Mason Rebeland, An Agent Based Model Of Politics, Environment, And Insurgency

Mason Rebeland, An Agent Based Model Of Politics, Environment, And Insurgency - Transcript

International Studies Review 2010 12 31 52

MASON RebeLand An Agent Based Model of Politics Environment and Insurgency1
Claudio Ciof Revilla and Mark Rouleau George Mason University
Social simulation models from computational social science are beginning to provide signi cant advances in terms of implementing more complex social human and natural dynamics that are characteristic of how countries operate in the real world In particular increasingly realistic agent based models can improve capacity for early warning understanding and prediction The MASON RebeLand model presents three innovations over earlier models i an explicit polity model with politically complete structure and processes ii social and natural model components within an integrated socio natural system and iii generative dynamics where insurgency and the state of the polity stable unstable failing failed and recovering occur as emergent phenomena under a range of social and environmental conditions Three scenarios are demonstrated showing stable unstable and failing polity conditions The MASON computational system for agent based and network modeling also permits additional experiments and extensions

The state of any given polity in every country ancient or contemporary universally depends on multiple internal and external factors Exactly how is the state of a polity or its political stability affected by internal endogenous or environmental exogenous processes such as changing conditions in its economy demography culture natural environment ecosystem climate or combined socio natural pressures This general question and its many variations represent a major unsolved challenge in social science in general and computational social science in particular It is also a question of signi cant interest in policy circles Answering such a question and related versions is also necessary for improving applied research such as for early warning and prediction This article presents an agent based model of a polity with structure and process based on contemporary political science concepts and principles that is a political system speci cally composed of a geographically situated society with a system of government that produces policies for addressing public issues Moreover aggregate or collective features such as public moods domestic political instability insurgency and state failure are generated endogenously as emergent phenomena from individual actor level and institutional behaviors Both the polity and its natural environment in this model can be extended in a number of research directions
1 Funding for this study was provided by the Center for Social Complexity of George Mason University and by ONR MURI grant no N00014 08 1 0921 Thanks to members of the Mason HRAF Joint Project on Eastern Africa MURI Team and to Armando Geller Rebecca Goolsby Gerald Schneider John Tagney Peter Turchin Nils B Weidmann and anonymous reviewers for comments The opinions ndings and conclusions or recommendations expressed in this work are those of the authors and do not necessarily re ect the views of the sponsors

2010 International Studies Association

32

MASON RebeLand

Background The terms stability and instability have two different meanings in social and political science for example compare Bhatia and Szego 2002 with Choucri Goldsmith Madnick Morrison and Siegel 2007 1 Political stability in the classical dynamical sense of Ljapunov and systems theory means that a polity is stable if and only if it can withstand changes under a given range of various forms of stress such as social economic political or environmental stress that may or may not include violence In this sense a polity may or may not be violent when it is unstable and vice versa because stability and violence are not always directly correlated Stability in this sense is a deep dynamical property 2 Political stability also means lack of political violence so conversely political instability in this second sense means that is is synonymous with occurrence of public violence in a given polity Accordingly stability in this second common sense is an observed behavior unrelated to deeper dynamical properties Clarifying these meanings is important for basic theory understanding and applied purposes early warning prediction For model building purposes we use the terms political stability and instability in the former sense of systems theory although the latter type political violence is obviously also relevant Earlier Empirical Studies Prior empirical studies have identi ed several variables responsible for political instability Early research uncovered relative deprivation as an impetus to civil unrest Gurr 1970 showing that the probability of political instability increases as the gap between perceived needs and actual gains of a given state s population grows Subsequently others argued that revolutionary movements serve as an outlet for the general population s frustration with the current social structure of the state leading citizens to use violence to realign this structure to attain expected social bene ts Morrison 1971 Political instability as violence is still relevant despite some criticisms Gurney and Tierney 1982 as being too individualistic that is not explaining why citizens would respond to their frustrations collectively not separating cause from effect that is not distinguishing between relative deprivation as a structural as opposed to perceptual feature of society and a number of its assumptions rest on weak psychological foundations that is the aggressive response as the only response to social frustration Weaknesses in the theory relative deprivation have led to development of the culturalist argument positing ethnic or religious differences as sources of political instability While researchers continued to maintain the classic economic argument of the haves versus the have nots other studies for example Horowitz 1985 Connor 1994 Huntington 1996 have investigated the role of nationalism and identity as proposed by culturalists Deutsch 1953 Anderson 1983 Gellner 1983 This group hypothesized that political instability was higher not just when relative deprivation was present or perceived to be present but when deep seated cultural differences latent unobservables served as the de ning characteristic of competing groups The salience of this argument grew at the end of the Cold War as religious and ethnic con icts began to make headlines Some important points should be noted about this research First it has not entirely replaced relative deprivation theory as it was understood that differences alone provide little impetus for con ict Second culturalism seemed to

Claudio Ciof Revilla and Mark Rouleau

33

claim that con ict was inevitable Huntington 1996 although historical circumstances might indicate otherwise Some recent empirical studies provide con icting support for earlier work leaving culturalist claims open to debate Two independent studies have recently cast doubt on the culturalist argument Fearon and Laitin 2003 argue that the opportunity to rebel is as important as the motivation or willingness for rebellion consistent with Ciof Revilla and Starr 2003 1995 showing that the probability of civil unrest does not depend on cultural fractionalization or economic disparity but rather on particular features of a state that favor insurgency that is different opportunity structures such as rough terrain large populations and nancial and or bureaucratic weakness Others have extended this opportunity argument by focusing on the nancing of insurgency activity through lootable resources Collier and Hoef er 1998 2004 Political violence is more likely in states with weak institutional structures and highly accessible natural resource endowments with high marketability such as diamonds or gold These two studies have been important in emphasizing the dichotomy between state and rebel capacity in regard to political instability Recent work by Cederman and Girardin 2007 despite questioning the research methods of Fearon and Laitin further supports an opportunity based process for political instability albeit with the presence of ethnic con ict The failure of institutions is also highlighted by studies on the political instability of partial or weak democracies King and Zeng 2001 In summary previous empirical work has highlighted a set of variables that should underlie a model of civil unrest This research typically maps some combination of motivation and opportunity to the onset of insurgency and political instability Researchers have de ned motivation primarily as a relationship between the perceived needs of the general population and the ability of the state to ful ll these needs One can measure capacity in this manner by identifying the state s ability to provide collective goods in times of need Fortin 2008 an approach that is also consistent with the dynamical meaning of stability not the former based just on violence Yet this literature has also shown that motivation is meaningless without the means to execute insurgencies This has led researchers to de ne a set of enabling and constraining features of states that increase the likelihood of political instability In terms of enabling insurgency empirical studies highlight the importance of nancial resources and social alliances These are then buffered by constraints such as the strength of the institutional and military capacity of the state or the cover offered by its social political terrain that is population demographics or geographic features Finally if an insurgency were to get off the ground success is often tied to a mixed strategy using violence to disrupt government legitimacy and funding social programs to win over the general population Kalyvas 2006 Mousseau and Mousseau 2007 Some of these features have already made their way into existing computational models of civil unrest as mentioned in the next section However as demonstrated below our model provides a wider treatment of these phenomena for understanding and eventually predicting O Brien 2009 how micro level dynamics between society and government affect overall polity stability Earlier Computational Modeling The rst computational models of civil unrest appeared during the Cold War in parallel to but methodologically distinct from empirical studies These early models focused speci cally on the dynamics of asymmetric con ict that is irregular war guerilla war insurgency and other forms of domestic political violence that may induce instability in a polity for both pure theoretical and applied prediction purposes From a formal methodological perspective these earlier models

34

MASON RebeLand

used systems of ordinary differential or difference equations ODEs to represent the main macro structural relations among belligerents and consequently were typically implemented using the system dynamics approach Forrester 1968 Hanneman 1988 These models attained their greatest success when representing asymmetric con ict at the most aggregate level Gilbert and Troitzsch 2005 53 These models were also often empirically specialized to represent prominent insurgency cases at the time such as the Vietnam War or the Soviet invasion of Afghanistan Milstein and Mitchell 1969 Ruloff 1975 Allan and Stahel 1983 Stahel 1985 In fact from the birth of computational modeling until the more recent introduction of object oriented methodology of agent based modeling ABM the dominant approach to computational exploration of civil unrest was through the use of system dynamics models Choucri et al 2007 or through large N studies for example King and Zeng 2001 Bates 2008 Most computational research in civil unrest is now undertaken using the ABM approach with relatively fewer models using the system dynamics approach Choucri et al 2007 The turn toward ABM is due to the recognition of the importance of bottom up causal processes in the development of civil unrest Fearon and Laitin 2003 de Rouen and Sobek 2004 Kalyvas 2006 Cederman and Girardin 2007 ABM permits researchers to move beyond the limitations of the aggregate approach underlying system dynamics models encouraging researchers to identify and experiment with the micro processes involved in the production of emergent macro scale social patterns or distributions Epstein and Axtell 1996 This has led researchers to attempt to understand the complex relationships of critical actors involved in socio political instability that is civilians government at various organizational levels rebels and government forces military or police The Sugarscape model of Epstein and Axtell was one of the rst ABM simulations to demonstrate the signi cance of this emergence method Epstein and Axtell 1996 Bigbee et al 2007 With a few simple rules Sugarscape was able to reproduce macro social patterns in the abstract permitting experimental investigation of micro dynamics capable of driving con ict as well as analysis of other economic and social dynamics for instance wealth distributions In subsequent work Epstein 2002 has applied Sugarscape s con ict mechanisms to produce one of the rst civil violence ABM simulations investigating the emergence of rebellion and ethnic cleansing behavior as a product of the perception of police force numbers and intervention tactics at the level of individual agents Unlike early state level ABM simulations incorporating concepts of nationalism and domestic culture such as the GeoSim model Cederman 2003 Ciof Revilla and Gotts 2003 or collective identity models Lustick 2000 the Sugarscape Civil War model Epstein 2002 takes a highly magni ed approach to examine the citizen based impetus for rebellion as opposed to simply outlining the effects of domestic unrest upon international relations The ISAAC and EINSTein models Ilachinski 2004 use a similar technique to model combat at the level of individual soldiers These models were also the rst to rigorously apply insights from complex adaptive systems such as the notion of self organization to replicate a range of nonlinear dynamics commonly found in irregular warfare These early models set examples that led to the development of a small suite of civil unrest models in subsequent years Current models of civil unrest extend earlier models through the introduction of some form of either natural or social complexity For example the IRUBA model Doran 2005 a meso scale replication of Epstein s civil violence model at the level of provincial con ict uses simple geographic features such as terrain and spatial distribution of rebel resources and forces to test various insurgency and counterinsurgency tactics Bennett s 2008 most recent work in this same area omits variations in terrain but instead includes social emotions such

Claudio Ciof Revilla and Mark Rouleau

35

as the distinction between the perception of fear and anger in domestic populations resulting in a favorable evaluation of the US military s new hearts and minds strategy The REsCape model Bhavnani Miodownik and Nart 2008 also omits geographic features but provides the rst representation of selected institutions within a repressive state These models use social networks for instance news media weather events tsunami social identities ethnicities and political economic freedom to explore the development of civil unrest The MASON RebeLand model builds on these extant efforts in the most recent generation of computational models The MASON RebeLand Model RebeLand is a model of a simple abstract polity designed to highlight essential and recognizable features of socio natural complexity required to generate bottom up civil unrest that can although not always lead to state failure Similar to REsCape Bhavnani et al 2008 and in a more specialized way than GeoSim events in RebeLand are driven by spatio temporal interactions of abstract civilian agents whose social support is needed to maintain governmental legitimacy and capacity in an effort to either quell from the point of view of the government or foment from the perspective of the rebel rebellion possibly leading to regime change RebeLand builds on previous meso scale simulations of civil unrest such as REsCape as well as micro scale simulation models such as IRUBA Bennett s insurgency model Epstein s civil violence model ISAAC and EINSTein particularly with respect to emphasis on emergent or generative collective behaviors and their impact on the state At the level of the individual RebeLand models movement patterns and attack strategies of both rebel and government forces the success of which in uences resulting struggles for state power In sum RebeLand builds on the meso and micro level strengths of its predecessors aiming to model socio political behaviors that are recognizable to a political scientist while maintaining minimalist parsimonious socio natural complexity The basic research questions addressed by RebeLand are two 1 2 How does a polity respond to various levels and combinations of societal stress and governmental performance How can insurgency domestic political instability or even state failure in some extreme cases emerge as a bottom up phenomenon in the life of a polity

Both questions have pure and applied motivation linked to theory and policy respectively The simulated model polity must be spatially situated geography is a constituent not optional feature of every country in the world and possess a system of government that conforms to canons of political science The latter requirement means that the governmental architecture must be recognizable to a political scientist not implicit or arbitrary Additionally following a complexity perspective emergent phenomena must be generated by the interaction of agents in a bottom up way it must not be hard wired The RebeLand model is written in MASON Multi Agent Simulator of Networks or Neighborhoods Luke Ciof Revilla Panait and Sullivan 2005 an agent based modeling simulation toolkit written in the Java programming language The MASON Project URL is http cs gmu edu eclab projects mason Examples of other agent based modeling toolkits include Netlogo Repast and Cormas Gilbert 2008 Nikolai and Madey 2009 An important design feature of the MASON system for developing social simulations is its original architecture in terms of completely separating computation

36

MASON RebeLand

from visualization thus increasing speed and other desirable features needed in this investigation for example data elds for individual cognitive structures in subsequent versions see Discussion Another MASON feature not yet used in current versions of RebeLand is its ability to run in combination with ECJ a widely used evolutionary computation software Other related MASON models developed thus far include the Wetlands HouseholdsWorld Hierarchies and AfriLand models all of which are also spatially oriented agent based models with interactive societal and natural environments to re ect geographically situated polities Collier and Hoef er 2004 Ciof Revilla Luke Parker Daniel Rogers Fitzhugh Honeychurch Frohlich Depriest and Amartuvshin 2007 Ciof Revilla et al 2008 Ciof et al 2008 Ciof Revilla and Rouleau 2009 RebeLand is a single country model whereas other MASON models thus far represent other forms of social organization Wetlands represents a society of simple hunter gatherers with minimal social complexity at one end of the spectrum while the international system in AfriLand represents another end of the spectrum as a multi country model with a substantial level of social complexity Model Structure Figure 1 shows a map view of RebeLand as a polity or country that is situated in a natural environment with a basic socio natural system or socio ecological ontology Simon 1996 Liu Dietz Carpenter Alberti Folke Moran Pell Deadman Kratz Lubchenco Ostrom Ouyang Provencher Redman Schneider and Taylor 2007 Ostrom 2009 The country itself consists of an island surrounded by water therefore omitting external or neighboring interactions with other countries that is lacking any neighboring international relations in this initial

FIG 1 Map of RebeLand Island showing its main natural and social features Cities are shown in green natural resources in yellow and rebel and government forces in red and blue respectively Roads and provincial boundaries are in gray and yellow respectively Physical topography is shown on a green tone scale and the island is surrounded by ocean Source Prepared by the authors

Claudio Ciof Revilla and Mark Rouleau

37

version Both polity and environment are simple but grounded in each relevant domain The socio natural interaction is also intentionally simple re ecting mostly environmental effects on the polity Subsequent research will also examine long term anthropogenic effects on the environment The environmental and political components of RebeLand are as follows Environmental Component The RebeLand environment consists of terrain and a simple weather system that simulates climate dynamics For example climate change events such as prolonged droughts climate variability and other climatological features can be simulated in RebeLand Research on societal effects or consequences of climate change and variability was a design requirement In turn the terrain consists of physical topography and land cover Additionally generic natural resources for example oil diamonds gold or similar yellow triangles in Figure 1 are distributed over the terrain Additional features such as hydrology and other ecosystemic components can also be added Political Component The political component of RebeLand is speci cally designed to answer the basic research questions stated earlier Following a political science orientation the RebeLand polity consists of a society and a system of government for dealing with public issues through public policies see Figures 2 and 3 as in the standard model of a basic polity Almond Bingham Powell Dalton and Strom 2006 Ciof Revilla 2009 Initially and as we demonstrate in the subsequent section the government formulates policies to address issues that affect society Later in the simulation under some conditions the society can also generate insurgents that interact with government forces as well as other emergent phenomena There are three provinces in RebeLand see Figure 1 each with a subset of population centers resources roads and other features of the country Provincial boundaries yellow were drawn to produce approximately equal size but this and other features can be easily modi ed Following a social geography orientation the population society is not arbitrarily located but rather is concentrated in centers a capital city provincial cities towns and villages and the set of population centers is distributed throughout the island according to a Zipf distribution with some noise A system of roads connects population centers with natural resource locations The roads

FIG 2 High level UML class diagram of a simple polity such as that represented in RebeLand Government manages public issues through policies A more detailed UML class diagram of RebeLand omitted here for space considerations would include levels of public administration across the country State and Cities in the three provinces as well as heterogenous agents general population rebels and military Source Adapted from Ciof Revilla 2009

38

MASON RebeLand

FIG 3 Graphical interpretation of the main simulation loop from the perspective of the RebeLand socio natural environment Issues enter the environment with a user de ned issue onset rate a lognormal decay rate and a power law distributed magnitude This allows users to de ne the level of stress a government will probably face in a given simulation run The agent activation state Activate Agents is detailed Figures 4 6

were drawn using a gradient driven algorithm that takes into consideration distance and terrain as detailed below Given the ontology just described there are two classes of RebeLand agents Primary agents consist of the general population cities and the state Cities represent local public administration organizations whereas the state represents the system of government of the overall polity national government Secondary agents consist of rebels generated from the general population under a range of conditions discussed below rebel groups and government forces representing police and military units Rebels are supported by rebel group organizations horizontal polities Ferguson and Mansbach 1996 that support them and fund alternative policies that rival of cial state policies Thus rebel groups represent horizontal or alternative polities or organizations such as Hamas Hizbullah or Sendero Luminoso that compete with the state the of cial vertical polity regarding the provision of public goods to the population Government forces seek to destroy insurgents by attacking them and guarding the home city Dynamics Besides the formalization of entities in a structural sense a key strength of agent based models lies in the ability to formalize dynamics in complex social systems such as in a polity From a high level perspective RebeLand aims to operate as all polities do At any given time a public issue affects the population which causes societal stress In response government formulates and implements policies that aim to eliminate or mitigate stress on the population Government operates with capacity derived from revenues produced by taxes public nance and taxes are paid by

Claudio Ciof Revilla and Mark Rouleau

39

FIG 4 Graphical representation of the main simulation loop from the perspective of the general population When activated by the scheduler a general population agent will perform its day to day functions update its satisfaction assess its support for the government based on the government s ability to address pending social issues and assess its support for a rebel faction based on the rebel group s funding of alternative policies and the rebel use of violence as it affects the general population

the population based on disposable income derived from labor Normally state capacity is suf cient to deal with public issues but various factors can contribute to instability and even political change just like in real world polities More speci cally initialization and the main simulation loop in RebeLand are as follows Understanding these dynamics is important for purposes of both analysis and applications early warning prediction as lack of familiarity with the underlying computational processes of the simulation model can be a source of misinterpretation Initialization The RebeLand simulation begins by generating a random island environment Starting from the center most cell a greedy algorithm selects a pre de ned number of contiguous grid cells that will serve as the island s land cells Next a terrain generation algorithm makes a series of random elevation changes conditioned by a user de ned parameter of three possible natural environments grassland hills and mountains The simulation then randomly distributes resources oil gold and diamonds throughout the island and calculates their pro tability based on current market value amount of resource present distance to nearest shoreline tradeability and distance to the next nearest resource Following a Zipf distribution of cities sizes Berry Grif th and Tiefelsdorf 2008 the city center of the largest city is placed on the cell of the most pro table resources One thing to note is that each cell has a maximum general population capacity Thus if the city center cell is overpopulated a random neighboring cell is chosen and the remaining population inhabits this

40

MASON RebeLand

cell This process continues until all general population agents belonging to this city have been placed onto the map The city placement algorithm is repeated with the city center of the next largest city being placed on the cell of the next most pro table resource and so on until all cities have been placed on the island Finally an A algorithm is used to draw a shortest path road network between the cities accounting for both elevation changes and linear distances Main Simulation Loop and Typical Simulation Run The RebeLand main simulation loop conforms to the known system and processes of a polity based on contemporary political science theory Ciof Revilla 2009 In a typical model run primary agents general population cities the state perform day to day tasks including a basic subsistence cycle for the general population see Figures 3 6 Issues affect the population in the cities or more generally the national population as a whole Issues can originate within the society itself such as in ation increasing an issue endogenous to the society or in the environment drought or other ecological change or disaster affecting agricultural productivity exogenous to the society Local city governments produce policies to deal with public issues If population dissatisfaction increases due to stress insurgents may be generated from the population If so then government generates counterinsurgency units to combat and manage the insurgency if and when it is capable Insurgency is also a source of public issues and stress on the population along with other sources such as the environment and or endogenous social processes

FIG 5 Graphical representation of the main simulation loop from the perspective of the city agent Each city will tax its population to gain a source of revenue This revenue is then used to implement policies as a means of offsetting the effects of a pending social issue in ation ood etc The city is also responsible for generating military units to defend itself from rebel attacks

Claudio Ciof Revilla and Mark Rouleau

41

FIG 6 Graphical representation of the main simulation loop from the perspective of the state The state is a high level decision maker responsible for revenue redistribution in the case where some cities have an excess of funding while other cities lack the funding needed to support necessary policies

More speci cally once the RebeLand island environment and society are de ned a typical simulation run involves a number of potential agent to agent and agent to environment interactions The simulation schedule activates agents at random and once activated each agent both performs expected tasks and responds to its current environment For example general population agents attempt to earn income purchase goods and maintain their subsistence needs as expected tasks Depending on their social condition general population agents will adjust their satisfaction and regime support values appropriately City government agents tax the general population to build revenue and distribute welfare bene ts If economic or environmental issues are present the city agent will attempt to use its available revenue to generate policies which then dampens the effects of these issues on their citizens that is mitigates social stress a key function of government and public policy in a stable polity If rebel forces are present city agents use conscription to create a military force to defend the city Finally the last of the non mobile agents the state agent simply directs a share of the remaining city revenue toward the most pressing issues left unsupported Both general population agents and city agents are also capable of producing mobile agents when conditions necessitate For example general population agents can create rebel groups or individual rebel agents when their satisfaction level drops below a minimum threshold and their risk propensity is relatively high On the other hand city agents create military units in response to the presence of these rebels Both these rebel and military agents are then added to the simulation schedule and the main simulation loop changes to include their actions Upon activation rebel agents pick a target for terrorism and attempt to move toward this objective If no military units are present the rebel agent moves toward its target and when reaching the target commits the terrorist act If military units are present

42

MASON RebeLand

FIG 7 RebeLand simulation run under a stable scenario where government has suf cient capacity to manage public issues that arise in the normal life of the polity and public support remains favorable General population needs under a stable scenario Perceived wealth green available food blue perceived security red Support for government blue and support for rebels red under a stable scenario Trends in state capacity under a stable scenario Number of concurrent issues red concurrent policies green

rebels simply attempt to move away Military units on the other hand seek out rebels and engage in combat on sight The success of these combats is then determined by a random draw weighted by the size and resource endowment of each combatant force In sum the main simulation loop of RebeLand consists of a number of expected and reactive actions that result in the development of recognizable patterns of civil unrest as an emergent phenomenon

Claudio Ciof Revilla and Mark Rouleau

43

Computational Demonstration Results In this section we report results from three experimental scenarios simulated in RebeLand to illustrate its potential for basic and applied research Each scenario was run numerous times with similar results and the results reported here are representative of each scenario The rst scenario represents a stable political situation in the sense de ned above where government successfully manages public issues and society is satis ed Switzerland The second scenario is a signi cantly more contentious political situation where public issues produce extensive societal discontent dissent and incipient insurgency which however is managed by government Mexico Russia and Columbia The third scenario is an unstable political situation where the insurgency manages to topple the government bringing about state failure Somalia For each scenario we report three sets of simulation results using the timeseries graphs shown in Figures 7 9 a General population needs in terms of average perceived wealth green average food level blue and average security value red b Population support for government blue versus support for rebels red representing key dimensions of public opinion and c Trends in state capacity in terms of number of current public issues being managed red number of non security issues blue and number of current polices green

Stable Scenario Figure 7 shows results from the rst scenario or stable regime This baseline case shows the general population quite happy in terms of average perceived wealth eating well and feeling secure Figure 7a as in a stable and prosperous country Concurrently levels of support for government are high and conversely support for rebels remains low Figure 7b As well governmental policies keep track of emerging public issues Figure 7c so the polity is operating in a stable mode Some minor support for rebels can exist even in this stable scenario because the state has suf cient capacity to deal with issues Contentious Scenario Results from the second contentious scenario are shown in Figure 8 This case shows the general population as being less satis ed in terms of average perceived wealth eating less well and feeling somewhat less secure Figure 8a as in a less stable but still mostly viable country In this case levels of support for government can remain high but conversely support for rebels is ocassionally spiked when hunger and poverty begin to take hold Figure 8b In this scenario governmental policies have a harder time keeping up with emerging public issues Figure 8c so the polity is operating in a decreasingly stable increasingly unstable mode Note that support for rebels 8b red drops back down after each spike but resettles at slowly rising levels indicative of a fundamental diachronic rift Unstable Scenario State Failure Results from the third unstable scenario are shown in Figure 9 This case shows a more severe decrease in the average perceived wealth of the general population in addition to eating less well and feeling signi cantly less secure

44

MASON RebeLand

FIG 8 RebeLand simulation run under a contentious scenario where government still has suf cient capacity to manage public issues that arise in the normal life of the polity but insurgency emerges as a consequence of popular dissatisfaction Legend is the same as in Figure 7 General population needs under a contentious scenario Support for government and support for rebels under a contentious scenario Trends in state capacity under a contentious scenario

Claudio Ciof Revilla and Mark Rouleau

45

FIG 9 RebeLand simulation run under an unstable scenario where government has insuf cient capacity to manage public issues and the emergent insurgency manages to topple the government Legend is the same as in Figure 7 plus non security issues blue General population needs under an unstable scenario Support for government and support for rebels under an unstable scenario Trends in state capacity under a contentious scenario

46

MASON RebeLand

FIG 10 Individual agent distributions in RebeLand Income and satisfaction These simulated results resemble known empirical patterns Source Prepared by the authors Pareto distribution power law of individual agent income Bimodal distribution of agent satisfaction

Figure 9a as in an increasingly unstable polity with declining prosperity In this case levels of support for government plummet catastrophically around time 325 with a corresponding rise in support for rebels Figure 9b In this unstable scenario governmental policies do not keep track of emerging public issues Figure 9c even when such issues no longer concern security as shown by the blue non security issues metric on Figure 9c This scenario produces state failure as an emergent phenomenon a consequence of the government being overwhelmed by public issues relative to state capacity and concomitant popular support for rebels The interesting political phenomena in RebeLand namely shifts in public moods onset of insurgency and its subsequent development and governmental crises and state failure episodes always occur as emergent phenomena not as directly hard wired processes or events and consistently across all three scenarios In other words RebeLand is universally capable of producing these

Claudio Ciof Revilla and Mark Rouleau

47

phenomena as a result of its own endogenous dynamics as a generative computational theory should Epstein 2006 These properties are as signi cant for theoretical purposes as they are for practical policy and analysis purposes Interventions on complex systems based on poorly understood dynamics can lead to undesirable consequences In addition to these results RebeLand is also able to replicate two signi cant patterns commonly found in real world polities a Pareto power law distribution of income Figure 7a and a bimodal distribution of popular satisfaction 7b Both results lend face validity to the model Empirical features such as these and others are necessary in models intended for applied purposes early warning prediction whereas they are less necessary in purely theoretical models for developing basic understanding or insights Discussion Our results so far demonstrate two important points related to the initial research questions that RebeLand was designed to investigate i the state of the polity as a whole does indeed respond to various levels of societal stress and governmental performance by remaining stable scenarios 1 and 2 or in some extreme situations scenario 3 undergoing regime failure consistent with contemporary political theory and ii the overall political dynamics insurgency political instability and state failure emerge as bottom up phenomena they are not hard wired through complex mechanisms and feedback loops that can be traced down to the perceptions decisions and actions of many individual agents in the populations in interaction with government and a natural environment Although many features and processes would still have to be included to approximate a real world country these basic results provide signi cant face validity and encouraging prospects for further analysis and development In reference to the extant empirical literature mentioned in the introduction RebeLand resembles a semi democracy that is somewhere between a democracy and autocracy This is because the polity is designed to respond to societal stress with governmental policies aimed at mitigating or eliminating stress as in a democracy that responds to citizen s needs However RebeLand lacks a constitutional speci cation for governmental turnover and accountability institutional design Landman 2008 218 in this respect resembling more an autocracy Interestingly as a semi democracy state failure in RebeLand scenario 3 shows the effects of stress and insuf cient governmental capacity recently reported by the empirical literature For example very high levels of infant mortality re ecting a high level of stress on the population and legislative inef ciency correlate with high risk of state failure King and Zeng 2001 Similarly predation on governmental resources rent seeking and corruption which are common features in many countries on a global basis can similarly bring about state failure in semi democracies that once stood a reasonable chance of becoming more stable democratic polities Bates 2008 In sum results from RebeLand are consistent with the empirical nding that partial democracies have a higher risk of failure than either full democracies or autocracies King and Zeng 2001 650 This is mostly because partial democracies have many of the same structural and processual features of full democracies mainly the issue stress policy response cycle of a basic open polity but lack the ef ciencies and capabilities of a welldeveloped democratic polity due to rent seeking and other pathologies Further computational analysis with very low and very high levels of democracy are needed to simulate lower levels of state fragility in those extreme cases The MASON RebeLand model along with other similar agent based models that inspired it for example Bennett 2008 Bhavnani et al 2008 contributes to the scienti c literature on state stability and failure analysis in several ways

48

MASON RebeLand

1

2

3

The ABM approach enables the formalization of an entire polity in a way that goes beyond what other types of modeling approaches statistical or mathematical game theoretic or dynamical system can formalize or model This feature also contributes to social science interdisciplinary integration and the implementation of a whole system approach Computational simulation modeling enables the experimental analysis of policies and other features that cannot be implemented through earlier approaches For example the agent based modeling approach lends itself to asking many what if questions of theoretical and practical interest for example for early warning and prediction well beyond what can be accomplished through earlier approaches such as econometric models Spatial agent based models also permit a viable implementation of coupled socio natural systems and the complex interaction that arise in the normal life of a geographically situated polity Additionally this can be done without excluding other non spatial features such as social networks social organizations or institutions

The agent based modeling approach in social science such as but not limited to areas of interest in political science and international relations can be used for early warning and prediction if and only if the simulation system passes a variety of rigorous veri cation and validation tests also known as formal tests of internal and external validity Taber and Timpone 1996 Ciof Revilla 2002 In the case of RebeLand this would require the speci cation of a target system some real world country to be represented by the simulation system in such a way that the early warning and prediction questions to be investigated by the latter would be within the range of accuracy provided by the speci cation of the former For example in RebeLand a single public issue is usually insuf cient to bring about state failure even in less than optimal situations when government corruption exist or policy implementation is not very ef cient By contrast the concurrent stress of several major public issues affecting society such as the conjunction of in ation with climate change with resource depletion and with insurgency can cause state failure even when government corruption and policy inef ciencies are low In terms of early warning the agent based model offers the opportunity to look under the hood and examine precursory dynamics before the onset of major changes such as is typical of complex systems that can fail catastrophically through continuous drift failure in critical subsystems at lower micro levels In RebeLand the catastrophic collapse of support for government around time T 320 is detectable by earlier warning signs that begin to drift Figure 9 such declines in perceived wealth food availability and increasing inability of government policies in tracking emergent public issues The timing of sudden change will become better understood as we mark progress in the understanding of these precedent anticipatory dynamics and stresses The key difference between the agent based modeling approach to predicting state stability failure and earlier methodological approaches that were less integrative lies in the ABM focus on antecedent processes and causal dynamics that lead to emergent phenomena coups rebellions and large scale violence can occur through a variety of modes or complex combinations of state variables that can be understood through simulation but are dif cult or impossible to discover let alone explain through statistical models that lack causal processes The representation of causal processes attitudes and behaviors is explicit in ABMs just as they are in the real world when decision makers estimate risk individuals assess their well being and populations experience collective dynamics The signi cance and potential contribution of ABMs in social science is comparable

Claudio Ciof Revilla and Mark Rouleau

49

with that of other object based computational models in other areas of science such as seismology or ecology population biology both areas where macrophenomena emerges from the interaction of many individual micro agents Improvements in early warning and prediction of social and international events will become more feasible through future advances in ABM veri cation and validation In future research we plan to experiment with RebeLand in several directions 1 Additional experiments need to be conducted to gain a better understanding of the relationship between societal stresses state policy responses and polity stability For example a better understanding of societal versus natural stress is desirable A much more detailed understanding of precedent dynamics and aggregate change is necessary for early warning and prediction purposes Such applied goals are feasible but can only be attained through greater emphasis on the conditions under which sudden changes and other surprises can occur in addition to advances in empirical validation veri cation Norms are not explicit in this initial version of RebeLand In the referent target system norms are used by population agents government agents and rebel agents so the simulation system could make greater use of norms as explicit entities that govern agent decisions and behaviors perhaps within a BDI framework beliefs desired and intentions RebeLand can be used to populate a region or continent an international system on some scale not just a single country with appropriate variations in polity features Efforts in this direction are already under way under this same project MASON AfriLand model Ciof Revilla and Rouleau 2009 Greater emphasis on cognitive agent architecture is desirable especially for scenarios where the polity becomes unstable or failing when perceptions also play an increasing role

2

3

4

5

Conclusions Computational models of polities and insurgency civil unrest guerrilla violent rebellion and internal war have a research tradition dating back several decades across the social sciences That earlier generation of models was written in procedural programming languages as opposed to the current generation of computational models that are written primarily in object oriented programing languages such as Java Earlier agent based models ABMs on similar topics of domestic politics have been useful in covering parts of RebeLand s scope just as we hope RebeLand will inspire further developments The Java based MASON RebeLand model presented here offers three innovations with respect to earlier models of polities and insurgency i an explicit polity model with politically complete structure and processes for society government and policy making ii social and natural model components within an integrated socio natural system that captures ecosystemic complexity and iii generative dynamics where insurgency and the state of the polity stable unstable failing failed and recovering occur as emergent phenomena under a range of social and environmental conditions not as predetermined aggregate outcomes Taken together these three innovations allow researchers to conduct a variety of new computational or virtual experiments with RebeLand Such experiments can be particularly valuable in terms of providing new

50

MASON RebeLand

insights into and understanding of the complex dynamics between society government violent opposition policy alternatives and environmental ecosystems the later included but simpli ed in RebeLand Three scenarios were demonstrated in this article showing stable unstable and failing polity conditions An important result pertains to the general overall resiliency of a polity normally requiring not just one or a few stressful issues to experience polity failure but a large set in combination such as in ation plus insurgency plus environmental stress Although more research is needed in terms of model validation the overall behavior of RebeLand has face validity including the generation of realistic distributions such as those for income Pareto distribution and public opinion bi modal when under stress The MASON computational toolkit also permits additional experiments and extensions beyond the initial set reported in this article In the area of early warning and forecasting models like RebeLand are more likely to be useful in terms of revealing unforeseen dynamics or consequences in a qualitative sense rather than exact quantitative predictions concerning the precise timing and intensity of events From empirical research we know that the frequency of political events of many kinds coups crises wars and revolutions occurs with approximately Poissonian random distribution However the intensity or severity of events is more often of a different kind such as a Weibull distribution Experiments conducted with agent based models such as RebeLand and others should attempt to replicate such distributions and provide new insights for developing a better understanding of the underlying socio natural dynamics For example improvements in governmental capacity implementation reliability and lower levels of corruption should show signi cant improvements on polity stability an important prospect for stabilization and reconstruction operations Such results may not immediately yield actionable policy recommendations in terms of speci c programs but at the very least they may offer new insights of value to both researchers and policy analysts References
Allan Pierre and Albert A Stahel 1983 Tribal Guerrilla Warfare Against a Colonial Power Analyzing the War in Afghanistan Journal of Con ict Resolution 27 590 617 Almond Gabriel A Jr G Bingham Powell Russel J Dalton and Kaare Strom 2006 Comparative Politics Today A World View New York Pearson Longman Anderson Benedict 1983 Imagined Communities New York Verso Bates Robert H 2008 When Things Fell Apart State Failure in Late Century Africa New York Cambridge University Press Bennett Scott D 2008 Governments Civilians and the Evolution of Insurgency Modeling the Early Dynamics of Insurgencies Journal of Arti cial Societies and Social Simulation 11 4 Available online Berry Brian J L Daniel A Grif th and Michael R Tiefelsdorf 2008 From Spatial Analysis to Geospatial Science Geographical Analysis 40 229 238 Bhatia N P and Giorgio P Szego 2002 Stability Theory of Dynamical Systems Berlin Springer Bhavnani Ravi Dan Miodownik and Jonas Nart 2008 REsCape an Agent Based Framework for Modeling Resources Ethnicity and Con ict Journal of Arti cial Societies and Social Simulation 11 2 Available online Bigbee Anthony Claudio Ciof Revilla and Sean Luke 2007 Replication of Sugarscape Using MASON In Agent Based Approaches in Economic and Social Complex Systems IV Post Proceedings of The AESCS International Workshop 2005 edited by T Terano H Kita H Deguchi and K Kijima Tokyo Springer Cederman Lars Erik 2003 Modeling the Size of Wars From Billiard Balls to Sandpiles American Political Science Review 97 135 150 Cederman Lars Erik and Luc Girardin 2007 Beyond Fractionalization Mapping Ethnicity on Nationalist Insurgencies American Political Science Review 101 173 185

Claudio Ciof Revilla and Mark Rouleau

51

Choucri Nazli D Goldsmith Stuart Madnick J B Morrison and Mark Siegel 2007 Using System Dynamics to Model and Better Understand State Stability Paper presented at the 25th International Conference of the System Dynamics Society Boston MA Ciof Revilla Claudio 2002 Invariance and Universality in Social Agent Based Simulations Proceedings of the National Academy of Science of the U S A 99 Suppl 3 14 7314 7316 Available online Ciof Revilla Claudio 2009 Simplicity and Reality in Computational Modeling of Politics Computational and Mathematical Organization Theory 15 26 46 Ciof Revilla Claudio and Nicholas M Gotts 2003 Comparative Analysis of Agent Based Social Simulations GeoSim and FEARLUS Models Journal of Arti cial Societies and Social Systems 6 4 Available online Ciof Revilla Claudio J Daniel Rogers and Maciej Latek 2008 The MASON HouseholdWorld Model of Pastoral Nomad Societies In The Science of Social Simulation Selected Post Proceedings from the Second World Congress in Social Simulation Tokyo Springer Ciof Revilla Claudio and J Daniel Rogers 2008 Computing the Steppes Data Analysis for Agent Based Modeling of Polities in Inner Asia Paper presented at the Annual Conference of the American Political Science Association August 28 31 2008 Boston MA Ciof Revilla Claudio and Mark Rouleau 2009 MASON AfriLand A Regional Multi Country Agent Based Model with Cultural and Environmental Dynamics Proceedings of the Human Behavior Computational Modeling and Interoperability Conference 2009 HB CMI 09 Joint Institute for Computational Science Oak Ridge National Laboratory Oak Ridge TN USA June 23 24 2009 Ciof Revilla Claudio and Harvey Starr 2003 1995 Opportunity Willingness and Political Uncertainty Theoretical Foundations of Politics In Necessary Conditions Theory Methodology and Applications edited by G Goertz and H Starr Lanham MD Rowman and Little eld Publishers Inc Ciof Revilla Claudio Sean Luke Dawn C Parker J Daniel Rogers William W Fitzhugh William Honeychurch Bruno Frohlich Paula Depriest and Chunag Amartuvshin 2007 Agent based Modeling Simulation of Social Adaptation and Long Term Change in Inner Asia In Advancing Social Simulation The First World Congress in Social Simulation edited by T Terano and D Sallach Tokyo New York and Heidelberg Springer Verlag Collier Paul and Anke Hoef er 1998 On Economic Causes of Civil War Oxford Economic Papers 50 563 573 Collier Paul and Anke Hoef er 2004 Greed and Grievance in Civil War Oxford Economic Papers 56 563 595 Connor Walker 1994 Ethnonationalism Princeton NJ Princeton University Press Deutsch Karl W 1953 Nationalism and Social Communication Cambridge MA The MIT Press Doran Jim 2005 Iruba An Agent Based Model of Guerrilla War Process In Representing Social Reality Pre proceedings of the Third Conference of the European Social Simulation Association ESSA 198 205 Epstein Joshua 2002 Modeling Civil Violence An Agent Based Computational Approach Proceedings of the National Academy of Science of the U S A 99 Suppl 3 7243 7250 Available online Epstein Joshua 2006 Generative Social Science Studies in Agent Based Computational Modeling Princeton NJ Princeton University Press Epstein Joshua M and Robert Axtell 1996 Growing Arti cial Societies Social Science from the Bottom Up Cambridge MA The MIT Press Fearon James D and David D Laitin 2003 Ethnicity Insurgency and Civil War American Political Science Review 97 1 75 90 Ferguson Yale H and Richard W Mansbach 1996 Polities Authority Identities and Change Columbia SC University of South Carolina Press Forrester Jay W 1968 Principles of Systems Cambridge MA Wright Allen Press Fortin Jessica 2008 Measuring State Capacity in Postcommunist Countries Paper presented at the American Political Science Association Annual Meeting Boston MA August 28 31 2008 Gellner Ernest 1983 Nations and Nationalism Ithaca NY Cornell University Press Gilbert Nigel 2008 Agent Based Models Thousand Okas CA Sage Publishers Gilbert Nigel and Klaus Troitzsch 2005 Simulation for the Social Scientist Second edition Buckingham and Philadelphia Open University Press Gurney Joan Neff and Kathleen J Tierney 1982 Relative Deprivation and Social Movements A Critical Look at Twenty Years of Theory and Research Sociological Quarterly 23 33 47 Gurr Ted Robert 1970 Why Men Rebel Princeton NJ Princeton University Press Hanneman Robert A 1988 Computer Assisted Theory Building Modeling Dynamic Social Systems Newbury Park CA Sage

52

MASON RebeLand

Horowitz Douglas A 1985 Ethnic Groups in Con ict Berkley CA University of California Press Huntington Samuel 1996 The Clash of Civilizations and the Remaking of World Order New York Simon and Shuster Ilachinski A 2004 Arti cial War Multi Agent Based Simulation of Combat Singapore World Scienti c Kalyvas Stathis N 2006 The Logic of Violence in Civil War Cambridge Cambridge University Press King Gary and Langche Zeng 2001 Improving Forecasts of State Failure World Politics 53 623 658 Landman Todd 2008 Issues and Methods in Comparative Politics An Introduction Third ed London and New York Routledge Liu Jianguo Thomas Dietz Stephen R Carpenter Marina Alberti Carl Folke Emilio Moran Alice N Pell Peter Deadman Timothy Kratz Jane Lubchenco Elinor Ostrom Zhiyun Ouyang William Provencher Charles L Redman Stephen H Schneider and William W A Taylor 2007 Complexity of Coupled Human and Natural Systems Science 317 1513 1516 Luke Sean Claudio Ciof Revilla Liviu Panait and Keith Sullivan 2005 MASON A Java Multi Agent Simulation Environment Simulation Transactions of the Society for Modeling and Simulation International 81 517 527 Available online at http cs gmu edu eclab projects mason Lustick Ian S 2000 Agent Based Modelling of Collective Identity Testing Constructivist Theory Journal of Arti cial Societies and Social Simulation 3 Available online Milstein Jeffrey S and W C Mitchell 1969 Computer Simulation of International Processes The Vietnam War and the Pre World War I Naval Race Peace Research Society International Papers 12 117 136 Morrison Denton E 1971 Some Notes Toward Theory on Relative Deprivation Social Movements and Social Change American Behavioral Scientist 14 675 692 Mousseau Michael and Demet Yalcin Mousseau 2007 How the Evolution of Markets Reduces the Risk of Civil War Paper presented at 4th Annual General Conference of the European Consortium for Political Research ECPR University of Pisa Italy September 6 7 2007 Nikolai Cynthia and Gregory Madey 2009 Tools of the Trade A Survey of Various Agent Based Modeling Platforms Journal of Arti cial Societies and Social Simulations 12 22 Available online O brien Sean P 2009 Crisis Early Warning and Decision Support Contemporary Approaches and Thoughts on Future Research International Studies Review 12 87 104 Ostrom Elinor 2009 A General Framework for Analyzing Sustainability of Socio Ecological Systems Science 325 419 422 de Rouen Karl R Jr and David Sobek 2004 The Dynamics of Civil War Duration and Outcome Journal of Peace Research 41 303 320 Ruloff Dieter 1975 Kon iktlosung durch Vermittlung Basel Switzerland Birkhauserverlag Simon Herbert A 1996 The Sciences of the Arti cial 3rd edition Cambridge MA MIT Press Stahel Albert A 1985 Dynamic Models of Guerilla Warfare In Dynamic Models of International Con ict edited by U Luterbacher and M D Ward Boulder CO Lynne Rienner Taber Charles S and Richard J Timpone 1996 Computational Modeling Thousand Oaks CA London and New Dehli Sage Publications