Model Selection Methods for Categorical Regression

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Improving model selection methods for analyzing epidemiological and clinical datasets using categorical regression would be invaluable to the medical and health care research communities. Categorical regression (logistic, multinomial logistic) models are used extensively to identify patterns of alcohol-related symptoms, define criteria of psychiatric disorders, and assess policies regulating alcohol. However, many such models are developed with inadequate support to fully explore the complex relationships between predictor and outcome variables. Commercially available statistical software provides automated procedures (e.g. stepwise, best subsets regression) that largely ignore model misspecification, experimentwise error rate, and multicollinearity. Furthermore, these procedures are severely limited by the number of models that can be considered and thus have no theoretical basis for global convergence. Martingale Research will develop a new model selection method that supports automated model-building using categorical regression. This Phase I study will demonstrate, using datasets representative of NIAAA databases, that the proposed statistical approach will: 1) support correct model selection decisions in the presence of model misspecification, 2) converge to an optimal model under fairly general conditions, and 3) mitigate experimentwise error rate. These results will demonstrate the essential technical feasibility required for further Phase II investigations and provide the foundation for developing commercially available software. PROPOSED COMMERCIAL APPLICATION: Martingale Research Corporation intends to incorporate the proposed model selection method into an advanced categorical regression software package that addresses model misspecification, experimental-wise error, and global convergence. This will provide Phase III commercial software that supports improved solutions to a wide range of problems encountered in alcohol-related epidemiological research, as well as other aspects of database interpretation and management.