Fractional Multinomial Logit Available?

[Jimmy Charité]

Hi everyone,

I'm looking through the documentation for MNLogit and it looks like estimating the fractional variant of the multinomial logit isn't an option (i.e. similar to fmlogit). Is that the case? If so, is there any planned support for it?

Thanks! 

[josef...@gmail.com]

It's on the wishlist but no immediate plans, unless someone else works on it.

I don't find a direct issue, related issue is

https://github.com/statsmodels/statsmodels/issues/8380

There are several issues for extending binary models, Logit, Probit to QMLE with continuous data, which is already covered by GLM.

Multivariate/multinomial extension is still open.

there are 3 possible version to cover this area

- QMLE (quasi-maximum likelihood with misspecified likelihood) for continuous, fractional data, i.e. allow endog to be continuous in simplex for MNLogit

- MLE for multinomial count data, i.e. sum of counts/trials per observation can be larger than one (as in GLM-Binomial)

- MLE for simplex data, this would be a multivariate extension of the BetaModel, similar to the QMLE but with properly specified likelihood.

However, currently there is no draft or prototype version for any of those, and I don't have it on my short term roadmap.

MNLogit could easily be expanded to allow continuous endog, but it would be easier in a subclass of it that overrides the computation that use computational shortcuts for the case that exactly one binary endog dummy is 1 and all others are zeros. Using a robust sandwich cov_type would produce correct standard errors and inference. (as in Stata)

Josef

--
You received this message because you are subscribed to the Google Groups "pystatsmodels" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pystatsmodel...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pystatsmodels/0355715b-c74b-4fdf-b868-0f7a52d3cc1bn%40googlegroups.com.

[Jimmy Charité]

Thank you!

You received this message because you are subscribed to a topic in the Google Groups "pystatsmodels" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/pystatsmodels/SU9yAlFWtrU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to pystatsmodel...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pystatsmodels/CAMMTP%2BDVMtSRCyhRfrmsDkNEwFMsw%3D%2BSpP_cpuYOVsuAYHhLig%40mail.gmail.com.