Multiple imputation then delete

[Delia DDKan]

Hi I am trying to work on a dataset with missing outcome data. I would like to use the method proposed by von Hippel (2007) which is to do multiple imputation first on all participants including those with missing data, then dropping the participants that had their outcomes imputed before conducting regression analysis. I want to ask how can I manipulate each of the imputed dataset to drop the participants that had the outcome variable imputed and subsequently use the new imputed datasets to run the regression analysis? would love any advice thank you!

[daniel klein]

If missing values are only in the outcome, then imputing them only to delete the imputations is utterly pointless.

If you have missing values in the predictors as well, then the general approach is to create a marker/indicator variable for missing outcomes before imputation. This is easiest in flong datasets. We need more details, especially in terms of used syntax, to give more specific advice.

[Rich Goldstein]

in addition to the helpful reply from Daniel Klein, you might want to read the following: Sullivan, TR, et al. (2015), "Bias and precision of the "Multiple imputation, then delete" method for dealing with missing outcome data", American Journal of Epidemiology, 182(6): 528-534