Toll pricing: Computational tests for capturing heterogeneity of user preferences

Because of the increasing interest in implementation and exploration of a wider range of pricing alternatives, a study was done to compare approaches for capturing the heterogeneity of user preferences in forecasting the demand for tolled facilities resulting from user responses to pricing schemes. Existing tools used in practice typically deal with users’ value of time (VOT) heterogeneity in one of two ways: (a) ignore VOT and use a constant value of time and (b) recognize VOT by defining discrete user classes, each corresponding to a given VOT range, represented by its midpoint or average VOT. The first approach is fundamentally incorrect and would lead to highly biased estimates of network performance; the second is only a coarse approximation. A third approach, which recognizes that VOT is a continuous variable that is distributed probabilistically across the user population, is gaining greater attention. The following questions are examined: (a) What are the impacts of different VOT assumptions on prediction biases in forecasts of toll road usage under different pricing schemes? (b) Do the discrete treatments provide a good approximation of models with continuously distributed VOT, and if so how should VOT be assigned to user classes? (c) What are the relative computational implications of different VOT assumptions, and what is the role of efficient implementation techniques for large-scale network applications? The results suggest that recognizing the continuous nature of the VOT distribution is justified both with regard to accuracy and on computational grounds, especially in light of recent algorithmic and implementation advances.