Multiple imputation (in case of a complete dataset)

[reynae...@gmail.com]

Normally Multiple imputation is used by missing data (of course). 

But what would happen iif you do multiple imputation when you have a complete dataset?

Would you get x times the same dataset? Or x times a different dataset with the same mean as in the original dataset?

And what happens with the precision?

thanks in advance

[Jonathan Bartlett]

If there are no missing data, there is nothing to impute. I suppose you could construct x "imputed" datasets, which are identical to the original, complete data. You then run your analysis on each dataset, which will give you an identical set of results on each, since the datasets are identical. If you then apply Rubin's rules, the overall point estimate, as the average of the x estimates, will be equal to the estimate obtained on the original data. The between imputation variance is then zero, so Rubin's variance estimator just equals the variance estimate obtained from the analysis of the original data - so, correctly, you haven't gained or lost any precision.

So in summary, you will just get back the same answer as analysing the original data, and there isn't any point in doing this!

[Hanne Reynaert]

Thank you! 

Op 2 feb. 2017 3:48 p.m. schreef "'Jonathan Bartlett' via Missing Data" <missin...@googlegroups.com>:

--
You received this message because you are subscribed to the Google Groups "Missing Data" group.
To unsubscribe from this group and stop receiving emails from it, send an email to missing-data+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.