@article{oai:chuo-u.repo.nii.ac.jp:00001217,
 author = {藤越, 康祝 and 青木, 誠 and 櫻井, 哲朗 and 杉山, 髙一},
 journal = {中央大学理工学研究所論文集},
 month = {Mar},
 note = {application/pdf, This paper examines tests for independence of p variables which were proposed under normality, focussing on a high-dimensional case. The tests considered are based on（1）likelihood ratio test with large sample approximation,（2）the sum of squarred correlations with large sample approximation, （3）the sum of squarred correlations with high-dimensional approximation, and（4）the sum of squarred covariances with high-dimensional approximation. First, by numerical experiments we point some tendency on whether the actual rejection probabilities of the tests are near to the nominal rejection probabilities. In particular, the actual rejection probabilities of the test（2）are almost near to the nominal rejection probabilites for large sample situations as well as high-dimensional situations. Next we examine whether such property of the test（2）holds for discrete data. It is shown that the test（2）is fairly robust, though the result depends on the type of discrete distributions. Finally we extend the test（2）to one for independence of the variables after the effects of the other variables are removed., 【査読有】},
 pages = {1--10},
 title = {高次元における独立性の検定と頑健性},
 volume = {13},
 year = {2008},
 yomi = {フジコシ, ヤスノリ and アオキ, マコト and サクライ, テツロウ and スギヤマ, タカカズ}
}