LCA Interpretation of Output
Message/Author
 Sarah Dauber posted on Wednesday, February 28, 2007 - 9:07 am
Hello,
I am running an LCA with ordinal indicator variables. Each variable has between 3 and 5 response categories. I am looking at the part of the output titled "results in probability scale." My understanding is that the estimate for each response category of each variable is the probability of endorsing that response, is this correct?
Can this also be interpreted as the percent of people in each class that endorsed that response?
For ex, one of my variables is mental health problems with the following responses and estimated probabilities for class 1:
No problems .914
Meds only .080
Hospitalization and meds .010
Is it correct to say that 91% of people in class1 have no mental health problems, 8% have meds only, and 10% have hospitalization and meds?

Thanks so much for your help!

Sarah Dauber
 Linda K. Muthen posted on Wednesday, February 28, 2007 - 11:39 am
That sounds correct.
 Sarah Dauber posted on Monday, March 12, 2007 - 8:11 am
Hello,
I just ran a 2-class LCA model. When I checked the percent of respondents in each category of each indicator against the frequencies of each item printed from SPSS, they were slightly off. What could be the reason for this?

Thanks,
Sarah Dauber
 Linda K. Muthen posted on Saturday, March 17, 2007 - 5:53 am
I think it is probably that a different sample size is being used in each case. Check the number of observations.
 Jungeun Lee posted on Tuesday, June 23, 2009 - 3:40 pm
Hello,

I am running a LCA with count variables. Mplus gave me means of each indicator per a class. I am thinking, exponentiated means of each indicator = average expected count of each indicator in a given class. Is it correct?

Thank you so much!
 Bengt O. Muthen posted on Wednesday, June 24, 2009 - 9:00 am
Mplus estimates the log rate (because the model is linear in this quantity). So if you want the rate (mean), yes, you would exponentiate it.