Bayesian Estimation of Restricted Latent Class Models for Ordinal and Nominal Response Data

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

This research project will advance statistical models and machine learning algorithms used to describe the structure underlying human response data. Decision makers and researchers in industry and academia collect and use large-scale datasets to understand human behavior. Current methods extract profiles underlying response data where the latent structure is known. However, the latent structure is rarely known, so existing methods are appropriate in only a few narrowly defined cases. The project will develop exploratory methods for identifying common patterns and profiles when the latent structure is unknown. The research will have broad implications in healthy populations and in clinical populations, particularly in providing decision makers with fine-grained information to accelerate human development and improve quality of life. The methods to be developed will be applied to data involving individual differences in cognitive performance, psychological factors that influence college student academic behavior, and patient health outcomes. Graduate students will be involved in the conduct of the research. Publicly available software will be developed to provide researchers and decision makers with cutting-edge tools. This research project will develop and apply statistical models that will enable precise, powerful modeling and testing of hypotheses about systematic patterns in human response data. Restricted latent class models (RLCMs) will be investigated for a broad range of human response data, including ordinal ratings, rankings, and nominal choices. Methods for estimating the latent structure for general RLCMs for ordinal and nominal response data will be developed. Bayesian methods will be used to improve upon existing models by offering a flexible framework that explores the underlying structure. Project innovations will include establishing new identifiability theory and investigating ordinal specifications in the latent structure. Insights gained from this research could lead to significant developments in the underlying statistical theory for estimation of latent structure models. Advances in this theory will have implications for a broad spectrum of applications beyond education and psychology, including other social science disciplines and machine learning applications that cluster multivariate nominal data according to underlying features. Researchers will be provided with tools to relate the latent structure to other criteria, such as academic success, effective decision-making, and health outcomes. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.