Message/Author |
|
Adrian Byrne posted on Wednesday, February 05, 2014 - 8:27 am
|
|
|
Dear Professor Muthen, How can one tell Mplus to distinguish between latent class indicators at time 1 and observed categorical distal outcomes at time 2? We do not wish for the distal outcome to affect the classes, which occurred previously in time, in a cross-sectional way. We want to investigate how the latent class membership at time 1 affects observed category membership at time 2. Many thanks. |
|
|
Please see the stepwise mixture modeling techniques discussed in Asparouhov & Muthén (2013). Auxiliary Variables in Mixture Modeling: 3-Step Approaches Using Mplus. Accepted for publication in Structural Equation Modeling. An earlier version of this paper is posted as web note 15. Appendices with Mplus scripts are available here. which is on our website. |
|
|
Dear Professor Muthén, Thank you for your quick response and suggested solution. I have been following the method and Mplus code (Appendices E and F) to conduct the 3-step procedure with an arbitrary secondary model within your paper. However, I encounter a problem at step 3 as I receive the following ERROR message: *** ERROR in MODEL command Variances for categorical outcomes can only be specified using PARAMETERIZATION=THETA with estimators WLS, WLSM, or WLSMV. I am using a binary outcome variable (which I am treating as categorical). I am aware that PARAMETERIZATION=THETA is not allowed when TYPE = mixture. Is it simply best to treat the outcome variable as continuous (if possible) or is there another solution to allow for categorical outcome variables at step 3 of the suggested procedure? Many thanks. |
|
|
You should treat the variable as categorical but not mention the variance term. Variances are not estimated for categorical variables. |
|
|
Dear Professor Muthén, Thanks again for your help. I have another query regarding the 3-step procedure (Appendices E and F); can I still apply the same procedure when I want to regress the latent class membership at time 2 on observed predictors at time 1? This time I want to treat the latent variable as a response rather than a predictor. If this 3-step procedure is not appropriate, can you suggest a better alternative? Many thanks. |
|
|
You can use R3STEP for this. See the first couple of appendices. |
|
|
Hi, I am doing a LCGA with two binary indicators assessed at five measurement points starting 1 year after treatment. Is it possible to predict class membership and to investigate the stability of these classes by taking the trajectories during therapy (three assessment points) into account, e.g., by another LCGA? Thank you for your response. |
|
|
It sounds like you want to consider two sequential LCGAs. This is doable and you can have the latent class variable of the second process regressed on the latent class variable of the first process. So your input would look something like: Model: %overall% c2 on c1; Model c1: give the LCGA for c1 give the intercepts/threshholds that are class-varying with c1 classes Model c2: give the LCGA for c2 give the intercepts/threshholds that are class-varying with c2 classes and within each of those |
|
|
Hi Bengt, thank you very much for your response. Do you have/know any example for this sequential LCGA? Best, Stefanie |
|
|
Maybe UG ex 8.7 could be a useful starting point. I discussed an example in the paper below which you find on my UCLA website, accessed via the Mplus website through "Contact us/About Us": 77) Muthén, B., Khoo, S.T., Francis, D. & Kim Boscardin, C. (2003). Analysis of reading skills development from Kindergarten through first grade: An application of growth mixture modeling to sequential processes. Multilevel Modeling: Methodological Advances, Issues, and Applications (in press). S.R. Reise & N. Duan (Eds). Mahaw, NJ: Lawrence Erlbaum Associates, pp.71-89. [Available as PDF] |
|
|
See also the handout for Topic 6, slides 158-168. |
|
Back to top |