Collective violence spreads through social learning among autonomous agents, especially when decisions of the observing agents are not simultaneous, but vicariously yet rationally made on the basis of information they receive from the actions of other agents on in-group. These actions and decisions can be expressed as a sequential game in which the outcome of prior actions influence the likelihood of subsequent actions as beliefs are updated through Bayseian updates at each time step. The model presented below hence underscores the dynamic underlying process at play during various episodes of collective violence. This model specifies the mechanism through which a local episode of collective violence may amplify the global likelihood such events, and specifies conditions under which violence waxes and wanes.
June 15, 2020