

Extreme values in LCA results 

Message/Author 


The output of a mixture model states: IN THE OPTIMIZATION, ONE OR MORE LOGIT THRESHOLDS APPROACHED AND WERE SET AT THE EXTREME VALUES. I used random starting values (not specified by the user) and also increased the MITERATIONS, but still receive the above message. I get the message when fitting a 3 or 4 class model but not a 2 class. 1) Is there anything I can do to avoid the extreme value? 2) If I get extreme values does that invalidate the mixture model results? Thanks 


This message is not a problem. It is telling you that for some clssses a latent class indicator has a probability of zero or one. This can be helpful is defining a class. 


Dear Bengt and Linda, I am running an LCA with 50 DVs and 358 observations. All DVs are binary, ordinal or counts and are declared as such with the 'categorical' and 'count' keywords. There is a lot of missing data (much of it meaningful missing) and some of the DVs are very highly intercorrelated (> .9). I get 2 Extreme values messages: IN THE OPTIMIZATION, ONE OR MORE LOGIT THRESHOLDS APPROACHED AND WERE SET AT THE EXTREME VALUES. EXTREME VALUES ARE 15.000 AND 15.000. IN THE OPTIMIZATION, ONE OR MORE LOGIT SCALE PARAMETERS APPROACHED AND WERE SET AT THE EXTREME VALUES. EXTREME VALUES ARE 15.000 AND 15.000. The first one (Logit threshold) only lists one ordinal indicator (DV) (7 times). The second one (Logit scale parameters) mentions 11 DVs, all counts. In my case, are the 2 messages not a problem as you write above? 


Most likely. But to be sure we have to see the full output; so if you like, send to Support along with your license number. 

Back to top 

