I am using the Windows version of Amelia II for imputations on a seven-category nominal variable (multiparty vote choice). The observed distribution is typical of vote shares in multiparty elections (some parties are small, others big) but the distribution of imputations looks suspiciously normal (when seen in the density plot obtained from the diagnostics window). To check, I recoded the variable's categories, but the imputations for the recoded variable still look almost as Gaussian-normal as before recoding, which I think cannot be. Is there a possibility that Amelia doesn't recognize the nominal nature of the variable although I set the appropriate option in step 2?