The PI identifies five persistent problems in dealing with categorical data analysis, and proposes new approaches to solve them. Because much of social science data is categorical (such as race, gender, party membership, employed/unemployed, and state of hypothetical latent structure models) instances of the prob- lems appear in all the social sciences. Most statistics assume continuous variables, though it is known that violating that assumption can lead to erroneous conclusions for categorical data. Theoretical constructs, such as those in Markov models, which appear widely in social science theories also often use categorical 'data,' or discrete states. Thus these problems affect everyday data analysis problems faced by most social sciences researchers. PI proposes to attack the five problems identified, and moreover, to do so in a way which minimizes assumptions about the distributions of the data (parametric assumptions). Success with any of the five problems promises a significant improvement in data analysis and theory building for the majority of practicing sociologists.
