Bones of Contention: Territorial, Maritime, and River Issues

Florida State University, Tallahassee FL

Investigators

Abstract

Consider a researcher who must choose between two rival models, based on how well those models explain the behavior in a given set of data. Suppose one is a strategic choice model and the other is a nonstrategic model -that is, the models have different functional forms. How would the researcher discriminate between these models? In other words, how would she determine which of the models is "better?" In this project, Principal Investigators answer this question by merging two promising areas of methodological research, and then use those methods to analyze international conflict. The first line of research, by Signorino and various coauthors, has demonstrated that traditional specifications of statistical models are generally inconsistent with strategic theories of political behavior. The main message of this research has been that the functional form of one's statistical model must be consistent with the relationships implied by the behavior under analysis, or the statistical inferences will be invalid. However, aside from comparisons of likelihood values and percent correctly predicted, Signorino has not provided a more rigorous framework for comparing (nonnested) strategic models against one another, or against nonstrategic models. Clarke, on the other hand, has engaged in a second line of research, addressing comparative model testing for nonnested models. The main message of his research has been that nonnested model testing requires special techniques that are unknown to most political scientists. New and existing nonnested testing methods have been developed and adapted in prior research to deal with competing limited dependent variable models. The competing models in this literature, however, have only been nonnested in terms of their covanates. Testing strategic choice models against each other or against nonstrategic models requires discriminating between models that are nonnested in terms of their functional forms. The research conducted by Signorino and by Clarke is clearly complimentary. By developing nonnested tests for strategic choice models, Signorino and Clarke will provide the necessary tools for assessing whether strategic models outperform their nonstrategic rivals, or whether certain strategic specifications outperform other strategic specifications. By addressing strategic choice models, Signorino and Clarke will expand the set of functional forms beyond that currently available in the nonnested testing literature. As in previous work on nonnested model testing, the proposed research will make use of classical, Bayesian, and nonparametric statistical techniques. The project goals include: * Adapting absolute discrimination tests (such as the Cox test) for strategic models * Adapting and developing relative discrimination tests (such as the Vuong test, Clarke's non- parametric test, and Bayes factors) for strategic models. * Refining current Bayesian techniques for use with strategic models, and conducting Bayesian estimation of strategic models (to conduct tests based on Bayes factors). * Applying these methods to models of deterrence, alliance politics, and the effect of domestic politics on militarized interstate disputes. * Incorporating these techniques into currently available software (e.g. STRAT and WinBUGS). By combining these lines of research, future scholars will be able not only to develop statistical models that are truly consistent with their theories, but also to test those theories against each other. The attainment of these goals will allow the scientific study of international relations to move significantly beyond its current state.