We are interested in estimating mixture models looking at changes in sex intentions, attitudes, and norms over time (grades 8, 9 , and 10) on the part of adolescents. (a separate model for each variable). our first choice would be growth mixture modeling, but are aware that it's best to have 4 or more time points (we have 3). would it make sense to estimate these models in a latent profile analysis instead (so, thinking of it as profiles of people based on, for example, their intentions at different time points)? or would it still be preferable to approach this as a growth model? i'm not sure if this is useful information, but we did an lpa for intentions, and it would appear that the profiles for that variable suggests pretty linear (a straight line) change over time within the classes.
bmuthen posted on Monday, February 09, 2004 - 8:10 am
An LPA would seem ok, but note that is assumes uncorrelated variables within classes. In contrast, a growth mixture model allows within-class variation due to the continuous growth factors. Sounds as if a growth mixture model could simply use a random intercept, no slope, growth factor - I think I'd go with that even though you have only 3 time points.