Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Physica A 528 (2019) 121385

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Change agents and internal communications in organizational


networks

Evangelos Ioannidis a , , Nikos Varsakelis b , Ioannis Antoniou a
a
School of Mathematics, Faculty of Sciences, Complex Systems Analysis Laboratory (COSAL), Aristotle University of
Thessaloniki, 541 24 Thessaloniki, Greece
b
School of Economics, Faculty of Economic and Political Sciences, Complex Systems Analysis Laboratory (COSAL), Aristotle
University of Thessaloniki, 541 24 Thessaloniki, Greece

highlights

• Modeling change adoption in real organizational networks, engaging change agents.


• Agents with high degree (hubs) are most suitable to act as change agents.
• Senior employees are not always the most suitable to act as change agents.
• Communicating with ‘‘local hubs’’, results in faster change adoption.
• Network-based ‘‘Planning’’ reduces the change adoption time by 80%.

article info a b s t r a c t

Article history: The adoption of change in organizational networks is conditioned by the engagement
Received 4 July 2018 of the so-called ‘‘change agents’’ initiating ‘‘cascades of change’’, as well as by the
Received in revised form 30 March 2019 internal communication among the members of the network. We investigate how the
Available online 15 May 2019
dynamics of the adoption of change is influenced by the engagement policy of the change
Keywords: agents and by the internal communication. In this perspective, we compare central
Organizational networks engagements (high degree, high closeness, high betweenness, high eigen-centrality) with
Change adoption dynamics random engagement. We also compare three selection rules for communication, namely:
Change agents selection by randomness, selection by high link weight, and selection by high link weight
Centralities and degree centrality. The dynamics of change adoption is modeled by generalizing the
Internal communications discrete diffusion equation in order to incorporate both the engagement policies of the
Change management strategy change agents and the internal communication rules. Results are obtained by simulating
the solutions (change management scenaria) on 4 real organizational networks. Agents
with high degree (hubs) are found to be most suitable to act as change agents. The most
suitable change agents however, are not always the senior employees. The adoption
of change is much faster, if agents communicate with ‘‘local hubs’’, avoiding random
contacts. Change agents feed the members of the network with ‘‘thin slices’’ of influence,
in order to avoid crossing the ‘‘confidence bound’’. Small increase of the size of ‘‘slices’’,
results in significant acceleration of the adoption of change. We also estimate the value
of ‘‘planning’’ versus ‘‘no planning’’, in the context of Change Management.
© 2019 Elsevier B.V. All rights reserved.

∗ Corresponding author.
E-mail addresses: ioannidek@math.auth.gr (E. Ioannidis), barsak@econ.auth.gr (N. Varsakelis), iantonio@math.auth.gr (I. Antoniou).

https://doi.org/10.1016/j.physa.2019.121385
0378-4371/© 2019 Elsevier B.V. All rights reserved.
2 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

1. Introduction

Teams, organizations, and societies are formed by individuals who are characterized by their own opinions, emotions,
behaviors and attitudes. Social Dynamics is the field of study that investigates how the local interactions among
individuals influence the emerging dynamics of all these ‘‘activations’’, namely opinions, emotions, behaviors and attitudes.
The application of the concepts and methodologies of Statistical Physics and Agent-based Modeling on Social Dynamics [1–
5] has led to both discrete and continuous models of opinion dynamics. In discrete opinion models, the opinion
of each individual is a two-state or three-state variable. The states represent the possible choices, often as few as
two: yes/no, support/opposition, Windows/Linux, Android/iOS, buying/selling [6–8], or three: increase/decrease/invariant,
agree/disagree/don’t care [9]. The main discrete opinion models are the Voter model [6,7] and the Majority model [8].
Continuous opinion models are applicable when the opinions of individuals range over a continuous spectrum of values,
from one extreme to the other, as for example the degree of satisfaction, desire or preference regarding a politician or
a product. The main continuous opinion models are the Deffuant–Weisbuch (DW) model [10,11] and the Hegselmann–
Krause (HK) model [12]. In order to model the empirical fact that opinion influence is realized only if the opinions of
the communicating individuals are close enough, both continuous opinion models (DW and HK) assume that influence
is possible only if the opinions are closer than the so-called ‘‘tolerance’’ value ε [2]. This is also known as ‘‘confidence
bound’’ [2] or ‘‘relative agreement’’ [13–15]. The key distinction between these two continuous opinion models is that
HK model assumes that each individual simultaneously communicates with all ‘‘compatible’’ neighbors, while DW model
assumes that each individual communicates randomly with only one neighbor at a time. In this paper, we shall deal with
continuous models of social influence, following the modeling approach of the DW model.
Concerning opinion dynamics in networks, 2 key questions emerge:
(Q1) How opinion dynamics is influenced by specific network features?
(Q2) How to achieve the desired opinion influence?
Concerning question (Q1), opinion dynamics has been studied in relation to several network features, as for example:
the degree of agents [16,17], the directedness of the influence relationships [18], the presence of negative influence
relationships [15,19–23], the presence of mass media [19,21,23,24], the extent of the confidence bound [10,11,24–27],
the presence of extremists [13,14,22,25], the initial opinion distribution [15,27]. Concerning question (Q2), guiding public
opinion towards a pre-set target can be implemented using the so-called ‘‘informed agents’’. Informed agents are the
engaged individuals who intentionally and invisibly pull the opinions of all agents towards a pre-set target [28]. It
was found [28] that only a small number of informed agents may successfully influence the public opinion towards
the pre-set target, and that increasing the average degree of individuals can help the informed agents in decreasing
the consensus time. In case where the informed agents are attached to a pre-existing network, it was found [29] that
attaching the informed agents to agents of ‘‘high out-degree and low in-degree’’ is a highly effective strategy. Watts and
Dodds also found earlier [30] that large cascades of influence may not be driven by key influentials, but by a critical mass
of ‘‘easily influenced individuals’’. In case where agents are able to judge the communicated opinions, it was found [31]
that in order to maximize the influence of the informed agents, their guidance should be started when agents have little
knowledge about the subject under consideration. This finding reveals the significance of ‘‘timing’’ for effective influence.
In other words, the intentional and invisible influence of the informed agents should start before agents stop being ‘‘easily
influenced’’.
Companies and organizations willing to stay competitive in the modern socio-economic environment of rapid
change [32] should manage to change effectively in short time intervals [33]. People’s natural tendency for inertia has
been identified as the main factor, hindering the change process [33]. Just as in Newton’s First Law of Motion, people
resist to change, and this is a challenge for managers and leaders. The role of organizational networks in managing
change has already been pointed out [34–36]. As the social adoption of innovation is usually hard [33,37,38], the goal of
Change Management is to promote attitude change towards innovation adoption, aiming to sustainable organizational
change [39–44]. Therefore, successful organizational change cannot be realized and understood without taking into
account the attitude of individuals (‘‘positivity’’ or ‘‘negativity’’) towards the adoption of innovation. The significance
of individuals’ attitude towards organizational change has already been pointed out [45,46]. The influence of other forces
(management restructuring, technology, economy, socio-political change) [39–44] conditioning organizational change is
beyond the scope of this first exploratory work. Neglecting these forces is the main limitation of our modeling context.
We restrict our analysis to the emerging dynamics of individuals’ attitude resulting from agent-to-agent interactions, for
three reasons: (a) we want to understand the dynamics of individuals’ attitude, unconditioned by other forces, (b) the
time scales of attitude change is usually smaller compared to the time scales of other forces, (c) simplification of the
model.
As modeling the dynamics of opinion influence is also applicable for the dynamics of attitude change [47], we
intent to investigate the emerging change adoption dynamics for several scenaria of attitude change in organizational
networks. To address this challenge, ‘‘change agents’’ are engaged to initiate ‘‘change cascades’’, directing the agents
towards some desired attitude [48], as ‘‘informed agents’’ are engaged to pull opinions towards some desired opinion [28].
In order to avoid crossing the ‘‘confidence bound’’, the change agents feed the members of the network with ‘‘thin
slices’’ of influence, directing their attitudes accordingly. ‘‘Thin slicing’’ guarantees that influence is not ‘‘blocked’’ due
to ‘‘tolerance’’ limitations. However, too ‘‘thin slicing’’ delays the adoption of change. In addition to change agents, the
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 3

Table 1
Weight values and attitude change.
wλκ = 1 No resistance to change: Agent κ adopts fully the attitude of agent λ.
0 < wλκ < 1 Resistance to change: Agent κ adopts partially the attitude of agent λ.
wλκ = 0 No influence: There is no influence from agent λ to agent κ .

internal communication policy of the agents may also matter significantly to the successful implementation of a change
program, resulting in faster or slower change adoption rates. It was found that communicating change via agents of high
credibility results in greater change of beliefs, opinions, behaviors or attitudes [49]. The estimation of agents’ credibility
may depend on several factors, such as trustworthiness and expertise [50].
Concerning change adoption dynamics in organizational networks, 2 key questions emerge:
(Q3) How is change adoption dynamics influenced by the position of change agents (engagement policy of change
agents)?
(Q4) How is change adoption dynamics influenced by the way agents select other agents for communication (internal
communication policy)?
The goal of this work is to address questions (Q3) and (Q4). Concerning the influence of the position of change agents
(Q3), we address the issue whether central nodes are more advantageous over random nodes. The centrality of the initiator
nodes may have significant impact to the emerging dynamics of the system [51]. Different centralities (degree, closeness,
betweenness, eigen-centrality) may be more appropriate for different types of network flows [52]. In this perspective,
we compare in this first exploratory work the resulting change adoption dynamics for 5 representative engagement
policies of change agents, namely: engagements with high centrality (degree, closeness, betweenness, eigen-centrality),
as well as random engagement with uniform probability distribution. Other methods for identifying influential spreaders
in complex networks may include the k-shell decomposition, for unweighted [53] or weighted [54] networks. Concerning
the influence of the communication policy among the agents (Q4), we investigate three different selection rules, namely:
selection by randomness, selection by high link weight, and selection by high link weight and degree centrality.
The paper is organized as follows. In Section 2, we present the relevant definitions and assumptions of our model. In
Section 3, we specify the strategy implemented by the change agents and the real organizational networks to be analyzed.
Questions (Q3) and (Q4) are discussed in Sections 4 and 5, correspondingly. As random engagement of change agents
and random communication is the least efficient scenario, we estimate the value of ‘‘planning’’ versus ‘‘no planning’’, in
terms of change adoption time (Section 6). The impact of the ‘‘slicing’’ size is addressed in Section 7. The key findings and
real-world implications are summarized in Section 8.

2. The model

We formulate attitude dynamics in networks of agents influencing each other, assuming that each agent selects another
agent to communicate at each time t = 0, 1, . . .. Our formulation allows to implement several communication policies,
depending on the awareness and the preferences of the selecting agents.

Definition 1 (Attitude). Attitude is the individual’s degree of like or dislike for an object. The objects of attitude include
persons, events, places, things, issues. Attitudes are judgments involving affection, cognition, culture [55]. Carl Jung defined
attitude as ‘‘the readiness of the psyche to act or react in a certain way’’ [56]. We denote the attitude of agent κ at time t
by ψκ (t), taking values in the interval [−1, 1] assessing the degree of positiveness or negativeness [13–15,20,57].

Definition 2 (Attitude Networks). Attitude networks consist of agents κ = 1, 2, . . . , N with attitudes ψκ (t) at times
t = 0, 1, . . ., linked via communication channels. We denote by wλκ the weight of influence of agent λ on agent κ , for
attitude change, taking values in the interval [0, 1] (Table 1) [10,11,13,14,16–18,25].

Assumption 1 (One Agent is Selected at a Time). The attitude change of each agent κ during the time interval (t , t + 1] is
the difference: ψκ (t + 1) − ψκ (t). The durations (t , t + 1] are selected-specified so that the attitude of agents may change
by one ‘‘link event’’ only among their in-neighbors. If agent κ selects agent λ for communication at time t, the agent λ may
select or may not select agent κ at the same time t. Therefore, at each time t, the communication between two agents may
be unilateral or bilateral, depending on the selections of the agents. Moreover, although each agent κ may be influenced
by only one agent at each time t, the agent κ may influence more than one ‘‘followers’’ at the same time t. In continuous
social influence models in networks, there are two modeling approaches for the durations (t , t + 1], namely: ‘‘only-one-
agent-selection’’ modeling [10,11,13–19,25] (models based on the Deffuant–Weisbuch model) and ‘‘many-agents-selection’’
modeling [12,58,59] (models based on the Hegselmann–Krause model). According to the above, our model is classified into
the first category. In this way, the reduction of efficiency and reliability associated with multi-tasking is avoided [60–62].
We consider small durations of influence, taking into account that, at each duration (t , t + 1], attention is focused on one
source of communication only. Coarser descriptions involving large durations, where there can be many ‘‘simultaneous’’
influencers, can be easily discussed by summing successive influences. However, many successive influencing contacts
4 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

Table 2
Selection rules.
Selection rule s Selection probability psλκ Awareness and preference

pr
Jwλκ ̸ =0K
Random λκ = ∑N Awareness of incoming links.
ν=1 Jwνκ ̸ =0K
s =r All incoming links have equal selection probability.
pw
wλκ
Incoming weight λκ = ∑N Awareness of incoming weights.
ν=1 wνκ
s =w Higher incoming weights have higher selection
probability.
pw
wλκ ·dλ out
d
Incoming weight λκ = ∑N Awareness of both weights and centrality.
( out
ν=1 wνκ ·dν )
and out-strength s = wd Preference of high value of the product:
weight · centrality

may be necessary before individual attitude change takes place. For example, radical novelties or eccentric behaviors
may spread as ‘‘complex contagions’’ requiring multiple sources of reinforcement to induce adoption [63,64]. This case is
beyond the scope of our model and is the key limitation of Assumption 1.

Definition 3 (Internal Communications — Selection Rules). The selection of the in-neighbor λ by agent κ for commu-
nication is specified by the value of the Selection (Decision) Function Dκ (t ). Each Selection Function, implements some
Selection Rule s, specified by the corresponding Selection Probability psλκ . For example: Dsκ (t ) = λ means that agent λ was
selected by agent κ at time t with probability psλκ . Selection rules indicate both the awareness and the preference of the
selecting agent κ . In this first exploratory work we consider 3 representative probabilistic selection rules (Table 2). The
selection rule s which is adopted is assumed to be the same for all agents and for all times t.
The term [[Q ]] is the Iverson bracket [65] which converts Boolean values to numbers 0,1:
1, if Q is True
{
[[Q ]] =
0, if Q is False

λ is the out-strength centrality, taking values in the interval [0, 1], defined as:
The term dout
∑N
degλout µ = 1 wλµ
dout
λ = { }= { }
maxϕ=1,2,...,N degϕout ∑N
maxϕ=1,2,...,N w
µ = 1 ϕµ

Assumption 2 (Bound for Attitude Change). Agents resisting attitude change [33] may ‘‘filter’’ relevant social influences
of their in-neighbors, applying a ‘‘Filtering Rule’’, formulated in terms of the distance |ψλ (t ) − ψκ (t )| of the attitudes. The
selecting agent κ may be influenced by the selected in-neighbor λ, only if the distance of their attitudes is not greater than
the ‘‘tolerance’’ value ελκ : |ψλ (t ) − ψκ (t )| ≤ ελκ . This criterion for social influence is known as ‘‘bounded confidence’’ [2].
We assume ελκ = 2 · wλκ , taking values 0 ≤ ελκ ≤ 2 for weight values 0 ≤ wλκ ≤ 1 (Definition 2, Table 1). There are
two reasons underlying this assumption: (a) the empirical fact that if the influence weight wλκ is low, then the tolerance
value ελκ for attitude change should also be low, (b) as the range of values of the difference |ψλ (t ) − ψκ (t )| is the interval
[0, 2], the ‘‘tolerance’’ value ελκ should also have the same range.

Definition 4 (Attitude Dynamics Equation). The attitude dynamics equation is a generalization of the discrete diffusion
equation in networks, taking into account the above discussion. The attitude dynamics equation for each agent κ , during
the time interval (t , t + 1], is:
N

ψκ (t + 1) = ψκ (t ) + Φλκ (t ) (1)
λ=1

where Φλκ (t ) is the attitude influence of agent λ on the selecting agent κ :


Φλκ (t ) = Dsκ (t ) = λ · wλκ · [[|ψλ (t ) − ψκ (t )| ≤ ελκ ]] · (ψλ (t ) − ψκ (t ))
[[ ]]
(2)
The bracket [[|ψλ (t ) − ψκ (t )| ≤ ελκ ]] guarantees that attitude influence may take place only if |ψλ (t ) − ψκ (t )| ≤ ελκ
(Assumption 2). The assumption that attitude influence is proportional to the difference ψλ (t ) − ψκ (t ), is common in
social influence models, as for example: formulas (1) and (2) in [10], formulas (1) and (2) in [11], formulas (5) and (6)
in [13], formulas (5) and (6) in [14], formula (3) in [25], formula (1) in [18], formula (1) in [19], formulas (3) and (4) in [15],
formula (3) in [16], formula (8) in [17], formula (1) in [20]. Eq. (1) is a generalization of the above models, incorporating
moreover the implementation of several selection rules (Definition 3, Table 2). The analogous equation for knowledge
diffusion incorporates the implementation of selection rules and prioritizations [66–68].
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 5

Table 3
Simulations.
Conditioning factor Case
Engagement by ‘‘Randomness’’
Engagement by high ‘‘Degree’’
Engagement policy
Engagement by high ‘‘Closeness’’
of change agents
Engagement by high ‘‘Betweenness’’
Engagement by high ‘‘Eigen-centrality’’
Selection rule s Selecting by ‘‘randomness’’ (s = r)
for communication Selecting by ‘‘incoming weight’’ (s = w)
among agents Selecting by ‘‘incoming weight and out-strength’’ (s = w d)
Consulting Company (CC)
Real organizational Research Team of a Manufacturing Company (MC)
network Partnership network of a Corporate Law Firm (LF)
IT Department of a Fortune 500 Company (IT)
10%
Percentage of change agents
20%
in the network
30%
5%
Slicing parameter β 10%
15%

3. Estimating change adoption

We aim to change the attitude of all agents of the network. Initially (t = 0), all agents have more or less the
same attitude, taking random values in a narrow interval, for example [−0.85, −0.75]. As weights are non-negative
(Definition 2), only ‘‘attractive’’ interactions are possible. Therefore, if there is no intervention, the final distribution
of attitudes will lie within the same interval [−0.85, −0.75]. We simulate and evaluate selected Change Management
Strategies, aiming to influence agents towards a ‘‘desired’’ attitude ψ target . The Change Management Strategies are
implemented by engaging certain agents initiating ‘‘change’’ by purposely disseminating messages, opinions, values and
behaviors. These professionals are known as ‘‘change agents’’ [48]. The objective of change agents is to influence the
agents of the network, so that their attitude becomes eventually close to the desired attitude ψ target . As change agents are
in charge of bringing about change within the network, their dynamics (Definition 5) is distinguished from the dynamics
of the other (regular) agents (Definition 4).

Definition 5 (Change Agents Dynamics Equation). Change agents ω adjust their attitudes intentionally towards the
desired attitude ψ target , taking into account the attitude of their out-neighbors. The attitude of each change agent ω
should remain ‘‘close enough’’ to the average attitude ψ̄ωout (t ) of his/her out-neighbors, due to their ‘‘bounded confidence’’
(Assumption 2). The average attitude of the out-neighbors of the change agent ω at time t is:
∑N
ν=1[ψν (t ) · [[wων > 0]]]
ψ̄ωout (t ) = ∑N (3)
ν=1 [[wων > 0]]

The attitude of each change agent ω at time t + 1 is a shift of the average attitude ψ̄ωout (t ) towards the desired attitude
ψ target , weighted by a ‘‘slicing’’ parameter βω (t ) with values 0 ≤ βω (t ) ≤ 1. Therefore, the attitude dynamics equation
for each change agent ω is:

ψω (t + 1) = ψ̄ωout (t ) + βω (t ) · ψ target − ψ̄ωout (t )


( )
(4)

Change agents are actually triggering a ‘‘change cascade’’ in the network. ‘‘Thin( slicing’’ guarantees ) that influence is not
‘‘blocked’’ due to the ‘‘tolerance’’ limitations (Assumption 2). The term βω (t ) · ψ target − ψ̄ωout (t ) in Eq. (4) is the ‘‘slice’’
of influence, used by the change agent ω, to adapt to ‘‘tolerance’’ limitations of out-neighbors. Therefore, the metaphor
‘‘thin slicing’’ means that the ‘‘slicing’’ parameter βω is small enough (here, up to 15%, see Table 3). From Eq. (4) , we
observe that the attitude evolution of change agents does not depend on their previous attitude stage ψω (t ), but on
the previous average attitude of out-neighbors ψ̄ωout (t ). Change agents ‘‘adapt’’ their attitude to their local environment
as ‘‘chameleons’’ for maximal influence. Summarizing Eq. (4), the attitude of each change agent ω, at each time t, is
determined by the target attitude ψ target and the average attitude ψ̄ωout (t ) of the out-neighbors to be influenced. The
action of change agents may be viewed as the action of driver nodes for network control [69,70]. The action of change
agents, in this perspective, is internal locally adaptive control. In this first exploratory work, we shall assume stationary
‘‘thin slicing’’, identical for all change agents: βω (t ) = β , for all ω = 1, 2, 3, . . . and for all times t = 0, 1, 2, . . . The
selected values of β in the simulations are presented in Table 3.
6 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

Definition 6 (Change Adoption Time (CAT)). The CAT T of the network is the minimal time required for all agents
⏐ to
change their initial attitude to a new attitude which is approximately the desired attitude ψ target : ⏐ψκ (T ) − ψ target ⏐ < δ

for κ = 1, 2, . . . , N. In this work we shall assume ψ target = 0.8 and δ = 0.05. The average CAT of the network is estimated
as the average T̄ , considering several different realizations.

Definition 7 (Change Adoption Diagram (CAD)). The change adoption percentage p (t ) at each time t = 0, 1, 2, .. is the
percentage of agents who have approximately the desired attitude ψ target at time t:
∑N ⏐ψκ (t ) − ψ target ⏐ < δ
[[⏐ ⏐ ]]
κ=1
p (t ) =
N
The CAD of the network is the temporal plot of the change adoption percentage p (t ). In all CADs, for every time t,
we estimate the average value and the standard error of the change adoption percentage, considering several different
realizations.
The change adoption dynamics is assessed in terms of change adoption times (Definition 6) and change adoption
diagrams (Definition 7). In this work, the average CATs and the CADs are estimated considering 100 different realizations.
At time t = 0, some agents are engaged to act as change agents. As change agents act as ‘‘initiators of change’’
for the network, their identification is a strategic decision of high significance. In this first exploratory work we shall
compare the resulting change adoption dynamics for 5 representative engagement policies of change agents, namely:
engagements with high centrality (degree, closeness, betweenness, eigen-centrality), as well as random engagement with
uniform probability distribution. The engagement policy of change agents, as well as the number of change agents in the
network (as percentage of all agents), selected in each simulation, are presented in Table 3.
The definitions of the centrality indices [71] are given below for completeness. We consider only out-centralities,
because change agents are engaged to influence other agents.

Definition 8 (Weighted Out-degree). The weighted out-degree degκout of agent κ is defined as the sum of the outgoing
weights wκν to all its first neighbors:
N

degκout = wκν
ν=1

Definition 9 (Weighted Out-closeness). The weighted out-closeness cκout of agent κ is defined as the inverse of the sum
of the distances dκν to all other agents:
1
cκout = ∑N
ν = 1 dκν
ν ̸= κ
where dκν is the weighted directed distance from agent κ to agent ν . As in our work weights represent communication
channels (Definition 2), the distance dκν is estimated on the transformed weight matrix: − ln (wκλ ) [72].

Definition 10 (Weighted Betweenness). The weighted betweenness bκ of agent κ is defined as the sum of proportions of
all shortest weighted paths (geodesics) between pairs of other agents (except agent κ ) that pass through agent κ :
N N
∑ ∑ Mλ(κ )µ
bκ =
Mλµ
λ=1 µ=1
λ̸=κ µ̸=κ

where Mλ(κ )µ is the number of shortest weighted paths (geodesics) between agents λ and µ (with κ ̸ = λ and κ ̸ = µ)
that pass through agent κ , while Mλµ is the total of shortest weighted paths (geodesics) between agents λ and µ. As for
the estimation of Closeness index (Definition 9), the shortest weighted paths (geodesics) are identified on the transformed
weight matrix: − ln (wκλ ) [72].

κ of agent κ is the κ -component


Definition 11 (Weighted Out-eigen-centrality). The weighted out-eigen-centrality eout
of the right normalized eigenvector (Perron–Frobenius), associated with the right dominant eigenvalue z0 of the weight
matrix wκλ :
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
w11 ··· w1N e1 e1
⎜ .. .. .. ⎟ · ⎜ .. ⎟ = z · ⎜ .. ⎟
⎝ . . . ⎠ ⎝ .⎠ 0 ⎝ .⎠
wN1 ··· wNN eN eN
We consider the following 4 real networks:
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 7

Fig. 1. CADs for the IT network with 10% change agents and slicing parameter β = 5%. The 5 engagement policies of change agents (random, degree,
closeness, betweenness, eigen-centrality) are realized with ‘‘random’’ selection rule (s = r). The corresponding CADs with the ‘‘incoming weight’’
selection rule (s = w) and the ‘‘incoming weight and out-strength’’ selection rule (s = wd) are qualitatively the same (Appendix SM II).

• An organizational network of a ‘‘Consulting Company’’ (CC) of size 46 [73].


• An organizational network of a ‘‘Research Team of a Manufacturing Company’’ (MC) of size 77 [73].
• A partnership network of a ‘‘Corporate Law Firm’’ (LF) of size 71 [74].
• An organizational network of a ‘‘IT Department of a Fortune 500 Company’’ (IT) of size 56 [75].

All networks are both weighted and directed, with off-diagonal, non-negative and constant in time weights. The weights
are constructed from the relevant datasets: [76] for CC and MC, [77] for LF, and [75] for IT (Appendix, SM I).

4. Change agents: Central versus random engagement

We investigate the impact of the engagement policy of change agents on the dynamics of change adoption (Q3). For
each selection rule s, we compare the emerging CADs for the 5 representative engagement policies of change agents
(Table 3). The CADs for the IT network with 10% change agents and slicing parameter β = 5%, for random selection rule
(s = r) are presented in Fig. 1. The CADs for all cases explored (Table 3) are presented in the Appendix (SM II). The average
CATs for the IT network with 10% change agents are presented in Fig. 2. The relative decrease (%) of the average CATs,
comparing engagements of change agents based on high centrality with respect to random engagement are presented in
Fig. 3. The average CATs as well as the relative decrease (%) of the average CATs, due to different engagement policies of
the change agents, are presented in the Appendix (SM III), for all cases explored (Table 3).

Remark 1 (Agents with High Degree (Hubs) are Most Suitable to Act as Change Agents). Change adoption is usually
significantly faster when change agents are engaged based on high centrality, compared to random engagements (for
the CADs see Appendix SM II, for the average CATs and relative decrease (%) of the average CATs see Appendix SM
III). Moreover, for most cases, change agents with high degree centrality perform better, compared to other central
engagements. This fact is clearly illustrated in the CADs (Fig. 1), in the average CATs (Fig. 2) and in the relative decrease
(%) of the average CATs (Fig. 3).

Remark 2 (The Advantage of High Degree Over Random Engagement of Change Agents is Significantly Enhanced, if
Agents Communicate by Preferential Selection). If change agents are engaged based on high degree (the best engagement
policy, Remark 1), the advantage (decrease of the average CAT) over random engagement is about 50% (Appendix SM III),
in case agents communicate randomly (s = r). However, in practice, agents do not communicate randomly. Instead, they
select other agents for communication by some rule s (Definition 3, Table 2). For instance, according to the ‘‘incoming
weight’’ (s = w) or according to the ‘‘incoming weight and out-strength’’ (s = wd). In these cases, the advantage of high
degree engagement over random engagement is even higher (Appendix SM III). This fact is clearly illustrated in Fig. 3. More
specifically, for random selection (s = r), the decrease of average CATs is about 55%, while in case of preferential selection
the decrease is even higher: about 65% for the ‘‘incoming weight’’ (s = w), and about 80% for the ‘‘incoming weight and
out-strength’’ (s = wd). Generally speaking, the advantage of central over random engagement, is significantly enhanced
in the case of the ‘‘incoming weight and out-strength’’ selection rule (s = wd) (Appendix SM III). This advantage is higher
if there are few change agents in the network (Appendix SM III).
8 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

Fig. 2. Average CATs for the IT network with 10% change agents. The 5 engagement policies of change agents (random, degree, closeness, betweenness,
eigen-centrality) are realized with ‘‘random’’ selection rule (s = r) (left), ‘‘incoming weight’’ selection rule (s = w) (middle) and ‘‘incoming weight
and out-strength’’ selection rule (s = wd) (right). Results are presented for slicing parameter β = 5%, 10%, 15%.

Fig. 3. Relative decrease (%) of the average CATs for the IT network with 10% change agents, when change agents are engaged based on high
centrality (degree, closeness, betweenness, eigen-centrality) with respect to random engagement. Results are presented for ‘‘random’’ selection rule
(s = r) (left), ‘‘incoming weight’’ selection rule (s = w) (middle) and ‘‘incoming weight and out-strength’’ selection rule (s = wd) (right), and for
slicing parameter β = 5%, 10%, 15%.

We have seen [66] that agents in lower hierarchical classes, may outperform agents in higher hierarchical classes, in
terms of knowledge acquisition speed. We investigate the analogous question by comparing the social influence capability
of employees of different hierarchical classes. The social influence capability is estimated in terms of centralities. We
present the results for degree centrality (the best engagement policy, Remark 1) for the four real organizational networks
in Fig. 4. The results for other centralities (Table 3) are presented in the Appendix (SM IV).

Remark 3 (Employees in Senior Positions do not Always Serve as the Best Change Agents). Employees with high centrality
might not be in senior positions (Appendix SM IV). Therefore, the capability of employees for initiating change (estimated
by centralities) is not always ‘‘positively correlated’’ with their hierarchical position in the formal organizational chart. For
example, we observe in Fig. 3 that all ‘‘Executives’’ in the IT network have very low degree centrality compared to other
employees, even from the ‘‘Administrative Staff’’. In addition, many ‘‘Researchers’’ in the MC network are more suitable to
act as change agents compared to ‘‘Project Leaders’’. On the contrary, several senior employees in both CC and LF networks
have also high degree centrality.

5. Internal communications: Preferential versus random selection

We investigate the impact of the selection rule s on the dynamics of change adoption (Q4). For each engagement
policy of the change agents, we compare the emerging CADs for the 3 selection rules s (Table 3). The CADs for the LF
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 9

Fig. 4. Employee hierarchical class versus employee centrality in four real organizational networks. Dots represent employees.

Fig. 5. CADs for the LF network with 10% change agents and slicing parameter β = 5%. The 3 selection rules for communication among agents
(‘‘random’’ (s = r), ‘‘incoming weight’’ (s = w) and ‘‘incoming weight and out-strength’’ (s = wd)), when change agents are engaged randomly.

network with 10% change agents and slicing parameter β = 5%, for random and high degree engagement policy of the
change agents are presented in Figs. 5 and 6, correspondingly. The CADs for all cases explored (Table 3) are presented
in the Appendix (SM II). The relative decrease (%) of the average CATs, comparing the selection rules ‘‘incoming weight’’
(s = w) and ‘‘incoming weight and out-strength’’ (s = wd) with respect to ‘‘random’’ selection rule (s = r) are presented
in Fig. 7. The relative decrease (%) of the average CATs, due to different selection rules s for communication among agents,
are presented in the Appendix (SM III), for all cases explored (Table 3).

Remark 4 (The Adoption of Change is Significantly Faster When Agents Communicate by Preferential Selection Instead
of Random Selection). Change adoption is usually significantly faster when agents communicate by preferential selection
s = w or s = wd, compared to random selection s = r (for the CADs see Appendix SM II, for the average CATs and
relative decrease (%) of the average CATs see Appendix SM III). This fact is clearly illustrated in the CADs (Fig. 6) and in the
relative decrease (%) of the average CATs (Fig. 7). The advantage of preferential over random selection for communication
is less significant if the change agents are randomly engaged (Figs. 5 and 7). More specifically, this advantage (decrease
of the average CAT) is about 10% if change agents are randomly engaged, and may rise up to 40%–50% if change agents
engaged by high centrality (Fig. 7). We observe a ranking of the three selection rules. The order from the most efficient to
the least efficient selection rule is: ‘‘incoming weight and out-strength’’ (s = wd), ‘‘incoming weight’’ (s = w), ‘‘random’’
(s = r). This ranking is persistent when the change agents are centrally engaged (Fig. 6) and not always persistent when
change agents are randomly engaged (Fig. 5). Generally speaking, the advantage of preferential over random selection for
communication, is significantly enhanced in the case where change agents are engaged based on high centrality (Appendix
SM III). This advantage is higher if there are few change agents in the network (Appendix SM III).
10 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

Fig. 6. CADs for the LF network with 10% change agents and slicing parameter β = 5%. The 3 selection rules for communication among agents
(‘‘random’’ (s = r), ‘‘incoming weight’’ (s = w) and ‘‘incoming weight and out-strength’’ (s = wd)), when change agents are engaged based on
high degree. In case change agents are engaged based on high closeness, or high betweenness, or high eigen-centrality, the CADs are qualitatively
the same (Appendix SM II).

Fig. 7. Relative decrease (%) of the average CATs for the LF network with 10% change agents, when the selection rule s for communication among
agents is the ‘‘incoming weight’’ (s = w) (left) or the ‘‘incoming weight and out-strength’’ (s = wd) (right) with respect to ‘‘random’’ selection
rule (s = r). Results are presented for the 5 engagement policies of change agents (random, degree, closeness, betweenness, eigen-centrality), and
for slicing parameter β = 5%, 10%, 15%.

6. Estimating the value of planning

From the analysis of the impact of the engagement policy of change agents (Q3, Section 4) and of the selection
rule s (Q4, Section 5), we concluded that the least efficient combination is random engagement of change agents with
random selection rule (s = r). In other words, the worst performance corresponds to lack of planning for initiating and
communicating change. It is remarkable however, that change agents ‘‘achieve results’’ (change adoption percentage is
100% eventually) even in this case of ‘‘no planning’’ (random engagement of change agents and random communication
among agents), as observed in Figs. 5 and 7, as well as in Appendix SM II for all cases explored (Table 3). We consider this
least efficient combination of ‘‘randomness’’ (random engagement of change agents and random communication among
agents), as ‘‘reference case’’ in order to estimate the value of planning by different engagement policies and selection
rules. The estimation is assessed by the relative decrease (%) of the average CAT. We present in Fig. 8 the results for the IT
network with 10% change agents. The corresponding results for all cases explored (Table 3) are presented in the Appendix
(SM III).

Remark 5 (Planning the Engagement of Change Agents and the Internal Communications can Accelerate the Adoption
of Change by 80%, Compared to ‘‘No Planning’’). Engaging change agents based on high centrality, combined with
preferential selection for communication, results in significantly faster adoption of change, compared to lack of any plan
for initiating and communicating change. More specifically, the best change management strategy is engagement of change
agents based on degree centrality and preferential communication based on the ‘‘incoming weight and out-strength’’
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 11

Fig. 8. Relative decrease (%) of the average CATs with respect to random engagement of change agents and ‘‘random’’ selection rule (s = r), for
the IT network with 10% change agents. The 4 engagement policies of change agents based on high centrality (degree, closeness, betweenness,
eigen-centrality) are realized with ‘‘incoming weight’’ selection rule (s = w) (left) and with ‘‘incoming weight and out-strength’’ selection rule
(s = wd) (right). Results are presented for slicing parameter β = 5%, 10%, 15%.

Fig. 9. Relative decrease (%) of the average CATs for slicing parameter values β = 10% and β = 15%, with respect to the reference value β = 5%.
The 5 engagement policies of change agents (random, degree, closeness, betweenness, eigen-centrality) are realized with ‘‘random’’ selection rule
(s = r) (left), ‘‘incoming weight’’ selection rule (s = w) (middle) and ‘‘incoming weight and out-strength’’ selection rule (s = wd) (right). Results
are presented for the IT network with 10% change agents.

selection rule (s = wd) (Appendix SM III). This fact is clearly illustrated in the relative decrease (%) of the average CATs
of Fig. 8, where the decrease can be up to 80%. The value of planning (decrease of the average CAT) is higher if there are
few change agents in the network (Appendix SM III).

7. The impact of ‘‘slicing’’ size

We investigate the conditioning of change adoption dynamics from the value of slicing parameter β of change agents.
The influence of change agents is stronger for higher values of slicing parameter β , given the ‘‘tolerance’’ limitations
(Definition 5). We present in Fig. 9 the relative decrease (%) of the average CATs for β = 10% and β = 15%, with respect
to the reference value β = 5%, for the IT network with 10% change agents. The corresponding results for all cases explored
(Table 3) are presented in the Appendix (SM III).

Remark 6 (Increasing the Slicing Parameter Results in Significant Decrease of the Time for the Adoption of Change.
This Decrease is Persistent, Even if the Conditioning Factors (Table 3) Vary). From Fig. 9, we observe that average CAT
can be reduced almost to the half, if the slicing parameter β is increased from 5% to 10%. Increasing further the slicing
parameter β to 15%, results in 65% decrease of the average CATs. For all cases examined (Table 3), significant drop of
average CAT is observed (Appendix SM III) for the first 5% increase of the slicing parameter β (from 5% to 10%). It is
remarkable, that the decrease of the average CATs (Fig. 9) is not influenced by the engagement policy of change agents
or by the selection rule of agents. By comparing the corresponding diagrams for different percentages of change agents
and different organizational networks (Appendix SM III), we observe also that the decrease of the average CATs is not
influenced by the percentage of change agents or by the structure of the organizational network examined.
12 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

8. Conclusion

We investigated how change adoption dynamics is influenced by the engagement policy of change agents (Q3), and by
the selection rule s of agents for communication (Q4). The results of our analysis are based on simulations of the solutions
of the attitude change Eqs. (1) and (4). The novelty of this generalized discrete diffusion equation is the incorporation of
selection rules (Definition 3, Table 2) and engagement policies of change agents (Definition 5). The analogous equation
for knowledge diffusion incorporates selection rules and prioritizations, as well as positioning policies of the ‘‘experts’’ (highly
knowledgeable agents) [66–68]. These diffusion processes are in general non-uniform and non-isotropic, as diffusion in
disordered systems [78] and anomalous biased diffusion [79]. Our model is expected to be applicable to several modes
of attitude spread, as for example the diffusion of organizational citizenship behavior though an organization [80]. The
results of the simulations of the solutions of the attitude change equation, conditioned by the influence of the change
agents, are summarized as follows:

• Agents with high degree (Hubs) are most suitable to act as Change Agents (Remark 1). This is reasonable because
agents with high degree is likely to have more friends than their friends have [81]. Thus, the targeted attitude
spread, originated by high degree change agents, is enhanced. Therefore, the identification of influential positions
is necessary for the successful implementation of Change Programs. Hence, Organizational Network Analysis (ONA)
should be incorporated in Change Management. In fact, the decision for the engagement of change agents should be
based on network analysis. Network-based theories for Organizational Change, as for example the structural leverage
theory [82], look for key opinion leaders to act as leverage points for change. We found that employees in senior
positions do not always serve as the best change agents (Remark 3). This finding is in line with earlier studies [83],
indicating that sociometric and self-reported measures of leadership are weakly correlated. The action of change
agents may be viewed as the action of driver nodes for network control. In this context, it was found [69,70], that
the ‘‘driver nodes are usually not nodes with high degree’’. This result appears to challenge our finding that agents
with high degree (hubs) are most suitable to act as change agents. If we view our change agents as control agents,
our ‘‘control strategy’’ is completely different from the classical external control, adapted for networks [69,70].
Therefore, there is no challenge nor contradiction. Our ‘‘control strategy’’ is internal and continuously adaptive
to the local environment of each control agent (change agent). Summarizing, from the point of view of network
control, internal locally adaptive control is more effective, if the control agents are nodes with high degree, while
classical external control is more effective when the control agents are not nodes with high degree.
• Preferential Selection for Communication results in significantly faster Adoption of Change (Remark 4). Commu-
nicating with ‘‘local hubs’’, avoiding random contacts, results in much faster adoption of change. This finding reveals
that the adoption of change depends, to a large extent, on the ‘‘local-micro’’ social interactions. On the other hand,
communities where agents are communicating randomly are the most resistant to change, therefore less vulnerable
to manipulation.
• Planning the engagement of Change Agents and the Internal Communications can accelerate the Adoption of
Change by 80%, compared to ‘‘no planning’’ (Remark 5). This estimation of the value of planning, reveals the ‘‘size
of improvement’’ using net-centric strategies for the adoption of change, within the frame of Organizational Network
Analysis (ONA). On the other hand, in case no plan is available (change agents are randomly engaged and agents
communicate randomly), what is the most beneficial choice: Engaging the change agents centrally or communicating
with preferential selection? The answer follows from Figs. 3 and 7. Central engagement of change agents may result
in up to 55% improvement (Fig. 3), while communicating with preferential selection may result in up to 15%
improvement (Fig. 7). The advantage of central engagement over preferential communication in this context, is
stronger if there are few change agents in the network (Appendix SM III).
• Small Increase of the Slicing Parameter results in significant acceleration of the Adoption of Change. Moreover,
this fact is ‘‘robust’’ to variations of the other conditioning factors (Remark 6). However, increasing the slicing size
should not result in crossing the ‘‘confidence bound’’ (Assumption 2). Therefore, the change agents should ‘‘adapt’’
the slice size to the ‘‘confidence bound’’, so that Change is ‘‘digested’’ (Definition 5).

Real-world Implications for Change management:

• As highly influential employees are not detectable from the formal organizational chart alone, Organizational Network
Analysis (ONA) is useful to identify the most suitable change agents, and moreover to uncover the communication
patterns resisting the adoption of change.
• As the realization of change takes place locally at the micro level, the ‘‘Bottom-Up’’ agent-based modeling approach
is recommended.
• Network Theory is expected to improve the inner workings of socio-economic organizations in a measurable
way [84–87]. Hence, the analysis of Social Interactions [5] in relation to Organizational Behavior and Attitude Change
has a clear added value and offers competitive advantage.

Future Work
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 13

• The temporal order of interactions-selections among the agents, may condition significantly the actual influence
paths within the network [88–92]. In this first exploratory work, this information is included in ‘‘aggregated
form’’ in the weights wλκ . In case we have the information cornering when and in which order the interactions-
selections occurred, we may not discard the timing and order of interactions-selections. In such a case, our model
should be extended to a higher-order network model [93–95], capturing not only who is connected with whom
(network topology) and the frequency of interactions (weights), but moreover the temporal order of interactions (temporal
networks).
• In case of heterodox beliefs or radical novelties or eccentric behavior, as for example the spread of high-risk social
movements or avant garde fashions or unproven technologies, the bounded confidence of the agents can be extremely
low. The key point here is that the adoption of unusual behaviors requires social reinforcement. Hence, the spread
of the targeted attitude may require social affirmation from multiple sources, due to the weak ties of influence,
which impede the diffusion process significantly [64,96].

Acknowledgments

We thank Professors Eugenia Petridou, George Metakides, Dimitris Kugiumtzis, Paul Spirakis, Simon Bensasson, Chronis
Moyssiadis, Nikos Farmakis and Doctor Christos Varsakelis for interesting discussions. We thank Professor H. Eugene
Stanley and the anonymous reviewers, whose constructive criticism have significantly improved the presentation of this
work. Results presented in this work have been produced using the AUTH Compute Infrastructure and Resources. We
appreciate the highly efficient technical support for the computations by the members of the Scientific Computing Office
(IT AUTH).

Appendix A. Supplementary data

Supplementary material related to this article can be found online at https://doi.org/10.1016/j.physa.2019.121385.

References

[1] S. Galam, Sociophysics: a personal testimony, Physica A 336 (1–2) (2004) 49–55, http://dx.doi.org/10.1016/j.physa.2004.01.009.
[2] C. Castellano, S. Fortunato, V. Loreto, Statistical physics of social dynamics, Rev. Modern Phys. 81 (2) (2009) 591–646, http://dx.doi.org/10.1103/
RevModPhys.81.591.
[3] J. Lorenz, H. Rauhut, F. Schweitzer, D. Helbing, How social influence can undermine the wisdom of crowd effect, Proc. Natl. Acad. Sci. USA 108
(22) (2011) 9020–9025, http://dx.doi.org/10.1073/pnas.1008636108.
[4] P. Mavrodiev, C.J. Tessone, F. Schweitzer, Quantifying the effects of social influence, Sci. Rep. 3 (1360) (2013) 1–6, http://dx.doi.org/10.1038/
srep01360.
[5] F. Schweitzer, Sociophysics, Phys. Today 71 (2) (2018) 40–46, http://dx.doi.org/10.1063/PT.3.3845.
[6] P. Clifford, A. Sudbury, A model for spatial conflict, Biometrika 60 (3) (1973) 581–588, http://dx.doi.org/10.1093/biomet/60.3.581.
[7] R. Holley, T. Liggett, Ergodic theorems for weakly interacting infinite systems and the voter model, Ann. Probab. 3 (4) (1975) 643–663,
http://www.jstor.org/stable/2959329 (day accessed: 15-May-2018.
[8] S. Galam, Minority opinion spreading in random geometry, Eur. Phys. J. B 25 (4) (2002) 403–406, http://dx.doi.org/10.1140/epjb/e20020045.
[9] X. Castelló, V.M. Eguíluz, M.S. Miguel, Ordering dynamics with two non-excluding options: bilingualism in language competition, New J. Phys.
8 (12) (2006) 308, http://dx.doi.org/10.1088/1367-2630/8/12/308.
[10] G. Deffuant, D. Neau, F. Amblard, G. Weisbuch, Mixing beliefs among interacting agents, Adv. Complex Syst. 3 (01n04) (2000) 87–98,
http://dx.doi.org/10.1142/S0219525900000078.
[11] G. Weisbuch, G. Deffuant, F. Amblard, J.-P. Nadal, Meet, discuss, and segregate!, Complexity 7 (3) (2002) 55–63, http://dx.doi.org/10.1002/cplx.
10031.
[12] R. Hegselmann, U. Krause, Opinion dynamics and bounded confidence models, analysis, and simulation, J. Artifical Soc. Soc. Simul. 5 (3) (2002)
http://jasss.soc.surrey.ac.uk/5/3/2html (day accessed: 15-May-2018.
[13] G. Deffuant, F. Amblard, G. Weisbuch, T. Faure, How can extremism prevail? A study based on the relative agreement interaction model, J.
Artif. Soc. Soc. Simul. 5 (4) (2002) http://jasss.soc.surrey.ac.uk/5/4/1html (day accessed: 15-May-2018.
[14] F. Amblard, G. Deffuant, The role of network topology on extremism propagation with the relative agreement opinion dynamics, Physica A 343
(2004) 725–738, http://dx.doi.org/10.1016/j.physa.2004.06.102.
[15] E. Kurmyshev, H. Juárez, R. González-Silva, Dynamics of bounded confidence opinion in heterogeneous social networks: Concord against partial
antagonism, Physica A 390 (16) (2011) 2945–2955, http://dx.doi.org/10.1016/j.physa.2011.03.037.
[16] M. Jalili, Social power and opinion formation in complex networks, Physica A 392 (4) (2013) 959–966, http://dx.doi.org/10.1016/j.physa.2012.
10.013.
[17] M. Jalili, Effects of leaders and social power on opinion formation in complex networks, Simulation 89 (5) (2013) 578–588, http://dx.doi.org/
10.1177/0037549712462621.
[18] Y. Gandica, M. del Castillo-Mussot, G.J. Vázquez, S. Rojas, Continuous opinion model in small-world directed networks, Physica A 389 (24)
(2010) 5864–5870, http://dx.doi.org/10.1016/j.physa.2010.08.025.
[19] T.V. Martins, M. Pineda, R. Toral, Mass media and repulsive interactions in continuous-opinion dynamics, Europhys. Lett. 91 (4) (2010)
48003–48008, http://dx.doi.org/10.1209/0295-5075/91/48003.
[20] P. Fan, H. Wang, W. Yu, P. Li, Y. Hu, Opinion dynamics in social networks with heterogeneous relationships, J. Inf. Comput. Sci. 11 (6) (2014)
1831–1841, http://dx.doi.org/10.12733/jics20103170.
[21] W. Quattrociocchi, G. Caldarelli, A. Scala, Opinion dynamics on interacting networks: media competition and social, Sci. Rep. 4 (4938) (2014)
http://dx.doi.org/10.1038/srep04938.
[22] K. Fan, W. Pedrycz, Emergence and spread of extremist opinions, Physica A 436 (2015) 87–97, http://dx.doi.org/10.1016/j.physa.2015.05.056.
[23] K. Fan, W. Pedryczb, Evolution of public opinions in closed societies influenced by broadcast media, Physica A 472 (2017) 53–66, http:
//dx.doi.org/10.1016/j.physa.2017.01.027.
14 E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385

[24] M. Pineda, G.M. Buendía, Mass media and heterogeneous bounds of confidence in continuous opinion dynamics, Physica A 420 (2015) 73–84,
http://dx.doi.org/10.1016/j.physa.2014.10.089.
[25] G. Weisbuch, G. Deffuant, F. Amblard, Persuasion dynamics, Physica A 353 (2005) 555–575, http://dx.doi.org/10.1016/j.physa.2005.01.054.
[26] X. Song, W. Shi, G. Tan, Y. Ma, Multi-level tolerance opinion dynamics in military command and control networks, Physica A 437 (2015)
322–332, http://dx.doi.org/10.1016/j.physa.2015.05.082.
[27] C. Jin, Y. Li, X. Jin, Political opinion formation: Initial opinion distribution and individual heterogeneity of tolerance, Physica A 467 (2017)
257–266, http://dx.doi.org/10.1016/j.physa.2016.10.025.
[28] M. Afshar, M. Asadpour, Opinion formation by informed agents, J. Artif. Soc. Soc. Simul. 13 (4) (2010) http://dx.doi.org/10.18564/jasss.1665.
[29] O. AskariSichani, M. Jalili, Influence maximization of informed agents in social networks, Appl. Math. Comput. 254 (2015) 229–239, http:
//dx.doi.org/10.1016/j.amc.2014.12.139.
[30] D. Watts, P. Dodds, Influentials, networks, and public opinion formation, J. Consum. Res. 34 (4) (2007) 441–458, http://dx.doi.org/10.1086/518527.
[31] K. Fan, W. Pedrycz, Opinion evolution influenced by informed agents, Physica A 462 (2016) 431–441, http://dx.doi.org/10.1016/j.physa.2016.06.
110.
[32] I. Antoniou, M. Reeve, V. Stenning, The information society as a complex system, J. UCS 6 (3) (2000) 272–288.
[33] J. Kotter, L. Schlesinger, Choosing strategies for change, Harv. Bus. Rev. (2008).
[34] R. Cross, S. Parise, L. Weiss, The role of networks in organizational change, McKinsey Q. (2007) https://www.mckinsey.com/business-functions/
organization/our-insights/the-role-of-networks-in-organizational-change (day accessed: 15-May-2018.
[35] M. Johnson-Cramer, S. Parise, R. Cross, Managing change through networks and values, Calif. Manage. Rev. 49 (3) (2007) 85–109.
[36] R. Cross, C. Ernst, B. Pasmore, A bridge too far? how boundary spanning networks drive organizational change and effectiveness, Organ. Dyn.
42 (2) (2013) 81–91, http://dx.doi.org/10.1016/j.orgdyn.2013.03.001.
[37] E.M. Rogers, Diffusion of Innovations, Free Press, 1962.
[38] R. Galavan, J. Murray, C. Markides, Strategy, Innovation, and Change: Challenges for Management, Oxford University Press, 2008.
[39] N. Dalton, J. Benson, Innovation and change in Japanese human resource management, Asia Pac. J. Human Resour. 40 (3) (2002) 345–362,
http://dx.doi.org/10.1177/1038411102040003257.
[40] J. Kenny, Effective project management for strategic innovation and change in an organizational context, Proj. Manage. J. 34 (1) (2003) 43–53,
http://dx.doi.org/10.1177/875697280303400106.
[41] D. Hegger, J. van Vliet, B. van Vliet, Niche management and its contribution to regime change: The Case of innovation in sanitation, Technol.
Anal. Strateg. Manage. 19 (6) (2007) 729–746, http://dx.doi.org/10.1080/09537320701711215.
[42] B. Michaelis, R. Stegmaier, K. Sonntag, Affective commitment to change and innovation implementation behavior: The role of charismatic
leadership and employees’ trust in top management, J. Change Manage. 9 (4) (2009) 399–417, http://dx.doi.org/10.1080/14697010903360608.
[43] M. Musteen, V. Barker, V. Baeten, The influence of CEO tenure and attitude toward change on organizational approaches to innovation, J. Appl.
Behav. Sci. 46 (3) (2010) 360–387, http://dx.doi.org/10.1177/0021886310361870.
[44] H. Volberda, F. Van Den Bosch, O. Mihalache, Advancing management innovation: Synthesizing processes, levels of analysis, and change agents,
Organ. Stud. 35 (9) (2014) 1245–1264, http://dx.doi.org/10.1177/0170840614546155.
[45] M. Vakola, l. Nikolaou, Attitudes towards organizational change: What is the role of employees’ stress and commitment?, Employee Relations
27 (2) (2005) 160–174, http://dx.doi.org/10.1108/01425450510572685.
[46] D. Giauque, Attitudes toward organizational change among public middle managers, Public Pers. Manage. 44 (1) (2014) 70–98, http://dx.doi.
org/10.1177/0091026014556512.
[47] W. Jager, F. Amblard, A dynamical perspective on attitude change (2004) https://pdfs.semanticscholar.org/77f4/
142ef3fd9bd33c96a5ef72391151c4a6b6c7.pdf (day accessed: 15-May-2018).
[48] J. Battilana, T. Casciaro, Change. Agents, Networks, Change agents networks and institutions: A contingency theory of organizational change,
Acad. Manage. J. 55 (2) (2012) 381–398, http://dx.doi.org/10.5465/amj.2009.0891.
[49] E. Aronson, J. Turner, M. Carlsmith, Communicator credibility and communication discrepancy as determinants of opinion change, J. Abnorm.
Soc. Psychol. 67 (1) (1963) 31–36, http://dx.doi.org/10.1037/h0045513.
[50] E. McGinnies, C. Ward, Better liked than right: Trustworthiness and expertise as factors in credibility, Pers. Soc. Psych. Bull. 6 (3) (1980)
467–472, http://dx.doi.org/10.1177/014616728063023.
[51] M. Jalili, M. Perc, Information cascades in complex networks, J. Complex Networks (2017) http://dx.doi.org/10.1093/comnet/cnx019.
[52] S. Borgatti, Centrality and network flow, Social Networks 27 (1) (2005) 55–71, http://dx.doi.org/10.1016/j.socnet.2004.11.008.
[53] M. Kitsak, L.K. Gallos, S. Havlin, F. Liljeros, L. Muchnik, H.E. Stanley, H.A. Makse, Identification of influential spreaders in complex networks,
Nat. Phys. 6 (2010) 888–893, http://dx.doi.org/10.1038/nphys1746.
[54] A. Garas, F. Schweitzer, S. Havlin, A k-shell decomposition method for weighted networks, New J. Phys. 14 (2012) 1–14, http://dx.doi.org/10.
1088/1367-2630/14/8/083030.
[55] T. Vogel, M. Wanke, Attitudes and Attitude Change, Psychology Press, 2002.
[56] C. Jung, Collected Works of CG Jung, Volume 6: Psychological Types, Princeton University Press, 1971.
[57] J. Golbeck, D. Hansen, A method for computing political preference among twitter followers, Social Networks 36 (Special Issue on Political
Networks) (2014) 177–184, http://dx.doi.org/10.1016/j.socnet.2013.07.004.
[58] J. Hendrickx, Order preservation in a generalized version of krause’s opinion dynamics model, Physica A 387 (21) (2008) 5255–5262,
http://dx.doi.org/10.1016/j.physa.2008.05.018.
[59] H. Liang, Y. Yang, X. Wang, Opinion dynamics in networks with heterogeneous confidence and influence, Physica A 392 (9) (2013) 2248–2256,
http://dx.doi.org/10.1016/j.physa.2013.01.008.
[60] W. Gladstones, M. Regan, R. Lee, Division of attention: The single-channel hypothesis revisited, Q. J. Exp. Psychol. [A] 41 (1) (1989) 1–17,
http://dx.doi.org/10.1080/14640748908402350.
[61] H. Pashler, Dual-task interference in simple tasks: Data and theory, Psychol. Bull. 116 (2) (1994) 220–244, http://dx.doi.org/10.1037/0033-
2909.116.2.220.
[62] R. Indermühle, S. Troche, T. Rammsayer, Personality and the psychological refractory period: No evidence for an extraversion-or
intelligence-related effect, Can. J. Behav. Sci./Revue Can. sci. Comportement 43 (3) (2011) 214–221, http://dx.doi.org/10.1037/a0024156.
[63] D. Centola, V. Eguíluz, M. Macy, CaScade dynamics of complex propagation, Physica A 374 (1) (2007) 449–456, http://dx.doi.org/10.1016/j.physa.
2006.06.018.
[64] D. Centola, M. Macy, Complex contagions and the weakness of long ties, Am. J. Sociol. 113 (3) (2007) 702–734, http://dx.doi.org/10.1086/521848.
[65] D. Knuth, Two notes on notation, Amer. Math. Monthly 99 (5) (1992) 403–422, http://dx.doi.org/10.2307/2325085.
[66] E. Ioannidis, N. Varsakelis, I. Antoniou, False beliefs in unreliable knowledge networks, Physica A 470 (2017) 275–295, http://dx.doi.org/10.
1016/j.physa.2016.11.070.
[67] E. Ioannidis, N. Varsakelis, I. Antoniou, Communication policies in knowledge networks, Physica A 492 (2018) 360–374, http://dx.doi.org/10.
1016/j.physa.2017.09.078.
E. Ioannidis, N. Varsakelis and I. Antoniou / Physica A 528 (2019) 121385 15

[68] E. Ioannidis, N. Varsakelis, I. Antoniou, Experts in knowledge networks: Central positioning and intelligent selections, Physica A 509 (2018)
890–905, http://dx.doi.org/10.1016/j.physa.2018.06.003.
[69] Y.-Y. Liu, J.-J. Slotine, A.-L. Barabási, Controllability of complex networks, Nature 473 (2011) 167–173, http://dx.doi.org/10.1038/nature10011.
[70] Y. Zhang, A. Garas, F. Schweitzer, Value of peripheral nodes in controlling multilayer scale-free networks, Phys. Rev. E 93 (1) (2016) 012309,
http://dx.doi.org/10.1103/PhysRevE.93.012309.
[71] S. Borgatti, M. Everett, J. Johnson, Analyzing Social Networks, SAGE Publications, 2013.
[72] J. Goñi, M.P. van den Heuvel, A. Avena-Koenigsberger, N.V. de Mendizabal, R.F. Betzel, A. Griffa, P. Hagmann, B. Corominas-Murtra, J.-P. Thiran,
O. Sporns, Resting-brain functional connectivity predicted by analytic measures of network communication, Proc. Natl. Acad. Sci. USA 111 (2)
(2014) 833–838, http://dx.doi.org/10.1073/pnas.1315529111.
[73] R. Cross, A. Parker, The Hidden Power of Social Networks: UnderstandIng how Work Really Gets Done in Organizations, Harv. Bus. Press, Boston
MA, 2004.
[74] E. Lazega, The Collegial Phenomenon: The Social Mechanisms of Cooperation Among Peers in a Corporate Law Partnership, Oxford University
Press, 2001.
[75] L.C. Freeman, Datasets (2018) http://moreno.ss.uci.edu/data.html#krebs (day accessed: 15-May-2018).
[76] T. Opsahl, Datasets (2018) http://toreopsahl.com/datasets/#Cross_Parker (day accessed: 15-May-2018).
[77] L.C. Freeman, Datasets (2018) http://moreno.ss.uci.edu/data.html#lazega (day accessed: 15-May-2018).
[78] S. Havlin, D. Ben-Avraham, Diffusion in disordered media, Adv. Phys. 36 (6) (1987) 695–798, http://dx.doi.org/10.1080/00018738700101072.
[79] L. Skarpalezos, A. Kittas, P. Argyrakis, R. Cohen, S. Havlin, Anomalous biased diffusion in networks, Phys. Rev. E 88 (1) (2013) 012817,
http://dx.doi.org/10.1103/PhysRevE.88.012817.
[80] D. Brass, A Social Network Perspective on Organizational Citizenship Behavior, The Oxford Handbook of Organizational Citizenship Behavior,
Oxford University Press, 2018, pp. 317–330,
[81] S. Feld, Why your friends have more friends than you do, Am. J. Sociol. 96 (6) (1991) 1464–1477, http://dx.doi.org/10.1086/229693.
[82] C. McGrath, D. Krackhardt, Network conditions for organizational change, J. Appl. Behav. Sci. 39 (3) (2003) 324–336, http://dx.doi.org/10.1177/
0021886303258267.
[83] R. Iyengar, C. Van den Bulte, Opinion leadership and social contagion in new product diffusion, Mark. Sci. 30 (2) (2010) 195–212, http:
//dx.doi.org/10.1287/mksc.1100.0566.
[84] D. Brass, Being in the right place: A structural analysis of individual influence in an organization, Adm. Sci. Q. 29 (4) (1984) 518–539,
http://dx.doi.org/10.2307/2392937.
[85] D. Brass, M. Burkhardt, Potential power and power use: An investigation of structure and behavior, Acad. Manage. J. 36 (3) (1993) 441–470,
http://dx.doi.org/10.5465/256588.
[86] D. Brass, J. Galaskiewicz, H. Greve, W. Tsai, Taking stock of networks and organizations: A multilevel perspective, Acad. Manage. J. 47 (6) (2004)
795–817, http://dx.doi.org/10.5465/20159624.
[87] S. Borgatti, A. Mehra, D. Brass, G. Labianca, Network analysis in the social sciences, Science 323 (5916) (2009) 892–895, http://dx.doi.org/10.
1126/science.1165821.
[88] H. Lentz, T. Selhorst, I. Sokolov, Unfolding accessibility provides a macroscopic approach to temporal networks, Phys. Rev. Lett. 110 (11) (2013)
118701, http://dx.doi.org/10.1103/PhysRevLett.110.118701.
[89] I. Scholtes, N. Wider, R. Pfitzner, A. Garas, C. Tessone, F. Schweitzer, CaUsality-driven slow-down and speed-up of diffusion in non-Markovian
temporal networks, Nature Commun. 5 (5024) (2014) 1–9, http://dx.doi.org/10.1038/ncomms6024.
[90] M. Rosvall, A. Esquivel, A. Lancichinetti, J. West, R. Lambiotte, Memory in network flows and its effects on spreading dynamics and community
detection, Nature Commun. 5 (4630) (2014) 1–13, http://dx.doi.org/10.1038/ncomms5630.
[91] R. Lambiotte, V. Salnikov, M. Rosvall, Effect of memory on the dynamics of random walks on networks, J. Complex Networks 3 (2) (2015)
177–188, http://dx.doi.org/10.1093/comnet/cnu017.
[92] T. Peixoto, M. Rosvall, Modelling sequences and temporal networks with dynamic community structures, Nature Commun. 8 (582) (2017) 1–12,
http://dx.doi.org/10.1038/s41467-017-00148-9.
[93] I. Scholtes, N. Wider, A. Garas, Higher-order aggregate networks in the analysis of temporal networks: path structures and centralities, Eur.
Phys. J. B 89 (91) (2016) 1–15, http://dx.doi.org/10.1140/epjb/e2016-60663-0.
[94] I. Scholtes, When is a network a network?: Multi-order graphical model selection in pathways and temporal networks, in: KDD 2017:
Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Halifax, Nova Scotia, Canada, 2017,
http://dx.doi.org/10.1145/3097983.3098145.
[95] R. Lambiotte, M. Rosvall, I. Scholtes, Understanding complex systems: From networks to optimal higher-order models, Phys. Soc. (2018) 1–8,
arXiv:1806.05977.
[96] D. Centola, The spread of behavior in an online social network experiment, Science 329 (5996) (2010) 1194–1197, http://dx.doi.org/10.1126/
science.1185231.

You might also like