Biostatistics Seminar Series - Jing Huang, PhD

Abstract:Biomedical research is facilitated with massive healthcare data generated from complex data generating mechanism from different sources. With larger and larger data, complex models are of increasing interest to scientists. Traditional likelihood methods for complex models sometimes fail and estimators often scale poorly due to mis-specifications of data generating mechanism. Compared to traditional likelihood methods, composite likelihood methods may be less statistically efficient, but they can be more computationally convenient and robust to model misspecifications. In this talk, I will present a serial of work to show how statistical tools, e.g., composite likelihood functions, can be used to make robust inference from large-scale healthcare data. I will explain the research gap in statistical theory of composite likelihood, and demonstrate recent works on evaluation of the goodness-of-fit for composite likelihood inference, and on the hypothesis testing with prior knowledge using the composite likelihood