This question is regarding the use of imputation methods, such as Multiple Imputation, for multiple regression models.
It is often suggested that one should compare the regression model created using the imputed data to a regression model created with the original data using the complete case analysis method to take care of missing values.
However, I am concerned with whether the imputed model/data should be used for further analysis or the complete-case analysis model/original data?
By further analysis I'm referring to analysis such as:
- Multicollinearity using VIF
- Checking for interaction terms
- Checking whether regression assumptions have been met:
- Such as normality of residuals