Apologies for a potentially dumb question: are there any examples of how to change the class order in LCA (conditional on covariates)? I.e., to make the last class first (for instance). I know the SVALUES option is used to provide input statements with starting values, but how exactly is this done? I.e., a script would be helpful.
I am working on a LCA analysis with covariates. I'm wanting to apply the boostrap likelihood ratio test to compare k with k-1 classes. According to the manual, and informaiton on this site, there is a requirement to reorder the classes to get the last class as the largest.
Based on seed values from stable solutions from random start models (i.e. with repeated highest loglikehoods), I'm using OPTSEED command to set the seed. I'm also introducing parameter start values based on the solutions in the random start models.
I'm finding, however, that the order in which I try to introduce classes, according to the starting values, produces different solutions. If I keep the order the same as the random starts model, I get exactly the same solution. However, if I try to order the classes differently (e.g. from smallest to largest according to random starts model), my class solutions are very different. Sizes of classes are very different, and they don't follow from small to large.
Am I doing something wrong, or is it possible that I just have very unstable/multiple solutions for my data? I know that even with the random starts model, I was having to use a lot of random starts to get stable solutions if that helps.