Comparing models and profiles (including significance) using LPA (continous variables)

[Serge Damottaveiga]

Hello everyone,

Thanks for all the helpful information on this site.

I am trying to compare models and profiles in LPA (continous variable), and obtain the LMR (AIC and BIC do not tell me whether the models are significantly different). I know this is easy in Mplus, so wondering why it is so difficult in STATA :-).

If there is another way (than LMR) to compare model/profile significance, that would be much appreciated.

Thanks for your help!

Serge

p.s. I can get the tables comparing the models and profiles with AIC, BIC, and df, just missing the significance for comparing models (two at a time). Thank you!

[Phil Bromiley]

Please follow the FAQ on asking questions.

Try not to use acronyms unless you are sure they are used in all areas (like AIC). What a LMR is is not clear to me. You talk about profiles, but don't explain what that means. It sounds like you ran a regression but even that is not clear.

[Serge Damottaveiga]

My apologies for the use of acronyms and for not being clear in my question.

I have a few variables, for which I am trying to obtain a number of profiles that fit the data (thus taking a person-centered approach). As such, I am using Latent Profile Analysis (a variant of Latent Class Analysis). I am trying to obtain the number of profiles that fits the data best, and thus need to compare whether 2 profiles is better than one, three is better than two, and so on. I could just use AIC and BIC, but these do not capture whether the profiles are significant different from one another.

To do so, I would need to conduct a LMR (Lo-Mendell-Rubin adjusted likelihood ratio test), and also ideally calculate entropy. Would someone have some syntax to calculate these?

Furthermore, I would also like to compare models (and profiles) between different time periods. For example, if I get 3 latent profiles for my variables of interest at time 2 and then 3 latent profiles again for these variables at time 4, I would like to compare whether these profiles are similar or different. To do so, I would need to do Latent Transition Analysis, and was wondering again if STATA was able to run these, and if someone had syntax that could help.

Thank you!

Serge

[Radhouene DOGGUI]

Serge Damottaveiga please, did you find how to run the Lo-Mendell-Rubin adjusted likelihood ratio test on STATA ?