

Retrospective classification 

Message/Author 

Daniel posted on Monday, September 08, 2003  12:25 pm



Is it possible in your opinion to classify growth trajectories retrospectively by regressing the latent class variable from say time 14 on a variable measured at time 4? What I'd like to say is that individuals high on this covariate at time 4 were more likely to follow one trajectory over a comparison. Does this make sense? The idea is a retrospective cohort study, where individuals high and low on a risk factor are compared retrospectively on growth. 

bmuthen posted on Tuesday, September 09, 2003  10:03 am



Typically, the latent class variable ("c" say) is thought to be a preexisting condition so that c predicts the growth curve. So, regressing c on the time 4 covariate would seem to reverse the time ordering. Even if you center at time 4 so that the intercept is defined at time 4, the slope is defined as earlier change, so having c be predicted by the time 4 covariate and at the same time predict the slope would be strange. Perhaps instead you could let c predict the time 4 covariate, which in turn predicts the outcome at time 4. 

Back to top 

