报告题目：Statistical methods for meta-analysis with very few studies

报 告 人：童铁军 教授  博士生导师

报告时间：2020年12月15日（周二）下午15:00-18:00

报告平台：腾讯会议， ID: 989 662 816

摘要：Meta-analysis is a statistical technique to synthesize the research findings from multiple studies for decision making. It has been increasingly popular in the past several decades, mainly due to its wide application in evidence-based practice. In meta-analysis, the two commonly used models include the common-effect model (CEM) and the random-effects model (REM). Recently, it is recognized that the fixed-effects model (FEM) is also essential for meta-analysis, especially when the number of studies is very small. FEM not only avoids the unrealistic assumption on a common effect in CEM when the heterogeneity exists, but also avoids the low estimation accuracy of the between-study variance in REM when there are only few studies. With this new model, the existing methods are no longer sufficient for conducting meta-analysis. In view of the demand, we have recently proposed a few new methods for estimation and model selection related to FEM to further advance the literature. As an example, to choose a proper model between CEM and REM, the Q statistic and the I2 statistic are routinely used as the criteria for model selection. Yet to the best of our knowledge, there is no existing method for model selection between FEM and REM in the literature. To solve this problem, we applied the Akaike information criterion (AIC) for model selection between FEM and REM, and we expect that our new criterion will have potential to be widely applied in meta-analysis and evidence-based practice.

报告人简介：童铁军，香港浸会大学数学系教授，博士生导师，副系主任，国际统计协会当选会员。主要科研方向包括非参数回归模型、高维数据分析、Meta分析和循证医学。2005年于美国加州大学圣巴巴拉分校获得统计学博士学位，2005-2007年在美国耶鲁大学从事生物统计博士后研究，2007-2010年在美国科罗拉多大学博尔德分校担任助理教授，2010年至今任职于香港浸会大学。已在国际知名的学术期刊Biometrika、JASA、JMLR、Statistical Science等发表学术论文70余篇，包括两篇ESI热点文章和三篇ESI高被引文章，单篇论文最高引用1800余次。

欢迎广大师生参加！