I have a post using Cholesky decomposition algorithm

https://bbs.pinggu.org/thread-1129334-1-1.html

The algorithm can be write into a function call with FCMP and directly use in a data step.

The tough part for this problem is to find a valid covariance/correlation matrix.

A valid covariance/correlation matrix should be positive semi-definite.

For example 3x3 correlation (covariance ~ correlation scale by variance components), it has 3 rhos, but 3 rhos need to satisfy certain conditions to be positive semi-definite. Higher dimensions will have more conditions. In other words rhos cannot arbitrary defined.

If you have a real problem, at lease you can define a sample covariance/correlation matrix and start from there.

等等熟悉IML 的高手。

你需要知道v1~v7之间相互的协方差，以及v1~v7各自的方差，这也是协方差阵包含的信息~
我在这篇帖子里，用的是randnormal函数生成多元正态，
对比cholesky分解协方差阵，从多元正态的定义生成多元正态，
https://bbs.pinggu.org/thread-2157211-1-1.html

可能比较草率，我用一篇文献中的协方差矩阵和设置进行模拟，然后解释结果。

《Multiple Imputation Approaches for the Analysis of Dichotomized Responses in Longitudinal Studies with Missing Data》

感谢回复，请问你这则帖子中的协方差矩阵是怎么算得的？需要考虑哪些因素呢？

我只是随机生成一个协方差阵来生成多元正态分布；
只需要保持协方差阵正定即可。
至于怎么算协方差阵，应该是和实际问题有关，你希望v1-v7之间有怎样的
相关关系，就应该设计什么样的协方差阵