Testing the assumption of multivariate normality.
Testing the assumption of multivariate normality.
复制标题
检验多元正态性假设。
DOI：
--
复制
发表时间：
2004
期刊：
影响因子：
--
通讯作者：
G. Bogat
中科院分区：
文献类型：
--
作者：
A. Eye;G. Bogat
讨论了多元数据偏离多正态性的程度的评估方法。这些方法中最著名的是Mardia的多变量偏度和峰度测试，它允许人们测试与多正态性假设兼容的零假设。然而，如果这些零假设被拒绝，研究人员就不知道特定部门是否存在过多的违规行为。提出了一种扇区测试和综合测试。扇区测试允许识别包含与多正态性假设所预期的不同数量的情况的子空间。综合检验允许人们检验总体上多正态分布的假设是否成立。Mardia的测试和新的部门和综合测试适用于一个项目的数据对青少年的养育能力。
Methods of assessing the degree to which multivariate data deviate from multinormality are discussed. The best known of these methods, Mardia’s tests of multivariate skewness and kurtosis, allow one to test null hypotheses that are compatible with the assumption of multinormality. However, if these null hypotheses are rejected, researchers do not know whether particular sectors carry inordinate amounts of the violations. A sector test and an omnibus test are proposed. The sector test allows one to identify subspaces that contain different numbers of cases than expected on the assumption of multinormality. The omnibus test allows one to test whether, overall, the hypothesis of a multinormal distribution is tenable. Mardia’s tests and the new sector and omnibus tests are applied to data from a project on parenting abilities of adolescents.