Abstract
Recent efforts have dedicated significant attention to agent-based simulation models in a spectrum of fields, including modern warfare, with less attention paid to nonviolent campaigns. A type of computer experiment used to study complex scenarios, these models simulate agent actions subject to nonlinear dynamics, generating nondeterministic outcomes. An agent-based simulation model developed by Chenoweth of Harvard Kennedy School to study the dynamics of nonviolent resistance was reconstructed, and sensitivity analysis was conducted with the intent of identifying conditions most beneficial to its success. However, the Chenoweth model involves 25 factors, making interpretation difficult. Each iteration of the model produces a pass/fail/draw outcome, effectively censoring stalemates and requiring that each case be run sufficient times to estimate the nonviolent success probability for those conditions. Indiana University's high-performance supercomputer was available for computation, but an efficient test plan was needed. A series of Latin hypercube designs was constructed for this wargaming application, leveraging their space-filling qualities to obtain better visibility into factor effects and interrelationships. The study plan evolved to address findings about model functionality and within-case repeatability and in order to locate viable regions.