This paper addresses model design and data-related issues pertaining to the use of risk assessment models in support of conflict early warning systems. The paper first examines the data needed for risk assessment models. It argues that there are actually two types of data that we must collect and use: data to determine the theories that best explain the outbreak or escalation of violent conflict and data corresponding to variables that have been demonstrated to have predictive power and thus are appropriate for use in a risk assessment model. With respect to intrastate conflicts there remains a dearth of data of the first type that is only slowly being corrected. Collecting data of the second type will continue to be an inefficient process until better theories have been formulated and tested with the first type of data. The paper then examines a number of models germane to risk assessment and conflict early warning. It finds that except for one class of models, so-called correlational models, there has been only limited work done even though the alternatives appear to have promise.
