Experimental Design / Python solution

36 views
Skip to first unread message

André Duarte

unread,
Jul 20, 2022, 3:47:45 AMJul 20
to Biogeme
Dear all, 
Dear Prof. Bierlaire, 

Hope everything is going fine. 
My name is André Duarte, I was a Transportnet fellow, worked for some time with prof. Amalia  Polydoropoulou, and attended your Discrete Choice course at EPFL 2008/2009. Probably you don´t remember me but it was a pleasure to meet you.  

After some years of not dealing with transport modeling, I have now a challenge to make a stated preferences exercise. It´s a traditional one, 3 alternatives car (time, cost), bus (time, cost, headway), and train (time, cost, headway and transfer time), with 5 choice sets per survey, with O/D and car ownership dependent availability. 
It is focused on a feasibility study for improvements to the train connection.

I have read on the forum your suggestion/preference for using a random experimental design, instead of optimal designs. I also discarded the optimal design option, because I do not have any a priori idea of coefficients and so on. 
However, due to sample limitations, probably 200 surveys (+- 800 choices), with low train choice probability,   I´m considering using a traditional fractional design (like the ones we get from SPSS). Can I have an opinion on the best approach? 

Another question is regarding the experimental design generator. I have found The Orthogonal Array package (pietereendebak.nl)  for python. Are you aware of being used for SP´s and if it´s reliable? 

I will have to deal with implementing the blocking factor in this solution, for the panel observations, but I will research it further afterward. 

Thank you in advance, 
My best regards, 
André

Bierlaire Michel

unread,
Jul 20, 2022, 3:55:08 AMJul 20
to andrecantoe...@gmail.com, Bierlaire Michel, Biogeme
Hi Andre,

On 15 Jul 2022, at 15:15, André Duarte <andrecantoe...@gmail.com> wrote:

Dear all, 
Dear Prof. Bierlaire, 

Hope everything is going fine. 
My name is André Duarte, I was a Transportnet fellow, worked for some time with prof. Amalia  Polydoropoulou, and attended your Discrete Choice course at EPFL 2008/2009. Probably you don´t remember me but it was a pleasure to meet you.  

After some years of not dealing with transport modeling, I have now a challenge to make a stated preferences exercise. It´s a traditional one, 3 alternatives car (time, cost), bus (time, cost, headway), and train (time, cost, headway and transfer time), with 5 choice sets per survey, with O/D and car ownership dependent availability. 
It is focused on a feasibility study for improvements to the train connection.

I have read on the forum your suggestion/preference for using a random experimental design, instead of optimal designs. I also discarded the optimal design option, because I do not have any a priori idea of coefficients and so on. 
However, due to sample limitations, probably 200 surveys (+- 800 choices), with low train choice probability,   I´m considering using a traditional fractional design (like the ones we get from SPSS). Can I have an opinion on the best approach? 

What is important is to check and “clean” the design afterwards. Remove questions that involve dominance or irrealistic alternatives. Simulate your design to check that you are not missing anything important.
Also, perform a pilot survey to test the design in real life.

I find this document interesting: https://b.kobe-u.ac.jp/~sanko/pub/Sanko2001_1.pdf


Another question is regarding the experimental design generator. I have found The Orthogonal Array package (pietereendebak.nl)  for python. Are you aware of being used for SP´s and if it´s reliable? 

I don’t know. 


I will have to deal with implementing the blocking factor in this solution, for the panel observations, but I will research it further afterward. 

Thank you in advance, 
My best regards, 
André

--
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/bddc7aec-19fc-49a1-898f-0a3bb39fe825n%40googlegroups.com.

André Duarte

unread,
Jul 20, 2022, 8:08:52 AMJul 20
to Biogeme
Thank you Professor, 

I will do what you suggest. I´m already simulating choicesets and will do an internal pilot during the summer.  Thanks for sharing the document, i´ll take a look. 

Best regards
André 

Reply all
Reply to author
Forward
0 new messages