Message/Author 

Anonymous posted on Friday, September 05, 2003  11:28 am



I am new to latent class analysis and I'm trying to running a 2 class model with 1 dichotomous and 3 3category variables. I'm not sure how to specify the starting values for the thresholds. I tried putting in some numbers ranging from 2 to 2. However, I keep getting an error message that says: "IN THE OPTIMIZATION, ONE OR MORE LOGIT THRESHOLDS APPROACHED AND WERE SET AT THE EXTREME VALUES. EXTREME VALUES ARE 15.000 AND 15.000." I've tried putting in different values, but have not been able to fix the problem. I'd appreciate any suggestions about what might be the source of the problem and how to fix it. Thank you! 


This is not a problem. It just means that in certain classes, certain items have either a probability of zero or one. This can help define the classes. 


We tried to change the reference class in a latent class analysis by entering starting values for the thresholds, but the reference class did not change. Even when we tried using extreme starting values the reference class did not change. We only succeeded by constraining the thresholds with the @ symbol. Do you know why this might have happened? Thank you very much for your help! 


Do you have STARTS=0; You may find using the STARTS option of the OUTPUT command useful. This gives the input with starting values and you can simply change the class numbers. 


We had random starts, but using STARTS = 0 solved the problem. Thanks for your help! 

DavidBoyda posted on Sunday, March 04, 2018  7:34 am



Hi Professor, I have regressed my LCA classes onto two distal outcomes, however, one of the variables has low observations (<30 cases). When examining the thresholds for said distal outcomes one of the variables across two classes has high fixed thresholds: Y2$1 15.000 0.000 999.000 999.000 I understand this is because there is (almost) nobody in the highest category of Y2 in that class. Is there any function within Mplus that i can call to get around this? Or is it normal to report this? 


You don't want to get around this; just report it. It is useful to know that membership in a certain class determines the distal outcome category. A comment on nomenclature  you don't regress classes onto a distal. You regress a distal onto the classes. 

DavidBoyda posted on Tuesday, March 13, 2018  9:56 am



Dear Bengt, in reference to my enquiry above your answer (March 04) Y2$1 15.000 0.000 999.000 999.000 could you provide a little more clarity on what you said: "...It is useful to know that membership in a certain class determines the distal outcome category". My classes were mental health and distal outcome is suicide. So, are you saying, there were suicide attempts in class 1 but no attempts in the other classes and this is a function of class endorsement? kind regards, D 


If Y2 is a binary distal, the Y2$1 15.000 result for a certain class implies that when a person is in this class there is zero probability of being in the highest category of the distal. The fact that the probability is zero and not for instance 0.25 (greater than zero, not ignorable, but not close to 1 or even 0.5) makes the relationship between the latent class and the distal stronger. 

Back to top 