I am relatively new to multiple imputation (and statistical analysis in general), so I apologize if my question seems naïve to more experienced users.

I am dealing with a somewhat large dataset (about 40 relevant variables and about 8000 observations) based on survey responses. Most variables in the dataset suffer from missing values, so I used Amelia II to impute the data. Now I have five imputed datasets (Stata 14 format) with no missing values.

My question is: How do I perform statistical analyses across these five datasets? My dependent variables are ordinal, so I will probably do ordered probit. However, I am not quite sure how to do this with multiple datasets. Do I have to somehow combine the five imputed datasets? Any help (especially the relevant syntax/commands) would be much appreciated.

  • $\begingroup$ Please see advice on software-specific questions in the Help Center. The statistical questions here are, in my view, on-topic but requests for Stata syntax are not. There is an entire manual volume on multiple imputation in Stata at stata.com/manuals14/mi.pdf $\endgroup$
    – Nick Cox
    Commented Nov 7, 2015 at 18:00

1 Answer 1


Basically, what you need to do is called Rubin's Rules. You run your ordered probit model on each of the datasets individually, treating them as if they were fully observed real data. Now, you have five estimates, and you pool them down in to one. Here is a good graphic to explain

Also, I think perhaps you would be better off reading up on MI before you go any further. 5 multiply imputed datasets is kind of the old way of thinking. Now, many researchers suggest that 5 is the minimum number of imputations. MI is a powerful tool, but we should make sure that we are using it correctly before we use it.

There are so many tiny details we need to pay attention to in order to do MI. I learned about MI by reading this book, and I thought it was very good and thorough.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.