Hi, I'm conducting a LGMM on response patterns to a discrete event where I have baseline data prior to the event and then data at 6, 12, and 18 months post event. These three time points are obviously consistently spaced by subject, however, my baseline data is variable in terms of it's distance by subject from the event. How can I account for this in my model? I have data on the amount of time between the baseline measure and the event. I was thinking of freely estimating the 5-month data point so that i'm not fixing the distance between baseline and 6months, but I'm not sure if that makes sense. Any thoughts? Thanks for your help.
It sounds like you might want a combination of a survival model and a growth model. The individually-varying time between baseline and the event fits a survival analysis. Are you saying that the survival time plays a role in the subsequent growth model? Otherwise, the 2 steps could be analyzed separately.
Hi Bengt, Interesting, I hadn't thought of that at all. I think that I should run them together as time may play a role.The analysis is bereavement trajectories in older adults. The unconditional model built from depression scores from before the event to 3 post-loss time points. As proximity to the loss could effect these scores as many spouses may see the loss coming, I think it would be important to account for this in the model rather then running a separate analysis. Does that seem reasonable to you? P.S. thanks for getting back to me. I know Friday after thanksgiving is only a pseudo-holiday, but still. Best, Isaac
I see. Perhaps then one could let baseline depression and time to event from baseline be time-invariant covariates for the growth factors of growth for 6, 12, and 18 month outcomes. Perhaps also some interaction of the two - for instance, it would seem that a high baseline depression score has a different meaning if there is a short time to the bereavement event as compared to a longer time to event.
I see, so rather then including baseline depression scores as a point in the unconditional model, use it and time to event and perhaps the interaction as as covariates. That makes a lot of sense to me and exactly addresses my concern. Thanks a lot Bengt.