报告人：Song Xinyuan (宋心远)
邀请人：潘灯
报告人简介：宋心远，香港中文大学统计系教授，系主任。宋心远教授的研究方向是潜变量模型，贝叶斯方法，统计计算和生存分析等。同时还担任多个国际期刊包括《Psychometrika》，《Biometrics》，《Computational Statistics & Data Analysis》和《Structural Equation Modeling: A Multidisciplinary Journal》的副主编或编委。已在国际期刊发表超过100篇论文，近期论文主要发表于《Journal of the American Statistical Association》，《Biometrika》，《Biometrics》，《Bioinformatics》，《Psychometrika》，《Quantitative Finance》等期刊。
（一）
报告题目：Parametric and Nonparametric Causal Mediation Analysis
报告摘要：This study develops a joint modeling approach incorporating latent traits into causal mediation analysis with multiple mediators and a survival outcome. We first consider a parametric model, which consists of a set of linear models for investigating the relationships among various parallel or causally ordered mediators and the exposure, and a proportional hazards model for deriving the path-specific causal effects on the scale of hazard ratio under the counterfactual framework. A Bayesian approach with Markov chain Monte Carlo algorithm is developed to estimate the causal effects efficiently. Then, we extend the parametric model to a tree-based nonparametric model to address the heterogeneous mediating mechanism for time-to-event data. The proposed tree-based model has two components: a set of Bayesian additive regression trees for characterizing the marginal and joint distributions of the mediators and a nonparametric accelerated failure time model for formulating the expected survival time given the treatment, mediators, and confounders. Finally, we consider a conditional generative adversarial network-based approach to conduct mediation analysis from a deep learning perspective. The proposed model is applied to a study on the Alzheimer’s Disease Neuroimaging Initiative dataset to investigate the causal effects of the APOE-ε4 allele on the disease progression, either directly or indirectly.
报告时间：2021年12月10日（星期五）14:30-16:30
报告地点：腾讯会议：245 418 352
（二）
报告题目：Two-part Models for Cross-sectional and Longitudinal Semicontinuous Data
报告摘要：The semicontinuous variable is characterized by a mixture of zero values and continuously distributed positive values. A two-part model manages this semicontinuous variable by splitting it into two random variables; one is a binary indicator to determine whether the response is zero, another is a continuous variable to determine the actual level of the positive response. This study introduces a two-part model with latent variables to analyze cross-sectional semicontinuous data with latent variables and a two-part hidden Markov model to analyze longitudinal semicontinuous data with nonignorable missing covariates. Bayesian approaches coupled with spike-and-slab and conditional Laplace lasso prior are developed for simultaneous variable selection and parameter estimation. The proposed methodologies are applied to the Chinese General Social Survey and Alzheimer's Disease Neuroimaging Initiative datasets. New insights into the non-cognitive ability and the pathology of Alzheimer's disease are obtained.
报告时间：2021年12月15日（星期三）14:30-16:30
报告地点：腾讯会议：645 240 567
