I'm trying to perform a two-way-ANOVA out of a dataset containing missing values. My principal approach is
Proc GLM data=...;
class group block;
model response=block group block*group;
How can I get the corresponding ANOVA-table after running PROC MI? All the examples I found so far seem to simply address the imputation process itself rather than the original problem. I'd like to compare my F- and p-values from the data with missing values to the results of the multiple imputation step. I'd be very glad if you could help me.
You can run GLM with missing values. However, your GLM model is probably incorrect. If you have one observation of each level of group in each block, then block*group is the residual error term. Thus, your model is overparameterized. Take out the block*group term. If you have multiple independent replicates of each level of group in each block, then you can separate the interaction and the residual. But most people consider block, and especially block*group as random-effect terms. You have them as fixed-effect terms. You will need to read more about use of GLM for random effects before addressing the missing-value issue. If you do decide to explore block or the interaction as random effects, you should definitely consider use of PROC MIXED, especially with missing data, instead of use of GLM.