In addition, I am not clear on how to identify the classes in the Latent Transition matrix. I am concerned about class switching and subsequently ran the model with start values for the items in each model (model at Time 1 and model at T2),resulting in a slightly different transition matrix than when run without the starting values - yet, I am still not able to determine which classes are which. Any help would be greatly appreciated.
You have a total of 3 x 3 = 9 classes. Call the class variables c1 and c2. The first 2 digits give the categories for c1 and for c2.
The means/probabilities of the outcomes for different latent class variable categories is what give the categories meaning/interpretation. Using start values from the final estimates of a run with SVALUES is a good idea to avoid class switching - combine that with Starts=0 but check that you get the same best loglikelihood value.
Anonymous posted on Tuesday, December 05, 2017 - 1:24 pm
Thank you, Dr. Muthen. 1. Just to be clear about the interpretation of the classes from the Pattern table, 1 1 for example, is prevalent at ~17.9%, which suggests that 17.9% of the sample is in category 1 at time 1 and in category 1 at time 2 (if the class variables are time)?
2. I have run the LTA with SVALUES, and now have start values. To avoid class switching then, do I need to include the start values for each indicator (item), and each category (3) for both class variables? I am not sure if this makes sense, but I am just not sure how to use the start values in the second run to avoid class switching.
General comment: Be sure to listen to Topics 5 and 6 of our Short Courses on the website.
2. Yes in your first question.
If the classes at time 2 are in a different order than the classes at time 1, you can use the SVALUES values for the indicators and re-arrange them in the new order that you desire in order to give starting values that don't need Starts (but instead Starts=0).