I am using multiple imputation to deal with around 49 missing observations for my outcome variable from my 324 observation panel dataset. I used Stata to perform 10 imputations for this, using
mi impute regress for this. Now my panel has 490 more observations. A conventional approach is to perform analysis on each imputed dataset and then pool the results.

What if I take the imputed datasets, and average each of the imputed values, and use it as some sort of mean imputation for the imputed data? Let's say for year 2010 in country A (which was missing), I have now imputed 10 sets of possible observations. I am tempted to just take an average value and use it in my regression.

What are the flaws associated with this approach? Is it a better approach than simple mean imputation?

  • 3
    $\begingroup$ The point of multiple imputation is to simulate variation of your data. If you are just taking average of all datasets, you would do it better by doing simple mean imputation. $\endgroup$ – Pere Feb 6 '18 at 17:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.