Calibrated Labeling Method for Generating Bicyclist Route Choice Sets Incorporating Unbiased Attribute Variation

Abstract

Discrete choice model estimation requires specification of the alternatives considered for each observed choice. In route choice problems based on real-world travel observations, generally only the chosen route is observed, and the rest of the choice set remains hidden from the analyst. In dense travel networks, thousands of paths may connect a given origin or destination, necessitating methods for generating a reasonable subset of options. A new method is proposed for generating deterministic route choice sets. The technique modifies the labeled routes method, in which multiple criteria are optimized individually to generate attractive routes. The proposed method offers two potential improvements: (a) multiple routes are generated for each label by allowing a sensitivity parameter to vary and (b) a calibration step fits the alternative shortest path deviation distribution to observed behavior. The resulting process is more flexible than traditional labeled routes, yet it maintains strong links to behavior and reduces potential attribute bias. The proposed calibrated labeling method is applied to bicyclist route choice in a dense urban network. Results suggest that the proposed technique outperforms existing methods on several key criteria. In addition, explicitly linking choice set generation to observed travel patterns creates a more intuitive behavioral link than existing strategies. The proposed method should be immediately useful for route choice modeling in similar contexts. Furthermore, the basic framework could be more broadly applicable for route choice set generation.