ASC constant

[jean luca]

hello,

I'm a beginner with discrete choice models. I'm not sure I understand why we estimate the ASC if the logit model only contains V_in and V_jn which are the deterministic parts. Since we have assumed that the error term follows a logistic distribution, so:

Pr( i  |  { i , j } ) = Pr(epsilon < V_in - V_jn ) = F(V_in - V_jn )

Here, the error term (which is modeled with the ASC) is not in the formula anymore.

So why do we estimate it?

[Michel Bierlaire]

The ASC plays the same role as an intercept in linear regression.
By including it in the V as a parameter to estimate from data, we can assume that the mean of the error term is zero.

> --
> 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 view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/cf336656-e380-4434-98e3-f3402cae4e19n%40googlegroups.com.
> <Untitled picture.png>

Michel Bierlaire
Transport and Mobility Laboratory
School of Architecture, Civil and Environmental Engineering
EPFL - Ecole Polytechnique Fédérale de Lausanne
http://transp-or.epfl.ch
http://people.epfl.ch/michel.bierlaire