Seeking Advice on Improving MNL Model Fit with Low Rho-square

[Suppharutchaya Chongutsah]

Hi Professor,

I've recently run a Multinomial Logit (MNL) model using my dataset. This initial model, without any interaction terms (just the base model), resulted in a disappointing model fit, with a Rho-square of less than 0.1. Even after increasing the sample size, there has been no improvement in this model.

1. Is it typical for the base model to show such a low Rho-square (especially lower than 0.1)?

2. Could you suggest any possible reasons for this outcome?

Moreover, I would appreciate any advice on how to enhance the model's fit.

Thank you in advance for your guidance.

[Michel Bierlaire]

> On 9 May 2024, at 15:50, Suppharutchaya Chongutsah <s.chon...@gmail.com> wrote:
>
> Hi Professor,
> I've recently run a Multinomial Logit (MNL) model using my dataset. This initial model, without any interaction terms (just the base model), resulted in a disappointing model fit, with a Rho-square of less than 0.1. Even after increasing the sample size, there has been no improvement in this model.
> 1. Is it typical for the base model to show such a low Rho-square (especially lower than 0.1)?

The value of rho square or rho-bar square are irrelevant. They can only be used to compare two models.
Therefore, the value that you report is not a problem.

By the way, did you calculate the null loglikelihood?
the_biogeme.calculateNullLoglikelihood(av)

If not, the rho square is calculated based on the initial log likelihood, not the null log likelihood.

> 2. Could you suggest any possible reasons for this outcome?
> Moreover, I would appreciate any advice on how to enhance the model's fit.
> Thank you in advance for your guidance.
>
>

> --
> 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/cfc25911-ccc0-4a0f-b28a-1fbfcb70e686n%40googlegroups.com.

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