[Tiaga Falcao]

Tom Scott Thank you for the link to the paper; it's an interesting and complex approach but definitely worth trying. However, according to stata's documentation 'estat gof' doesn't work after 'gsem' (proposed 'xtmlogit' alternative), so the goodness-of-fit still remains a challenge.

[Tom Scott]

Oh I see. Yes, it looks like you can use 'estat ic' to compare models but you cannot get an overall goodness of fit test. Does comparing predicted to observed values suffice? Sorry I could not be of more help but it sounds like a cross-lagged panel model using gsem might still be a good option for you.

[Tiaga Falcao]

Thanks. I am studying gsem and mlogit; so far cross-lagged panel data looks very promising.

[Tom Scott]

I'm pretty sure it's the exact same thing Jeff Wooldridge suggested in your other post, except that it's in a sem framework. I think they both use ML estimation

[Siege Taker]

This discussion has gone way from where it started, so I am trying to pick up from somewhere.
Thanks @ Tiaga, for bringing this up and thanks to Tom for the contribution. I am new to Stata and until lately didn't realise I need another code for Panel data multinomial logistic regression other than mlogit. Then there is the problem of endogeniety and unobserved hetrogeneity. Anyways, it appears gsem is the only way to go or maybe there are other options within xt- code family? xtmlogit is not recognised. I am still checking these out.

[Tiaga Falcao]

This discussion has gone way from where it started, so I am trying to pick up from somewhere.
Thanks @ Tiaga, for bringing this up and thanks to Tom for the contribution. I am new to Stata and until lately didn't realise I need another code for Panel data multinomial logistic regression other than mlogit. Then there is the problem of endogeniety and unobserved hetrogeneity. Anyways, it appears gsem is the only way to go or maybe there are other options within xt- code family? xtmlogit is not recognised. I am still checking these out.

Siege Taker If you can restructure your data in the choice models setting then cmxtmixlogit seems is the best solution for the multinomial panel data models (https://www.stata.com/manuals/cm.pdf) otherwise, according to Tom and Dr. Wooldridge, gsem is suitable when one has heterogeneity consideration and mlogit should be fine without heterogeneity concerns if panel variable was controlled.

[Siege Taker]

Thanks Tiaga Falcao, I am struggling to fit a model with gsem. I have also tried cmp but met more complications. The research and model appears to be going good with mlogit except for an endogenoues explanatory variable I need to control for. I am wondering if there any other way around controlling for endogeneity around mlogit without dropping it completely