报告人：北京工业大学理学部谢田法教授
时间：2023年7月21日(周五)上午10:00-11:00
地点：本部维格堂319
摘要: The relative risk and odds ratio are widely used in many fields, including biomedical research, to compare two treatments. Extensive research has been done to infer the two parameters through approximate or exact confidence intervals. However, these intervals may be liberal or conservative. A natural question is whether the intervals can be further improved in maintaining the correct confidence coefficient of an approximate interval or shortening an exact but conservative interval. In this paper, when two independent binomials are observed we offer an effort to improve any of the existing intervals by applying the h-function method. In particular, if the given interval is approximate, then the improved interval is exact; if the given interval is exact, then the improved interval is a subset of the given interval. This method is also applied multiple times to the improved intervals until the final resultant interval cannot be shortened any further. To demonstrate the effectiveness of the method, we use two real datasets to illustrate in detail how several good intervals in practice are improved. Two exact intervals are then recommended for estimating each of the two parameters in different scenarios. If possible, we will also introduce the optimal confidence intervals for other two parameters when double sampling subjected to false-positive misclassification and so on.
作者简介：谢田法，博士、教授，北京工业大学理学部副主任。主要研究方向为稳健统计、复杂数据建模与模型检验、统计方法应用及标准化。已在国内外核心期刊发表论文30余篇。参与制修订国家标准多项（其中已发布22项）、国际标准2项（其中作为项目负责人发布国际标准1项）。2014年获“中国标准创新贡献奖”二等奖。主持了国家自然科学青年基金、国家社科基金、北京市自然科学基金、国家地质实验测试中心等单位委托的横向课题等多个项目。现为全国统计方法应用标准化技术委员会副秘书长。2014年1月成为ISO/TC69下多个分技术委员会的成员。2015年10月起成为北京应用统计学会常务理事。
邀请人：唐煜