gsem model fit indices for manuscript reporting

[Samantha Brown]

I conducted a mediation analysis using gsem with survey data because my mediators are count data and I set them as negative binomial variables. Consequently, gsem does not allow for the same model-fit commands as sem and so I am curious how people publish their findings utilizing gsem. Below is my syntax. I obtained estimates of the indirect and total effects and now just need model-fit information. I have read several threads on gsem post estimation commands without much luck and would greatly appreciate feedback.

svy linearized, subpop(domain_interest): gsem (Child_Gender -> Wave3_Ext, ) (chdAgeY -> Wave3_Ext, ) (CG_Gender -> Wave3_Ext, ) (Wave1_Ext -> Wave3_Ext, ) ///
(PPH_MCS -> DS3_NGPR, family(nbinomial constant) link(log)) (PPH_MCS -> DS3_AGPR, family(nbinomial constant) link(log)) /// (PPH_MCS -> Wave3_Ext ) ///
(DS3_NGPR -> Wave3_Ext, ) (DS3_AGPR -> Wave3_Ext, ) [=], nocapslatent

gsem, coeflegend
nlcom _b[Wave3_Ext:DS3_NGPR]*_b[DS3_NGPR:PPH_MCS]
nlcom _b[Wave3_Ext:PPH_MCS]+_b[Wave3_Ext:DS3_NGPR]*_b[DS3_NGPR:PPH_MCS]

nlcom _b[Wave3_Ext:DS3_AGPR]* _b[DS3_AGPR:PPH_MCS]
nlcom _b[Wave3_Ext:PPH_MCS]+_b[Wave3_Ext:DS3_AGPR]* _b[DS3_AGPR:PPH_MCS]

[Andrew Musau]

I am curious how people publish their findings utilizing gsem. Below is my syntax

I do not use gsem, but I would be surprised if it is not possible to export results from a standard Stata estimation command. The issue with your post is that you do not present a reproducible example. You can refer to a similar example in the manual and state exactly what it is that you want to extract.

[Samantha Brown]

That is the issue I have survey data and a subpopulation, so I am using svy in combination with having to use the gsem command because I have negative binomial mediators. I have not come across a specific example that suggests model fit indices that support both svy and gsem that would be reportable in a manuscript. I was hoping that someone has conducted a similar model that would be willing to share their process. The postestimation command examples are found on page 544 https://www.stata.com/manuals/sem.pdf. The Wald test seems to be the only test that could provide some index of model fit, but I don't have latent variables so I am uncertain what paths I would notate.

[Andrew Musau]

#7 in this thread gives a possible answer as to why sem reports goodness of fit (gof) indices and gsem does not. I guess that you can report the results without including any gof indices and in a footnote (or if anyone asks), explain why the gof indices are not reported.

[Samantha Brown]

Thank you! I appreciate it.