

Longitudinal patterns of multivariate... 

Message/Author 

Liang posted on Friday, April 08, 2011  8:38 am



Dear Dr. Muthen & Muthen, I have a question about how to use Mplus to define longitudinal patterns of multivariate psychological variables. We have three observation time points (j=1,2,3). In jth observation time we observe distress score (Y_j1), somatization score (Y_j2), and anxiety score (Y_j3). If at each observation time we use a global score (Y_j) for the three variables, it is easy to write the code as following, suppose we want 4 classes. Usevariables are Y_1 Y_2 Y_3; Classes=c(4); But the investigator insists to use three variables at each time point to classify 4 longitudinal patterns (not three independent models for distress, somatization, anxiety, separately). Then can I write as the following? Usevariables are Y_11 Y_12 Y_13 Y_21 Y_22 Y_23 Y_31 Y_32 Y_33; Classes=c(4); If it is wrong to write in this way, what should I do? Do you have any similar example for me to refer? Thank you very much for your answer! 


You can write it the way you propose. Your model becomes a parallel process model (although with 3 instead of 2 processes). See the UG index for such a model. To which it sounds like you want to add one latent class variable that is determined by all 3 processes. 

MCA posted on Thursday, January 16, 2014  7:57 pm



Hello, I am wondering if it's possible to estimate a series of such models with increasing numbers of categorical latent variables, and then test how many categorical latent variables are needed. If so, do you have examples/guidance on how to perform such model comparisons? Finally, is it possible to combine categorical and continuous manifest variables in this setting (like you can do for crosssectional LCA)? If so, is there an example anywhere of this? Thanks in advance! 


Are you referring to the above discussion from April 2008? 

MCA posted on Friday, January 17, 2014  9:13 am



Hi, Sorry for not being clearer  I was referring to the discussion from April 08, 2011. The original question was: "I have a question about how to use Mplus to define longitudinal patterns of multivariate psychological variables. We have three observation time points (j=1,2,3). In jth observation time we observe distress score (Y_j1), somatization score (Y_j2), and anxiety score (Y_j3). If at each observation time we use a global score (Y_j) for the three variables, it is easy to write the code as following, suppose we want 4 classes. Usevariables are Y_1 Y_2 Y_3; Classes=c(4); But the investigator insists to use three variables at each time point to classify 4 longitudinal patterns (not three independent models for distress, somatization, anxiety, separately). Then can I write as the following? Usevariables are Y_11 Y_12 Y_13 Y_21 Y_22 Y_23 Y_31 Y_32 Y_33; Classes=c(4); If it is wrong to write in this way, what should I do? Do you have any similar example for me to refer? Thank you very much for your answer!" Thanks so much! 


When I read this message now, I interpret it as a multiple (distress, somati, anx) indicator model for a single factor at 3 time points rather than a parallel process model. But BIC can be used in both cases to decide on the number of latent classes that are needed. You can combine any types of observed variables. 

Back to top 

