感谢神：新年伊始便让我有论坛币来象大家请教！
It is well-known that the inter-subject cv of a crossover design can be derived from the variance estimate of the SUBJECT effect. For example,

The estimated intersubject-cv = sqrt(exp(0.009933)-1), where 0.009933 is obtained from the ODS output table 'Covariance Parameter Estimate'.
Note that the estimate is consistent regardless of the METHOD= option in the PROC MIXED statement for a balanced dataset.
However, different METHOD= options lead to different answers when the dataset is NOT balanced. For example, if you delete the last observation from the sample dataset, then methods TYPE1/TYPE3/ML/REML/MIVQUE will have different variance estimates.
Question: Is there a preferred METHOD= option in PROC MIXED for intersubject CV estimation? Why?