

LTA: Regress trans probs on categoric... 

Message/Author 

Brian Segal posted on Wednesday, March 19, 2014  2:13 pm



I am working on an LTA and would like to regress the transition probabilities on a categorical covariate (a binary indicator for gender). Example 8.14 in the user's guide shows how to regress the transition probabilities on a continuous covariate, and example 8.13 shows how to obtain transition probabilities, conditional on a categorical covariate. Unfortunately, I have not been able to make 8.14 work for categorical covariates, and 8.13 is not quite what I want. To make this problem more concrete, I am modeling latent health behaviors and I want to test whether transition probabilities differ between men and women. Under example 8.13, Mplus estimates transition probabilities separately for men and women. You can do a Wald test to evaluate the difference between the two estimates, but that requires a few extra calculations after obtaining the Mplus output. Given that I would like to run several models with multiple covariates, it would be convenient if Mplus could parameterize the model so that it directly estimates the effects of the covariates on transition probabilities. Is that possible in Mplus with categorical covariates? I believe it is the default parameterization in proc LTA. Thank you for your help. 


You can use the ex8.14 approach also with a categorical covariate (you don't specify it as categorical). If you have problems with this, please send output and license number to Support. See also our web note 13. 

Brian Segal posted on Thursday, March 20, 2014  3:12 pm



Thank you very much for your help. Coding the categorical covariates as indicator variables and then treating them as continuous seems to work well. 


In regression, covariates are either binary or continuous. In both cases they are treated as continuous. The scale of covariates is not as issue. 

Back to top 

