Model averaging: revisiting our approach to decision rule heterogeneity and improving our travel behaviour models

Despite the frequent use of model averaging in many disciplines, it is not yet an idea often considered in transport behaviour modelling. The idea behind model averaging is that a single model can be created by calculating contribution weights for a set of candidate models, depending on their relative performance, thus creating an ’average’. In this paper, the authors demonstrate that this idea can be used to investigate sources of heterogeneity, to improve model fit in both estimation and forecasting and to average measures obtained from welfare analysis across different models. Firstly, whilst many applications have viewed latent class models as an ideal way to capture heterogeneity in the decision rule used by decision-makers, results from model averaging suggest that the increased flexibility in latent class models may only capture taste heterogeneity. Secondly, the authors test model averaging across latent class models, mixed logit models and models with a combination of linear and logarithmic transformations for the attributes. They find that model averaging across the candidate models results in a consistent improvement in model fit for both estimation and in forecasting with subsets of validation samples. Crucially, the authors additionally find that model averaging can improve model fit for models ran on large-scale revealed preference datasets, meaning that it could prove a particularly useful tool if typical complex choice models such as mixed logit cannot be implemented. Finally, they demonstrate that model averaging can be used to obtain welfare measures by averaging across measures obtained from the set of candidate models.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ADB40 Standing Committee on Transportation Demand Forecasting.
  • Corporate Authors:

    Transportation Research Board

    ,    
  • Authors:
    • Hancock, Thomas O
    • Hess, Stephane
    • Daly, Andrew
    • Fox, James
  • Conference:
  • Date: 2019

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References; Tables;
  • Pagination: 21p

Subject/Index Terms

Filing Info

  • Accession Number: 01697541
  • Record Type: Publication
  • Report/Paper Numbers: 19-04675
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 7 2018 9:31AM