That doesn't work for GLM although it does for reg. Paul
Paul R. Swank, Ph.D., Professor Health Promotions and Behavioral Sciences School of Public Health University of Texas Health Science Center Houston ________________________________ From: Impute -- Imputations in Data Analysis [[email protected]] On Behalf Of Trivellore Raghunathan [[email protected]] Sent: Thursday, October 25, 2012 1:07 PM To: [email protected] Subject: Re: Proc MIANALYZE I think it is better to use COVB or EST type data set as input into PROC MIANALYZE. Have you tried that? Raghu On Thu, Oct 25, 2012 at 1:22 PM, Swank, Paul R <[email protected]<mailto:[email protected]>> wrote: I recently ran an imputation analysis in SAS using Procs MI and MIANALYZE. The imputations worked fine. The problem is a general linear model with multiple predictors and interactions. The problem is when I read the documentation for using MIANALYZE with GLM, it said to output the parameters and the inverse of X’X. I did this but MIANALYZE kept saying that I couldn’t use inverse(X’X) with categorical variables, although I did not include a class statement in my GLM. When I looked at the output files created by ODS, the parameter estimates had all the variable names listed without problem but the inverse X’X file had renamed all the interaction terms as dummy001, dummy002, etc. I don’t know why that is or why it wouldn’t work. I do know that if I dropped the inverse X’X statement from MIANALYZE, it worked. Does anyone know what is going on here. Paul Dr. Paul R. Swank, Professor Health Promotion and Behavioral Sciences School of Public Health University of Texas Health Science Center Houston -- Trivellore Raghunathan (Raghu) Chair and Professor of Biostatistics School of Public Health Room M4208 1415 Washington Heights University of Michigan Ann Arbor, MI 48109 Phone: (734) 615-9832 Fax: (734) 615-7068 "A good life is filled with selfless actions full of compassion knowing well that we are all one"
