Pseudo r2 in probit

statistics

#1

Hi

One of the things I need in the code that I’m writting is to run a probit regression… I do it with GLM

I would like to obtain the Pseudo R2.

In the StatBase documentation there is a function for provinding r2

r2(obj::StatisticalModel, variant::Symbol)

With a linear model there is no problem, it does not need the variant. My problem is that I am unable to make it work in a probit.

It needs the variant. It offers three possible variants:

  • :MacFadden (a.k.a. likelihood ratio index), defined as 1 - log L/log L0.
  • :CoxSnell, defined as 1 - (L0/L)^(2/n)
  • :Nagelkerke, defined as (1 - (L0/L)^(2/n))/(1 - L0^(2/n)), with n the number

But nothing that I have tried seems to make it work.

Perhaps the question is siilly (I am rather new at this):

How do you input the “variant::Symbol” argument?

Thanks so much!


#2

Just as it suggests in the documentation, eg

r2(mymodel, :CoxSnell)

Note that in the source, the docstring says :MacFadden but the source has :McFadden, please open an issue.