positions. This
paper examines the power of alters as an attempt to explain these variances. It is expected that
power positively moderates the effect of degree centrality and performance. The Nobel prize
network between 1901-1905 is used as dataset. Exploratory analysis and centrality
calculations is done by UCINET which is good accredited network analysis software. The
results reaffirmed the positive effect of degree centrality on performance, but show no
5 1. Introduction
The field of (social) network studies has grown extensively in terms of articles concerning
different academic disciplines (Borgatti et al., 2009; Brass et al., 2004). Starting in the field
of sociology, the network approach is now also used in social sciences, political science and
even economics (Borgatti et al., 2009). These studies prove that there is a wide variety of
interesting phenomena one could explain by taking a network perspective. An explanation for
its popularity is the focus on relationships among social entities, and on the patterns and
implications of these relationships (Wasserman and Faust, 1994), instead of traditionally
focusing on the observations only. Among other things, these social entities can be
individuals, (work) groups or organizations. The relationships they share can take many
forms, and not limited to, for example friendship, influence or financial transaction.
According to the network perspective, social entities are embedded within networks of
interconnected relationships that provide opportunities for and constraints on behavior (Brass et al., 2004). This makes a network perspective exceptional for it doesn’t examine an
individual in isolation like most traditional perspectives do, however it focusses on relations
and structured patterns of interactions to explain certain phenomena (Borgatti et al., 2009).
These logics of social networks as described by Borgatti et al. (2009) have significant bearing
on individual behavior and on individual outcomes like performance. For example, taking a
network perspective helped researchers explain topics as group problem solving, diffusion
and adoption of innovations, consensus and social influence and exchange and power among
others (Wasserman and Faust, 1994). One can say that the thoughts behind network
perspective are intuitively correct when we think of our own lives and the effect others may
have on it.
This thesis will address the concept of network advantage and its relationship to
6 through investigation social capital and more specifically centrality (Burt et al., 2000;
Coleman, 1988 Freeman 1978). Correspondently this means, when thinking of social
networks, that a person can occupy a certain position which is only advantageous for that
particular person. Advantage is therefore only related to that particular position, which means
that others in the network don’t have (the same form, or amount of) advantage. These positions each have their advantages in terms of information for example. However, an
interesting phenomenon is that not everyone makes use of his or her advantage (Burt el al.,
2013; Casciaro, 1998). Why, in some cases, this happens is still the topic of a wide debate.
These debates mostly focus on individual differences, mostly touching upon topics like
personal characteristics or the ability to perceive one’s network correctly. These studies on the
individual level indeed proof to be fruitful (Casciaro, 1998). However, to a lesser extend has
there been a focus at the properties of one’s connections. It might be that the difference in
characteristics of people one is connected to, explain part of the variance in performance.
Analyzing characteristics of one’s connections is different from the prevalent literature which
focuses has been on the consequence of network position on a person’s outcomes.
The idea that your connections have an influence on you has its foundations in both
network theory and also in social capital theory (Burt, 2000). Even though social capital
theory is on itself complicated because of scientific disagreement of what it should entail
Sobel (2002), one commonly accepted aspect of this theory sees social connection as a
resource to achieve an individual his or her needs. Translating this concept to network theory is that one’s network predicts a certain outcome. With this in mind and in an organizational perspective, one can imagine that those connected to powerful, or influential people, may
hope to achieve more and therefore perform better than those people who don’t have powerful
7 network advantage and performance. These thoughts will lay a new route in explaining
network advantage variances and is the particular reason for conducting this research.
To find the moderating effect of powerful connections this thesis will make use of a
database consistent of previous Nobel prize nominees and their respective nominators for the
years 1901-1937. This implies that our data is derived from an extensive amount of archival
data stored by the Nobel prize institution. The reason why the Nobel prize network is useful
for this study is that it is a large closed network where there is a clear link between who is
connected to who. The fact that the network is closed will allow for a relatively clean analysis
since there will be almost no externalities from outside the network. Another point is that all
nominators have initially been ordered by a number ranking 1 to 6 which indicates the
authority this person had within the network. This will later be argued to be a fair proxy indicating one’s power.
Studying a highly specialized network like the Nobel Prize network can also give new
insights into the mechanisms behind winning or losing a competition. The Nobel Prize
network can be described as an elite network (Crawford, 2001) where actors have to vote
upon notable scientific work for winning the Nobel Prize. Not often realized is that the people
within this (and in fact every) social network influence each other through their relationships.
So winning or losing may also be a result of this interplay. It becomes interesting when
comparing these mechanism of the Nobel competition to an organizational setting. One which
immediately comes to mind is the board of directors. This group of people have as one of
their main responsibilities the selection of a CEO. And just like winning the Nobel Prize,
becoming CEO may very much be, at least partly, the result of the relationships and
influences people in the network have with each other. To a lesser extend has social network
theory been used to explain selection of CEO’s, so it gives a great new perspective to an
8 network intuitively makes sense and one can understand that personal connections of a
candidate-CEO can be of great use. In particular the influence your connection can have on
the selection process, thus on others, is an interesting factor to examine. These influences can
derive from power connections have within the network. Using social network theory to help
explain leadership selection (and in this particular case CEO selections) can be of interest for
it gives another perspective to the dynamics behind selecting a leader.
The contribution of this thesis will be threefold: firstly the positive effect of network
advantage on performance will be investigated once more. The study will look at the most
widely studied concept of network advantage namely Centrality (Opshal et al., 2010), and if
this increases the chance of performing better. It will add to the existing literature on this
topic. Secondly, light will be shed on the moderating effect of power of those one is
connected to. The main question will be if the positive effect stated in the first part of the
study will change. Based on the literature one may predict an increased positive effect on
performance. This part contributes to the ongoing search of explaining different performance
outcomes people in similar network position have. Overall the study will do this by looking at
a select group of scientist within the Nobel Prize network. This network, which is distinct
from most traditional networks, can be seen as an elite network and therefore the study will
contribute to the marginal literature on elite network and performance. It opens the door to the
third contribution of this study. The elite network under review, and the competition it entails,
has similarities to Directors/CEO-selection in corporate board rooms. Before someone has “won” the position, a network of managers of great knowhow have discussed and voted on the nominees. Who wins could not only be the matter of who is the most suitable, but may
also be a result of the power of his or her connections. The discussion explains how power
comes to those sitting in multiple boardrooms from different organizations. Something which
9 performance, while also investigating the effect of powerful connections gives the following
research question:
What is the effect of an alter’s power on the relationship between network advantage and performance?
Figure 1. The conceptual model depicting Power as moderator
This paper will continue with a literature review starting with a brief introduction to social
networks. If one is new to social network theory it helps to read this section. Furthermore one
will notice that there a couple authors particular prominent in the existing literature on social
capital. Especially studies of Burt (1997; 2000), Lin (2001) and Coleman (2001) are seen as
influential papers in the field. Even though centrality is the main focus of this paper, it is
worthwhile knowing what Burt and Coleman have attributed significantly to network
advantage literature. Furthermore a more comprehensive explanation of power and how it has
been linked to social network theory is given. It will become clear that this is not a
straightforward construct even though it has been widely studied in social sciences. The
methodology section will provide more details on the Nobel prize dataset and explains how to
retrieve our desired data from the network. The network analysis done in this paper will stay
limited to calculating the measures for degree centrality as put forward by Freeman (1978). Power
10 Afterwards a linear and logistic regression will be conducted. After the results a thorough
discussion of the implications of these results will follow.
2. Literature review
The research done in this paper has its foundation from social capital and network theory.
Therefore, the literature review will begin with describing the important elements from these
streams which are deemed relevant for further understanding of the paper. Afterwards the
main literature on network advantage and (network) power will be described, including the
hypotheses flowing from this literature.
2.1 Basics network definitions and Social Capital
Networks, and in particular social networks, are likely known from platforms such as Facebook. What a wider audience probably doesn’t know is that, historically speaking, networks have been used by researchers for decennia to explain social phenomena (Freeman,
2004). Interest in social network structure and the interplay of social relations sparked many
sociologists attention ever since 1930 (Scott, 2012). Sociologist Moreno (1934) is credited for
doing the first social network experiment at Brooklyn Public and private schools. Asked to
investigate why girls were running away from a boarding school, he was the first to examine
friendship patterns between the girls. A hand-drawn imagine of these relationship is what
Moreno called a sociogram and which still is the term used today. These kind of phenomena
are explained based on the assumption that we all are embedded in local communities, in
which we connect with others through relationships (Prell, 2012). These relationship ties can
be either weak or strong, compare for instance your colleague you once casually met at the
coffee machine to your close family. And to a certain degree our ties (or actually the actions
11 thus grounded in the importance of relationships among interacting units (Wasserman and
Faust, 1994).
In the appendix I one can find a table of terminology frequently used in social
network theory and in social network analysis. Feel free to use this table as a reference when
there is some confusion of what a term means. Please note that this table does not contain all
definitions, but only those deemed helpful in understand this research paper. Due
specialization within the field of social network analysis, many different terminology have
emerged (Scott, 2012). The appendix will therefore, for example, also contain terms from graph theory (which emerged parallel to social networks and is similar to Moreno’s sociogram). To discuss all of these goes beyond the scope of this paper. Brass (2002) and
Borgatti and Halgin (2011) explained how, due the increasing interest of social networks and
specializations, there now exists considerable confusion about network theorizing. Especially
in social network analysis theorizing there is a distinction between “network theory” and “theory of networks”. Network theory refers to the mechanisms and processes that interact with network structures to end up with certain outcomes for individuals and groups (Borgatti
and Halgin, 2011). On the other hand, theory of networks tries to explain how particular
network structures come to be. This means asking questions as who forms ties with whom
and which actor becomes central or a broker in a structure. So it is good to notice that this
thesis will work with concepts of network theory.
Almost inherently to network theory is the concept of social capital. It is best
explained in contrast to other better known forms of capital. Were physical capital is
embodied in tools and machines. And human capital is created by developing your own skills
and capabilities to be able to act in (new) ways. Social capital exists through relations among
people that facilitate actions (Coleman, 1988). Burt (2004, p. 4-5) gives a clear explanation of
12
advantage. Social capital explains how people do better because they are somehow better connected with other people. Certain people are connected to certain others, trusting certain others, obligated to support certain others, dependent on exchange with certain others. One's position in the structure of these exchanges can be an asset in its own right. That asset is social capital, in essence, a concept of location effects in differentiated markets.” So the
advantage created by a person's location in social structure is known as social capital (Burt,
2004). As Burt (2004) clearly described these advantages can come from trust or the
obligation to support each other and this defines that these dynamics stem from human
interactions. Besides, and in more general terms, there are two explanations as to why
embedded resources in social networks enhance the outcomes of actions (Lin, 2001). Firstly
information about opportunities comes to those tied to specific individuals. Giving these
actors the chance to voice opinion qualitative better than others. In turn this can lead to faster
promotion (Lin, 2001). Secondly, ties may exercise certain influence on people in charge of
decisions (for example hiring or promotion) involving ego (Lin, 2001). This is explained
through the fact that some alters due to their formal positions enjoy higher authority, which in
turn make them carry more resources that can influence outcomes of important organizational
events like promotion or CEO-hiring (Lin, 2001). In general, organizational outcomes can
take many forms. A literature review conducted by Adler & Kwon (2002) shows how social
capital influences career success and executive compensation (based on Burt, 1992; Gabbay
& Zuckerman, 1998; Podolny & Baron, 1997; Belliveau, O’Reilly & Wade, 1996; Burt,
1997). There is still quite some debate of whether we should call social capital even capital,
and hence there exist different definitions in the academic literature (Adler & Kwon, 2002;
Sobel, 2002). However, our working definition will be in line with Burt (2004) since his work
13 following paragraph. Furthermore, Burt (1997, 2000, 2005) are recognized to be highly
influential in the social network field.
Some well-known academic work in the field of social networks and social capital is
done by Granovetter (1973), Coleman (1988) and Burt (1992; 1997). First mentioned, in his
paper The Strength of Weak Ties, research was done on the concept of interpersonal ties.
These ties can either be weak, strong or absent and, except for the latter, transmit information
between two people. Granovetter (1973) argued that novel information flows often through
people’s weak ties rather than strong ties. This happens since close friends move in the same circles as we do and therefore information flows often overlap with what we already know.
What is also interesting to mention is that these circles often overlap, meaning that person’s A
strong tie with B and C will give a higher chance that B and C are also connected through a
weak tie. This is known as the weak tie hypothesis originally stated from probability
mathematics by Rapoport (1957). In later work Granovetter (1983) explains how in this
situation clumps or cliques of social structure will form being bound predominately by strong
ties, and that weak ties will function as the crucial bridge between any two densely knit
clumps of close friends, organizations or any type of social actor. Since this bridge is the
connection to novel information his work proved that the chance of finding work increases
when one contacts his weak ties instead of his strong ties. This assumption holds even though
close friends may be more interested than acquaintances in helping us; social structure can
dominate motivation (Granovetter, 1973, 1983). In sum, work of Granovetter (1983) proves that success/advantage stems from bridging clusters through one’s weak ties.
Coleman’s (1988) work on network closure illustrated that social capital can come
from trust and social support within a high density network. He recognizes that there are two
broad intellectual streams of looking at an individual: 1) the rational, based from a more
14 governed by social norms, rules and obligations (Coleman, 1988). Both streams are correct
and he argues that social capital is a tool that can combine these two streams together. Social
capital is defined by its function, for what it is used for (Coleman, 1988). No matter what the
function is, there are always two elements in common. According to Coleman (1988) it
consists of social structures, and it facilitates certain actions of actors within this structure.
Remember the working definition of social capital as defined by Burt (2004)? It is evident
that many aspects of his definition stem from work done by Coleman (1988). In effect,
closure is a form of social capital where the highly dense network causes social norms, trust
and the diffusion of information enabling the network to achieve positive outcomes. An
important assumption for closure to hold, is that little “holes” should exist in the network. Holes can be seen as simple gaps, or missing ties in the network. Coleman (1988), calls these
the reason why not always social norms starts to exists and simply calls it the lack of closure.
The reason why in his opinion these holes are bad for the network as a whole, is the fact that
some actors in the network can profit from these at the expense of others. Resulting in less
optimal social capital for the whole network.
In continuation of this research paper the notion of social capital will often reoccur,
but only on the individual level. Examining network advantage involves examining the
individual advantage stemming from a particular network position. As one will notice from
social capital theory, these two concepts (of advantage and position) go hand in hand. Firstly
the concept of centrality will be explained which will the main concept used in this paper
positively effecting performance. Afterwards power will be explained and hypothesized to
15 2.2 Centrality and performance
Starting in the late 40’s with work of Bavelas (1950) centrality became a widely studied concept in network communication (Freeman, 1978). Later the ideas were used in many
different academic fields (Burgess, 1968; Snadowsky, 1972; Rogers and Agarwala-Rogers,
1976). So what is centrality? The centrality of nodes, or the identification of which nodes are
more “central” than others, has been a key issue in network analysis (Freeman, 1978;
Bonacich, 1987; Borgatti, 2005; Borgatti et al., 2006). According to Freeman (1978), who is
famous for developing centrality measures still used today, one can best think of a typical star
shaped network. This particular shape of network is depicted in figure 1 below.
Figure 1. A star-shaped graph with five points
Person A in the middle of the graph is universally assumed to be structurally more central
than any other node (B, C, D or E) in this network (Freeman, 1978). So as A is the most
central it would have the highest centrality. In relation to Lin’s (2001) discussed earlier, a
central position can draw on resources like information to enable this actor to perform better.
Quantifying centrality then became the next step. Freeman (1978) names three distinct
structural properties that are uniquely possessed by the center of the star. Firstly there is
degree centrality which numbers the amount of ties connected to. Secondly, betweenness centrality which entails that node A falls on the geodesics between the largest possible
16 number of other points. And lastly, Closeness centrality which counts how close a node is to
others and compared to others.
Figure 1 is of course a simplification of reality, but it does intuitively grasp the
consequences of having a high centrality. The positive effects stemming from a high
centrality position are linked to the properties discussed above: it has more ties, it can reach
all the others more quickly, and it controls the flow between the others (Opsahl, Agneessens,
Skvoretz, 2010). Because they have many ties, they may have alternative ways to satisfy their
needs, and hence are less dependent on other individuals (Hanneman & Riddle, 2005).
Furthermore, because of these ties they may have access to, and are able to call on more of the
resources of the network as a whole (Lin, 2001). These resources can be anything, from
borrowing a car of your neighbor to endorsements of your peers. However, what is frequently
seen as a valuable resource is information. Access to better information makes that a central
node can perform better.
So the more central a person is, the better he or she can draw on information, the
higher the level of performance. Within a study of Ahuja, Galletta and Carley (2003) on
virtual R&D groups, an individual’s centrality was measured as an indirect effect, but had the strongest positive effect on performance. But how to define performance? Different authors
positively linked centrality to for example job satisfaction (Sparrowe et al., 2001), power
(Brass, 1984), influence in decision making (Marden and Friedkin, 1993) and innovation
(Ibarra, 1993). Interesting to see is how centrality leads to promotion or appointment as CEO
or director. This implies a competition setting were multiple actors in the network strive to
“win” the promotion or appointment. As discussed earlier, this study will examine the Nobel prize network which is in competition for winning the Nobel prize. The first hypothesis will
17
H1a) Having a high centrality position will significantly increase the chances of winning the competition.
Even though in a competition setting the obvious performance outcome would be winning or
losing, performance can also take a different, more commonly seen, form. Namely,
performance can be the amount of work you finish. Within the academic community, in
which the Nobel actors resides, published articles is often seen as determinant of how a
scientist is performing. Since the Nobel prize network consist of mostly well know
researchers, these papers can form a good indication of their performance. Interesting to see is
if centrality leads to higher productivity of papers published. And therefore:
H1b) Having a high centrality position will positively effect productivity.
2.3 Power and powerful connections
Power has been linked to Network Advantage as an outcome variable (Burt, 1997; Brass and
Burkhardt, 1993; Freeman, 1978; Hanneman and Riddle, 2005), resulting from centrality, however never as a moderator in the link between Network Advantage and Performance. The
latter speculating on the effect of powerful connections. Meaning that an individual with
powerful connections will have different outcomes than someone who hasn’t have these connections. How to define powerful connections, and hence how do describe power, is
answered in the following section.
A variable that is researched widely, with respect to network analysis on an individual
basis, is power. That some people have more power than others is one of the most palpable
facts of human existence (Dahl, 1975). It has been there since the very beginning. And still,
the concept of power remains elusive despite the many attempts to properly conceptualize the
18 (1962) the elusive notion of power would have hold. Work at that time by Emerson (1962)
started to conceptualize power for the first time as the property of a social relation. According
to him, saying that someone has power over someone else is not a clear statement. When X
has influence over Y, signifying that person has power, it may as well be that X is subject to
Z’s influence (Emerson, 1962). With these two authors an example is made of the hardship of that time to conceptualize power, although this is still true today when looking at the wide
array of definitions in the literature. Mostly focusing on organizational literature Ibarra (1993)
sums up authors that define power as the ability to overcome resistance to achieve a desired
result (Astley & Sachdeva, 1984; Dahl, 1957; House, 1988; Pfeffer, 1981). This definition is
a good statement when defining the consequences of power. Other definition stemming from
social sciences defines power as the ability to influence or control the behavior of people
(Schein, Greiner, Virginia, 1988). Or from international politics “Power is the production, in
and through social relations, of effects that shape the capacities of actors to determine their circumstances and fate” (Barnett and Duvall, 2005, p. 39).
In light of social capital, power can best be approached by examining social
relationships and the effect these have on any actor in the network structure. It relates to
Emerson’s (1962) theory that power resides implicitly in other’s dependency. Meaning that
power comes to those who have the control or influence over things (which can be anything!) that one’s connection wants. Without someone’s dependency you could say that there exists no power (at least over others). Thinking back of the previous discussed forms of network
advantage, one starts to grasp that these concepts relate. Centrality, by definition, is having
the most important position within the network, thus it makes sense that studies found that
this position positively links to control or influence over others connected with. Knoke and Burt’s (1983) research on centrality and prestige states that high prestige is acquired by receiving unreciprocated choices from others. So what is this “thing” that is controlled that
19 enables power? Well, particularly important for this research paper is the control of
information. Network advantage from one’s centrality position means that there is control over information that flows from node to node through a specific actor.
Having power is the ability to control information flows. So is being connected to
someone who owns that ability more beneficial than being connected to someone who has
less power or none at all? To understand the effects of the power of alters it helps to
understand two distinct types of power. Ibarra (1993) notices that there is usually a difference
made between on the one hand individuals having a base of power, being a personal or
positional attribute of power. And on the other hand, being able to enact or use that power for
one’s purpose. Think of it as a passive and active way of using power. In which exercising power enables someone to affect outcomes (Mintzberg, 1983; Salancik & Pfeffer, 1977).
Fairly similar to the passive notion of power is authority. Authority is a form of power
which goes with a position and is legitimated by the social (or official) norms (Thompson,
1956). So in fact, an actor can have authority when this is perceived as such by the social
structure. Believing or trusting someone with authority is a subject on his own, but that this
happens is proved multiple times within the field of social sciences. Take for example a
simple situation of trusting the doctor, only because he has the job (take a look at Milgram
(1963), also known as the Milgram study if obedience sparks your interest). Another example
is a respected scientist who is named the authority on the subject he is affiliated with. Or, as is
seen in formal structures, authority is derived from the “rank” or hierarchical position established in the structure’s norms.
As to explain reasons why actors in similar centrality positions differentiate in their
performance it may be that their alters have a role in it. Those connected to high power alters
may receive better information which enables them to perform better than those who don’t
20 performing better, however since between-central nodes differentiation literature is scarce,
information is a good starting point. Within the Nobel prize structure we identity alters as the
nominators. These nominators all have different authority credentials which will be explained
later on. It forms a good setting for investigating the following two hypothesis:
2a) High power alters will positively influence the chances between centrality and winning the Nobel prize.
2b) High power alters will positively affect the relationship between centrality and productivity.
As is evident from previous academic works, network studies often use terms as
actors, ties, position, relations and information. And in case of confusing the terms, please
refer back to Appendix I which give all relevant definition based on the work of Wasserman
and Faust (1994) and Prell (2012) on social network analysis.
3. Methodology
Based on the book The Nobel Population 1901-1937 by Crawford and Ullrich (1987) a
complete overview exists of all nominees and nominators between the indicated time period.
It should be noted that this census applies only to the Nobel prize for Physics and Chemistry.
These records have been made public by the Alfred Nobel Memorial Foundation. Based on
this archival data a network was constructed from the years 1901-1905 with all Physics
laureates. One may critically observe why not for the entire period of 36 years? Because the
data would simply be too large which would consume time that goes beyond the scope of a
Master thesis. Secondly, this time window in particular contain scientists that paved the
21 radioactivity by Becquerel (1896) and further developed by Pierre and Marie Curie (1898), or
X-rays discovered by Röntgen (1895). This makes these five years, in my opinion, extra
interesting. Their discovered occurred before the Nobel prize even existed and so they are part
of the very first in the long list of winners of the prestigious prize.
Like in most social networks, each year consist of actors, scientists in fact, who have
been nominated for the Nobel prize (nominees). Furthermore, there are also those who
nominated the candidates (nominators). In total the network consist of 103 actors over this
five year period. As was part of the nomination process, only certain people were allowed to
nominate candidates. Therefore, nominators were ranked by classes. A class specified one out
of six official authorities that were allowed to nominate. Officially described as “authority” in
the Nobel Foundations statutes it also forms the power variable we will discuss later in this
section. Before it comes to that a brief introduction will be given of the Nobel prize network
including steps that had to be taken before ending up with a ready-to-use dataset. Afterwards
the variables will be discussed.
1901 Physics Nobel network 1
1. Mean 1,81
2. Standard Deviation 7,232
3. Sum of Ties 67
4. Variance 1,54
5. N of Observations 37 Table 1. Descriptive table for Physics Nobel network in 1901
3.1 The Nobel Prize Network
Table 1 above shows the amount of actors and ties for the Physics 1901 network. An
overview of the remaining years can be found in appendix II. One may wonder why the
descriptive data is meager compared to more common research data where there is a
22 that social network analysis (SNA) does not take this type of monadic attributes of individuals
into account. Here one studies dyadic attributes of pairs of individuals (Borgatti and Everett,
1997). In 1901 there were 37 actors involved with a total number of 67 ties. The mean amount
of ties was 1,81. Please note that these kind of descriptives give a very general overview of
what happens within the network. When examining relationships on the individual level, it are
precisely the individual uniqueness of the amount of ties, and the strengths of them (although
strength is not discussed in this paper), that really matters.
It helps to briefly explain network graph notations commonly used in social network
research and establish what type of network we are dealing with. The data under review is a
network called G=(V,E), where V is the set of vertices (also known as nodes/actors) and E is
the set of edges (also known as ties). Since multiple years are examined, a distinction between
the graphs will be made so that G₁=1901; G₂=1902; […] G₅=1905. Note how these notations
are derived from graph theory. Elaborating on the Nobel network data, one may notice that
when a nominator nominates someone this implies a directed link from V→V’. This means
that we are working with a directed network in which it is know that the nominator is the
‘sender’ and the nominee the ‘receiver’. Furthermore, one could make the distinction between nominees on the one hand and nominators on the other hand which indicates that one is
dealing with a two-mode network (also known as affiliation or bipartite networks; Borgatti and Everett, 1997; Latapy et al., 2008). Again in simple terms, a nominators is affiliating
himself with a nominee.
However, in this study the data will be treated as a one-mode network. This is
important for conducting a proper study on centrality, in fact, turning two-mode into
one-mode networks is a common practice in social network analysis (Opshal, 2003). Also when
desiring to measure centrality as developed by Freeman (1978) only dichotomous ties are
23 one of the main critics of Freeman’s measurements (Bonacich, 1987). In the case of the Nobel network, it is argued that all actors are of the same (Nobel prize) social bubble. Also all actors
are from the same academic environment. As described in the literature review this constitutes
an elite network of well renowned scientists and can therefore be put in a one-mode network.
It simplifies the data to a great extend which in turn helps with analyzing. Figure 2 shows the
graph from 1901. These graphs were made to get preliminary insights in to the data to help
visualize and understand what the network looks like. Graphs are made of every year and can
be found in the appendix if one is interested. The red nodes are the nominators and the blue
are nominees. Notice how all ties are dichotomous so that they either are present or absent.
Furthermore, there are predominately clusters around each nominee, which are star-shaped.
And if one thinks about it this is perfectly correct since the only action within this network
consists of nominators voting for their respective nominee. These stars already imply that the
nominee has a certain amount of centrality. As is allowed by the Nobel foundation,
nominators were allowed to select two and sometimes even three candidates.
Figure 2. Nobel network graph 1901
Each year constitutes a new network so in preparation for the centrality measures
24 adjacency matrix is a square actor-by-actor matrix as depicted in appendix IV. The same
actors stand in the columns as well as in the rows. Anything that can be presented in a graph
can be presented in a matrix. Let the matrix be called X and the content of any cell be Xij. As one can see, the matrix only has binary data (1’s and 0’s). This can only mean two possible things, either a tie exists which means that an actor has nominated a nominee. Or there is no
tie which mean that this particular action has never occurred. Within SNA this type of tie is
defined as a unidirectional dyadic tie and basically has only two options, which is either the
tie is there of it isn’t. That’s why this type of tie is coded as either 1 (the ties is there) or a 0
(no tie). This type of binary matrix is a characteristic detail of a one-mode network. Notice
how the graph in figure 2 is the visual depiction of the matrix in appendix IV.
3.2 Independent variables: Centrality
As discussed earlier, network advantage is an advantage a particular node in the network can
have stemming from his or her network position. This thesis will look at network advantage
by examining degree centrality as conceptualized by Freeman (1978). For binary networks
like the Nobel Prize network Freeman’s measurements are good enough (Bonacich, 2007). Degree centrality is the term used for the amount of ties ego has to other nodes in the network.
This is turn means that actors who have more ties to other actors may be located at
advantaged positions. Because they have many ties, they may have alternative ways to satisfy
their needs, and hence are less dependent on other individuals (Hanneman & Riddle, 2005).
Because they have many ties, they may have access to, and be able to call on more of the
resources of the network as a whole. So, a very simple, but often very effective measure of an
actor's centrality and performance potential is their degree. In undirected data, as is this
dataset, actors differ from one another primarily in how many connections they have. If an
25 to direct ties to them, and this may indicate their importance (and in this case their chance of
winning). A high degree centrality means many ties connected to that node, while for low
degree the opposite is true.
The measure of network advantage are calculated with the help of Borgatti, Everett
and Freeman (2002) social network analysis software UCINET 6. This free software is a
helpful tool and offers many more measures commonly used in SNA. The calculations of
centrality degree stem, like many of Freeman’s (1978) work, from graph theory. Individual
node degree centrality is not difficult to calculate. It is simply the sum of each tie in the
adjacency matrix representing the Nobel network. Getting the total graph degree centrality,
one could compare the variation in the different nodes from the whole. Fortunately, UCINET
gives the normalized degree centrality measures so no manual calculations have to be
performed. Of course it is interesting to know how to calculate degree centrality by hand in
case such fantastic software would not be available. The formula is derived from Freeman
(1978) and is used to measure relative centrality compared to others in the network. 𝑝𝑘 being
the node you want to measure and 𝑎(𝑝𝑖, 𝑝𝑘) = 1 if and only if 𝑝𝑖 and 𝑝𝑘 are connected by a line (Freeman, 1978). So when 𝐶`𝐷(𝑝𝑘) is large, many ties are connected to that node, and the opposite is true for a low 𝐶`𝐷(𝑝𝑘). Remember that the notations used in the formula stem from graph theory (for notation definitions, please refer to paragraph 3.1):
𝐶`𝐷(𝑝𝑘) =
∑𝑛𝑖=1𝑎(𝑝𝑖, 𝑝𝑘) 𝑛 − 1
26 3.3 Dependent variable: performance
As has mentioned before, performance will be measured by examining two performance
related outcomes: winning the Nobel prize and productivity. The first performance variable
seems to be clear. Either winning or losing the prize is a binary way of studying the outcome
and will require a different approach to derive at the results (see paragraph 3.5). Information
on this straightforward outcome is found through the work of Crawford, Heillbron and Ullrich
(1987). Interesting to mention for those not well acquainted with the Nobel prize: it is
common that the prize is shared by two or more laureates. In our 1901-1905 data there were a
total of eight winners.
The second performance measurement will be derived from publication productivity.
Productivity being a good quantitative way of looking at performance. This simply implies
that we recover the amount of publications the authors in our network have published for at
least the periods between 1901 and 1905. This, in combination with the centrality measures,
will allow to test for the relationship with productivity. However, to get an even better view of
the impact, we have also included the productivity from approximately 10 years before and to
10 years after the five year observation window. This way we can control for the period when
the Nobel competition was over for the authors. It is believed that adding 10 years is a good
time-frame to check for these differences. Publications were recovered from the free public
search engine for academic publications and literature Microsoft Academic 2.0, previously
known as Microsoft Academic Search. The tool features semantic search technologies and it
currently indexes over 150 million entities. This means that (almost) all publications from the
27 3.4 Moderator variable: Power (authority)
As has been described in the literature, is power still a hard to grasp concept. Fortunately,
Crawford, Heillbron and Ullrich (1987) provides us with rules established by the Nobel
Foundation itself. Everyone who is allowed to nominate a candidate is ranked on the basis of
their authority code. This code runs from 1 to 6 and implies a category a nominator belongs
to. However, when coding all the actors in our network there had to be made adjustments to
the official ranking. Numbers had to be attached to nominees as well and this has not
originally been done in the dataset. Therefore the authority codes are now as follow:
[1]=Nominee; [2]=Specially invited individuals; [3]=Chairholders at invited universities;
[4]=Physics and Chemistry professors at the Nordic universities listed in the special
regulations of 1900; [5]=Previous Winners in physics and/or chemistry; [6]=; Members of the
Academy's Nobel Committees; [7]= Members of the Academy of Science; [8]=Previous
winners and members of the Academy of Sciences. These codes are ordinal in nature and
therefore preferred to be adjusted so that it fits within regression analysis. More on this matter
in the next paragraph.
3.5 Analysis
As read from the previous paragraphs the independent variable centrality and the moderating
variable power are both continuous. The dependent variables are continuous and binary for
respectively productivity and winning. Since the hypotheses imply a prediction of what would
happen with the outcome, a multiple and a logit regression analysis are conducted. Running
these tests require two datasets. One for the winning as an outcome and the other dataset for
testing the change in productivity as an outcome. Of course, there should be no reason to use
28 winners, after preliminary analysis it became clear that the outcome winning got inflated
when leaving nominees in the dataset after they had won. STATA would believe that winners
would still be winners after they had won, therefore affecting the results in a wrong way. Hence, the creation of a “winners” dataset and a “productivity” dataset.
This thesis will run a logistic regression for the winners dataset since winning is a
binary outcome. Like in any regression searching for a moderation effect the model will first
test the effect of centrality on winning the Nobel prize. Afterwards authority will be
incorporated and see if the changed explanatory power of the model (R-change) is significant.
Finally the interaction effect between centrality and authority is added and again the
interesting question is if this positively (or negatively) adds to the initial relationship. The
same steps will be conducted for the productivity dataset, however here a multiple regression
analysis is used. One note of attention for logistic regression, the output generated only tells
us the likelihood of winning the Nobel prize. In other words, what is the chance of winning
and how does an extra independent variable help increase (or decrease) this chance.
Even though the Nobel prize network is highly specialized and closed, it is still
important to realize that the analysis forms a mixture between SNA and conventional statistics
(in the form of regression analysis). Using both is highly common (Hanneman and Riddle,
2005) and, in the case of finding an interaction effect, even necessary. What also becomes
relevant is checking the assumptions for regression. The most important one is to check for
multicollinearity, which is true when independent variables correlate with each other. For
example when the interaction effect (X*Y) and X and Y correlate to a certain high degree.
When variables correlate with each other too much there is no chance in getting a reliable
regression coefficient estimates. To see if this is true the variance inflation factor (VIF) becomes important. All VIF’s are below 3 which indicates that there is little multicollinearity. To achieve the right results, a two-way interaction will be executed between centrality and
29 authority. The interacting variables will be calculated by multiplying the variables with each
other. Afterwards, the logistical and the multiple regression can be executed.
4. Results
This chapter starts with describing the (significant) correlations regarding winning and
productivity. The first correlation matrix introduces these correlations for the winning dataset
(N=1341), while the second matrix shows them for the productivity dataset (N=1404).
Furthermore this chapter will explain the results from the regression analysis by examining
the direct effects, as well as the moderation effects, of the variables on winning and
productivity.
4.1 Significant correlations
Table 1 illustrates the correlations within the winning dataset, while table 2 shows these for
the productivity dataset. Both tables start by showing the mean and standard deviations of the
variables. Notice how the centrality and authority differ in both tables. This is due the
different amount of N in both datasets (Winning, N=1341; Productivity, N=1404). What also
stands out is that centrality has a very low mean this is since the centrality measures could
only be calculated for the years the actors participated in the Nobel prize competition. When
examining the significant correlations for table 1 one notices that centrality correlates
positively (r = 0.5402, p < .01) to winning the Nobel prize. Thinking back of the star-shaped
clusters surrounding the nominees this result makes sense. The more indegree centrality (i.e.
the more ties connected to you), the higher the chance of winning. This trail of thought seems
not to be true when examining the relationship to productivity seen in table 2. Here a negative
30 direction had been found? With respect to the data it could be that actors high in centrality
were actually busy with winning the Nobel prize, busy with that single breakthrough paper
and therefore having less time to produce more. Another reason might be that centrality
reflects past levels of performance. Before they entered the Nobel prize competition they may
have produced many papers, and during the time of the competition they may have
transitioned into a phase of their career that they don’t produce much anymore. That alters vote on ego is thus based on its past performance. Appendix V shows a scatterplot done
during exploratory analysis of the data. Here one sees that winners are less productive than
those who do not win the Nobel prize. In respectively the winning as the productivity dataset
authority is positively correlates (r = 0,0810, p < .01; r = 0,0761, p < .01) to centrality. This
means that those higher in authority seemed to enjoy higher centrality. Even though these two
independent variables correlate, the multicollinearity check done in the analysis before
31
Table 1. Means, Standard Deviations and Correlations
Variables M SD 1 2 3 4
1. Winning 0.0045 0.0018 -
2. Centrality 0.0040 0.0006 0,5402** -
3. Authority 0.5570 0.0210 -0,0203 0,0810** -
4. Centrality X Authority 0.0035 0.0007 0,3356** 0,7717** 0,1708** -
Note, N=1341. Correlations are based on the Winner dataset **= Correlation is significant at the 0,01 level
Table 2. Means, Standard Deviations and Correlations
Variables M SD 1 2 3 4
1. Productivity 1,3006 0,0857 -
2. Centrality 0,0039 0,0005 -0,0220 -
3. Authority 0,5909 0,0209 -0,0263 0,0761** -
4. Centrality X Authority 0,0035 0,0008 -0,0112 0,7727** 0,1671** -
Note, N=1404. Correlations are based on the Productivity dataset **=Correlation is significant at the 0,01 level
32 4.2 Direct effects on winning
Table 3 shows the results of the logistic regression and the direct effect of our independent
variables. With a total of N=1341 observations the model in general has a significant Wald
chi2 (=37,20, p < 0,01). And quiet high explanatory power (Pseudo-𝛥𝑅2 = 0,4576). So the model as a whole explain substantially the outcome of winning. Model 1 tests the effects of centrality on the independent variable winning and shows that centrality (β = 0,375, p < .01) significantly affects the chances of winning the Nobel prize. These results indicate that
hypothesis 1a is supported. Scoring high on centrality does increase the chances of winning
the Nobel prize. An explanation for this result can be the fact that more nominators means
more votes which in turns mean a higher chance of winning. Adding authority in model 2 as
independent variable does not significantly explain more variance in the model (Pseudo-𝛥𝑅2
= 0,0641). This would mean that the people with power are not necessarily the people that
win. However it might also be explained that higher authority are given to nominators while
nominees all have an authority of zero.
4.3 Moderation effects on winning
To test the moderation effect of power as described in hypothesis 2a, an interaction variable
between centrality and power has been made. This can be seen in model 3 of table 3. Based
on the literature a positive influence of power was predicted since high power alters would
provide the ego with more information. The results show that β = 0.092 and the increased
explanatory power of the model Pseudo-𝛥𝑅2 = 0,0014 are both not significant. Although they
are positive which implies that the direction was correct. Alas, hypothesis 2a was not
33 4.4 Direct effects and moderation effect on productivity
In table 4 the results from the linear regression are depicted. The direct effect of centrality on
productivity is negative (β = -0,0220, P < .01) as well as significant. This signals that the
higher central actors are in the Nobel network, the less productive they are in terms of
published articles. This conclusion was already clear when looking at the correlations. Being
more central might causes a person to be too busy which affects the amount of work that
could be done. It seems that the authority in model 2 does not add any explanatory power to
the model. It shows as well a negative direction, which means that actors higher in authority
produce less. However note that this result was not significant which means that statistical
support still has to be given.
Model 3 shows the interaction variable of centrality and authority, similar to the one used in the logistic regression. Here the model also doesn’t signals any statistical significant explanation of the variance (𝛥𝑅2 = 0,0002). Which means that the interaction effect is not there and a moderation of power between the centrality and performance couldn’t be found.
Hypothesis 2b is not supported. An explanation for this might be that those actors high in
authority produce less since they are too busy with other tasks. Or that authority simply
doesn’t help. Another explanation might be that the authority linked to a nominator based on the statutes from the Nobel Foundation, is not a good measurement for the real authority that
34
Table 3. Summary of Logistic Regression Analysis Results: Authority as a moderator of the relationship between Centrality and Winning
Model B SE B β z
𝛥
Pseudo R2 Model 1 0,4576* Centrality 36,0176 5,9050 0,375** 6,10 Model 2 0,0641 Centrality 39,7611 8,6826 0,215** 4,58 Authority -4,2449 3,3617 -0,867 -1,26 Model 3 0,0014 Centrality 39,3625 8,850 0,041** 4,45 Authority -25,9115 4194,455 -1,007 -0,01 Centrality X Authority 62,3226 13194,09 0,092 0,00Note, N=1341. Correlations are based on the Winner dataset **= Correlation is significant at the 0,01 level
Table 4. Summary of Regression Analysis Results: Authority as a moderator of the relationship between Centrality and Productivity
Model B SE B β t
𝛥𝑅
2 Model 1 0,0005* Centrality -3,5399 1,7133 -0,0220** -2,07 Model 2 0,0006 Centrality -3,2366 1,6872 -0,0201 -1,92 Authority -0,1013 0,1202 -0,0247 -0,84 Model 3 0,0002 Centrality -5,9007 2,4482 -0,0366** -2,41 Authority -0,1110 0,1229 -0,0271 -0,91 Centrality X Authority 2,4270 1,8498 0,0216 1,32Note, N=1404. Correlations are based on the Productivity dataset *=Correlation is significant at the 0,01 level
35 5. Discussion
This chapter will shed light on the results as seen in the previous chapter. The results show
interesting and, for the moderation hypothesis, contradictory findings. This thesis shows one
significant result, however it does not show how this happens. Based on the current literature
an explanation of these results will be given together with managerial and practical
implications. All in the hope that this sparks new interest in the influences of ego’s alters, and
particularly in explaining performance differences between actors in similar positions. To do
this the chapter will end with limitations and future directions to explore.
5.1 Managerial and Practical implications
The conceptual model and its hypothesis contribute to the ever growing body of social
network theory. Hypothesis 1a reaffirms the positive effects of network centrality on
performance. Confirming the hypothesis that centrality degree positively influences one’s chances of winning a competition. The Nobel prize is a very specific competition and shows
to be a very specific network structure. The actors in this network are all scientist and to an
extend important in their fields (Zuckerman, 1977). They are experts and their network has
therefore been described as elite (Crawford, 2001; Zuckerman, 1977). One may think that this
network is too specific, it couldn’t possibly show characteristics compared to the internal
networks of a company? On the contrary, the scientific community in general (where the
Nobel network actors originate from) have similar network characteristics as seen in firms:
with scientific communication, reward systems and social controls (Mulkay, 1976).
Interesting may be the new insights this research can give on the matter of
CEO-selection. Network theory is a perspective that also plays an important role in understanding
how individuals within an organization make decisions as well as how organizations are run
36 many social dynamics which are not easily seen. That these dynamics exists has first been
addressed by Zeitlin (1974) who reviewed if separation of ownership and control of the firm
really existed. A year later Benson (1975) described the interorganizational network and how
companies interact and control resources and power within their select network. They may be
linked directly or indirectly with each other and this is when board interlocks becomes
evident. Simply put, board interlock occurs when a board director sits at two or more
corporate boards (Scott, 1991). Besides having internal executives, large enterprises also have
outside non-executives which can be public or political figures. But most are hired from
banking, insurance and investment companies (Scott, 1991). Scott (1991) states how some
executives can have three or more directorships and that this comes with great power and
influence. Their directorships spread throughout the economy, and forms a corporate inner
circle of corporate decision-makers with power and influence across the business system as a
whole (Useem, 1984; Davis, Yoo and Baker, 2003). These directors exercise control and
power through two of the three distinct kinds of incorporate relation (Scott, 1991). Personal
relations are direct connections between people, these can come from kinship, friendship or
acquaintances. Furthermore capital relations are the links that result from shareholdings and
from the granting or withholding of credit.
The Nobel prize network described in this thesis is a proper benchmark of the often
hard to access data on CEO/director selection and the effects that board interlock ties have on
the selection process. Although not yet widely studied, power and authority may have a
significant effect on who will be appointed as new director or CEO. From the CEO’s
perspective the wish to select directors who have similar background characteristics to
themselves has been one of the major findings in directors selection literature (Kaczmarek,
Kimino and Pye, 2012; O’Reilly and Main, 2010). However, recently CEO’s have been
37 to appoint new directors who have different backgrounds from them (Filatotchev and Toms,
2003; Zhu and Westphal, 2014). One can already begin to grasp the different dynamics that
are at play. The influence from different stakeholders may be far greater than anticipated from
the existing literature. The power and authority from interlock board networks are very
different from the construct used in the Nobel structure. Using the authority codes described
from the Nobel foundation statutes has its implications on the results. And although the
results derived from the Nobel prize laureates does not show any significant results
concerning the moderating effect of authority it may very well be the case in the case of board
interlock networks.
5.2 Limitations and future research
This thesis had its academic foundations in network centrality as described by Freeman
(1978). The centrality construct is composed of three elements, being degree, betweenness
and closeness. One of the limitations of this thesis is that only the investigation of degree
centrality made sense when looking at the Nobel network. The reason being that the dynamic
of the network was only nominating, which resulted in ties to one (or two) specific nominees.
Betweenness, which is the position between two (or more) different nodes implies that this
node has some sort of role in transferring information between these two actors. However, nominees most often didn’t even know by whom they were nominated (Crawford, 2001; Zuckerman, 1977). Lastly, closeness would imply the closeness of ego to others through the
shortest geodesic path, however preliminary analysis of the data showed only exclusive
star-like clusters, because of the same network dynamic discussed earlier. Although it is widely
accepted that centrality has positive effects on performance outcomes, it might be interesting
to test the other centrality measurements that were not applicable to the current dataset as
38 different measurements, leaving more complexity in to the network would require a weighted
network analysis. Considering a complex network like the CEO/directors network, a weighted
network would show the differences in nodes and the implications of these differences better.
By dichotomizing the data, what has been done often to ease the analysis done, much info
contained in the dataset is lost. And consequently the complexity cannot be described as
richly (Opsahl et al., 2010).
There are however other well established network advantage measurements derived
from Burt’s structural holes and Coleman’s closure. Work of Burt (1992; 1997) focusses on structural holes in social networks. One can simply put that a structural hole is a missing tie
between two actors from different clusters in the network. When someone positions itself
where it can use many structural holes, a brokerage position emerges and this person
theoretically has better access to diverse information (Burt, 1997). The term broker or
brokerage reminds of the job a stock-broker has, and it depicts a good metaphor to what Burt
sees as brokerage. There are a couple of good definitions found in academic research. A
recent one is Stovel and Shawn’s (2012) definition of brokerage as “the process of connecting
actors in systems of social, economic or political relations in order to facilitate access to
valued resources”. The “process of connecting actors” is done by being the link between actors that else wouldn’t exist. A broker is therefore positioned within a gap between actors. “Facilitating access to valued resources” is the act of helping goods, information,
opportunities or knowledge flow across that gap (Stovel and Shawn, 2012). In that respect a
network broker is, just as a stock exchange broker, a node that trades on gaps in social
structures (Burt, 2000; Marsden, 1982). Similarities between Granovetter’s weak ties
hypothesis and Burt’s brokerage across structural holes are evident. Being the link between
social structures and facilitating transfer of valued resources is what gives the broker network
39 Instead of making bridges through weak ties between groups, closure is about
strengthening ties within the group (Burt et al., 2013). Closure results from cohesion in social
networks. Yielding alignment, closure enhances collective action (Comet, 2007). An actor
who has a closure position will most likely be involved in strengthening connections to gain
advantage by getting better what we already do (Burt et al., 2013). An illustrative way to
visualize a closure position is to think about a closed cluster of nodes. The node that has the most central position, something which is called the node’s centrality, falls on the shortest path between others and therefore has a potential for control of communication (Freeman,
1977). What is evident in this situation is that, in time, the regular interaction between the
relationships generates trust and increased reputation of this hub (Granovetter, 1983). The
dense ego-network leads to ego’s alters to coordinate with each other to help ego (Borgatti
and Foster, 2003). It is this combined aspect that is beneficial for the closure position. Early
work measuring the effect of closure has proved that it relates to increased group efficiency in
problem solving, perception of leadership and the personal satisfaction of participants
(Bavelas, 1950; Bavelas and Barret, 1951).
Brokerage and closure are two well established constructs for network advantage,
however both were not applicable to the Nobel dataset since measures for these constructs
apply better to more complex networks. For example, Burt’s concept of structural holes is
mostly applied to examine brokerage between clusters. This makes brokerage better
applicable on bigger and complex datasets. Since board interlocks are very complex social
structures these two measurements of network advantages might give an interesting
40 6. Conclusion
This thesis has tried to inspire future research to focus on the effects of alters on ego’s outcomes. It hoped to show that power moderates the effect of network advantage and
performance. Although the moderating effect was not supported, network advantage in the
form of centrality did give a significant change in performance. Looking into power of alters
as a trail of thought, together with different interpretations of it, is still worthwhile to explore
in the future. That alters have an effect on you (as ego) feel intuitively correct when we think
of this based from our daily lives. In fact, interlock boards literature state that stakeholders
have influence on organizational outcomes. It might be very likely that their influence also
ensures the selection of their nominee. Social capital and social networks have already
addressed these kind of relationships, however very little has been done with respect to
explaining variances between identical positions. Explaining why actors in identical network
positions differ in performance is a new direction social network analysis can help with.