摘要
通常的极大似然估计对反应y方向和设计空间x方向的突出值都很敏感,容易引起参数估计值的偏差。本文提出一种用于Logistic回归的稳健估计方法。该法的基点是弧立少数突出值,配合大部分非突出值,使截尾平均对数似然达到最大,因而最终参数估计值不受突出值影响。文章阐述了所用稳健估计的原理,并对有关问题进行了讨论。
Usual maximum likelihood estimators are sensitive to outliers either in response y or in design space x, which cause the biased parameter estimates. A technique for resistant fit of logistic regression model was suggested and it is based on the maximizing the lower side trimmed mean of log-likelihood. The suggested method isolates the outliers from the majority of the observations, therefore the final estimates are not affected by the outliers.The principles were stated and the related topics were discussed.
出处
《中国卫生统计》
CSCD
北大核心
1990年第2期29-33,共5页
Chinese Journal of Health Statistics
关键词
LOGISTIC回归
稳健估计
突出值
Logistic regression Outliers Robust estimate Maximum likelihood
