学术报告(陈勇 11.20)
Statistical inference under non-regular conditions: Bartlett identity, boundary and identifiability issues
摘要:
In this talk, I will talk about a general asymptotic theory for statistical inference for situations with non-regular problems (Bartlett identity, boundary and non-identifiability problems) that arise in applications including variance component models, multivariate survival models, mixture models, and some partially specified models. When these non-regular problems occur, test statistics that overcome misspecification of Type I errors and substantial loss of statistical power are needed. I will cover a few ideas in tackling these problems in parametric and semiparametric inference. I will show that these considerations are critical in model robustness, statistical power, and validity. These ideas have important implications in biomedical studies including semiparametric tests of homogeneity in mixture models that are relevant to analysis of DNA methylation data, pseudo-likelihood methods to various studies in genetic quantitative trait locus analyses, longitudinal data with irregular observation times, pharmacovigilance studies, and genetic studies under population structure.
Bio: Yong Chen is tenured Professor of Biostatistics and the Founding Director of the Center for Health Analytics and Synthesis of Evidence (CHASE) at the University of Pennsylvania. He is an elected fellow of American Statistical Association, International Statistical Institute, Society for Research Synthesis Methodology, American College of Medical Informatics, and American Medical Informatics Association. He founded the Penn Computing, Inference and Learning (PennCIL) lab at the University of Pennsylvania, focusing on clinical evidence generation and evidence synthesis using clinical and real-world data. Dr. Chen’s group has pioneered innovative approaches to the sharing of aggregated data to advance multi-center clinical research and he has extensive experience with conductive research on clinical evidence generation using large scale observational data (including EHR and claims data).