PMM for categorical variables with a small number of positive answers

[Yaqun-Yuan]

Hi all,

I am trying to impute categorical variables (2 or 3 levels) with PMM in MICE. I have 80 times of iterations and 80 imputed datasets. Although the iteration plots look fine in terms of the convergence, I am not sure if the imputation is still stable or reliable when involving variables with small numbers of positive answers, e.g., for a binary variable, the proportion of positive answer ('yes') is less than 2% (n~10-30, <10 sometimes), and more than 80% are the negative answer ('No'). Because I am going to report the prevalence (proportion of the positive answer) based on the imputed data, so the validity of the method here is vital.

Looking forward to your advice!

Thank you!

Yaqun