In this paper I develop a theoretical toolkit how to choose connectivity weights in Spatial Autoregressive models (SAR). I show how to decompose connectivity weights to reflect interrelationships stemming from free-riding in public good provision and complementary technologies in private good production. The resulting weights are of the form bij=f(bi, bj), i.e. they are a function of actor-specific properties and not dyad-specific. This reduction in parameters allows the researcher to estimate the underlying actor-specific weights bi, along with the connectivity parameter rho, given sufficiently large numbers of panels across which bi are constant.
I use my new method to analyze the fighting efforts of opposition groups in conflict networks in civil wars in Africa covering the years 1991 to 2011. My method allows me to explore the interconnected nature of fighting, and systematically test which strategic dynamics give rise to conflict networks. Coalitions between opposition groups that seek to redress a grievance or that aim at secession should be subject to public good problems, as the benefits from stopping the cause of the grievance or achieving independence are non-excludable and non-rivalrous. In contrast, coalitions that seek rent extraction opportunities such as localized control of raw materials reap private spoils of war and therefore face few collective action problems. For the statistical analysis I draw on the geo-referenced Event Dataset (GED) published by the Uppsala Conflict Data Program (UCDP).
