Modeling Strategic Route Choice Behaviour: A Cumulative Prospect Theory Approach
M Razo, S Gao, University of Massachusetts Amherst, US
This research develops a modeling framework for strategic route choice using Cumulative Prospect Theory, a non-expected-utility discrete choice model. Models are validated with synthetic data and estimated on stated preference data.
With the increasing prevalence of Advanced Traveler Information Systems (ATIS), effective analysis and prediction of traffic demand requires proper accounting for the effects of real-time information. While route choice modeling has historically relied only on a priori information about traffic conditions, the next generation of modeling must account for information that travelers receive en route to their destinations.
Adaptive choice behavior allows for path alteration in response to information received en route. Strategic choice behavior extends this by anticipating en route information and including possible detours in the assessment of alternatives. Empirical evidence of such behavior is analyzed in the authors' previous work, but a generally applicable, predictive model is necessary for further advancement.
This research develops a discrete choice modeling framework for strategic route choice behavior. Since drivers are unlikely to be uniformly strategic, the framework includes a latent-class structure to account for both strategic and non-strategic behavior. Estimation is performed using both synthetic data and stated-preference (SP) data.
The models are based on Cumulative Prospect Theory (CPT), which accounts for choice behavior which does not maximize expected utility. Under CPT, outcomes are assessed relative to a reference point, or "status quo". A value function weighs each outcome according to the difference from the reference point. In addition, a weighting function models subjects' perception of probability, specifically the overweighting of low probabilities and underweighting of high probabilities.
CPT has been applied to the context of route choice in earlier work. This research builds upon a loss-domain-only strategic route choice model, which is defined and estimated with synthetic data in Gao et al. (2009). In addition to validating this model with SP data, the choice of reference point is explored, including gain-only and mixed-prospect models.
Current estimation results show that a loss-only latent class CPT model is estimable with the SP data, with parameters well within the expected ranges. Synthetic data estimation results confirm that the model parameters can be properly identified. Robustness of the estimates will be investigated further.
The choice of reference point is a critical aspect of the CPT model. The loss-only model assumes that the reference is the lowest possible travel time. Gain-only and mixed-prospect models must be considered to accomodate other reference points. Estimability of these models with the SP data will be assessed. The limited network design used in the SP experiment may be problematic, and synthetic estimation will be used to investigate this issue in detail.
The major contribution of this research will be a mixed-prospect CPT model, with a latent-class structure to account for both strategic and non-strategic behavior. The estimability and robustness of the model will be explored in detail, as will the choice of reference point. An additional latent-class structure will be proposed to accommodate variation of reference points across subjects or observations. Where possible, the models will be validated with the SP data, which will reflect upon their realism as well as their estimability.
Association for European Transport