Networks Coalitions and Revolutions
From Santa Fe Institute Events Wiki
Dynamics of insurgency
In the last decade, several attempts at modelling rebellion have been made. The most well-know is Epstein's model of rebellion. Whereas Epstein's attempt does not allow for agents to change their ideology, other attempts have modeled counterisurgency responses in such a great detail that the models lacked tractability. We add the spread of insurgency and counter-insurgency dynamics to Epsteins model taking into account historical evidence that inter-community relations (types of networks) can affect such dynamics. As a second step, we try to remove all superfluous detail from the model, arriving at a sociophysical model of insurgency.
Preliminary Setup
We want to study a situation of networks in which the dynamics of one (or more) networks influence the topology of the network that portrays the social stratification (directed, weighted network?) of a country. Agents can revolt and if a certain threshold is reached, the government is overthrown.
Team
Vessela Daskalova, Micael Ehn, John Paul Gonzales, Andreas Ligtvoet, Thomson McFarland, Sergey Melnik, Anna Pechenkina, Florian Sabou, Kang Zhao.
Model (Draft June 13, 2010)
The main goal is to create a simple model of when a revolution occurs, where revolution means a change in the current hierarchy of the network. We look at three forces connected to revolutions: the spread of insurgency, repression from the government, and feedback effects from repression on the spread of insurgency. We investigate how these three forces influence the occurrence of a revolution.
Connection to existing literature: the model is similar to a contagion-immunization model with an additional component, which we call feedback effect of repression.
To-do:
- Take a given network structure - scale free network
- Each node has three possible states - police, rebel, citizen
- Start with a population of police and citizens
- random contagion of the citizens nodes in the first period turns some into rebels
- introduce repression and feedback from repression in subsequent periods, where repression is random, but feedback from repression occurs for nodes immediately connected to the repressed node
- Types of interactions:
contagion:
R - C rebel infects citizen with prob p1
repression:
P - R police represses rebels with prob p2
P - C police represses citizens with prob p3
feedback from repression:
C - R citizen turns into rebel with prob p4
R - C rebel turns into citizen with prob p5
Vary the above-defined parameters p1-p5 to see under what conditions a revolution can occur.
Actually, I'm thinking we should simplify this some more:).
Relevant models from the NetLogo library:
Networks - Virus on a Network
Next meeting: Sunday, June 13, 8pm
End Model Draft (June 13, 2010)
Implementation in the RebellionOnANetwork model
- Take a given network structure - scale free network
* Actual implementation: random and scale free
- Each node has three possible states - police, rebel, citizen
* Actual implementation: also jailed
- Start with a population of police and citizens
* Actual implementation: there are X (=3) rebels to begin with
- random contagion of the citizens nodes in the first period turns some into rebels
* Actual implementation: rebellion spreads from rebels or outraged citizens
- introduce repression and feedback from repression in subsequent periods, where repression is random, but feedback from repression occurs for nodes immediately connected to the repressed node
* Actual implementation: implemented
- Types of interactions:
contagion:
R - C rebel infects citizen with prob p1 (rebellion-spread-chance)
repression:
P - R police represses rebels with prob p2 (detention-chance)
P - C police represses citizens with prob p3 (not implemented)
feedback from repression:
C - R citizen turns into rebel with prob p4 (outrage-chance)
R - C rebel turns into citizen with prob p5 (not implemented)
So basically, we only have three parameters that control spread and repression!! This should make Vasella happy.
Sweep parameters
Big sweep (17000) [NB: the runs after 17000 should not be used]
["initial-rebellion-size" [5 15 50]] ["rebellion-spread-chance" [1 2 5]] ["scalefree?" false] ["number-of-nodes" 300] ["initial-police-size" [10 25 110]] ["time-in-jail" [25 25 100]] ["detention-chance" [10 20 90]] ["outrage-chance" [10 20 90]] ["average-node-degree" 6]
Modest run (810)
["initial-rebellion-size" 5] ["rebellion-spread-chance" [1 2 5]] ["scalefree?" true false] ["number-of-nodes" 300] ["initial-police-size" [10 50 110]] ["time-in-jail" 50] ["detention-chance" [10 40 90]] ["outrage-chance" [10 40 90]] ["average-node-degree" 6]
Model 17 June (with fear and counter-movements)
Implementation in the RebellionOnANetwork model
- Take a given network structure - scale free network
* Actual implementation: random and scale free
- Each node has three possible states - police, rebel, citizen
* Actual implementation: also jailed
- Start with a population of police and citizens
* Actual implementation: there are X (=3) rebels to begin with
- random contagion of the citizens nodes in the first period turns some into rebels
* Actual implementation: rebellion spreads from rebels or outraged citizens
- introduce repression and feedback from repression in subsequent periods, where repression is random, but feedback from repression occurs for nodes immediately connected to the repressed node
* Actual implementation: implemented
- Types of interactions:
contagion:
R - C rebel infects citizen with prob p1 (rebellion-spread-chance)
repression:
P - R police represses rebels with prob p2 (detention-chance)
P - C police represses citizens with prob p3 (not implemented)
feedback from repression:
C - R citizen turns into rebel with prob p4 (outrage-chance)
R - C rebel turns into citizen with prob p5 (1 - outrage-chance)
additional parameter - return to normalcy:
R - C rebel turns into citizen with prob p5 (counter-spread-chance)
Possible networks
- Family
- Geography (neighbourhood)
- Job
- Recreation
- Political opinion, and of course
- Social stratficiation
In the first version of the model, we will only implement 2 networks (which essentially can be represented as 1 network with different tie-strengths).
Mechanisms
- Agents have certain political opinions which they transfer through their networks
- Political opinion should be multi-dimensional (but in the beginning possibly quite simple: pro-contra, violent-peaceful; things start to change if agents get contra+violent)
- At a certain threshold level the stratification is changed, due to agents changing the network (breaking ties with people they disagree with?)
- This will lead to new power structures
- Repression: some insurgents can be eliminated by government.
Approach
Andreas is in favour of a simple reasoning & exploratory approach. Once we have something going, we may find additional details on revolutionary dynamics to improve the model. (We can base this loosely on the literature below)
Thomson McFarland OK, to just to throw out some of the things that some of us were talking about at the SF Complex at a very high level and notation to be worked out (bear with me, it's been a while since I've been thinking about these things in an abstract way):
- Define a K-dimensional policy/ideological space
- Select N, the number of nodes/agents
- Draw N ideal points in the policy space (a vector with length k for each agent)-- Note that the distribution from which we draw these preferences is something to be manipulated
- Build a network (structure TBD) for these agents
- At time t:
- Each agent i looks in its neighborhood, calculates midpoint in the policy space between itself and all nodes in the neighborhood
- Each agent moves to the centroid of the midpoints
- Each agent looks in a parameterized radius around its new location, links to "new" agents whose location at t-1 is in the zone. Agents could also sever links with agents outside the radius (but note this could sever the "shortcuts" in a small world network). Radius can be manipulated.
- Iterate (until stable distribution of preferences is reached?)
This is a potentially rich baseline model of preference formation from which to work. We can manipulate aspects of the network generated in (4) in order to explore how network structure affects preference formation; we can also tune the distribution of preferences to see how that affects time to convergence, etc.
We can also build in a variety of power structures/institutions by changing how agents move, that is, by having the "midpoint" each agent calculates between itself and its neighbors be a weighted move along the distance between the two agents instead of a midpoint. The qualitative interpretation here is that some agents, as defined by the power structure, can "pull" others closer to their ideal point.
We can also build on decision rules for when coalitions, parties, or consensus is reached. If we could identify distributions of preferences and network structures that can generate such clusters in opposition to the established power structure, we could argue that a revolution is reached.
Just thinking/typing out loud here...
Literature
- Multiple networks and mobilization in the paris commune, 1871 - Roger V. Gould
- Structural opportunity and perceived opportunity in social movement theory: iranian revolution - Charles Kurzman
- Modeling civil violence - Epstein
Since code changes quite rapidly, this is the link to the latest code for RebellionOnANetwork
Original ideas
The Importance of Network Structure for the Formation of Coalitions (Vessela Daskalova)
Hi, I'm thinking of investigating which network structures are best for the formation of parties or coalitions under different conditions. One could assume that each agent has a list ranking different policies. If the majority of agents have the same policy as a first priority, a coalition or "party" can be formed. Initially, we could look at how the structure of given networks influences coalition formation. In the longer term this can be also extended to looking at when revolutions, defined as changes in the existing hierarchical structures, occur. Basically, is there a network structure, which is more conducive to being overthrown - why were the French and the Russian revolutions successful, and others not? Anyone interested?
- Hi guys - cool project - I suggest you take a look at Roger Gould's 1991 article on networks in the Paris Commune. Gould was one of the first people in Sociology to think about social networks in a systematic way, and he did think a lot about conflict, violence and revolutions.