I am working with a small data set (n=267) with mostly categorical variables. For each variable there is relatively low missingness (1-5%). Some of my measures must be imputed with chained multinomial regression (notoriously problematic in imputation). These measures are important to include in my final model, but do not successfully impute.

How do I discuss this in my manuscript? Do I state that the final estimates are biased due to missingness and could not be addressed with imputation?