Improving scientific cumulation in international relations: An application of prediction analysis to theories of war initiation

This dissertation presents a statistical technique, Prediction Analysis, as a possible solution to the lack of scientific cumulation in the field of international relations. An application to the area of war initiation is presented and the level of success of the technique assessed. Scientific cumulation requires, at a minimum, that hypotheses be tested and that the results of those tests allow an investigator to conditionally accept or reject the hypothesis. Through various examples in the literature, it is shown that the same hypotheses, in slightly modified form, have existed for several decades. In large part, this is due to the practice of testing hypotheses over varied times and regions, using a variety of non-comparable measures. This is exacerbated by the tendency to artificially raise the level of measurement to the interval level, so that more powerful mathematical techniques, such as regression analysis, can be applied. This is often done without regard to the reasonableness of the operationalization. A general logic for predictions and criteria for prediction success are presented, followed by an analysis of existing measures; while they fail, Prediction Analysis does not. Hypotheses were selected from three areas of research: polarity, hegemony and power transition, and the expected utility framework. These groupings were selected because they reflect increasing information. The first predicted war initiations, by any nation-state against any nation-state; the second, wars by any major nation-state against any nation-state; the third, wars by specific nation-state against specific nation-states. Each has been studied using a variety of statistical techniques; Prediction Analysis handles them all, within a single framework. The vast majority of studies in the literature are shown to focus on wars that actually occurred, ignoring the far large number of peaceful periods in history. A new dataset, based on war opportunities for each nation-state in each year with respect to the other members of the European system, for the period 1816 to 1974, was created. The results, for both bivariate and multivariate analyses, demonstrate both the weakness of existing theory and the very broad application of Prediction Analysis