Multiple imputation in Panel data with RE, non-normal variables

[Maha Natsheh]

I have the analysis of panel data, which I have 38% missing data in the dependent variable, other independent variables are complete and don't need any imputation.
Bu knowing that most variables don't follow normal distributing except one independent variable
after long research, PMM is the most useful way for MI the non-normal variables, but not in Panel data especially that I have just 3 years.

on the other hand, I knew that mi impute (chained) more consider time changing, is that true?
here are the results for mi estimate after chained imputation:

please note that PMM turns cash/assets and Eqcapital/assets to be insignificant even on 0.1

on the other hand, I want to refine the imputation model more than than, some literatures suggests that we must add the dependent variable to the imputation model to efficiently use the observed value in reproducing missing values, especially if we have large proportion of missing values and especially if they are in dependent variables, so I introduce new variable that represents dependent it self, but it has also missing values! and the process of imputation didn't proceed in execution !! any suggestions in this part or on the whole case? I know it is complicated some how that's why I asked for help.

Thanks for your help in advance!

Maha Natsheh

[Carlo Lazzaro]

Maha:
why not taking a look at -ipolate-?

[Maha Natsheh]

Hi Carlo,
Is it suitable for non-linear relations between dependent variable and independent variables (imputation's auxiliary variables)?

[Carlo Lazzaro]

Maha:
see the valuable Nick Cox 's article available from https://journals.sagepub.com/doi/10....6867X231196519