0

I would like to estimate a mixed logit model with some restrictions applied on some of the parameters. Is there a way of doing this in the mlogit package in R?

For example, in the model described below I would like to allow the coefficient on quality to be random normal but with a mean restricted to be zero (so that mlogit effectively only estimates the standard deviation). How can this be done?

Logit <- mlogit(Choice ~ Quality + Quality:FSM + Quality:TestScore,data = dt, panel = TRUE, rpar=c(Quality = "n"), R=100, method = "bhhh")
Ivan Olszak
  • 1
  • You should be able to pass a vector indicating which parameters should be fixed to the `constPar` argument. See `?mlogit.optim`. This also requires that you supply a vector of starting values where the mean of your distribution is equal to 0. See `?mlogit`. – edsandorf Jun 17 '20 at 05:23
  • If I try the solution with constPar, I get an error message indicating that the system is computationally singular. I assume that this is because I am specifying a model with alternative-specific constants, and each alternative is associated with one level of Quality. However, if I specify that the coefficient on Quality is random with mean zero, then in principle that should be irrelevant. Why am I still getting that error message? – Ivan Olszak Jun 26 '20 at 09:04

0 Answers0