An Agent-Based Model for Emergent Opponent Organisations

(1)

An Agent-Based Model for Emergent Opponent

Organisations

Koen van der Zwet

Student number: 11826053

University of Amsterdam Faculty of Science

Thesis Master Information Studies: Business Information Systems

Supervisor: prof. dr. T.M. van Engers

Examiner: dr. R. Quax

August 16, 2018

(2)

Abstract

Terrorists, insurgents and criminals are typical opponents of governments as they threat to destabi-lize societies and endanger democracy and peace. The emergent behaviour of opponent organisations is a challenging topic in academic research and practice. The covertness of illicit operations hinders the possibility of extensive and detailed empirical research in actual networks. Current efforts to grasp “collective” opponent behaviour perceive a static system or attempt to isolate the effects of a policy. In addition to current efforts to grasp the ongoing complexities qualitative development of a quantitative experimental design is required to understand the complexities that yield opponent behaviour.

The goal of this thesis is to acquire insight in the complexity of the emergent behaviour of opponent organisations by multidisciplinary analysis of the behaviour of individuals. To deal with the covertness of opponent activity, we created an agent-based model which provides an opportunity to experiment and uncover the effects by complex mechanisms of the system to law enforcement policies. Two re-search questions have been answered through literature study, modelling and experimentation. The first question is formulated with respect to the modelling of opponent behaviour: Which individual charac-teristics are essential to include in an agent model of opponent individuals? The second focuses on the purpose of this model: How does the behaviour of individuals yield emergent “collective” opponent behaviour?

The emergence of different opponent organisational structures are dependent on the feedback mech-anisms of opportunity, threat and competition. The conceptual model embodies the complexity. The psychological, physical and organisational aspects are essential to model agents that mimic the charac-teristics of opponent behaviour and are adaptive to the complex mechanisms.

Experiments were conducted to analyse the computational model and demonstrated the adaptive opponent behaviour. The feedback loop mechanisms of law enforcement activity, social and oppo-nent opportunity and competition simultaneously influence processes at the micro-level. The emergent behaviour can change as individuals adapt to changes in their environment. Detailed modelling and experimenting can help to understand the ongoing complexities that yield the emergent opponent be-haviour.

Keywords. Opponent behaviour, Opponent networks, Multidisciplinary, Complex adaptive sys-tems, Agent-based modelling

(3)

Preface

This thesis reports the results of a study into opponent behaviour as part of a graduation internship at TNO and a graduation trajectory for the Master Information Studies (Business Information Systems) at the University of Amsterdam.

I would like to express my gratitude to my two mentors at TNO, Ana Isabel Barros and Bob van der Vecht and my mentor at the University of Amsterdam, Tom van Engers. Their supervision inspired me to new insights and encouraged me to study various interesting domains.

Lastly, I would also like to thank my parents and girlfriend for their encouragement and support during my bachelor and master study and the graduation process.

(4)

List of Figures

1 Complex Adaptive System . . . 1

2 Radicalisation of Opinion and Action (McCauley & Moskalenko, 2014) . . . 4

3 Adaptive opponent behaviour. . . 6

4 Adaptive emergent opponent behaviour . . . 7

5 Small snapshot of the visualisation of the modelled environment . . . 11

6 The amount of organisations by different intensities of law enforcement activity . . . 15

7 Face validation . . . 24

8 BDI process . . . 28

9 Simulation process . . . 30

10 Opponent process . . . 31

11 Organisation plan . . . 35

12 Simulations with different amount of law enforcement intensity. . . 38

13 Simulations with different target strategies by law enforcement . . . 40

14 Simulations with different values for the growth of social opportunity . . . 43

15 Simulations with various amounts of opponent opportunity . . . 45

List of Tables

1 Modelling documentation. . . 23

2 Entities, state variables and scales of the model . . . 29

3 Input variables of the system to create scenarios. Default values are set to create an environment of competition and an incentive to yield emergent behaviour. Competition is created as the amount of people times the environment costs is set higher than the combined amount of jobs and opponent resources. The incentive of emergent behaviour is set by a positive benefit for cooperation and organisation.. . . 36

4 Initializing Experiment 1 . . . 37

(6)

Introduction

Terrorists, insurgents and criminals are typical opponents of governments as they threat to destabilize so-cieties and endanger democracy and peace (Makarenko, 2004). Many studies have associated the decrease of social, political and economic development with the occurrence of opponent behaviour (Boerman & Stoffers, 2017). Governments establish law enforcement agencies in order to counteract these unlawful activities. The way the terrorist, insurgents and criminals operate to cope with the policies of these law enforcement agencies typify opponent behaviour and causes an environment of conflict (Schelling, 1971). While policing actions can have a dissuasive effect on opponent groups, they can also cause undesired effects such as retaliation, escalation and displacement (Devia & Weber, 2013; Spapens, 2006). Effective policies to counter opponent behaviour should account these negative effects and reduce the efficacy of opponents. Some measures by law enforcement to disturb the effectiveness of opponent organisations and networks appear to be relatively ineffective (Duijn, 2016). In order to disrupt opponent behaviour it is essential to understand how these organisations behave and why they grow, decline, merge or split. As law enforcement demands effective policies, analysis methods for opponent behaviour are required.

Interaction between the emergence of opponent behaviour, law enforcement policies and the society carry several complexities (Spapens, 2010). This interdependency is best described as a complex adaptive system (CAS) (Figure 1). The core attributes of CAS are self-organising, emergence, feedback loops, adaptive realignment and non-linearity (Eidelson, 1997).

Figure 1: Complex Adaptive System

The parallels between these attributes and opponent behaviour indicate the on going complexity. On a micro-level individual people (from this point referred to as individuals) inter-act and inter-act relatively autonomous. The individu-als are embedded in networks of kinship, social connections, cooperation and financial relation-ships (D. A. Bright, Greenhill, Ritter, & Morselli, 2015). These structures enable the individuals to communicate and cooperate autonomously, which indicates the self-organising function of the sys-tem. The emergent behaviour that the individu-als yield through self-organisation are the second

attribute of CAS. Cooperative actions of individuals produce synergy (Michalak, Rahwan, Skibski, & Wooldridge, 2015). Opponent organisation structures enable their typical competencies such as aggres-sive and financial activities (Ozgul, 2014). Historically, studies depicted opponent organisations as hi-erarchical and well-structured (Sheptycki, Jaffel, & Bigo, 2011). Recently, opponent structures such as terrorist and criminal organisations are described as loose-connected networks, which are based on indi-vidual interactions (Morselli, Gigu`ere, & Petit, 2007; Bichler, Malm, & Cooper, 2017). Through synergy

(7)

the emergent behaviour of these opponent groups create oscillating effects, which are different than the sum of their individual actions (Duijn, 2016).

Through emergent behaviour, individuals interact with the environment. This environment con-sists of the society, economy and government (Spapens, 2010). The opponent behaviour yields phenom-ena that are observable in the society, such as a growth of criminality. The environment reacts upon these phenomena and poses positive and negative feedback loops towards the micro-level, which are the third attribute of CAS. Self-organisation enables adaptive responses by the individuals. Positive feedback loops trigger opponent behaviour as they offer opportunities (opportunities that encourage opponent behaviour are referred to as opponent opportunities). Eidelson (1997) describes this as the intrinsic dynamics of self-organisation, as they develop a motivation for specific actions. These dynamics will appear to be essential to understand opponent behaviour. The increase of law enforcement activities are a typical ex-ample of a negative feedback loop towards opponent behaviour (Duijn, 2016). These negative feedback loops constrain the linear growth of the system.

We argue that in order to understand opponent behaviour, it is essential to analyse the system from different perspectives with a multidisciplinary approach. Current efforts to grasp “collective” oppo-nent behaviour perceive a static system or attempt to isolate the effects of a policy. In addition to current efforts to grasp the ongoing complexities qualitative (Spapens, 2006), development of a quantitative ex-perimental design is required to understand the complexities that yield opponent behaviour.

Thesis focus

The goal of this thesis is to acquire insight in the complexity of the emergent behaviour of opponent organisations by multidisciplinary analysis of the behaviour of individuals. The retrieved insights should lead to better understanding of the effectiveness of different policies to counter opponent behaviour, by accounting the positive and negative feedback loops of the system. The covert and complex nature of opponent organisations hinders the possibility to conduct empirical based research, especially of currently active organisations. Therefore a simulation framework is proposed that focuses on the self-organising opponents, their connections through networks, the emergent opponent behaviour and the interaction with the environment. On purpose environment is very abstract, only with elements such as “law enforcement actions”, “social opportunity” and “opponent opportunity”.

The principles of the framework should be translatable to study specific cases and create insight into the effects of specific environmental aspects upon emergent behaviour. For example in studies to-wards the emergence of criminal activities or the development of insurgency groups in a society with specific characteristics. The framework is modeled using agent-based modelling (ABM). The thesis has a bottom-up approach that starts at the mind of individual opponents.

Two research questions are drafted to cover the described knowledge gaps. The first question is formulated with respect to the modelling of opponent behaviour. The second focuses on the purpose of this model:

(8)

• Which individual characteristics are essential to include in an agent model of opponent individuals?

• How does the behaviour of individuals yield emergent “collective” opponent behaviour?

This thesis consists of three sections. Section 1 describes the modelling of opponent behaviour and concentrates on the first research question. Within this section, a literature study leads to a conceptual framework that comprises the complexities towards opponent behaviour. Section2 focuses on model validation and experimental results. Finally, Section3 contains a discussion of the thesis results and recommendation for future work. Two protocols of Grimm et al. (2010, 2014) are applied to plan, perform and document the model development. The TRACE protocol is applied to plan and document the modelling elements (Grimm et al., 2014) (Appendix A). The agent-based model aspects are described following the ODD format (Grimm et al., 2010) (Appendix B).

1 Modelling Opponent Behaviour

This section describes an approach to develop a model of opponent behaviour. In Section1.1an extensive literature study of various disciplines lead to a conceptual framework focused on opponent behaviour. Section1.2 explains why the methods of multi-agent systems and agent-based modelling are used to computationally model the concepts of opponent behaviour. Finally in Section1.3the model is described by application of the ODD protocol.

1.1 Conceptual Framework

This literature study focuses on understanding opponent behaviour by analysis of different disciplines. Three different aspects will be discussed. Firstly, the psychological aspects that influence the motivations of individual behaviour are discussed. The next perspective are the differences of individuals in terms of social and personal capabilities. The third focus is the adaptive responses of opponent individuals. Finally we will consolidate these three aspects in a conceptual framework in order to derive insight in how these individual behaviours can influence emergent behaviour of opponent organisations.

1.1.1 Rational behaviour

Opponent activities serve a certain purpose. Makarenko (2004) distinguishes criminal, terrorist and in-surgent organisations by their goals. Whereas drug trafficking, fraud and money laundering are typified as criminal organisations motivated by the desire to obtain money, terrorism and insurgent organisations are typified as ideology driven (Morselli et al., 2007). Individual motivations to start a specific form of opponent behaviour can be similar, but also quite different (Borum, 2007). The coherence of motives, decisions and actions is complex and studied by many scholars in various disciplines. The different processes to develop of goals and motivations by individuals adds to this complexity.

(9)

Rational choice theory offers an economic approach to explain the selection of actions (Cornish & Clarke, 1987). According to this theory, actions of individuals are guided by deliberate decision making in every circumstance. The theory describes opponents as conscious people who weigh their actions by expected benefits and costs. These benefits and costs can be represented by tangible assets such as money, but also abstract results like the spread of fear (Enders & Su, 2007). The benefits and cost of actions pose a trade-off towards the individuals as they attempt to fulfill their desires. This trade-off enables individuals to select the actions with the highest utility.

However, the claim of complete rationality has been rejected by multiple other studies. Fore-most, bounded rationality limits the capacity of people to contemplate the possible consequences of every possible action (Simon, 1955). Also, situations in which emotions can overtake rationality can lead to impulsive acts (Bosse, Gerritsen, & Treur, 2007). Within terrorist organisations it appears that the moti-vations of those with desperate, irrational thoughts conflict with the motimoti-vations of others within in the organisation (Caplan, 2006). The repercussions by law enforcers towards opponent behaviour and the negative effects of the actions on the society strengthens the impression of irrational behaviour. Un-derstanding the development of motivations for opponent behaviour is essential in order to grasp the emergent behaviour.

With their study of radicalisation, McCauley and Moskalenko (2008) set out twelve, less impul-sive, emotional mechanisms that led to radical behaviour. They argue for a distinction between individual, group and mass mechanisms. On a individual level, personal victimization, political grievance, gradual attraction towards groups and affection for power are mechanisms that yield radical feelings. On a group level, an extremity shift of like-minded groups or extreme cohesion can develop an atmosphere in which radical opinions are shared easily. At a mass level, hate and conflict create a breeding ground for extreme opinions. On an individual level, these emotions can generate radical opinions that justify the activities of opponent organisations. McCauley et al. (2008) use a pyramid (Figure 2) as an analogy to describe this phenomenon of opinion change. Active members form the top of the pyramid and sympathizers form the lower level. Supporters can be attracted to the higher level by active members, typically referred to as recruiting, or choose to join the group themselves in order to reach personal desires.

Figure 2: Radicalisation of Opinion and Action (McCauley & Moskalenko, 2014)

The process of justification of action creates two separate processes of radicalisation; developing opinions and performing actions. The second process of the adoption of radical actions is embraced by

(10)

the second pyramid ofFigure 2. McCauley et al. (2017) argue that both processes are steered by rational choice behaviour. People can adopt radical opinions through negative experiences or through social connections. These radical opinions can make radical actions rational. The disadvantages (costs) in a previous stage are diminished as they perceive a protection by their by their group, or a decline of the potential backlash by law enforcement (Ganor, 2008). Additionally, radical opinions can mitigate the feeling of immorality of radical actions (Bonger, 2015).

Ganor et al. (2008) describe the extent of public support for an organisation as an important factor to ensure the existence of the organisation. Repercussions by the government towards radical actions can create a breeding ground for justification of radical opinions, which can lead to an increase of people with radical opinions (Anderson Jr, 2011). This is referred to as the boomerang effect (R´ıos, 2013). The complex nature of opponent behaviour is emphasized by these positive and negative feedback loops.

While the described radicalisation process originally focused on insurgent and terrorist groups, some interesting parallels can be made towards the process of individuals becoming opponents in gen-eral. Routine activity theory emphasizes the situational aspects of criminal activities (Clarke & Felson, 1993). The theory focuses on the likelihood of opponent activity in situations with a likely opponent, who encounters a valuable opportunity with a low risk of possible repercussions. This correlates with rational choice theory and the process of radicalisation. The trade-off between opportunity and threat is typical for rational choice theory. The lack of social opportunities, the presence of profitable opponent opportunities or connections with criminal organisations can develop a breeding ground for opponent opinions (Carrington & Scott, 2011). Once the opponent opportunity outweigh the social opportunity, the individual can “radicalise” into an opponent and adopt radical actions.

Radical individuals can also experience a process of disengagement of radical actions or opin-ions. Windisch (2016) conducted a systematic literature review of research into disengagement and de-radicalisation from violent extremism. Disengagement is the process in which individuals do not actively participate in opponent organisation activities. With deradicalisation, individuals change their opinion about violent actions. Windisch et al. (2016) indicated multiple factors for disengagement from violent, non-violent, ideological and non-ideological groups, typified as pull and push effects. Disillusion, vi-olence and infighting were found as important factors that push radicals out of opponent organisations. Pull factors are the effects that attract individuals out of the organisations. Most of the pull factors, like relationships, maturation and establishment of family, were found in the social environment. Pull factors like employment and education are typical economic and social opportunities. These are the reversed processes of isolation and disconnection from society. Threats of repercussions or social isolation can prevent individuals from leaving illicit organisations (Bovenkerk, 2010).

Burt (2013) describes personality characteristics such as thrill-seeking and self-control that can be linked to the motivation to commit opponent actions. These characteristics are difficult to observe or quantify in practice. Future studies could investigate the influence of other individual types and charac-teristics upon the expected results by rational choice behaviour. For example, trill-seeking individuals

(11)

could perform opponent actions more easily, or identity-seeking individuals could have a lower barrier to join opponent organisations.

The above literature research identified important elements for the modelling of opponent be-haviour. In particular the rational approach of behaviour and the adaptation of individuals to changes in the environment. The opportunities individuals face, their state of radicalisation and the way they handle trade-offs as they adapt their actions in order to maximize their utility, form the basis of the proposed modelling of individuals.

1.1.2 Individual characteristics

During their efforts to achieve their goals, individuals are dependent on their resources. These resources differentiate the observable situation of different individuals. Bichler et al. (2017) conducted a systematic literature review for social network analysis (SNA) of drug trade networks structures. The literature emphasized the importance of human and social capital to successful opponents. Human capital are the individual knowledge, skills and resources that enable capabilities. Social capital are the social ties between individuals that enable them to contact others, share information and initiate cooperation.

Both Carley (2003) and Bright et al. (2015) emphasized the importance of multiple networks that embed the relationships between individuals. Social network analysis (SNA) is a frequently applied methodology to identify important links and nodes by centrality and analyse the resilience of a network by different metrics (Hamers, Husslage, & Lindelauf, 2011; Michalak et al., 2015; Duijn, 2016; D. A. Bright et al., 2015).

The different positions in these networks cause that some individuals are better placed than others consequently play a more central role. Bichler et al. (2017) describe brokers and hubs as two im-portant roles in networks. Brokers possess a vital position, with regard to the flow of information between two networks. They have the ability to connect individuals that would be separated otherwise. Hubs are the individuals who maintain a lot of connections, that offer several opportunities for cooperation.

Figure 3: Adaptive opponent behaviour

As described, individuals intent to max-imize the utility of their actions according to ra-tional choice theory (Cornish & Clarke, 1987). Including social and human capital is essential to opponents for proper rational decision making (Enders & Su, 2007; Calv´o-Armengol & Jackson, 2007). Figure 3summarizes the reasoning of op-ponent individuals. Opop-ponents perceive opportu-nities and threats through events in their

environ-ment. Simultaneously they maintain a certain level of radical opinion towards those opportunities and threats. These influence their rational choice behaviour, which they use to select the actions that maxi-mize the value of their human and social capital.

(12)

The theories of social and human capital provide essential elements to understand the complexity of opponent behaviour through self-organisation and adaptive responses. Modelling autonomous and reactive agents is enabled by implementation of individual capabilities and network functions. Social capital is an important factor in the previously described process of radicalization and cooperation. As capabilities, communication and cooperation involve utility, they require rational decision making by individuals.

1.1.3 Dynamics of “collective” emergent behaviour

The underlying psychological, rational and physical processes are incorporated at the micro-level of the conceptual framework (Figure 4). Collectively opponent individuals yield emergent behaviour that con-sists of individual and organised opponent activities. Opponent organisations do not exist without a purpose as they face opposing forces (Spapens, 2010). In the light of rational choice theory, organisations aim to achieve specific goals effectively. Organisation theory focuses on the deliberate and emergent strategies of cooperation (Mintzberg & Waters, 1985). In order to grasp the function and structure of cooperative opponent activities, a perspective on organisation theory is required.

Figure 4: Adaptive emergent opponent behaviour

Many efforts to dissect different forms of opponent organisations can be found in lit-erature. A working definition by the European Commission defines that criminal organisations are groups of more than two people, that exercise serious criminal offenses over a prolonged period of time, motivated by the pursuit of profit and/or power (Abadinsky, 2012). Some speak of the term organisations as a carte blanche to oversimplify the complexity of the ongoing processes (Ligon, Simi, Harms, & Harris, 2013). Over the years, scientists used different analogies to describe the

typical opponent structures. Originally a corporate model was used to describe the functionality of organ-ised opponent activity (Cressey, 1972). It was considered that explicit ranks, rules and division of labour exists within pyramid shaped organisations. Currently, studies commonly refer to the term network to describe opponent organisations (Arquilla & Ronfeldt, 2001; Bichler et al., 2017).

With the study into opponent organisation structures, Framis (2011) and Ligon et al. (2013) found a continuum of organisational sophistication. Respectively this continuum from mechanistic to organic is characterized by a more hierarchical structure and a predictable design with a higher degree of formal rules and decision making, to a flatter structure and unpredictable design of cooperation described as a flexible network. Von Lampe (2008) emphasized different functions and purposes of opponent or-ganisational structures, with a distinction between social, economic and quasi-governmental capabilities.

(13)

Efficiency and security

Whereas a lot of similarities can be found between common and opponent organisations, criminal, ter-rorist and insurgent organisations and networks are fundamentally differentiated by the essential need for secrecy of operations (Morselli et al., 2007). Law enforcement can focus on the opponent organisational structure to detect vulnerabilities of networks and search for opportunities to disruption (Duijn, 2016; Michalak et al., 2015; D. Bright, Greenhill, Britz, Ritter, & Morselli, 2017). Intensifying cooperation by opponents increases the chance of infiltration by law enforcement agencies or leakage of information (Enders & Su, 2007). Opponent organisations limit their amount of communication to cope with this threat. This adaptive mechanism is embedded in the conceptual model with the interaction between the opponent individuals, the emergent behaviour and the feedback loop by law enforcement (Figure 4). This displays the ongoing complexity of law enforcement and opponent behaviour.

Loosely connected networks seem more adaptive, as they can replace individuals with vital roles more easily or new structures can emerge quickly (Arquilla & Ronfeldt, 2001; Spapens, 2006; Morselli et al., 2007). However, Ligon et al. (2013) pointed that larger organisations are more lethal and are equipped to execute more complex actions (Asal & Rethemeyer, 2008). The tension between the efficiency and security in this manner yields a trade-off while optimizing the effectiveness of opponent organisations.

Competition and trust

Furthermore, opponent organisations form a threat to each other by competition (Spapens, 2010). Rios (2013) described a self-reinforcing equilibrium between rivalry by opponent organisations, violence and law enforcement activities. This study centralised the escalation of drugs-related violence as a result of competition for illegal markets. Competition on illegal markets is unstable as they lack formal mecha-nisms and rules to cope with disputes. When opponents start to use violence to compete for opponent opportunities (Figure 4) the equilibrium (progressive state) starts by the creation of a power vacuum (R´ıos, 2013). When law enforcement agencies use violence in order to regain control in the area, the violence aggravates. Eliminating individuals and their relationships will create new power vacuums and unbalance the market even further. Radil et al. (2010) link competition to different opponent organisational struc-tures. Radil describes that once organisational forms on illicit markets change from mechanic to organic, opponents become more individualistic and adaptive. This increases the competition and potential for rivalry.

Due to violence and oppression, relationships within illicit environments are characterized by a lack of trust (Von Lampe, 2008). Individuals are dependent on relationships of trust in order to deal with the competition and law enforcement. Von Lampe et al. (2004) describe the ambiguity of trust in organized crime, as trust is a mechanism to deal with multiple forms of risk and uncertainty.

The emergence of different opponent organisational structures are dependent on the feedback mecha-nisms of opportunity, threat and competition. The conceptual model embodies the complexity. The psychological, physical and organisational aspects are essential to model agents that mimic the character-istics of opponent behaviour and are adaptive to the complex mechanisms. This answers the first research

(14)

question. In the next section, the interaction of these concepts are modelled computationally in order to experiment and analyse the complex effects towards policies to contain opponent behaviour.

1.2 Methodology

An agent-based model (ABM) was created to conduct scenario based experiments within a simulation environment. The covertness of illicit operations hinders the possibility of extensive and detailed em-pirical research in actual networks (Fielding, 2017). Simulations should provide insight in the effects of the complex mechanisms and create a better understanding of the development of particular oppo-nent behaviour in practise. The ABM creates an experimental environment to study oppooppo-nent behaviour. Deriving outcomes of emergent behaviour is impossible by a strict mathematical approach due to the amount of interactions, activity, decision rules, states and variables within the complex system (Gilbert, 2008). An ABM offers the possibility to experiment and estimate the impact of the causal relations and parameters on the behaviour of the system.

NetLogo was used to create a computational model (Thiele, Kurth, & Grimm, 2011; Wilensky, 1999). NetLogo provides extensions to analyse the network characteristics of opponent organisational structures. The observer functionality of NetLogo enables the modelling of the feedback mechanisms and observe the functioning of the modelled system. The accessibility of NetLogo ensures created models are reusable and adjustable to specific case studies.

Simulation requires a computational representation of the theoretical behavioural concepts. The psychological and economical theory was translated to an agent model that represents the individual behaviour. This enables the analysis of developments at the individual level according to the selected the bottom-up approach. Multi-agent systems (MAS) modelling was used to create the computational model of opponent individuals. MAS studies focus on autonomous decision making, distributed systems, learning and optimization (Shoham & Leyton-Brown, 2008). The MAS approach enables coding and parameterization of the utilitarian theory of opponent individuals and organisations. Application of game theory within MAS methods contributes to a representation of human behaviour and rational responses in simulations. Activity based interaction of the individuals and the environment was based on pay-off metrics that represent the feedback mechanisms.

The assumptions and complexity of the conceptual framework require a strict methodology for modelling and interpretation (Gal´an et al., 2013). The Overview, Design concepts and Details (ODD) protocol was applied to keep the model comprehensive and reproducible (Grimm et al., 2010). The overview section includes the purpose of the model, the included entities and their attributes and the process of the model (Appendix B). The design concepts provide insight in the dynamics of the model. Finally, model details such as detailed opponent behaviour and system functions are added to the model. Analysis and validation of the model was planned and documented according to the TRACE framework (Grimm et al., 2014) (Appendix A). First, verification ensures the model structure and output. Secondly multiple scenario simulations are conducted to create model output for analysis. Finally, the

(15)

output of the model is validated on a conceptual level and a dimension of usefulness. This analysis is used for the discussion of this thesis and implications for future research.

1.3 Agent-based model

The purpose of our model is to understand how the interactions of opponent individuals under different conditions yield emergent behaviour of the resulting opponent organisations. Within our model, compe-tition has been operationalised by implementation of the mechanisms of rational choice theory and the feedback loops of opportunities and threats at the micro-level. At micro-level, the individual opponents operate by self-interest, autonomously and utility driven to mimic the rational decision making of op-ponents. Opponents can initiate an individual cooperation or setup a hierarchical organisation to create synergy. From the literature the following hypotheses are formulated:

Hypothesis 1: The form of cooperation by opponents will shift from a hierarchical form to a network as law enforcement agencies intensify their actions, which is based on the study by Ligon et al. (2013) of the development of the KKK.

Hypothesis 2:Hierarchical organisations will be smaller and less dense in scenarios with higher law enforcement intensity, based on the study of Kenney et al. (2013) of the al-Muhajiroun group.

Hypothesis 3: Cooperation by individuals in networks will be less dense as law enforcement agencies focus on infiltration of networks. Based on the research of Enders and Su (2007), in their structural view upon al-Qaeda.

Hypothesis 4:The amount of opponents will decline by a growth of social opportunity, Based on the study by Faria and Arce (2012) on recruiting of opponent organisations and support by society.

Hypothesis 5: Individuals will deradicalize as the amount of opponent opportunity declines, based on motives to disengage and deradicalise from radical organisations by Windisch et al. (2016).

To monitor the evolution of the system different metrics are identified from the literature. The metrics are opponent density, ratio between opponent organisations and individuals and the average opponent organisation size (Boerman & Stoffers, 2017; D. A. Bright & Delaney, 2013; Emmet & Broers, 2008), the density of the cooperation by opponents (Bichler et al., 2017) and the effectiveness of opponents (Morselli et al., 2007; Spapens, 2010). A detailed overview of the steps of model development are provided by the TRACE documentation inAppendix A. The model implementation is documented extensively within the ODD documentation inAppendix B. The next two sections outline the general programmed model objects and functions to understand the model validation and experiments inSection 2.

(16)

1.3.1 Overview and Design Concepts

As any ABM, our model consists of agents and the environment where they are integrated (Gilbert, 2008). The environment represents a simplification of the complex mechanisms of the systems of society, economy and governments in reality. A discrete event simulation engine will be used. Appendix C

provides an detailed overview of the entities and their attributes in our computational model. The design concepts are the principles of the conceptual model, which are programmed as the objects of environment, networks, individuals and organisational structures.

The simplified Environment is a two-dimensional socio-spatial structure that hosts the agents. Within this environment, agents can move and conduct actions. The environment exerts the feedback mechanisms from the macro to the micro-level (Figure 4). Implementation of these structures is provided byAppendix Dand elaborated in this section.

The agents are embedded in underlying Networks, which feature different functions. These net-works mimic the relationships of people in reality (D. A. Bright et al., 2015). Netnet-works consist of nodes and edges which are bounded by spatial distance. Three social networks are differentiated (Figure 5).

Figure 5: Small snapshot of the visualisation of the modelled environment

Firstly, the (green) “Local Network” represents small distance relationships of agents that know each other. This network enables agents to initiate con-tact. This causes local and bounded rationality, as agents can only commu-nicate with local agents. The second network is the (red) “Communication Network”. This network enables agents to share information about social and opponent opportunity, negotiate over cooperation or initiate organisa-tion. Lastly, the (blue) “Cooperation Network” enables cooperation and creates synergy during opponent activity, which is called “crime”.

The Environment enforces adversity and competition towards the agents. The goal of this environment is to imitate situations in which indi-viduals behave competitive in order to succeed. At each simulation step the

environment taxes the agents with the environment costs. The environment provides social and opponent opportunity towards the agents as sources of income, which can be used to pay the environment tax. The social opportunities (“jobs & job reward”) are distributed by the agents through communication. This causes the social opportunity feedback loop mechanism of the conceptual model. The opponent opportunities (“crime reward”) are distributed over the agents that conduct “crime”.

The most important attributes of the Agents are their three states: radical state, desire state, intention state. The radical state represents the opinions of the agent towards opponent behaviour. Pa-rameterization is elaborated by themodel detailssection. With their desire state, agents determine their goal to obtain a reward. This is influenced by the radical state. Once opponent opportunity outweighs social opportunity, the goal of the agent switches from social to opponent (seemodel details). The inten-tion statedefines the plan towards obtaining a reward. This is influenced by the desire state and current potential of the agent. Agents need a job to conduct the “work” capability or a opponent opportunity

(17)

with a positive expected utility to conduct “crime”, this is explained by themodel detailssection. Movement capability by agents enables them to contact new agents by modifying their local-network and improve their situation. Agents are in an unfavourable situation once the environment costs exceed their current reward. With the movement capability, competition is created as agents continually improve their situation.

The Opponent Opportunity is rewarded to the agents that conduct crime activity. Individually, each agent has the capability to conduct crime with effectiveness 1. Agents can initiate cooperation to create synergy and increase their individual effectiveness. The effectiveness of the opponent agents is registered by the simulator at each simulation time step. The opponent opportunity is rewarded to the opponents at the ratio of their individual effectiveness to the collective effectiveness of crime activity. This mechanism yields the opponent opportunity and competition feedback loop of the conceptual model. The crime activity can also attract law enforcement repercussion, which will be explained.

Agents can initiate an Organisation to create additional synergy and extent cooperation. One structure of organisation is modelled, which is a hierarchy with centralised decision making. Once an agent has a higher crime reward than the opponent neighbors of the agent, the agent can initiate an or-ganisation. The initiator is the leader of the organisation and those that join the organisation are the members. The organisation yields additional synergy, according to the theory which states that organisa-tions gather more human and social capital (Ligon et al., 2013; Bichler et al., 2017). The organisation can recruit additional members to improve the collective synergy. The organisation also attracts additional law enforcement activity. The white star ofFigure 5displays an organisation. The light blue links are the organisation communication links. The organisations and cooperation links represent the “collective” Emergent Opponent Behaviour of the system.

The modeled law enforcement agency conducts Law Enforcement Activities, which can focus on specific agents that conduct crime activity. At each simulation time step, the modelled law enforcement agency detects the opponent activities at a certain rate. They decide the intensity of their counter actions and choose an action type: direct countering or infiltration. With a direct countering action, the most effective opponents of the detected will be countered. An infiltration action aims to detect additional opponents through cooperation and communication links of the previous detected opponents and attempt to disrupt those crime activities. The infiltration strategy poses an additional threat towards opponents with a high degree of cooperation. A successful law enforcement activity disrupts an opponent activity and punishes the opponent with a negative reward.

Stochastic parameters were added to mimic the uncertain success rate of “work”, “crime” or “law enforcement”actions. These parameters are adjustable for specific cases to investigate the influence of a higher or lower degree of uncertainty to the social dynamics in the specific environment. As a result, the outcome of the model has a certain noise and therefore requires multiple simulations to validate the model outcome.

(18)

1.3.2 Model details

A simulation run starts with an initialization of the input parameters of the system.Appendix Gprovides an overview of the input parameters and the default values of the ‘base’ scenario. The purpose of different input variables are explained in Appendix G, as the parameters of the computational model and concepts of the conceptual model are linked. The ‘base’ scenario is set to establish competition, an incentive to the emergence of opponent structures and feedback loop mechanisms of social and opponent opportunity and law enforcement activity. Scenarios can be created by modifying input variables.

The reasoning of the individual agents is modelled computationally to shape adaptive and au-tonomous behaviour. The agents reason utility driven to mimic self-interest based behaviour and im-plement the concept of rational choice theory (Schelling, 1971; Clarke & Felson, 1993). The agents in our model are constructed according the BDI-framework to enable adaptive and reactive agent responses (Deljoo, van Engers, Gommans, & de Laat, 2017; Georgeff & Lansky, 1987). An overview of this imple-mentation is provided byAppendix B. Agents yield adaptive responses by modifying their internal states and movement, as described in the model overview. Agents change states according to the formulas of the radical opinion state and the expected utility of crime activity. The formula of radical opinion state change is given by:

ri,t(K, G, w, dn) = min (max (ri,t−1+ rr (sd+ sla+ sG) , 0) , 100) (1)

In which the rational opinion state (r) of agent i at time t (ri,t) is given by r at the previous time step,

influenced by the radicalisation pace (rr) and the satisfaction on the return from activities (sd), the amount

of law enforcement activity (sla) and the average radical opinion level of the communication network of

agent i (sg). The radical opinion level is bounded between 0 and 100, which mimics the radicalisation

pyramid of McCauley et al. (2017). With a radical opinion level above 50, the agent ascends to the upper level of the pyramid (Figure 2), which implies justification of opponent behaviour.

The radical opinion state effects the “expected costs” of crime activity. The formula of rational choice behaviour for opponent behaviour is given by:

maxUi(H, xi) = yi(G, H, O, xi) − ci(G, H, O, xi) (2)

In which the utility (Ui) of opponent activity x by agent i (xi) is given by the expected benefits (yi) and costs

(ci). These benefits and costs are dependent on the cooperation network (H) and organisation network (O)

of the agent, which the agent can modify by negotiation with the agents in the communication network (G). The agents deliberately add or deduct cooperation links to increase the amount of synergy or to cope with expected law enforcement activity. The formulas for the benefit, cost calculations are provided in

Appendix E, including details and examples. Similar model assumptions were implemented to optimize the structure of opponent organisations (Appendix F).

(19)

2 Model analysis and results

Our computational model was subjected to verification and validation to analyse the accuracy and appli-cability of the computational model (Gilbert, 2008). To demonstrate the value of the model, experiments with multiple simulations of different scenarios were conducted to test the hypotheses of Section1.3. The design, results and analysis of the experiments are presented inAppendix H.

Verification of the computational model ensures accurate correspondence with conceptual model (Rand & Rust, 2011). The verification process consisted of three steps: documentation, code walkthrough and programmatic testing (Grimm et al., 2014).Appendix AandAppendix Bdocument and describe the translation from the conceptual model to the computational model and the conducted verification process. Social based computational models require validation to estimate the value of the model out-put (Bharathy & Silverman, 2010). Different definitions for validation of ABM emerge from literature (Grimm et al., 2014). Models of social systems are incomplete, imprecise and complex by definition (Bharathy & Silverman, 2010). Selection of proper validation methods is required to estimate the precise value of the developed model. Predictive ABM requires real-world data validation to test whether the model output is generalizable to situations in reality (Rand & Rust, 2011). ABM of social systems have been criticized frequently for their lack predictive power (Bharathy & Silverman, 2010). The intended application of our computation model is to explore the effects of complex adaptive systems mechanisms towards emergent opponent behaviour, rather than the discovery of precise correlations for predictive power. The current level of abstractness within our computational model prevents any tests with real-world data and limits the predictive power. However, any corroboration of model output requires valida-tion of model assumpvalida-tions (Grimm et al., 2014). Therefore a validavalida-tion process for our computavalida-tional model was conducted at the dimensions of internal validity and methodological validity (Bharathy & Silverman, 2010).

Life cycle validation has been conducted to determine conceptual correctness and coherence between input and output (Bharathy & Silverman, 2010). Life cycle validation is divided in four steps: input, micro-face, macro-face and output validation. The estimated validity concerns the value of input parameters to construct experiments, model concepts to explore complexities and sensitivity of the input parameters upon the output values. As a result, the model should be applicable and interpretable.

Abstract input parameters inhibit tests with exact empirical input (Appendix G) (Bharathy & Silverman, 2010). However, qualitative comparison of inputs with real world phenomena illustrates plau-sible experimental designs. The initialization of experiments demonstrate scenarios to explore the influ-ence of environmental aspectsAppendix H. The experiments respectively mimic abstractions of policies by law enforcement agency or economic development. Other possible experiments could include the radicalisation pace to explore psychological aspects, or cooperation and organisation benefits to explore organisation aspects.

Micro-face and macro-face validation is the process of showing that the mechanisms, properties and aggregated patterns follow real-world patterns (Rand & Rust, 2011). Face validation is especially

(20)

valuable to our computational model as it ensures the holistic perspective on the system by our con-ceptual model. For face validation purpose colors were added to identify different networks and the different states of the agents in the observer of NetLogo. Aggregated patterns were plotted to observe the macro-level. At the micro-level, interactions between agents establish new communication networks. In conditions of little law enforcement activity, agents with the goal to commit opponent behaviour gather to create synergy (Appendix A, Figure 7). Agents with a high radical opinion level tend to deradicalize in a neighborhood with a low average radical opinion level due to influence by their social network. At the macro-level, the amount of opponent effectiveness grows by the emergence of opponent organisations. Additionally, the overall opponent opinion level increases as the amount of opponents grows.

Figure 6: The amount of organisations by different in-tensities of law enforcement activity

To analyse the output of the system, two experiments were conducted. The experiments demonstrate the ability to design scenarios that mimic various societies. The output of the sys-tem reveals the non-linear behaviour caused by the adaptive micro-level. The results to test Hypoth-esis 1 are displayed byFigure 6. Scenarios were designed to increase the intensity of the activity

by the law enforcement agency. The amount of opponent organisations decline as the rate of law en-forcement activity to opponent activity increases from 0.075 to 0.1. Hypothesis 1 is supported by this observation as the amount of opponents and the average size of organisations remain stable (Appendix H).

Similar analysis was conducted to evaluate Hypothesis 2 to 5. The organisation size was eval-uated in scenarios with different intensities of law enforcement activity. We were unable to observe a significant difference in organisation size between these scenarios. This refutes the statement of Hypoth-esis 2. The opponents are not forced to adapt to the increase of law enforcement activity, as additional activities are not targeted at specific opponents. The ratio of organisations and their size were evalu-ated with scenarios with different strategies by law enforcement agencies to target effective opponents or infiltrate networks. The increase of organisations compared to loose networks shows the capability of individuals to adapt to the infiltration strategy. The insignificant difference in effectiveness of oppo-nents under scenarios with different law enforcement strategies does not support Hypothesis 3. Although the emergent behaviour changes, the effectiveness remains stable. This demonstrates the resilience of opponents to disruptive strategies. The second experiment varies the amount of social and opponent op-portunity. Hypothesis 4 is supported by the output of the model, as the ratio of opponents declines as the amount of social opportunity increases. However, as the average radical opinion is stable in scenar-ios with different amounts of opponent opportunity, Hypothesis 5 is not supported. These observations demonstrates the importance of alternative opportunities for opponents in order to mitigate the emergence and growth of opponent organisations within this system.

(21)

focused on the amount of organisations. The effectiveness of an organisation grows with the addition of new members. Due to competition, opponents are forced to leave ineffective organisations. Therefore, as a result of competition, the amount of organisations decreases over time. The second observation focused on the increase of competition. Two different ways to increase competition between opponents are ob-served. The first cause for an increase of competition is a growth of the amount of opponents compared to the amount of opponent opportunity. Additionally this increases the density of the opponents in the system. The second cause for an increase of competition is a decrease the amount of opponent opportu-nity compared to the amount of opponents. Within this scenario the density of the opponents is stable. These different ways to increase competition influence the emergent behaviour. Potentially organisations are more spread in a scenario with a higher density of opponents, as more opponents can contact each other despite the limitations of the length of communication links. These unexpected observations show the value of an experimental environment to study emergent opponent behaviour.

The analysis of the computational model demonstrates that individual opponents are adaptive as they yield emergent “collective” opponent behaviour. Individuals deliberate opponent activities, which will become rational once they radicalise. The feedback loop mechanisms of law enforcement activity, so-cial and opponent opportunity and competition simultaneously influence processes at the micro-level. The emergent behaviour can change as individuals adapt to changes in their environment. Detailed modelling and experimenting can help to understand the ongoing complexities that yield the emergent opponent behaviour.

3 Discussion and conclusion

The emergent behaviour of opponent organisations is a challenging topic in academic research and prac-tice. The focus of this thesis was to acquire insight in the complexity of emergent behaviour of opponent organisations. Two research questions have been answered through literature study, modelling and ex-perimentation. To deal with the covertness of opponent activity, our computational model provides an opportunity to experiment and uncover the effects by complex mechanisms of the system to law en-forcement policies. The method and results of the study are discussed to examine the limitations and implications for future studies. The ability to reveal complexities regarding opponent behaviour with an interdisciplinary approach and analyse specific environmental concepts, such as competition and law enforcement activity, computationally are the main contributions of this thesis.

Application of ABM provides merits and disadvantages. ABM holds difficulties concerning validity and programmability. External validity of social ABM has been criticized heavily (Bharathy & Silverman, 2010). The model assumptions and the interactions in our model have a high level of abstractness. Additionally the covertness of opponent activities limits the amount of data to validate ABM output or estimate parameters. Social system ABM are not generalizable beyond the instances that have been examined in the current study (Rand & Rust, 2011). Improvements upon current efforts should take these limitations into account. Additionally modelling of dynamics within social systems by

(22)

ABM is relatively computationally and programming intensive compared to other methods such as system dynamics or statistical modelling (Rand & Rust, 2011). Therefore, development of ABM is relatively expensive with regard to money and time. To cope with the issues of validity and costs, development and results of ABM should always be compared with other modelling approaches.

However, application of ABM enabled a computational translation of the concepts found by a interdisciplinary literature study. Compared to other approaches such as analytical modelling or system dynamics, this methodology provides tools to model fine-grained evolution from the level of an individual (Rand & Rust, 2011). The scope of this thesis was to model the essential characteristics that reflect opponent behaviour. Advanced agents or multiple agent types could yield more valid models. To extend the current modelling efforts, advanced ABM could include additional attributes of learning agents and advanced game theoretic based behaviour (Shoham & Leyton-Brown, 2008).

Adoption of NetLogo entails important implications. The integrated modelling environment of NetLogo prohibits integration of additional modelling methods or program packages, which limits the scalability and effectiveness of the computational model. Advanced models of opponent behaviour could require specific programming approaches or a distributed agent-based modelling platform. Yet the application of NetLogo provided an programming environment for rapid development and accessibility to “non-programmers” (Wilensky, 1999). This is especially valuable as most of the scholars of opponent behaviour have a non-technical background and could experiment with the current model relatively easy. Multiple leads for modelling advanced opponent behaviour were found in the literature. These focus on the operations, the characteristics and organisation of opponents. Firstly, individuals maintained a constant level of activity within the current model. Studies of opponent behaviour have revealed the possibility to conduct opponent activities with less impact to hide operations (Enders & Su, 2007; Das-tranj, Easton, & Karaivanov, 2013). Additionally the frequency of cooperation and action influences the covertness (Enders & Su, 2007; Morselli et al., 2007). Secondly, modelling different agent types would enable agents with different personalities or skills. Modelling different types of organisation would en-able comparison between the effectiveness of multiple organisation types. This would enen-able analysis of actors with essential roles within networks (Bichler et al., 2017).

Application of this model towards real-world data was out of the scope of this thesis. The devel-oped modelling approach could be applied to historical cases of emergent opponent behaviour. Specific characteristics could be examined to explain the ongoing complexities, this would increase the external validity of the model. Additionally application of the model to real world cases, would explore and val-idate the value of conducting experiments to explain opponent behaviour. For the purpose of evaluation of law enforcement policies, models could include multiple detailed law enforcement strategies, such as disruption, detaining or elimination of opponents.

(23)

References

Abadinsky, H. (2012). Organized crime. Cengage Learning.

Anderson Jr, E. G. (2011). A dynamic model of counterinsurgency policy including the effects of intelligence, public security, popular support, and insurgent experience. System Dynamics Review, 27(2), 111–141.

Arquilla, J., & Ronfeldt, D. (2001). Networks and netwars: The future of terror, crime, and militancy. Rand Corporation.

Asal, V., & Rethemeyer, R. K. (2008). The nature of the beast: Organizational structures and the lethality of terrorist attacks. The Journal of Politics, 70(2), 437–449.

Barsade, S. G. (2002). The ripple effect: Emotional contagion and its influence on group behavior. Administrative Science Quarterly, 47(4), 644–675.

Bharathy, G. K., & Silverman, B. (2010). Validating agent based social systems models. In Simulation conference (wsc), proceedings of the 2010 winter(pp. 441–453).

Bichler, G., Malm, A., & Cooper, T. (2017). Drug supply networks: a systematic review of the organiza-tional structure of illicit drug trade. Crime Science, 6(1), 2.

Boerman, G. M. N. F., F., & Stoffers, E. (2017). National threat assessment organised crime 2017. Bonger, W. A. (2015). An introduction to criminology. Routledge.

Borum, R. (2007). Psychology of terrorism (Tech. Rep.). University of South Florida Tampa Dept Of Mental Health Law And Policy.

Bosse, T., Gerritsen, C., & Treur, J. (2007). Cognitive and social simulation of criminal behaviour: the intermittent explosive disorder case. In Proceedings of the 6th international joint conference on autonomous agents and multiagent systems(p. 58).

Bovenkerk, F. (2010). Uittreden: over het verlaten van criminele organisaties (Vol. 364). Amsterdam University Press.

Bright, D., Greenhill, C., Britz, T., Ritter, A., & Morselli, C. (2017). Criminal network vulnerabilities and adaptations. Global Crime, 18(4), 424–441.

Bright, D. A., & Delaney, J. J. (2013). Evolution of a drug trafficking network: Mapping changes in network structure and function across time. Global Crime, 14(2-3), 238–260.

Bright, D. A., Greenhill, C., Ritter, A., & Morselli, C. (2015). Networks within networks: using multiple link types to examine network structure and identify key actors in a drug trafficking operation. Global Crime, 16(3), 219–237.

Burt, C. H., & Simons, R. L. (2013). Self-control, thrill seeking, and crime: Motivation matters. Criminal justice and behavior, 40(11), 1326–1348.

Calv´o-Armengol, A., & Jackson, M. O. (2007). Networks in labor markets: Wage and employment dynamics and inequality. Journal of economic theory, 132(1), 27–46.

Caplan, B. (2006). Terrorism: The relevance of the rational choice model. Public Choice, 128(1-2), 91–107.

(24)

Carley, K. M., Dombroski, M., Tsvetovat, M., Reminga, J., Kamneva, N., et al. (2003). Destabilizing dynamic covert networks. In Proceedings of the 8th international command and control research and technology symposium.

Carrington, P. J., & Scott, J. (2011). The sage handbook of social network analysis. Sage Publications. Clarke, R., & Felson, M. (1993). Routine activity and rational choice: Advances in criminological theory

(volume 5). Piscataway, NJ: Transaction.

Cornish, D. B., & Clarke, R. V. (1987). Understanding crime displacement: An application of rational choice theory. Criminology, 25(4), 933–948.

Cressey, D. R. (1972). Criminal organization: its elementary forms. Harper & Row New York. Dastranj, R., Easton, S., & Karaivanov, A. (2013). Economics of crime networks.

Deljoo, A., van Engers, T., Gommans, L., & de Laat, C. (2017). What is going on: Utility-based plan selection in bdi agents. In The aaai-17 workshop on knowledge-based techniques for problem solving and reasoning ws-17(Vol. 12).

Devia, N., & Weber, R. (2013). Generating crime data using agent-based simulation. Computers, Environment and Urban Systems, 42, 26–41.

Duijn, P. (2016). Detecting and disrupting criminal networks: A data driven approach.

Easton, S. T., & Karaivanov, A. K. (2009). Understanding optimal criminal networks. Global crime, 10(1-2), 41–65.

Eidelson, R. J. (1997). Complex adaptive systems in the behavioral and social sciences. Review of General Psychology, 1(1), 42.

Emmet, I., & Broers, J. (2008). The green gold. report of a study of the cannabis sector for the national threat assessment of organized crime. KLPD-IPOL, Zoetermeer.

Enders, W., & Su, X. (2007). Rational terrorists and optimal network structure. Journal of Conflict Resolution, 51(1), 33–57.

Faria, J. R., & Arce, D. (2012). Counterterrorism and its impact on terror support and recruitment: Accounting for backlash. Defence and Peace Economics, 23(5), 431–445.

Fielding, N. G. (2017). The shaping of covert social networks: isolating the effects of secrecy. Trends in Organized Crime, 20(1-2), 16–30.

Framis, A. G.-S. (2011). Illegal networks or criminal organizations: Power, roles and facilitators in four cocaine trafficking structures.

Gal´an, J. M., Izquierdo, L. R., Izquierdo, S. S., Santos, J. I., del Olmo, R., & L´opez-Paredes, A. (2013). Checking simulations: detecting and avoiding errors and artefacts. In Simulating social complexity (pp. 95–116). Springer.

Ganor, B. (2008, Apr). Terrorist organization typologies and the probability of a boomerang effect. Studies in Conflict Terrorism, 31(4), 269ˆa283. doi: 10.1080/10576100801925208

Georgeff, M. P., & Lansky, A. L. (1987). Reactive reasoning and planning. In Aaai (Vol. 87, pp. 677– 682).

(25)

doi: 10.4135/9781412983259

Grimm, V., Augusiak, J., Focks, A., Frank, B. M., Gabsi, F., Johnston, A. S., . . . others (2014). Towards better modelling and decision support: documenting model development, testing, and analysis using trace. Ecological modelling, 280, 129–139.

Grimm, V., Berger, U., DeAngelis, D. L., Polhill, J. G., Giske, J., & Railsback, S. F. (2010). The odd protocol: a review and first update. Ecological modelling, 221(23), 2760–2768.

Hamers, H., Husslage, B., & Lindelauf, R. (2011). Centraliteitsanalyses van terroristische netwerken (Tech. Rep.). Tilburg University.

Hubner, J. F., Sichman, J. S., & Boissier, O. (2007). Developing organised multiagent systems using the moise model: programming issues at the system and agent levels. International Journal of Agent-Oriented Software Engineering, 1(3/4), 370. doi: 10.1504/ijaose.2007.016266

Kenney, M., Horgan, J., Horne, C., Vining, P., Carley, K. M., Bigrigg, M. W., . . . Braddock, K. (2013). Organisational adaptation in an activist network: Social networks, leadership, and change in al-muhajiroun. Applied Ergonomics, 44(5), 739ˆa747. doi: 10.1016/j.apergo.2012.05.005

Ligon, G. S., Simi, P., Harms, M., & Harris, D. J. (2013). Putting the ˆaoˆa in veos: What makes an organization? Dynamics of Asymmetric Conflict, 6(1-3), 110–134.

Makarenko, T. (2004). The crime-terror continuum: tracing the interplay between transnational organised crime and terrorism. Global crime, 6(1), 129–145.

McCauley, C., & Moskalenko, S. (2008). Mechanisms of political radicalization: Pathways toward terrorism. Terrorism and political violence, 20(3), 415–433.

McCauley, C., & Moskalenko, S. (2017). Understanding political radicalization: The two-pyramids model. American Psychologist, 72(3), 205.

Michalak, T. P., Rahwan, T., Skibski, O., & Wooldridge, M. (2015). Defeating terrorist networks with game theory. IEEE Intelligent Systems, 30(1), 53–61.

Mintzberg, H., & Waters, J. A. (1985). Of strategies, deliberate and emergent. Strategic management journal, 6(3), 257–272.

Morselli, C., Gigu`ere, C., & Petit, K. (2007). The efficiency/security trade-off in criminal networks. Social Networks, 29(1), 143–153.

Ozgul, F. (2014). Classification of terrorist networks and their key players. In International conference, misnc(pp. 145–157).

Radil, S. M., Flint, C., & Tita, G. E. (2010). Spatializing social networks: Using social network analysis to investigate geographies of gang rivalry, territoriality, and violence in los angeles. Annals of the Association of American Geographers, 100(2), 307–326.

Rand, W., & Rust, R. T. (2011). Agent-based modeling in marketing: Guidelines for rigor. International Journal of Research in Marketing, 28(3), 181–193.

R´ıos, V. (2013). Why did mexico become so violent? a self-reinforcing violent equilibrium caused by competition and enforcement. Trends in organized crime, 16(2), 138–155.

(26)

Sheptycki, J., Jaffel, B., & Bigo, D. (2011). International organised crime in the european union. Pol-icy Department C: Citizensˆa Rights and Constitutional Affairs. Brussels: European Parliament. Google Scholar.

Shoham, Y., & Leyton-Brown, K. (2008). Multiagent systems: Algorithmic, game-theoretic, and logical foundations. Cambridge University Press.

Simon, H. A. (1955). A behavioral model of rational choice. The quarterly journal of economics, 69(1), 99–118.

Spapens, T. (2006). Interactie tussen criminaliteit en opsporing: De gevolgen van opsporingsactiviteiten voor de organisatie en afscherming van xtc-productie en-handel in nederland. Intersentia. Spapens, T. (2010). Macro networks, collectives, and business processes: An integrated approach to

organized crime. Eur. J. Crime Crim. L. & Crim. Just., 18, 185.

Thiele, J., Kurth, W., & Grimm, V. (2011). Agent-and individual-based modelling with netlogo: intro-duction and new netlogo extensions. Die Gr¨une Reihe, 22(68), e101.

Von Lampe, K. (2008). Organized crime in europe: conceptions and realities. Policing: A Journal of Policy and Practice, 2(1), 7–17.

Von Lampe, K., & Ole Johansen, P. (2004). Organized crime and trust:: On the conceptualization and empirical relevance of trust in the context of criminal networks. Global Crime, 6(2), 159–184. Wilensky, U. (1999). Netlogo. evanston, il: center for connected learning and computer-based modeling,

northwestern university.

Windisch, S., Simi, P., Ligon, G. S., & McNeel, H. (2016). Disengagement from ideologically-based and violent organizations: A systematic review of the literature.

(27)

A

Appendix: TRACE-protocol documentation: Overview of model

development and analysis

(28)

Table 1: Modelling documentation

TRACE element Description

1. Problem formulation The covertness of the opponent organisations hinders the possibility to conduct empirical based research, especially of currently active organ-isations. This model will be used to conduct experiments and estimate the impact of the complex adaptive system effects on law enforcement policies. The concepts of the model were derived from a multidisci-plinary literature research. Modelling multiple perceptions of opponent behaviour embraces study of the complex mechanisms within the op-ponent behaviour. This approach allows possibility to expand future models and specify for real world cases and validation with historical data. A bottom-up approach was used to develop the model from the individual level. Focusing on the psychological, economical and organ-isational theory.

2. Model description We present a description of the model with the ODD-protocol (Section 1.3.1 & 1.3.2) (Appendix B). The model provides a framework with three entities: individuals, organisations and environment. NetLogo was used to code the computational model.

3. Data evaluation The model was not calibrated or validated using experimental data. Mechanisms and parameters were based on sources from literature. They create an abstract and complex environment focusing on the covertness and competition of opponent organisation.Appendix E pro-vides an overview of the dependencies of the parameters, explaining the mechanisms that create the negative and positive feedback loop to the micro-level. Future studies could focus on real-world cases imple-menting specific parameters to compare different situations, with for example more poverty or adversity. Such models would provide a more valid the purpose of the model in regard of data validation and predictive modelling.

4. Conceptual model evaluation The conceptual model is represented by Figure 4 and explained by Sec-tion 1.1. The conceptual model was translated to the design concepts of the computational model.

5. Implementation verification Extensive structural verification was conducted to ensure proper imple-mentation of the model specification. After multiple model iterations the syntax of the model was reviewed and cleaned to create a clear code and ensure the speed of simulations. Each coded procedure was exe-cuted on its own in the observer of NetLogo. The calculations explained byAppendix Ewere implemented in Excel and compared with the out-put of executions of the procedures.

6. Model output verification Default values of inputs parameters were set by analysis of the model output. Simulations with extreme parameter values were conducted to verify the mechanisms of the model. For example, simulations without opponent opportunity resulted in output without opponent activity. Ex-tensive simulation runs were conducted to find an acceptable warm-up period and balance between the parameters in the default scenario. The NetLogo interface allows face verification, which was used for verifica-tion of the design concepts such as radicalisaverifica-tion and networks. Figure 7 displays the face validation.

7. Model analysis Five hypothesis based on literature were set before modelling. Based on sensitivity analysis the default values of the input parameters were set and scenarios were designed. Experiments were conducted to explore the behaviour of the model and test the hypothesis. This is elaborated inAppendix H.

8. Model output corroboration The patterns of covertness and effectiveness of opponent activity have been found in the literature. Additionally psychological and economi-cal theory has been implemented to define the individual behaviour. An abstract model demonstrates the complexities to evaluate the effective-ness to contain the emergent opponent behaviour. Economical data of specific or comparable environments or empirical data of psychological processes could expand and validate current model assumptions.

(29)

An Agent-Based Model for Emergent Opponent Organisations