Linear regression on estimation samlpe after multiple imputation

[Rune Lomholt]

I am conducting a multiple linear regression analysis on multiply imputed data.

I wish to do a stepwise procedure were i stepwise specify my independent variables over a series of different models on the basis of a full model containing all variables of interest.

Normally, when doing stepwise linear regression on incomplete data, I perform the analysis on the estimation sample of the full model using an command like -gen nmin=e(sample)- to make sure all the models have the same number of observations, so the coefficients are more comparable between models excluding or including variables with reference to the full model.

How can I perform this procedure when using the command -mi estimate: reg- ?

[Bree Wang]

Has this problem been solved? I encountered the same issue.

[daniel klein]

I do not understand the problem. Once the data is (multiply) imputed, sample sizes no longer vary because of missing values in different variables; there are no longer any missing values. If that is not true, there is probably something wrong with the imputation.