I need to impute a nominal categorical variable using both continuous and categorical covariates. Proc MI documentation suggests DISCRIM option for nominal variables but this procedure uses only continuous variables. Is there any other option to include both continuous and nominal covariates in the estimation ?
You could consider using a nominal logistic model. You can model a nominal (unordered) response in PROC LOGISTIC using the LINK=GLOGIT option, and you can use a mixture of categorical (CLASS) and continuous predictors. See the example titled "Nominal Response Data: Generalized Logits Model" in the LOGISTIC documentation:
Over on SAS-L, there was a post just this morning regarding imputation of categorical variables. The author of the post suggested use of a product called IVEWARE for imputing categorical data. The author of that post also indicated that IVEWARE is free and can be called from SAS. Sounds like it would be worth checking into.