Improving Prediction and Causal Inference with Graphical Methods and Models

Critical to reliable prediction and causal inference is understanding structural relationships in the social and political systems under study. Graphical models are naturally suited for con-ceptualizing and representing relationships. This paper introduces and synthesizes a large and disparate literature on different types of graphical models, with particular attention on recent developments in theories of causal graphs and models of random graphs for relational data, and discusses the adaptation, application, and extension of these graphical methods and models in political data analysis in general, and in the modeling of structural properties of international relations data in particular. Graphical models can improve prediction and causal inference in these data by facilitating systematic examination and modeling of characteristics of the interna-tional network, by guiding the identification of the causal structure in the system, and by adding flexibility in functional form approximation. Initial results from analyzing the 1947-1989 MID data provide strong evidence that properties of the system as a whole and that of individu-al states/dyads embedded in the system hold important explanatory and predictive power, and clearly reveal the interdependence among dyads sharing a common member. 1