Re: paired observations

Proc TTest would yield biased results. You should first diagnose the
missingness as MAR /MCAR before choosing the method to impute. Model
based imputation can be done using Proc MI and Proc MIAnalyze.

It is better than a naive mean substitution

good luck

vasilis wrote:
> Dear all,
>
> I'd be grateful if you can consult me on the following problem.
>
> I've got a sample of patients measures twice: before and after
> treatment.
> Therefore in order to compare the means of the two measurements I use a
> paired t-test.
>
> However, in many cases (patients) the second measurement (that of post
> treatment) is missing.
>
> My question is whether it is still ok to use a paired t-test to compare
> the two measurements, since the post-treatment mean based on say x
> number of patients will be compared to the pre-treatment mean of the
> same x number of patients and not to the initial one which was based on
> x+n number of patients.
>
> Thanks in advance for your advice,
>
> Vasilis

.