It depends on the missing data mechanism. Under MAR and MCAR, multiple imputation and full information maximum likelihood are most appropriate for many applications. See, e.g. Enders (2010), Applied missing data analysis, Guilford Press.

Hello, Mohamad,

Maybe it will be useful to think if it is worth imputing missing data, instead of just discarding it? First, you have too few cases to appreciate the degree of randomness of the missing responses (MCAR/MAR/MNAR) - I suppose they are not random at all in your case (thus MNAR) -, second, is hard to assume a distribution from your data (which is key in obtaining a valid mathematical imputation).

As you said - if I understood well -, most of the missing answers are from the control group, which is the more populous, so without those answers, the groups will become somehow balanced. So, you should weigh which approach will be less biased? My 2 Cents...

I recently had a similar problem (some people missing from the post-test) and I used the generalized estimating equations (GEE) procedure to solve it. It does not impute the missing data, but it allows for the analysis of repeated measurements without losing people with incomplete data. Maybe it is some solution for your problem too.

Christian Geiser Thank you,

I studied the names you mentioned and gained a better understanding of the mechanisms, and thank you for introducing that great reference.

However, I decided to omit the people with missing responses cuz they were more in the control group (3 ppl), and still, the experimental and control group numbers are somehow sufficient.

My other choice would be using generalized estimating equations (GEE) which is a little complicated and unnecessary.

Cristian Popa Thank you for offering this new perception that I can let it go and omit ppl with missing data.

After evaluating the situation I decided to do repeated measures without those ppl who were more in the control group.

I was so into finding a way to impute missing data that I lost the whole picture and see if it's really worth doing it.

Klara Austeja Buczel Thanks for introducing GEE,

I did not know about this option and certainly, I will use it in the future.

However, this time after evaluating the situation and seeing that in our field (human ecology) it is not common for this number of ppl to use GEE, I put it aside and I am going to use repeated measures without ppl with missing data.

Again thank you for informing me about this option that I did not know about.