Hi Statalist

Questions regarding issues encountered during -mi- have been asked many times in this forum. I note especially My question is somewhat different. Specifically, I have 1 continuous variable & 7 binary variables with missing data. The continuous variable has about 10% of its data missing whereas the proportions of missing data for the binary variables are trivial (i.e. no more than 1-2% in each).

My question is that I don't understand why mi impute mvn works but mi impute chained does not since - from my reading of the Stata documentation - chained equations are a much more flexible (accommodating?) method.

Furthermore, I narrowed down the problem to be my binary variables in the MICE approach and even when I only include a single binary variable like
Code:
mi impute chained (logit) binary_var = ... , augment
it still fails.
  1. Is mvn necessarily a more restrictive method than MICE with only continuous & binary variables?
  2. is it ok to rely on mvn when -mi impute chained- fails to converge or does it signify the data is such that multiple imputation is perhaps ill-advised?
Thank you.


​