关于数据非正态分布的处理

you'd better use component based SEM, such as smartPLS, PLS graph and other similar softwares to deal with the unnormalized distribution data. In additon, to effectively deal with formative construct issues, component based SEM is more suitable than covariance based SEM, such as AMOS.  

在AMOS中用Boootstrap方法估计。

quasimaster 发表于 2012-10-3 08:33
在amos中的估计
可以在analysis properties-->output-->将outlier and normality打勾
分析后就可以看到正 ...

请问，这个怎么理解“如p2小于0.001,前面的马氏距离又与别的观察值有差异”？我看到有些p2小于0.001的对应的Mahalanobis d-squared值是明显比表格中p2大于0.001的那些对应的Mahalanobis d-squared值要大，那是不是就可以直接删去这些样本了？谢谢！