When you use the REPEATED statement in PROC GLM to analyze repeated measures data, the data should be arranged in multivariate format. That is, the data should contain one observation for each subject and the repeated measurements at different time points for a subject are recorded in different variables. If there are missing values for one or more time points for a subject, PROC GLM drops the entire observation for that subject.

To include the nonmissing measurements for such subjects in the analysis, you need to analyze the data using univariate format. For data in univariate format, one observation corresponds to the measurement at one time point for one subject. In this format, subjects have multiple observations and a variable indicating the time (or occasion or similar) is included in the data set to indicate the time (or occasion) of the measurements. The univariate analysis can be performed in PROC GLM or PROC MIXED. The PROC GLM approach is described in "Repeated Measures Analysis of Variance" in the Details section of the GLM documentation. The PROC MIXED approach is illustrated in the "Repeated Measures Analysis" in the MIXED documentation. However, PROC MIXED is the recommended approach because there are several advantages to using PROC MIXED:

• PROC MIXED has over 30 covariance structures to model the correlation in your repeated measures data.
• PROC MIXED uses the error structure you specify in computing the standard errors for parameters, LSMEANS, and ESTIMATE statements. PROC GLM only uses the MSE in these standard error calculations.
• PROC MIXED allows you to fit random coefficients model, which is helpful for certain repeated measures data, or when a subject-specific model is desired.