讲座题目：Mixture regression for longitudinal data based on joint mean-covariance model

主 讲 人：Jianxin Pan, Beijing Normal University (Zhuhai) and BNU-HKBU United International College, China

讲座时间：2022年4月24日（周日）上午9:00-10:00

地点：腾讯会议  ID 170-427-119

https://meeting.tencent.com/dm/Kg2fHJE17kDt

主讲人简介：

Professor Jianxin Pan holds a joint Chair Professorship of Beijing Normal University and BNU-HKBU United International College, both in Zhuhai. He was Professor of Statistics in the University of Manchester, UK, between 2002 and 2021.

Professor Jianxin Pan’s research interests include statistical modeling, statistical learning, and data science, with application to medicine, public health, finance, and industry. He has published over 130 research articles in journals of statistical science and multidisciplinary research fields, and 3 research monographs with Springer and Science Press. He was awarded funding from various research councils of the UK and Europe. He is Fellow of the Royal Statistical Society, Elected Member of the International Statistical Institute, and Turing Fellow of the Alan Turing Institute for Data Science and Artificial Intelligence in London. He was the Chair of the Royal Statistical Society Manchester Group and has been serving as Associate Editor for statistical journals, including Biometrics (2008-2018), Biostatistics and Epidemiology (2013-), Biometrical Journal (2016-), Journal of Multivariate Analysis (2019-), and Electronic Journal of Statistics (2022-). 

讲座摘要：

When modeling longitudinal data, it is common that the studied population is comprised of several groups of individuals while individuals within the same group share the similar kind of mean progression trajectories. Finite mixture models are often used to address this kind of unobserved heterogeneity in terms of mean. Existing methods, such as parametric and semiparametric mixture regression, usually model the mean in each subpopulation with assumption that observations sharing a common trajectory are independent or correlated but with a pre-specified correlation structure. Little research considers modeling of covariance structures while accounting for heterogeneity. In this talk, a joint model which models the mean and covariance structures is proposed within the framework of finite normal mixture regression, demonstrating how important the within-subject correlation is when clustering longitudinal data. Model parameters are estimated with an iteratively re-weighted least squares EM algorithm. The resulting parameter estimators are shown to be consistent and asymptotically normal. The model is able to identify various mean trajectories and covariance structures in all clusters. Simulations show that the proposed method works well and gives accurate clustering results through introducing covariance modeling. Real data analysis is made to illustrate the usefulness of the proposed method, particularly, giving a good performance in clustering COVID-19 deaths for European countries in terms of progression mean-covariance trajectory.