Joint Modeling of Constrained Path Enumeration and Path Choice Behavior: a Semi-compensatory Approach
S Kapla, Israel Institute of Technology, IL; C Giacomo Prato, Technical University of Denmark, DK
This paper proposes the development and estimation of a semi-compensatory route choice model that jointly represents constrained path enumeration as a function of traveler characteristics and path choice behavior as a function of route attributes.
Modeling route choice behavior presumes that travelers maximize their utility by choosing the best alternative from a given set of paths connecting origin and destination of their trip. Currently, models are estimated and flows are predicted by relying on choice sets generated with path generation techniques that do not guarantee the inclusion of all relevant routes that travelers would contemplate and the exclusion of all irrelevant routes that no traveler would consider.
Explicit path generation is motivated by empirical and conceptual motives. Empirically, higher accuracy in flow predictions and computational benefits are results of the choice set formation prior to model estimation and flow prediction. Conceptually, choice set formation and choice from alternatives are distinct mental processes that call for separate modeling. Choice set formation is a non-compensatory process that is constraint-related, as non-desirable attributes may lead to route elimination, and individual-related, as personal characteristics may influence route consideration. Choice from alternatives is a compensatory process that is preference-driven, as trade-offs among route attributes are considered while maximizing utility.
Several solutions are proposed in literature to the problem of generating path sets and modeling path choice behavior, but none considers jointly the non-compensatory and compensatory nature of these two mental processes. This paper proposes the development and estimation of a semi-compensatory route choice model that jointly represents constrained path enumeration as a function of travelers? characteristics and path choice behavior as a function of routes? attributes. Specifically, the model assumes a two-stage cognitive process consisting of conjunctive heuristic and utility maximization. At the first stage, travelers constrain the universal choice to a viable path set according to thresholds related to path attributes (e.g., travel time and number of turns). At the second stage, travelers choose their preferred alternative from their retained viable path set.
The semi-compensatory route choice model answers the aforementioned conceptual motives for modeling choice set formation and choice behavior. The model is constraint-based, since thresholds allow constraining the path set, individual-related, since threshold formulation depends on traveler characteristics, and preference-driven, since utility maximization governs the choice. From the mathematical perspective, the conjunctive strategy is represented with joint hazard-based duration and ordered probit models for representing the correlated thresholds of travel time and number of turns. The utility maximization stage is represented with a path size correction logit accounting for similarities across alternatives. Joint estimation of the two-stage model is performed through a maximum likelihood routine written in GAUSS matrix language.
The model is applied to revealed preferences data. Data collection focused on individuals who move regularly from home to work in an urban network and participated in a web-based questionnaire. The first part of the questionnaire elicited information regarding the respondents? socio-economic characteristics, spatial abilities and driving attitudes. The second part of the questionnaire collected information regarding routes declared by the respondents as considered routes from home to the workplace, as well as the route actually chosen. Individual thresholds are inferred from the considered choice set, while choice outcomes are the routes actually chosen.
Within the proposed framework, the universal choice set for each respondent is approximated through the simulation approach to path generation, while the considered path set includes the set of generated routes that satisfy individual thresholds. The declared path set is not directly assumed as the considered choice set, given the likely bias related to tiredness in the response and the cognitive effort required to remember all the alternatives actually considered. Constrained path enumeration is represented as a function of individual characteristics, while route choice from the retained path set is represented as a function of route attributes. The advantage of the proposed approach versus traditional path enumeration techniques lies in the ability to account for observed heterogeneity across individuals in the path set formation stage, since each traveler has its own path set deter-mined by its own set of thresholds. This constitutes an improvement with respect to deterministic path generation techniques that generate the same choice set for all travelers sharing the same origin-destination pair, and also with respect to stochastic path generation techniques that express heterogeneity as random noise rather than as a function of individual characteristics.
Results illustrate the competitiveness of the suggested path enumeration approach versus traditional non-constrained path enumeration. In addition, results suggest the applicability and feasibility of the proposed semi-compensatory approach for modeling route choice and offer a unique insight into the comprehension of the determinants of constrained path enumeration alongside the interpretation of the determinants of path choice behavior.
Association for European Transport