Determining goodness of fit for a LCA using multiply imputed data

[Tristen Clifton]

Hello all,

I am using Stata 18.0 to perform a latent class analysis using gsem.
I have gathered from previous posts that postestimation commands following multiple imputation are quite limited, and because of this I cannot obtain goodness of fit statistics for the multiply imputed LCA.

I have not seen this particular problem addressed, so my question is, can I base my model selection on the goodness of fit statistics derived from a non-imputed model? I realize it is not the same model, so this may be a silly question.

Alternatively, is there another test in Stata that I can use to assess fit when using multiple imputation?

Thank you!

[Rich Goldstein]

your situation is not clear to me at all; e.g., please give examples of GOF stats;

note, however, that in general one can use -mi xeq- to estimate the model for each imputed data set and operate on those results, saving the results of your operation and then combining at the end; will that do what you want? if yes, there are examples on Statalist of doing this and a search should find them

[Tristen Clifton]

Hi Rich. I am trying to assess goodness of fit between latent class models, so I am looking to get AIC and BIC values.
I will try to find another post that uses -mi xeq- for AIC and BIC, thank you.

[Rich Goldstein]

an example for C statistic and pseudo R-squared can be found at https://www.statalist.org/forums/for...ng-mi-estimate

specific code is in #6 there

[Rich Goldstein]

note that I did not mention AIC or BIC above - I am not sure whether the code linked to above is correct for them but you could look at the following 2 cites which may discuss this:

Marshall, A, et al. (2009), "Combining estimates of interest in prognostic modeling studies after MI", BMC Medical Research Methodology, 9: 57

White, IR, Royston, P and Wood, A (2011), "MI using Chained equations", Statistics in Medicine, 30: 377-399