I am currently using multinomial logistic regression to model disease outcome based on genetic haplotype. I've included 4 haplotypes in my model along with covariates as dependent variables. I've coded the values at each haplotype to correspond with the number of copies of that haplotype each individual is predicted to have and used the most common haplotype as the reference. In addition I've created a dummy value if the person does not have any copies of the haplotype. Unfortunately the dummy variable is affecting my model. Does anyone know of a way to set this value to zero (df=0) in the logistic model?
This sounds more like a coding problem than a missing value problem. If the Dummy variable is coded 0/1, then it is never missing. If PROC LOGISTIC is showing the dummy as having df=0, then matrix is singular and either the variable is conceptually redundant with your other variables or operationally redundant because you don't have enough observations.