01-30-2013 06:26 PM
Depending on how you're simulating wouldn't you just be picking 'people' for the study from your data. I think this is a jackknife simulation.
I'd take a sample of the data I had, with repetition, if applicable and go from there. Your real data has missing values so there shouldn't be an issue.
01-31-2013 08:26 AM
I think you need to provide more information. What is the mechanism for generating missing values? Is there a pattern of missing values that you are trying to simulate (see http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_mi_sect018.h...)? For example, missing completely at random? Because you stated "longitudinal data," do the missing values representing censored observations?
01-31-2013 09:42 AM
Thanks a lot for you answers Reeza and Rick!
I am interested in different scenarii for generating missing values, including the three ones described in the link you provided Rick.