KHB with multiply imputed data

[Oliver Winkler]

Dear statalisters,

I am using the khb method that is implemented in the ado "khb" by Ulrich Kohler et al. (http://www.stata-journal.com/sjpdf.h...iclenum=st0236). My question is whether someone has experiences using khb for multiply imputed data. The command is not part of the supported mi estimation commands. I tried a combination with cmdok but it is seems as the disentangle option (used in case of more than one mediator) is not working.
I would be very interested in your remarks on this.

[winkhaing]

Oliver Winkler
Dear Oliver, May I trace back your question posted in 2016.
Did you get any solution for your question?
I am also facing with the same problem, not get solution yet.

Thanks.

[winkhaing]

Ulrich Kohler
Dear Profess Ulrich Kohler,
Could you please allow me to tag/mention your name in this post?
Because you're one of pioneer of KHB - Stata command.
Thanks.

[Ulrich Kohler]

No, the disentangle option of -khb- does not work with multiple imputation., and as it stands I do not have plans to implement that.

[Bram Hogendoorn]

Hi Oliver, this article may be interesting for you:

Brand, J., van Buuren, S., le Cessie, S., & van den Hout, W. (2019). Combining multiple imputation and bootstrap in the analysis of cost‐effectiveness trial data. Statistics in medicine, 38(2), 210-220.

You could write a small program in Stata that single-imputes missing values and runs khb on your singly imputed dataset, and then bootstrap the entire procedure. Depending on the size of the dataset this may be very computationally intensive, but it could give you what you want. I do not know how to obtain the "disentangle" estimates but it should be possible.

[Bram Hogendoorn]

Another option is to manually adjust the -khb- sourcode and post the disentangle results to e(b) and e(V), before using -mi estimate-.