Do you have this paper:

https://pdfs.semanticscholar.org/b1c6/217d654be6de94ed3b2acd58e63e6dd3cb33.pdf

Dear Mehdi Pazhoohesh,

I guess this reference is very helpful for you to answer the question:

[13] McKnight, P.E. and McKnight, K.M., Sidani, S. and Figueredo,

A.J. (2007). “Missing Data: A Gentle Introduction”.

Guilford Press, New York.

Mohamed I. Riffi

Hi Mehdi,

If you are taking a full bayesian approach to inference in your study, then missing values can easily be handled. You can use MCMC or any other sampler (Gibbs, HMC) to get the posterior.

You need to have a model to impute the missing values. However if the missing values are only on the outcome variable, then you don't need additional model; the main one will do the job.

MCAR means: the missingness is completely random and not related to observed or un-obsereved values of the the variable itself or any other measured variable.

If you truly believe this is the case or you simulated the data yourself, then there is no clear impuration model; because it is essentially random.

In anycase, the advantage in using a full bayesian imputation (using MCMC or other) for MCAR situation is that you may be able to reduce the width of the credible interval. At the same time propagating the uncertainty in the missing values. The point estimates do not change whether you impute or not.

Hope this helps

Sreenath Madathil