[Limdep Nlogit List] Getting AIC, BIC and CAIC from a latent class model output

[Mikołaj Czajkowski]

immaculate omondi wrote:
> I am running some Latent class models. I would like to know how many segments (classes) would give me the best model fit. I understand that the information criteria statistics (Akaike Information Criterion- AIC, bayesian Information Criteria- BIC and CAIC) can be helpful, in part, in giving an indication of the fit of the model as the segments are added.
> 
> However, the output results I get from my models does not give these information criteria statistics. How can I get them for my models? If not, is there another way of assessing the model fit as segments are added in a model?

Dear Immaculate,

I don't know which Limdep version you are on, but in general Limdep 
would give you these statistics. Look for something like:

| Info. Criterion: AIC =          1.91033     |
|   Finite Sample: AIC =          1.91097     |
| Info. Criterion: BIC =          1.99864     |
| Info. Criterion:HQIC =          1.94358     |

This should be close under the log-lokelihood function value.

I am not sure how useful they are as a proper model selection criterion, 
and if a latent class model with fewer classes may be considered a 
restricted version of a latent class with more classes. For this reason, 
for what you're looking for I would try the Vuong test or the 
Ben-Akiwa-Swait test.

Best regards,

-- 
     Mikołaj Czajkowski

     Warsaw Ecological Economics Center
     Warsaw University
     http://www.woee.pl/