

Why LCA affected by order of indicato... 

Message/Author 

Dinh Son Bui posted on Saturday, November 17, 2018  7:23 pm



Dear all. I am running a LCA model with 10 indicator variables. Two of them are 4level categorical variables. I have 2 questions: 1. When I change the order of indicator variable in the "usevariable" line. For example from v1 v2 v3 v4 to v1 v4 v2 v3, two LCA models give different results. It is unexpected. 2.It is a general question. If indicator variables are dependent, e.g. wheeze status assessed at 1, 2, 3 ,4 and 5 years, do we need to specify anything in the command to account for the dependence. Thanks for your help. 


1. This may be due to not using enough STARTS to get the best solution. 2. See the paper on our website: Asparouhov, T. & Muthen, B. (2015). Residual associations in latent class and latent transition analysis. Structural Equation Modeling: A Multidisciplinary Journal, 22:2, 169177, DOI: 10.1080/10705511.2014.935844. (Download scripts). 


Dear Dr Muthen, I have increase Starts to 400: 100. It takes too long to run the model but still not reproducible when the order of variables is changed. Also the is warning a bout loglikelihood: THE BEST LOGLIKELIHOOD VALUE HAS BEEN REPLICATED. RERUN WITH AT LEAST TWICE THE RANDOM STARTS TO CHECK THAT THE BEST LOGLIKELIHOOD IS STILL OBTAINED AND REPLICATED. IN THE OPTIMIZATION, ONE OR MORE LOGIT THRESHOLDS APPROACHED AND WERE SET AT THE EXTREME VALUES. EXTREME VALUES ARE 15.000 AND 15.000. THE FOLLOWING THRESHOLDS WERE SET AT THESE VALUES: * THRESHOLD 1 OF CLASS INDICATOR HEATER45 FOR CLASS 2 AT ITERATION 227 * THRESHOLD 2 OF CLASS INDICATOR HEATER45 FOR CLASS 2 AT ITERATION 227 ..etc The model include only 10 variables (8 are binary and 2 is order variables with 4 levels). It is not too complex isn't it?. Can you help? 


If you are saying that the best logL is the same for the two variable orderings, please send your 2 outputs and data to Support along with your license number. 

Back to top 

