Diagnosing influential observations in quadratic discriminant analysis

被引:7
|
作者
Fung, WK
机构
[1] Department of Statistics, University of Hong Kong, Hong Kong, Pokfulam Road
关键词
case deletion; diagnostic measures; outliers and influential observations; quadratic discriminant analysis; simulated confidence envelope;
D O I
10.2307/2532839
中图分类号
Q [生物科学];
学科分类号
07 ; 0710 ; 09 ;
摘要
We propose various diagnostic measures for the identification of observations that are influential for the estimation of probability density functions, probabilities, and log odds in quadratic discriminant analysis. The quantile-quantile (Q-Q) plots together with simulated confidence envelopes are found to be useful for detecting influential observations in a data set of two species of biting dies that are so similar that they are difficult to discriminate morphologically. We iind that the omission of even one case in the data set can have a very great effect on the estimated probabilities that the observations belong to a particular group.
引用
收藏
页码:1235 / 1241
页数:7
相关论文
共 50 条