I'm working on something involving multiple imputation, and am trying to figure out how I can get an overall p-value associated with a categorical variable in my model. So first I use PROC MI to generate 10 imputed datasets. Then I'm using PROC SURVEYREG to fit a model for each of the 10 imputed datasets, and then using PROC MIANALYZE to combine the results. The model I'm fitting in PROC SURVEYREG is roughly of the form Y = A + B + C + X, where Y is a continuous outcome; A, B, and C are covariates being controlled for; and X is a categorical variable with four levels (X1, X2, X3, X4). PROC MIANALYZE will give me results (parameter estimates, p-values, etc.) for three levels of X individually (using the other level as a reference), but what I want is the overall p-value for this variable (i.e., if X1 is the reference level, then the p-value for the test of X2=X3=X4=0). I think this can be accomplished using a test statement in PROC MIANALYZE, but when I try to do this, I get the warning: "The TEST statement cannot be used when the input PARMS= data set is specified without COVB= or XPXI= data set." Any suggestions of how to proceed? I've seen that a similar question has already been discussed here, but I'm still trying to figure it out.
... View more
