This research explores the application of artificial neural network models to empirical research in political science. There are two main objectives. The first is to conduct comparative studies of neural network models and a variety of novel and standard models used in political science to assess the utility of neural network models to the field. This will be done through both computer simulation and application to substantively important political data, including the unique and comprehensive ROAD (Record of American Democracy) database at the Harvard-MIT Data Center. The second objective is to adapt neural network models to apply to the special problems in political science. In particular, the investigator plans to develop and implement methods of interpretation and inference from neural network modeling, results to facilitate the use of these models. This will be of importance to nonlinear models in general, where interpretation of estimation results and significance tests are more complicated than in the linear models. Other related methodological issues of general importance, such as data fitting vs. generalizations, noise level and model performance, will also be examined. This work will contribute significantly to the field of political methodology in the increasingly important area of nonlinear, flexible modeling of political data. Political and social relationships are generally characterized by nonlinearity and complexity, and their exact functional forms are rarely known or implied by substantive/formal theories. The standard models in political science research such as the linear regression, logit, and probit models assume known functional forms for the data generating process, and are usually linear with or without a nonlinear link function. Artificial neural network models are able to approximate unknown functional relationships and allow for arbitrary patterns of nonlinearity. The planned work will therefore significantly contribute to the improvement of the quality of empirical research in political science. This POWRE Visiting Professorship award will allow the investigator to expand her own methodological and teaching skills as well as afford her the opportunity to develop numerous substantive applications for her methodological innovations. This will give her an important research and educational experience that will contribute to her visibility in her field and her subsequent academic advancement.
