Jan H posted on Wednesday, August 14, 2019 - 10:46 am
I have a longitudinal dataset and I want to do a LCGA. I know that Mplus automatically applies FIML to missing data, but the question is if this approach is adequate for deceased participants? I have three measurements and several participants die between the second and third assessment.
There are several questions: First, basically if it is correct to even estimate missings for deceased participants?
Second, if not, is there a way to include participants in the FIML estimation for the first and second assessment and exclude them from the FIML for the third assessment because they have died between the second and third assessment?
Third, if not, how would you handle the data? Just exclude the participants who die between the second and third assessment from the entire analysis like listwise deletion, even if I have data for the first two waves?
Thanks for any recommendation.
Jan H posted on Wednesday, August 14, 2019 - 10:53 am
Fourth, or should I do the model estimation for both situations, one with the entire sample and FIML and one only with participants who are still alive at the third assessment and FIML and see if there are differences in the results?
Take a look at modeling alternatives in our article which focuses on dropout:
Muthén, B., Asparouhov, T., Hunter, A. & Leuchter, A. (2011). Growth modeling with non-ignorable dropout: Alternative analyses of the STAR*D antidepressant trial. Psychological Methods, 16, 17-33. (Download scripts)
This sounds like "pattern-mixture" modeling which is one alternative - with extensions - that we discuss in the article above.