I have a cross-over study with 2 periods where a few subjects are missing the endpoint in the second period. PROC MIXED was used for the analysis and I am trying to understand exactly how PROC MIXED handles the missing values. One suggestion has been that the mixed model imputes the missing values somehow. I was under the impression that mixed used all available data (as compared to PROC GLM), but does it impute the missing values?
MIXED does not impute anything for the missing data. It uses the data for the periods were there is data and ignores the periods for which there is no data. In that sense it is like the repeated measures ANOVA approach (but without the restrictive assumptions).
If you want to impute the missing endpoints and meet the assumptions, you could use the MI and MIANALYSE procedures with MIXED.