interpretation of ICLV models

12 views
Skip to first unread message

kiran.t...@gmail.com

unread,
Aug 11, 2021, 2:18:41 AMAug 11
to Biogeme
Dear Prof. Bierlaire,

In the example code of ICLV models that is provided in the biogeme examples file https://biogeme.epfl.ch/examples/latent/05latentChoiceFull.py, the way latent variables are included are like this:

BETA_TIME_PT = BETA_TIME_PT_REF * exp(BETA_TIME_PT_CL * CARLOVERS)

 V0 = ASC_PT + \ 
 BETA_TIME_PT * TimePT_scaled + \ 
 BETA_WAITING_TIME * WaitingTimePT + \ 
 BETA_COST_HWH * MarginalCostPT_scaled * PurpHWH +\ 
 BETA_COST_OTHER * MarginalCostPT_scaled * PurpOther

In the documentation it is specified that the co-efficients of PT time vary with the latent variable "CARLOVERS". My question is that if the coefficient of PT time is estimated as negative, it will be interpreted that  latent attitude of CARLOVERS decreases with the increase in travel time of PT?

Secondly, why did you choose to specify using exponent: 
"exp(BETA_TIME_PT_CL * CARLOVERS) ?"
Can we add the coefficient of CARLOVERS directly in the utility equation of PT?
Like this: "Coefficient of CARLOVERS * CARLOVERS"
Is this specificiation fine?

I would appreciate a lot for your help

Kind Regards,
Kiran


Bierlaire Michel

unread,
Aug 11, 2021, 8:37:20 AMAug 11
to kiran.t...@gmail.com, Bierlaire Michel, Biogeme

On 11 Aug 2021, at 05:17, kiran.t...@gmail.com <kiran.t...@gmail.com> wrote:

Dear Prof. Bierlaire,

In the example code of ICLV models that is provided in the biogeme examples file https://biogeme.epfl.ch/examples/latent/05latentChoiceFull.py, the way latent variables are included are like this:

BETA_TIME_PT = BETA_TIME_PT_REF * exp(BETA_TIME_PT_CL * CARLOVERS)

 V0 = ASC_PT + \ 
 BETA_TIME_PT * TimePT_scaled + \ 
 BETA_WAITING_TIME * WaitingTimePT + \ 
 BETA_COST_HWH * MarginalCostPT_scaled * PurpHWH +\ 
 BETA_COST_OTHER * MarginalCostPT_scaled * PurpOther

In the documentation it is specified that the co-efficients of PT time vary with the latent variable "CARLOVERS". My question is that if the coefficient of PT time is estimated as negative, it will be interpreted that  latent attitude of CARLOVERS decreases with the increase in travel time of PT?


If BETA_TIME_PT_REF is negative, the longer the travel time, the lower the utility, as we expect. The exp specification is designed to avoid a change of sign with a change of value of CARLOVERS. 

Now, if BETA_TIME_PT_CL is negative, it means that the more the individual loves car, the less sensitive s/he is to travel time. 


Secondly, why did you choose to specify using exponent: 
"exp(BETA_TIME_PT_CL * CARLOVERS) ?"
Can we add the coefficient of CARLOVERS directly in the utility equation of PT?
Like this: "Coefficient of CARLOVERS * CARLOVERS"
Is this specificiation fine?

 The exp specification is designed to avoid a change of sign with a change of value of CARLOVERS. You want to enforce that BETA_TIME_PT < 0.





I would appreciate a lot for your help

Kind Regards,
Kiran



--
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/f65dda83-104b-4c9c-a515-5d79baa4f54bn%40googlegroups.com.

Reply all
Reply to author
Forward
0 new messages