[Hua Peng (StataCorp)]

The troubling part of this is that the first error message "����DB�ж�V���_�C�P��W�H� invalid name" indicates either the dataset has an issue (not likely) or Stata mishandled a large dataset in this situation.

[Jeff Pitblado (StataCorp)]

estat lcgof reports a likelihood-ratio statistic and information criteria based on the log-likelihood. In general (in Sata) statistics based on the fitted log-likelihood are generally not available when the svy prefix is used.

If you do not have stratification, you can fit the model without the svy prefix. Use pweights and vce(cluster ...) with your PSU variable, then estat lcgof will show the AIC and BIC values, but still not the LR test.

[Katelin Hudak]

Brilliant! Thank you everyone! I am able to estimate the gsem command using pweights and then obtain the AIC and BIC that I need. The code looks like this

gsem (pthink_par_qual_bin cthink_par_qual_bin know_datr felt_nlove talk_diff ///
resp_hpe wor_depr <-)[pw = Weight2_W5], logit lclass(C 2)

estimates store twoclass

estat lcgof

And Hua, it does seem to be the latter of your hypotheses, given that once I condensed my dataset, the gsem command ran without any issues.

Thank you again for your time and assistance!

[Richard Williams]

I always wonder if it is somehow cheating when you use pweights (and vce) instead of svy. And if it isn’t cheating, why doesn’t svy let you get these stats when they are legitimate? Would it be too hard to program svy to work correctly under these circumstances?