I have some longitudinal data with the following variables,
id, time, outcome, stress, and multiple measurements on
outcome over time for each
id. The question of interest would be, does the change over time (ie growth, trajectory, slope, whatever you want to term it) in
stress predict the change over time (growth, trajectory) in the
outcome. When I think about this question it almost sounds like, “do the
id level slopes in
stress predict the
id level slopes in
I guess I am daft, but I am having a really hard time trying to figure out how to model this question!! (I did try to simulate some data, but I don’t think I ended up simulating quite what I wanted.)
I think what I might want is something called a “multivariate growth model”? But I have had trouble seeing exactly how to formulate it.
Here is what my ideas were:
bf_stress <- bf(stress ~ 0 + (1 + time|p|id)) bf_outcome <- bf(outcome ~ 0 + (1 + time|p|id)) m1 <- brm(bf_outcome + bf_stress + set_rescor(FALSE), data=data)
This would give me the correlation between the varying slopes for
stress and the varying slopes for
outcome but that isn’t exactly what I was looking for.
Another idea was:
bf_stress <- bf(stress ~ time + (time|p|id)) bf_outcome <- bf(outcome ~ time*stress + (time|p|id)) m2 <- brm(bf_stress + bf_outcome + set_rescor(FALSE), data=data)
stress in the
outcome model, but I don’t see that it is really looking at the association of the growth in
stress over time with the growth in
outcome over time…
I have also thought about running a growth model on
stress and one on
outcome and then extracting the point estimates and sd’s for the slopes for each and then running a measurement error model using the slopes as data, like this:
m.s <- brm(stress ~ 0 + (1 + time|id), data=data) m.o <- brm(outcome ~ 0 + (1 + time|id), data=data) #insert code to extract means and sd's from posterior samples and create new dataframe that is length(id) long m.sl <- brm(slopes_outcome | mi(sd_outcome) ~ 1 + me(slopes_stress, sd_stress), data=data2)
But that just seems rather hackish and also weird…
Is there a way to answer my question in a single model? I’d be grateful if someone could help! My experience is limited in these sort of longitudinal growth models.
What it seems like I want is for the growth model for
stress to be the predictor for the growth model for
outcome… but how?