About boxcox transformation

[Coco Wang]

I used a boxcox transformation for the time variable in my MNL model, but the lambda variable was not significant (12%), but the time variable itself was significant. Is this significance significant in this case?
Or, do I need to perform a boxcox transformation with lambda parameters for the time variable and then re estimate the model?

[Michel Bierlaire]

Being significant does not mean much. See this video to understand why: https://youtu.be/GmNodW_pUHk?si=gA4BDYlGaC1jphVL
You use those statistics to perform some hypothesis testing.
With box-cox, the hypothesis that the true value of the exponent is 0 corresponds to the logarithm specification. If you decide not to reject it, just accept it.

> On 12 May 2024, at 09:12, Coco Wang <31901...@zju.edu.cn> wrote:
>
> I used a boxcox transformation for the time variable in my MNL model, but the lambda variable was not significant (12%), but the time variable itself was significant. Is this significance significant in this case?
> Or, do I need to perform a boxcox transformation with lambda parameters for the time variable and then re estimate the model?
>

> --
> 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/783dc686-daa8-467a-9857-622ec067ab85n%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