Computersimulation und Konfliktforschung am Beispiel von GeoSim

Agenda Einführung in die agenten-basierte Modellierung Schellings Segregations-Model Einführung in Geosim Anwendungen in der Konfliktforschung

Definition ABM ist eine computergestützte Forschungsmethode, die es dem Forscher erlaubt, künstliche Welten zu kreieren, diese zu analysieren und damit zu experimentieren. Bottom-up Basiert auf zellularen Automata und verteilter künstlicher Intelligenz

Disaggregierte Modellierung Wenn <Kondition> dann <Handlung 1> sonst <Handlung 2> Unbewegte Agenten Beobachter Bewegte Agenten Daten Organisationen von Agenten Künstliche Welt

Blick vom Berliner Fernsehturm

Ethnische Stadtteile Klein-Italien, San Diego Chinatown, New York

Nachbarschafts-Segregation Die Spielregeln Ein Agent bleibt, wenn min. 1/3 der Nachbarn „seiner Art“ ist < 1/3 Thomas C. Schelling Nobelpreis für Wirtschaft 2005 Ansonsten zieht er auf eine freie, angenehmere Position um

Schelling (Converted).mov Vorführung 1 Schelling (Converted).mov

Ergebnisse des Models Zahl der Nachbarschaften Zufriedenheit Zeit Zeit

Europa um 1500

Europa um 1900

“States made war and war made the state” – Charles Tilly

GeoSim GeoSim nutzt Repast, ein Java-Werkzeug Staaten sind hierarchische, begrenzte Akteure, die in einem dynamischen, gitter-basierten Netzwerk interagieren

Vorführung 2 GeoSim.mov

Emergente Ergebnisse Anzahl der Staaten Anteil der sicheren Gebiete Zeit Zeit

Mögliche Equilibria Multipolarität mit 15 Staaten (Vorführung) Unipolarität Bipolarität

Modeling conflict with ABM Configurations Processes Qualitative properties Example 3. Democratic peace Example 4. Emergence of the territorial state Distributional Example 2. State-size distributions Example 1. War-size distributions Cederman 2003. "Modeling the Size of Wars.” American Political Science Review 97:135-150.

Cumulative war-size plot, 1820-1997 To check whether Richardson's law still holds up, I used casualty data from the Correlates of War data set based on interstate wars in the last two centuries. (standard data set) If we plot the cumulative frequency that there is war of a larger size in a logarithmic diagram, the power law turns into a straight line. I.e. like the Richter scale of earthquakes, both axes are logarithmic, both size and cumulative prob. Obviously for small wars, this probability is close to one. For very large wars, however, the probability is going to be very small. As you can see, the linear fit is striking, in fact almost spooky! This result translates into a factor 2.6. The steeper the line, the more peaceful is the system. Robustness: see Levy + Hui But conventional IR theories have little to say about this finding. Thus it is a true puzzle. We need to turn elsewhere... Data Source: Correlates of War Project (COW)

Theory: Self-organized criticality log f f Input Output s-a log s Complex System s Slowly driven systems that fluctuate around state of marginal stability while generating non-linear output according to a power law. Examples: sandpiles, semi-conductors, earthquakes, extinction of species, forest fires, epidemics, traffic jams, city populations, stock market fluctuations, firm size In the late 1980s, physicists discovered a large family of complex systems that generate power-law regularities. Per Bak, coined the term self-organized criticality: Paradigmatic example: sandpile. If trickle sand slowly on a sandpile ==> pile will grow, but not in an even fashion. From time to time, there will be avalanches of various sizes. More abstractly put, there is a linear input building up tensions held back by thresholds. When disturbances happen, they follow a power law. (hyperbolic curve) Sandpile differs from Newtonian physics (e.g. billiard balls) scale invariance: avalanches of all sizes possible; no typical size (fractals) history matters: path dependence, butterfly effect, but unlike chaos because output is not just random threshold effects and positive feedback (friction holding back the sand, or like pushing a piano over a rugged floor, sometimes is slides) Simple analogy: applying SOC to our puzzle Linear input: technological change building up tension Thresholds: decisions to go to war Positive feedback: avalanches through interdependent decisions ==> ABM standard approach in non-equil. theory

Self-organized criticality Per Bak’s sand pile Power-law distributed avalanches in a rice pile

War clusters in Geosim t = 3,326 t = 10,000

Simulated cumulative war-size plot log P(S > s) (cumulative frequency) log s (severity) log P(S > s) = 1.68 – 0.64 log s N = 218 R2 = 0.991 Does the sample run generate a power-law distribution? Yes! Wit a R^2 at 0.991 it even surpasses the empirical distribution. Range over four orders of magnitude. The slope of –0.64 is also realistic. This looks very much like the distribution in the empirical case. But is this representative? I have chosen the sample run such that it generates median linear fit out of 15 replications: lowest at 0.975 and highest at 0.996. Look at histograms: See “Modeling the Size of Wars” American Political Science Review Feb. 2003

Modeling conflict with ABM Configurations Processes Qualitative properties Example 3. Democratic peace Example 4. Emergence of the territorial state Distributional Example 2. State-size distributions Example 1. War-size distributions Cederman 2003. “Explaining State Sizes: A Geopolitical Model.” Proceedings of Agent 2003, eds. Macal, North & Sallach. Argonne.

2. Modeling state sizes: Empirical data log Pr (S > s) (cumulative frequency) log S ~ N(5.31, 0.79) MAE = 0.028 1998 log s (state size) Data: Lake et al.

Simulating state size with terrain

Simulated state-size distribution log s (state size) log Pr (S > s) (cumulative frequency) log S ~ N(1.47, 0.53) MAE = 0.050

Modeling conflict with ABM Configurations Processes Qualitative properties Example 3. Democratic peace Example 4. Emergence of the territorial state Distributional Example 2. State-size distributions Example 1. War-size distributions Cederman, L-E & K S Gleditsch. 2004. "Conquest and Regime Change" International Studies Quarterly 48:603-629.

Simulating global democratization Source: Cederman & Gleditsch 2004

A simulated democratic outcome

Replications with regime change Democratic share of territory Without collective security Democratic share of territory With collective security 0 2000 4000 6000 8000 10000 Time 0 2000 4000 6000 8000 10000 Time

Modeling conflict with ABM Configurations Processes Qualitative properties Example 3. Democratic peace Example 4. Emergence of the territorial state Distributional Example 2. State-size distributions Example 1. War-size distributions Cederman & Girardin 2010. “Growing Sovereignty” International Studies Quarterly 54: 27-48.

Taxation in a linear state 1.576 Tax: k= .5 .576 Discount: d = .8 1.224 .4 0.4 0.4 0.6 0.6 0.6 1 2 3 4 5

Results from the linear model Proposition 1. Assuming exponential decay, direct rule is always more efficient Proposition 2. Assuming a step function, indirect rule is more efficient for low threshold values x* x* x‘ 1/(1-k) k(3-k) 1-k pIDR(n) pDR(n) n p(n) We can now do the math for the line state: Prop. 1: Given exponential decay, DR is always best Graph: thin curve IDR, thick curve DR Prop. 2: Given a step function x*, there is a region below 1/(1-k) where IDR is superior above x’ But if x* slides beyond this point, DR will always pay off Model technological change: simple shift from IDR  DR

The initial state of OrgForms

Modeling technological change

OrgForms: A dynamic network model Conquest Technological Progress Systems Change Organizational Bypass

Indirect rule in the “Middle Ages”

Direct rule in the modern system

Replications with moving threshold and slope

Exploring geopolitics using agent-based modeling OrgForms GeoSim 0 GeoSim 4 GeoContest

Toward more realistic models of civil wars Our strategy: Step I: extending Geosim framework Step II: conducting empirical research Step III: back to computational modeling

Step I: Nationalist insurgency model Use agent-based modeling to articulate identity-based mechanisms of insurgency In Order, Conflict, and Violence, eds. Kalyvas, Shapiro & Masoud. CUP, 2008.