Log pseudolikelihood – Deviance

[Chan Ge]

Hi, could someone help with a question about log pseudolikelihood?
I estimate conditional logit models with standard errors clustered by the fixed group id. Stata reports Log pseudolikelihood for CLM with clustered standard errors.
I intend to assess the model fit of nested models.
My questions are: Is the deviance statistic still applicable here? Could log pseudolikelihood be used to calculate the deviance comparing two models? If not, is there a way to compare model fit in STATA? If yes, could you please share a reference?
Thank you!

[Chan Ge]

Hi all, for those who might be interested in the same question. I found that we can replicate the analyses using xtset -data- and then xtlogit, fe. xtlogit reports "log likelihood". Based on my data and my model specification, the two outputs are the same regarding log likelihood and log pseudo-likelihood and coefficient table. However, xtlogit, fe does not allow for clustered standard errors. Is there a way to correct for standard errors in xtlogit? Thank you ahead.

[Maarten Buis]

As soon as you use clustered standard errors, you will automatically get pseudo likelihoods. You cannot have one without the other. If you want to test nested models against one another you have to use a Wald test, i.e. use the test command.

[Chan Ge]

As soon as you use clustered standard errors, you will automatically get pseudo likelihoods. You cannot have one without the other. If you want to test nested models against one another you have to use a Wald test, i.e. use the test command.

Dear Maarten, thank you a lot for the guidance.
Based on your suggestion, I found a relevant post for anyone who may come to this post for similar questions (please click here).

Best