# Impute with the Mean or Median? Instrumental Variables

I am using instrumental variables and I have missing data. In r, I don't believe you can use the MICE package with the AER package. Therefore, I am going to impute with either the mean or median values. My variable is heavily skewed, so I am incline to use the median value. Do researchers use the median-imputation? I am more familiar with researchers using mean-imputation.

Is there an alternative to imputation with the mean and median value? If you can use the AER and MICE packages together, could someone point me to code that uses it?

• Single imputation of the type you intend to use will result in seriously underestimated standard errors of your final parameter estimates, and false significance. I would only use that approach if the proportion of incomplete observations is tiny. – Frank Harrell Aug 14 '15 at 18:15

The hard part about multiple imputation is the imputation, which mice can do even if for some reason you can't use the additional pooling functionality it provides. A quick look at the AER package suggests that it is mostly data, with examples using many functions like lm that certainly are compatible with mice.
If you do find a particular analysis function that is not compatible with the pooling functions of mice, proceed as follows, following the general procedure for multiple imputation:
1. Do your multiple imputations with mice and keep each separate complete imputed data set.