Tour-based Mode Choice Modeling Technique: US Practices
P Vovsha, J Freedman, Parsons Brinckerhoff, Inc, US; M Bradley, MB Research & Consulting, US
The paper provides a systematic overview of the existing tour-based mode choice models applied in San-Francisco, New York, Columbus, Atlanta, Sacramento, and Montreal and identifies directions for further improvements.
Achieving consistency between modeled modes for different trips on the same tour has been one of the major reasons for a wide acceptance of the tour-based modeling paradigm. Experience with the tour-based models developed and applied in such metropolitan regions as San-Francisco, New York, Columbus, Atlanta, Sacramento, and Montrea has shown that mode choice is one of the most complicated and least transferable models. This paper provides a systematic overview of the existing tour-based mode choice models and identifies directions for further improvements.
There are two different levels at which the mode choice decision is modeled:
- Tour mode (upper-level choice),
- Trip mode (lower-level choice conditional upon the upper-level choice).
The tour level reflects the most important decisions that a traveler makes in terms of using private auto versus public transit, non-motorized, or any other mode. Most of the models include at least two auto modes (either driver vs. passenger or by single vs. high occupancy), at least two transit modes by access (auto vs. walk), non-motorized mode, and some special modes (school bus and/or taxi). In the transit-oriented regions like New York and San-Francisco, additional stratification of transit modes is applied (rail vs. bus, etc).
Trip-level decisions provide the details of exact modes for each trip. Trip modes in each region depend on the variety of transit sub-modes (intercity rail, commuter rail, LRT, BRT, express bus, local bus, etc) and variety of auto occupancy (single, shared ride 2, shared ride 3, etc) and road pricing categories (toll vs. non-toll).
The paper discusses details of choice set formation at both tour and trip levels as well as the linkage between them implemented through the matrix correspondence rules, by using trip mode choice logsums in the tour mode utilities, etc.
The mode choice model is normally estimated and applied sequentially, taking into account stop frequency and stop location choices. The following hierarchy of sub-models is discussed:
- Tour mode choice based on the LOS variables for both outbound and return directions on a tour. Mode combinations for outbound and inbound half-tours take into account potential modal symmetry. Asymmetric modes are excluded if they are not frequently observed. For some metropolitan regions, certain asymmetric combinations are modeled if found to be significant. For example, outbound HOV passenger combined with inbound transit proved to be a non-negligible share of commuters in San-Francisco.
- Intermediate stop frequency and location choice. Additional stops on each half-tour are modeled either as part of the daily activity pattern (as in San-Francisco and Montreal) or as a separate stop frequency choice (as in the New York, Columbus, and Atlanta models), in part conditional on the chosen mode. In the Sacramento model a more flexible structure is applied where the presence of any stops during the day for a given activity purpose is modeled as part of the daily pattern, while the allocation of stops to particular tours, as well as the exact number of stops, is modeled at the half-tour level.
- Trip mode choice, conditional upon the chosen entire tour and half-tour mode as well as stop location and activity type. The trip modes are modeled sequentially and not necessarily by order of trip implementation but in some models, by trip distance (from the longest to shortest).
- Explicit parking lot / station choice for park & ride (P&R) and kiss & ride (K&R) trips. A consistent network processing of P&R and K&R trips require explicit modeling of the parking location with subsequently breaking the trip into the auto and transit legs. P&R and K&R trips require a special approach for outbound and inbound (reversed) directions with additional consistency constraints between them with respect to the chosen parking lot.
- Parking lot choice for auto trips (that might be different from the final trip destination and assumes an additional walk leg).
Mode choice decisions are closely intertwined with destination choice and time-of-day (TOD) choice. Sequencing of these choices and linkage between them are still open questions with different approaches applied in different models. The paper describes the experience with different structures where TOD choice was applied before mode choice (San Francisco, Columbus, Sacramento, Atlanta) versus alternative structures with mode choice applied before TOD choice (New York, Montreal). Possible directions for further enhancement of the integrity and simultaneity in modeling these choices are outlined.
Association for European Transport