Path Choice Model: Commonality Factor or Path Size used as indicators

[Marco Baldini]

Good evening!
I am currently working on a Path Choice model for active users (cycling and walking). After setting up my network, network variables and map matching the gps routes, the next step is the creation of the choice set. In this regard I have tried different methods: labeled preference alternatives with up to 30-40 alternatives per path, or BFSLE method. Depending on the generation method, I ended up facing one of these two issues:
- The observed path is SEPARATED from the choice set: it may happen that the observed path performes small detours which are hard to explain, resulting in the longest path among alternatives (for example). Hence the model would ALWAYS identify the observed as the longest.
- Another issue I faced is that the beta associated with the commonality factor CF (with C-Logit models) or with the Path Size (with PS Logit models) was estimated with the wrong sign. With C-Logit models, the CF enters with - beta_CF * CF, to apply a disutility for shared links; hence the estimated beta_CF should be positive. With PS models, the PS enters with + beta_PS * PS; hence the estimated beta_PS should be negative. What happens is that the observed path is the one with most overlap with the alternatives, meaning it always has the highest CF or PS, and the model uses it to identify the observed.

I am stuck on how to avoid this problem, and obtain an estimated model with correct signs, at least with respect to CF or PS. Any help would be GREATLY appreciated. Thank you

[Michel Bierlaire]

I would recommend sampling of alternatives
https://doi.org/10.1016/j.trb.2009.03.001
but using this sampling method:
https://doi.org/10.1016/j.trb.2012.11.002

This has been applied in this paper:
https://doi.org/10.1016/j.ijtst.2018.11.002


Regarding the path-size, you should look at the literature on the topic. I vaguely remember several papers discussing the sign issue.
You may want to look at Emma Frejinger's thesis
https://doi.org/10.5075/epfl-thesis-4009
although there are more recent works. For instance...
https://doi.org/10.3141/2538-02
https://doi.org/10.1016/j.trb.2020.02.006

> --
> 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/ac52dbcb-4bdb-4ba2-83f2-6c4376cbcd48n%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