Multinomial Probit + Identification on Python Biogeme

[Michael]

Hello everyone,

I'm currently trying to learn how to do Probit models on Python Biogeme, specifically multinomial probit.

While the Swissmetro examples present the example for a binary probit model (linked here for your convenience), I have a few inquiries:

• How would the following CDFs need to be changed for the multinomial case? P = {1: bioNormalCdf(V1-V3), 3: bioNormalCdf(V3-V1)}
• How is identification handled by Biogeme for probit cases? I have been reading Train's chapter on probit and he discusses extensively the need for modifying the var-covar matrices of the error-term, and I wonder if us users have any control over that in Python Biogeme? Or does Biogeme have a standardized way of doing so?
  
• Finally, is there any publicly available example for MNP models on Python Biogeme that I could potentially look at and learn from? I search here and on the older Yahoo group without much luck.
  

Thank you very much in advance!

[Bierlaire Michel]

On 16 May 2018, at 21:34, Michael <mahers...@u.northwestern.edu> wrote:

Hello everyone,

I'm currently trying to learn how to do Probit models on Python Biogeme, specifically multinomial probit.

Biogeme supports only binary probit. You need special simulation tools for probit (vhttps://doi.org/10.1016/S0191-2615(98)00028-vv), or approximations of the normal https://pubsonline.informs.org/doi/abs/10.1287/trsc.23.4.253

While the Swissmetro examples present the example for a binary probit model (linked here for your convenience), I have a few inquiries:

• How would the following CDFs need to be changed for the multinomial case? P = {1: bioNormalCdf(V1-V3), 3: bioNormalCdf(V3-V1)}
• How is identification handled by Biogeme for probit cases? I have been reading Train's chapter on probit and he discusses extensively the need for modifying the var-covar matrices of the error-term, and I wonder if us users have any control over that in Python Biogeme? Or does Biogeme have a standardized way of doing so?
  

Biogeme does not perform any default normalization. It is the task of the user. For binary probit, it is the same as for binary logit. 

• Finally, is there any publicly available example for MNP models on Python Biogeme that I could potentially look at and learn from? I search here and on the older Yahoo group without much luck.
  

If you want to estimate MNP, I suggest reading this https://doi.org/10.1016/S0191-2615(98)00028-9

Stata proposes something: https://www.stata.com/manuals13/rmprobit.pdf

Thank you very much in advance!

--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To post to this group, send email to bio...@googlegroups.com.
Visit this group at https://groups.google.com/group/biogeme.
For more options, visit https://groups.google.com/d/optout.