Multiple Imputation and Checking Regression Assumptions - What Data Should We Use?

If you use multiple imputation for missing data in a regression analysis you have to check the regression assumptions (e.g. normality, homoskedasticity, linearity). This tutorial explains what data (cases with complete data or imputed datasets) you should use for the assumption checks in a linear regression with multiple imputation, depending on the missing data mechanism (missing completely at random, missing at random).

Here is a text version of this tutorial complete with references:
Regorz, A. (2022). Checking regression assumptions when using multiple imputation for missing
data (Issues in Applied Statistics 1/22). Regorz Statistik.

Here are two Youtube videos explaining the missing data mechanisms (MCAR, MAR, MNAR):

[fatimaah]

I wish you were by course instructor!!! Thank you so much for the immensely helpful videos!

[vufybyp]

Thanks a lot for the information! The suggestions are super helpful! May I please ask if there are any r packages you would suggest for checking the assumptions of regression models?

[christophelatrille]

Hello, thank you very much for this very informative video. I have some missing MAR data to process. I have imputed with the mice package (m=20). If I understand correctly, I need to test the hypotheses on 5 separate samples (taken at random). With a view to referencing this methodology in a research article, do you have a reference to support this way of proceeding, please? It would be very helpful.

[uzmaimam]

What if I had big data sets, how to handle