How to combine analysis results from multiple imputed categorical variables

Abeid — Wed, 12 Nov 2014 14:03:39 GMT

Hi all

I am doing multiple imputation (specifically PROC MI with fcs statement). My incomplete data is a replication of a data set (say, I have incomplete datasets = 1000).

The imputations run fine but when I try to pool for inference I get the following error

ERROR: Within-imputation Estimate missing for variable intercept1 in _Imputation_= 1 in the input

PARMS= data set.

The missingness is only on one variable, the dependent variable which is categorical (y = 1, 2, 3, 4).

So I have a linear predictor: eta = a_j + beta1*x1 + beta2*x2 + beta3*x3 + beta4*x4 + b , b ~ N(0, d) , j = 1, 2, 3, 4.

I use PROC GLIMMIX to fit a mixed effect model, which runs fine with no errors. The problem comes when I invoke MIANALYZE. I get the error there up appended for the SAS code below.

PROC MIANALYZE PARMS =glimparms;

MODELEFFECTS intercept1 intercept2 intercept3 x1 x2 x3 x4;

RUN;

If i don't include intercepts, I get the beta estimates for the x's. i.e., for

PROC MIANALYZE PARMS =glimparms;

MODELEFFECTS x1 x2 x3 x4;

RUN;

Please someone help.

Much regards

Re: How to combine analysis results from multiple imputed categorical variables

SteveDenham — Thu, 13 Nov 2014 12:34:49 GMT

MIANALYZE and GLIMMIX have notorious compatibility problems that still have not been worked out, at least to my satisfaction. I know there have been some posts in this forum on approaches, so searching here might turn up some keys. If you do find something to make this jump, please post it back here.

Steve Denham

topic Re: How to combine analysis results from multiple imputed categorical variables in Statistical Procedures

How to combine analysis results from multiple imputed categorical variables

Re: How to combine analysis results from multiple imputed categorical variables