0

in my research I have been for a while estimating simple models in mlogit with binary choices and linear utility functions in R. Right now I am trying to replicate a destination choice model that uses non linear utility functions.

In my example I have a choice set for each individual of 10 and around 300 observations. All the choices are available for everyone and each of the choices have the same set of attributes. There are not individual attributes or alternative specific coefficients. Therefore, I just want to estimate a model with one utility function with just one set of coefficients for all the alternatives.

The utility function should be something like 

V=\beta_1 * att_1+\beta_2 * att_2 + \beta_3*ln(exp(\beta_4)*att_3+att_4)

My question is whether you can specify a nonlinear function like that in mlogit? If not, that specification it is even plausible? in which software?

Thanks.

jogo
  • 12,469
  • 11
  • 37
  • 42
Jaime Orreg
  • 1
  • 1

1 Answers1

0

As far as I know mlogit can only do linear-in-the-parameters utility specifications. If you want to stick with R, I would suggest looking at the Apollo package. It is very flexible and does allow for non-linear utility specifications.

EDIT: As a side note, you will still need to specify one utility function for each alternative even when the parameters are generic (not alternative specific), and normalize one utility function for identification.

edsandorf
  • 757
  • 7
  • 17