Work with an individual imputation as a regular data set

[Tina Wielen]

Hi everyone,
I have a household data set with missing values. I run multiple imputation (MI) on the data set. I have 20 imputed data sets.

I need to merge the MI household data with different data sets for further analysis. This means I need to work with only one individual imputation as a regular data set.
I know not to reduce my 20 MI data sets into a mean data set, as it removes the benefit from using MI.

The tool for selecting a single imputation and turning it into a regular data set is mi extract, and the syntax is very simple: mi extract n
Is there a way to determine which imputation preformed best to determine which data set is best to use for further analysis?

Many thanks in advance!

[daniel klein]

Using only one dataset has the same effect as using a summarized version (a "mean" dataset): it renders the multiple imputation useless. There is no such thing as the one "best" imputed dataset. From this perspective it does not matter which one to extract.

Why not merge the original data and then run multiple imputation on the final dataset that should be used for analysis? It would have the additional advantage that relationships between the imputed variables and those in in the merged dataset would correctly be acknowledged.

Best
Daniel

[Tina Wielen]

Thanks Daniel.

[Jacob Thomas]

I have the same problem and I could understand what Daniel is saying about rendering MI useless by averaging, but I just want to have a data set without missing values. However I do not understand your suggestion, Daniel. What do you mean by "merging the original data and then run multiple imputation on the final dataset"? Are there specific commands for that?
Does this mean we have to embed every operation (e.g sem, anova, and logistic regression) within the mi combine or other command? (I ask because I heard that sem and some operations will not work in Stata (not sure about other programs) and I am worried about that....

[daniel klein]

I think it would be wise to stick to one thread. If you have another question or a follow up, please post it in the appropriate place so it is easier for people to follow along.

Best
Daniel

[Carlo Lazzaro]

Jacob:
please see Daniel's helpful reply at: https://www.statalist.org/forums/for...ombine-command
Please, do not post the same query twice (usually, reposting dcreases teh chances of getting helpful replies).

That said, I would recommend you to switch to -mi- command suite: you will find that -mi estimate- does what you're after.

PS: Crossed in the cyberspace with Daniel's reply, which has a similar flavour.