MI extract does not extract one of the imputed variables

[Vilma Antonov]

Hi!

I have a dataset with 60 variables that were imputed using MICE. Unfortunately, one variable were excluded by mistake from the imputation and hence I imputed it using MICE after the first imputation was finished. The imputation was finished and all well, but when using MI extract, var61 does not get its imputed values with it!

set more off
set matsize 11000
mi set wide
mi register imputed var1-var60 
mi impute chained (pmm, knn(3)) var1-var60, add(10)

* And then 

set more off
set matsize 11000
mi set wide
mi register imputed var61
mi impute chained (pmm, knn(3)) var61, add(10)

** Which gives me the following output **
        Observations per m               
            
Variable    Complete    Incomplete   Imputed    Total
            
var61    40375    3    3    40378


* And then I try with the mi extract
mi extract 1 

sum var61
    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
var61 |     40,375    1.755012    1.261333          1          6

What is the problem? It feels like something obvious I'm missing.

[daniel klein]

That's not how mi works. In your situation, var61 does not have imputed values in the first 10 imputed datasets. That is made evident by mi extract command. Instead, var61 does have imputed values in the 10 imputed datasets 11 to 20. Yes, your code has created 20 imputed datasets. Unfortunately, none of the other variables will have imputed values in the datasets 11 to 20.

Moreover, you impute var61 with no predictors at all. Thus, the imputed values will be uncorrelated with all other variables. You do not want to do that.

What you want to do is delete all imputations and start over from scratch, this time including var61 in the imputation model.

[Vilma Antonov]

Thank you, I understand what you mean and performed a new imputation!