Hi, I have longitudinal continuous data where each subject is measured at 6 time points. I started by doing growth models with linear and quadratic trajectories but it seemed that a less restrictive model would work better. I could just free the coefficient of the slope term and all of that but instead I decided to just run a LCA model with my measures and the results looked ok. I wonder what is the problem of LCA with longitudinal data? this is my model (my time points are not equally spaced):
VARIABLE: NAMES IS id y0 y1 y2 y4 y6 y8; USEVARIABLES IS y0 y1 y2 y4 y6 y8;; IDVARIABLE is id; MISSING are ALL (999); classes = c(4); ANALYSIS: TYPE is mixture; MODEL: %overall% y0 y1 y2 y4 y6 y8 with y0 y1 y2 y4 y6 y8; PLOT: Type = plot3; series = y0(0) y1(1) y2(2) y4(4) y6(6) y(8);
When I look at the plot for estimated means I get a nice plot with trajectories with no restriction on its format and they make sense. When I look at plot of estimated means with individual profiles it also makes sense. The only thing I am not comfortable with is that I dont find this kind of model in the literature, that is, whenever there is longitudinal data it seems that people specify random effects. SO I wonder what would be the problems with this sort of LCA model.
We talk about using LCA for longitudinal data on slides 75-88 of our Topic 6 handout and video on our website. We say that LCA is a good way to explore the growth shape for categorical outcomes where you can't plot individual curves to get ideas. But LCGA, using growth factors, gives more parsimony and better BIC as we show. And you can regress those growth factors on covariates. I have seen some writings on this, perhaps by Katherine Masyn, but can't pintpoint it - you could email her at Harvard School of Ed.