Abstract: This paper considers the diffusion of political violence as a strategic interaction between two or more combatants. If we conceptualize civil war as a process of competitive state-building, where each participant seeks to establish a monopoly of force in a given piece of territory, then it is unlikely that each actor independently decides to fight -- much less continue, expand or relocate the fighting to other battlefronts. Rather, we should expect combatants to make the best choices they can given the likely responses of their opponents. Conventional approaches to modeling the spatiotemporal dynamics of conflict overlook the joint, interactive nature of these decisions, impairing our ability to explain and predict the occurrence of both unilateral and two-sided violence. To facilitate more theoretically-meaningful inferences, stronger model fit and improved predictive accuracy, I present an empirical model for simultaneous binary outcomes with bivariate (or multivariate) conditionally autoregressive priors. I illustrate the approach using declassified micro-level event data on insurgency and counterinsurgency from the archives of the Soviet secret police.