06-24-2015 04:44 PM
I have a dataset with 116595 observations; there are missing values for many of the variables. I have used the PROC MI to do multiple imputation with nimpute=10. So, now I have a dataset with 1165950 observations, which is 10 times the size of my original dataset. It mentions the number of the imputations and the corresponding imputed values.
Now, let's say I want to use this imputed dataset in a different software (say Mplus). My question is: is there any way to get an imputed dataset from SAS, that is of the same size as my original dataset, that is 116595 observations? I want SAS to do the imputations, but ultimately output a dataset (say, B) which will have the same number of observations as my original dataset (say, A); but with imputed values for the variables that were missing. Then that new dataset B will be the one which I will use in any other software.
Is the above feasible? Or, is some other method, such as, maximum-likelihood estimates better for this purpose? Any suggestion is appreciated.
06-24-2015 05:04 PM
One way would be to use the _imputation_ variable to select one set of imputations. The question becomes then choosing which of those to use. Summary statistics such as proc means using _imputation_ as a by variable perhaps.
Basically use what ever method you prefer for exporting your data and use where _imputation_ =xx where xx is your choice of 1 through 10. If used as a data set option, such as for proc export then:
proc export data=yourdatasetname (where=(_imputation_=1))
<rest of code>
selects the first set of imputed values.