Re: Question about panel structure in hybrid choice model (Biogeme)

[Michel Bierlaire]

Dear Eleni,


Estimating such a complex model with a panel setting is almost hopeless. Ignore the panel structure. The consequence is that you have less precision on the estimator, and that you need to use the robust standard error. But at least, you have estimates. Moreover, 500 draws is certainly insufficient. Use halton or MHLS draws.

I assume that you are using the latest version of Biogeme, and that you are running on a real server with several cores running linux.

Michel

PS Please post your questions on the users group <bio...@googlegroups.com>.



> On 3 Nov 2025, at 16:55, Eleni Papageorgiou <s24...@student.dtu.dk> wrote:
>
> Dear Professor Bierlaire,
> I am working on a hybrid choice model using Biogeme that links self-reported emotions (valence and arousal) with physiological indicators (heart rate, EDA, etc) through latent variables. The dataset includes repeated observations for each participant, so I initially enabled the panel structure using:
> database.panel('Participant_ID')
> traj = PanelLikelihoodTrajectory(condlike)
> loglike = log(MonteCarlo(traj))
>
> However, when running the model with this specification, Biogeme did not converge even after several hours. Without the panel structure, the model runs successfully, though I understand that ignoring the panel structure is not ideal.
> I used 500 Monte Carlo draws, 1000 iterations, and a sample size of 13,848 observations. Could you please advise whether there is a recommended way to handle this type of panel hybrid model in Biogeme, or any best practices for improving convergence in this setting?
> I’d be happy to share the code and model specification if that would help diagnose the issue.
> Thank you very much for your time and help.
> Best regards,
> Eleni Papageorgiou
> MSc Student - Technical University of Denmark (DTU)


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