Proc MI FCS regression dummy coding

[ccoman]

Hello everyone,

Currently I'm trying to employ multiple imputation using PROC MI for missing categorical and continuous data. Do I need to dummy code my categorical predictors before using them in FCS regression? If, for example, I want to impute continuous variable 'weight' using categorical variable 'income' (low, medium high). When undertaking regular lineair regression it's not possible to use categorical predictors, but PROC MI does have a class statement, so that's why I am confused.

proc mi data=lib.test nimpute=5 seed=54321 out=mi
class income;
var bloodpressure income age weight height;
fcs discrim (income = bloodpressure income age weight height  / classeffects=include) nbiter =20 ;
fcs reg (weight = bloodpressure income age height) nbiter =20 ;
run;

[SAS_Rob · Accepted solution]

No, it is not necessary to use dummy variables.  The CLASS statement will do that automatically (using effects coding) for the FCS