Innovations in Travel Demand Modeling, Volume 2: Papers (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

that allow some sensitivity of time- of- day choice to net- work conditions. All the models have used at least four network assignment periods: a.m. peak, midday, p.m. peak, and off peak. In some cases, free- flow conditions are assumed for the off- peak period, so no traffic assign- ment is needed for it. In some models, a fifth period has been added by splitting the off- peak period into early morning and evening–night. The more recent models, beginning with Columbus, use more precise time win- dows so as to schedule each tour and trip consistently during the day. This scheduling involves keeping track of the available time windows remaining after blocking out the time taken by each activity and associated travel. The time windows can also be used in the activity generation models. The Sacramento model and perhaps other mod- els are moving to half- hour periods to provide even more detail. The main constraint on how small the time peri- ods can be is the adequacy of the self- reported times in the diary survey data. There is evidence that people often round clock times to 10-, 15-, or 30-min intervals. TOUR TIME- OF- DAY RELATIVE TO MODE AND DESTINATION CHOICE MODELS It is not obvious whether activity and departure times should be predicted before both mode and destination choices, between them, or after both. There is some empirical evidence that shifts in time of day occur at two levels: the choice between broad periods of the day (e.g., morning, afternoon, etc.) is made fairly independently of accessibility, while smaller shifts of up to an hour or two are more sensitive to travel times and costs— the peak- spreading effect. Because all the models use broad net- work time periods, the tendency has been to model the choice of these periods for tours at a fairly high level above mode and destination choices (although in most cases the usual destination for work and school tours has already been predicted). In some models, time- of- day choice is predicted between the destination and mode choice levels, which allows the use of destination- specific mode choice log sums in the time- of- day model but requires that the destination choice model assumes (or stochastically selects) a specific time of day for the impedance variables. DEPARTURE TIME CHOICE MODELED SEPARATELY AT TRIP LEVEL Perhaps the placement of the model that predicts the choice of times for the overall tour is not as crucial if there is a separate model that predicts the departure time for each trip to the more detailed periods, conditional on the mode and origin–destination of each trip. Some model systems include such a model as the “lowest” one in the system. It is also possible to include such a model for car trips only so as to predict the shape of the demand profile within the broader peak periods. ACCESSIBILITY MEASURES IN UPPER- LEVEL MODELS Last, but certainly not least, is the issue of how to include most accurately the accessibility and land use effects in the upper- level models. Calculation of full log sums across all possible nests of lower- level alternatives is clearly infeasible with so many levels of choices. The ear- liest Portland models came the closest to including “proper” individual- specific logsums, but the structure of that model was relatively simple and the effect on model run time severe. The San Francisco models include mode- specific measures with set boundaries, such as the number of jobs accessible within 30 min by transit. The rather arbitrary cutoff boundaries in such measures can result in unexpected sensitivities when the models are applied. The New York and Columbus models use mode- specific travel- time decay functions that approxi- mate the log sum from a simple destination choice model. Such measures perform better but still have the problem that they are mode specific and that automobile and transit accessibility tend to be correlated, so it is dif- ficult to estimate model parameters for both of them. A method that solves this problem and is more consistent with discrete choice theory is to approximate joint mode–destination choice logsums. However, the mode choice log sums tend to vary widely across the popula- tion, so it is best to calculate different accessibility mea- sures for different population segments. The Sacramento models use such an approach, with aggregate accessibil- ity logsums for each combination of seven travel pur- poses, four car availability segments, and three walk- to- transit access segments— as those tend to be the most important segmentation variables in the mode choice models. REFERENCES Bowman, J., and M. Ben- Akiva. Activity- Based Disaggregate Travel Demand Model System with Activity Schedules. Transportation Research A, Vol. 35, 2001, pp. 1–28. Bradley, M., J. Bowman, and K. Lawton. A Comparison of Sample Enumeration and Stochastic Microsimulation for Application of Tour- Based and Activity- Based Travel Demand Models. Presented at European Transport Confer- ence, Cambridge, United Kingdom, 1999. Bradley, M. A., J. L. Bowman, Y. Shiftan, K. Lawton, and M. E. Ben- Akiva. A System of Activity- Based Models for Port- 19DESIGN FEATURES OF ACTIVITY- BASED MICROSIMULATION MODELS