Epstein, Parket, and Steinbruner created a simple game to model revolutionary behavior.

Agent-based modeling assumes that individuals are driven by a sets of simple choices. This could be managing scarce resources or deciding to revolt against the government. A large number of individuals making these small choices in a dynamic game create emergent and complex results.


This working paper presents an agent-based computational model of civil violence. We present two variants of the Civil Violence Model. In the first, a central authority seeks to suppress decentralized rebellion. In the second, a central authority seeks to suppress communal violence between two warring ethnic groups.

This Civil Violence Model assumes two basic actors: Agents and Cops. Agents are the general civilian population who may be for or against the government. Cops are government protectors.

Agent opinion of the government is determined by two factors – Hardship and Perceived Legitimacy of Government.

Grievance = Hardship (1 – Legitimacy) or G = H(1-L)

Agents who enjoy economic wealth and perceive the government as legitimate have very low grievance levels. Hardship alone is not a critical factor. It must be combined with an illegitimate government to reach critical levels of grievance.

They also include Risk Preference and Vision. Some Agents are Risk Takers and others are Risk Adverse, even if the probable utility of either choice is neutral. Vision is the agent ability to “see” positions in their area and gather information. This determines their ability to see other agents and cops. They are risk averse around cops and risk favorable in the absence of cops. This models the probability of arrest.

The Net Risk = Risk aversion x Probability of Arrest or N=RP
T is the threshold where grievance outweighs net risk.

So Agent Rule A is G-N>T be Active; Otherwise be Quiet.

This sets up a basic rule in the game to model civil disobedience and rebellion.

Epstein et al create similar game models for the cops.
Cops use vision to detect Active Agents and ignore Quiet Agents. They arrest Active Agents and place them in the model prison for a period of time.

This model
1) Active Agents deceive Cops. They are risk adverse in the presence of Cops so they act like Quiet Agents. When the Cops leave, they act as Active Agents. This is because Net Risk outweighs Grievance near Cops.

2) Free Assembly Catalyzes Rebellion
A high density of Active Agents are undeterred by a low density of Cops. This simulates aggressive mob behavior. Rationally the net risk is low because they cops cannot arrest all agents. Even low levels of grievance outweigh the net risk, so otherwise Quiet Agents become Active Agents – they join the mob.

3) More Cops decrease rebellion even in cases of hardship and low legitimacy. Reduction of Cops decreases Net Risk and increases likelihood of rebellion. This can be the Police State rule.

4) Punctuated Equilibrium followed by Revolution
There are periodic outbursts of rebellion with indefinite quiet periods in between. Revolutions do not occur gradually.

Here are some of the interesting model results.
A large drop in government legitimacy in small increments over time does not produce a rebellion. After each small drop in legitimacy, a handful of Agents turn Active and Cops arrest this. This increases the size of the prison population and prevents a rebellion. Cops arrest the “tails” of Active Agents and prevent them from organizing large networks to rebel.

A small drop of legitimacy at once can lead to rebellion. A large number of Agents turn Active before Cops can arrest them. This triggers mobs of Active Agents.

This could explain seemingly small “Triggering” events that lead to revolution.

A few minor changes can make Epstein’s game more complex. Assume Cops are Loyal or Disloyal to the government in the same manner as Agents: G-N>T. Cops can join the rebellion if their grievance is high enough. Another minor change – Quiet Agents can “inform” Cops of Active Agents, like a Secret Police. This increases a Cops vision and reduces the ability of Active Agents to deceive Cops. We could add “Rebel Cops” who arrest Quiet Agents in times of unrest. This further changes the dynamics.

There is a second model that describes two rival groups of Agents (Green and Blue) at war, with the Cops acting as peacekeepers. This produces interesting results as well, including complex results like of genocide and safe havens. Peacekeepers are most effective in defending a perimeter around an ethnic cluster (Safe Haven) and least effective when dispersed.

These models can be used to help describe historical events. If they count all variable inputs into the system, the results should accurately describe the historical event. It also offers a way to better predict the probability of future events – similar to meteorology.