Socio-economic factors in choice models with latent variables

[Rachel Murray-Watson]

Hi there, 

In this example: https://biogeme.epfl.ch/sphinx/auto_examples/swissmetro/plot_b16panel_discrete_socio_eco.html

How does one interpret the coefficients for, e.g. CLASS_CTE or CLASS_INC?

How do they interact with the other class coefficients (class 1, class 0, etc). 

Additionally, how would one add a third latent class in here?

Best wishes, 

Rachel 

[Michel Bierlaire]

> On 13 Jan 2025, at 16:25, Rachel Murray-Watson <r.mu...@gmail.com> wrote:
>
> Hi there,
>
> In this example: https://biogeme.epfl.ch/sphinx/auto_examples/swissmetro/plot_b16panel_discrete_socio_eco.html
>
>
> How does one interpret the coefficients for, e.g. CLASS_CTE or CLASS_INC?

There is no real interpretation. It is not a behavioral model, just a class membership model.
It means that the class membership probability varies with income.

>
> How do they interact with the other class coefficients (class 1, class 0, etc).

They don't. They are parameters of the class membership model.

>
>
> Additionally, how would one add a third latent class in here?

NUMBER_OF_CLASSES = 2 --> 3

W_0 = CLASS_CTE_0 + CLASS_INC_0 * INCOME
W_1 = CLASS_CTE_1 + CLASS_INC_1 * INCOME
PROB_class0 = models.logit({0: W_0, 1: W_1, 2:0}, None, 0)
PROB_class1 = models.logit({0: W_0, 1: W_1, 2:0}, None, 1)
PROB_class2 = models.logit({0: W_0, 1: W_1, 2:0}, None, 2)

>
> Best wishes,
>
> Rachel
>
>
> --
> 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 visit https://groups.google.com/d/msgid/biogeme/2a98ac56-d43c-45c1-9d53-d8169eabec49n%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

[Michel Bierlaire]

No. But a large number of classes (that is, 3 or more) generate a likelihood function with many local optima, which is very difficult to estimate.
Also, the data need to contain sufficient information to allow discrimination among the classes.


> On 27 Jan 2025, at 13:07, Rachel Murray-Watson <r.mu...@gmail.com> wrote:
>
> Thank you so much for your responses. Is there a good rule of thumb or metric (e.g. AIC) when deciding on the number of classes?