Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 37
26 I N N O VAT I O N S I N T R AV E L D E M A N D M O D E L I N G , V O L U M E 2
research project on time-of-day models that included a MODEL VALIDATION
case study of a new time-of-day model (including peak
spreading) for SF-CHAMP. Plans call for this new time- Travel behavior was validated by comparing travel data
of-day model to be incorporated into the model. in a household travel survey to related travel data in the
· The approach to trip assignment included a tradi- travel demand forecasting model. For the validation of
tional aggregate assignment because there were too few the current 1998 SFCTA regional travel demand fore-
resources in the project to implement a microsimulation casting model, the trip data in the 1990 Census and the
assignment methodology. This approach has been used 1990 MTC household survey data were compared with
in all other tour-based model applications in the United the same data in the model (2).
States to date (except Transims). Nonetheless, it intro- The model components were calibrated individually
duces aggregation bias and fails to take advantage of the by using various observed data sources. This effort
disaggregate information on each traveler during route involved calibrating each model separately and then
choice. reviewing highway and transit assignment results for
· SF-CHAMP combines trip tables from the MTC each of the five periods to make additional adjustments
regional trip-based model with trips generated from the in the model components. The adjustments were all
San Francisco tour-based model. As a result, only San made to constants within the models; there were no
Francisco residents are represented by the tour-based adjustments to model coefficients. Highlights of results
model and its advantages. of the calibration are summarized below for each model
component.
These limitations were known at the outset and
accepted as lesser priorities than the core objective of · Vehicle Availability: The vehicle availability model
building a tour-based model. In some cases, these limita- was calibrated primarily on two key variables--number
tions are already undergoing change in the update of the of workers per household and superdistrict--by using
SF-CHAMP model. the 1990 Census as the primary source of observed data.
There was one additional innovative aspect of the A second validation test was used to evaluate the total
mode choice model: the inclusion of reliability and number of vehicles estimated by the vehicle availability
crowding as explicit variables in the transit utility func- model compared with Department of Motor Vehicles
tions; this aspect was tested and then not included in the estimates of auto registrations. These data were different
final models. These variables were included in a stated- by 5%. Unfortunately, the 1990 MTC survey, which was
preference telephone survey of 407 transit users in San used to estimate the model, contained different results
Francisco. Logit analysis was used to estimate trade-offs for vehicle availability than the 1990 Census. Because
between in-vehicle time, frequency of service, reliability the 1990 Census has a much larger sample size, these
(defined as the percentage of days that the vehicle data were used to calibrate the vehicle availability model.
arrives five or more minutes late), and crowding (low The results, therefore, have indirect effects on the market
plenty of seats available; medium few seats available, segmentation of automobiles and workers that was car-
but plenty of room to stand; high no seats available ried out in the mode split model.
and standing room is crowded). It was estimated that · Full-Day Pattern Tour Models: The full-day pattern
improving the percentage of vehicles arriving on time by tour models were calibrated by converting tours to trips
10% (e.g., once every 2 weeks) is equivalent to reducing and comparing these to the 1996 MTC survey expanded
the typical wait time (half the headway) by 4 min for to match the 1998 population. The 1996 MTC survey
commuters or 3 min for noncommuters. It was also esti- was used because the number of trips within San Fran-
mated that improving the level of crowding from high cisco County was very low in the 1990 MTC survey due
to low is equivalent to reducing the typical wait time by to underreporting of trips. The underreporting of trips is
5 min for commuters and 9 min for noncommuters. not consistent across time periods or across trip purposes,
Thus, relative to commuters, noncommuters are, on conditions which may have influenced model estimation
average, less sensitive to delay but more sensitive to that was based on the 1990 MTC survey. The differences
crowding. In application, the reliability and crowding between trips by period were confirmed with initial
was coded in the transit network by means of observed assignments by periods with the uncalibrated San Fran-
system data collected by SFCTA. The trade-offs esti- cisco model revealing that the off-peak time periods were
mated between these variables and wait time were significantly underestimated compared with traffic
applied in performing transit assignment and found not counts. The vast majority of underreporting of trips in
to be coincident with the observed boardings. As a the 1990 MTC survey was in other tours.
result, these variables were not used in model · Destination (Primary-and-Intermediate Stop)
application. Choice Models: The destination choice models were cal-
