proc mi

[SMartin_82]

Hello,

I used the proc mi procedure for imputing missing data.

First I looked after the pattern, then I generated an output with 10 multiple imputations in the imputation phase, proc glm for the analysis phase and proc mianalyze for the poolig phase.

So, now I have a dataset with this 10 multiple imputations and an origin dataset with my missing data. But what happend next? How can I replace my missings in the original data? Or have I been working with the imputed dataset further?

Thanks for help...

BR Silke

[Rick_SAS · Accepted solution]

The multiple imputation method (which is implemented in PROC MI) is not designed to come up with a single value that you can use to replace the missing values in your data. "Replacing" is known as "single imputation" and leads to biased estimates for parameters such as regression coefficients. For more background and references, see the "Overview of the MI Procedure."

So what happens next?  Well, you can continue working with the 10-fold imputed data set, which will give you good estimates of uncertainty. That would mean repeating Stage 2 (analysis) and Stage 3 (PROC MIANALYZE) for subsequent analyses.  However, you might want to conduct an analysis that is not supported by PROC MIANALYZE. In that case, some analysts return to the original data and let the procedure drop the incomplete cases.