August 2006 www. camsys. com
Bay Area/ California High- Speed Rail Ridership
and Revenue Forecasting Study
Interregional Model System Development
prepared for
Metropolitan Transportation Commission and the California High-
Speed Rail Authority
prepared by
Cambridge Systematics, Inc.
with
Mark Bradley Research and Consulting
draft
report
draft report
Bay Area/ California High- Speed
Rail Ridership and Revenue
Forecasting Study
Interregional Model System Development
prepared for
Metropolitan Transportation Commission and the California High- Speed Rail Authority
prepared by
Cambridge Systematics, Inc.
555 12th Street, Suite 1600
Oakland, CA 94607
with
Mark Bradley Research and Consulting
date
August 2006
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. i
7530.005
Table of Contents
1.0 Introduction ......................................................................................................... 1- 1
1.1 Purpose of The Report ............................................................................... 1- 1
1.2 Overall Model Design ................................................................................ 1- 1
1.3 Contents of The Report .............................................................................. 1- 4
2.0 Data for Model Estimation................................................................................ 2- 1
2.1 Travel Surveys............................................................................................. 2- 1
Air, Rail, and Auto Passenger Surveys.................................................... 2- 1
Air Passenger Surveys........................................................................... 2- 1
Rail Passenger Surveys.......................................................................... 2- 2
Auto Passenger Surveys........................................................................ 2- 2
Caltrans Household Travel Survey.......................................................... 2- 4
Urban Area Household Travel Surveys .................................................. 2- 5
2.2 Highway and Transit Networks............................................................... 2- 8
Highway Network...................................................................................... 2- 8
Air Networks............................................................................................. 2- 10
Conventional Rail Networks................................................................... 2- 11
Urban Area Transit Networks ................................................................ 2- 12
Area Type................................................................................................... 2- 14
2.3 Socioeconomic Data.................................................................................. 2- 18
3.0 Interregional Models.......................................................................................... 3- 1
Model Component Linkages..................................................................... 3- 3
Accessibility Measures .......................................................................... 3- 5
Logsum Measures .................................................................................. 3- 6
3.1 Trip Frequency Models.............................................................................. 3- 7
Model Structure .......................................................................................... 3- 7
Model Specification .................................................................................... 3- 8
Accessibility .......................................................................................... 3- 10
Regional Dummy Variables................................................................ 3- 10
Estimation Results .................................................................................... 3- 11
3.2 Party Size Models ..................................................................................... 3- 15
Model Structure ........................................................................................ 3- 15
Model Specification .................................................................................. 3- 15
Alternative- Specific Constants ........................................................... 3- 15
Table of Contents, continued
ii Cambridge Systematics, Inc.
7530.005
Household Income............................................................................... 3- 15
Household Size..................................................................................... 3- 16
Number of Household Vehicles......................................................... 3- 16
Trip Purpose ......................................................................................... 3- 16
Estimation Results .................................................................................... 3- 16
3.3 Destination Choice Models ..................................................................... 3- 17
Model Structure ........................................................................................ 3- 17
Segmentation by Length ..................................................................... 3- 18
Trip Purposes........................................................................................ 3- 18
Model Specification .................................................................................. 3- 19
Travel Impedance................................................................................. 3- 19
Distance ................................................................................................. 3- 19
Area Type.............................................................................................. 3- 21
Location/ Region .................................................................................. 3- 21
Location Interaction Variables ........................................................... 3- 21
Size Functions ....................................................................................... 3- 23
Estimation Results .................................................................................... 3- 24
3.4 Access/ Egress Mode Choice Models..................................................... 3- 27
Model Structure ........................................................................................ 3- 27
Model Specification .................................................................................. 3- 27
Estimation Results .................................................................................... 3- 28
3.5 Main Mode Choice Models ..................................................................... 3- 32
Model Structure ........................................................................................ 3- 32
Model Specification .................................................................................. 3- 33
Estimation Results .................................................................................... 3- 34
3.6 Model Application.................................................................................... 3- 36
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. iii
List of Tables
Table 2.1 Air, Rail, and Auto Passenger Surveys by Mode, Distance,
and Purpose ............................................................................................. 2- 3
Table 2.2 Caltrans Travel Surveys of Interregional Trips by Mode,
Distance, and Purpose ............................................................................ 2- 4
Table 2.3 SCAG Travel Surveys of Interregional Trips by Mode,
Distance, and Purpose ............................................................................ 2- 6
Table 2.4 MTC Travel Surveys of Interregional Trips by Mode,
Distance, and Purpose ............................................................................ 2- 6
Table 2.5 SACOG Travel Surveys of Interregional Trips by Mode,
Distance, and Purpose ............................................................................ 2- 7
Table 2.6 Total of All Survey Interregional Trips by Mode, Distance,
and Purpose ............................................................................................. 2- 7
Table 2.7 Speeds ( Miles Per Hour) by Area Type and Functional
Classification............................................................................................ 2- 9
Table 2.8 Capacities ( Per Lane Per Hour) by Area Type and Functional
Classification.......................................................................................... 2- 10
Table 2.9 California Airport Demand for In- State Travel ................................ 2- 10
Table 2.10 Socioeconomic Data Classifications.................................................... 2- 18
Table 2.11 Traffic Analysis Zones.......................................................................... 2- 19
Table 3.1 Frequency of Trip Frequency in the Combined Surveys................... 3- 8
Table 3.2 Trip Frequency Models for Long Trips.............................................. 3- 13
Table 3.3 Trip Frequency Models for Short Trips.............................................. 3- 14
Table 3.4 Party Size Estimation Dataset.............................................................. 3- 15
Table 3.5 Business/ Commute Party Size Model ............................................... 3- 17
Table 3.6 Recreation/ Other Party Size Model ................................................... 3- 17
Table 3.7 Estimation Data by Purpose, Length, and Source ............................ 3- 18
Table 3.8 Destination Choice Models for Long Trips ....................................... 3- 25
Table 3.9 Destination Choice Models for Short Trips ....................................... 3- 26
Table 3.10 Access and Egress Mode Choice Shares ............................................ 3- 28
Table 3.11 Access Mode Choice Models ............................................................... 3- 29
List of Tables, continued
iv Cambridge Systematics, Inc.
7530.005
Table 3.12 Egress Mode Choice Models................................................................ 3- 30
Table 3.13 Overall Choice Shares in SP Data ....................................................... 3- 33
Table 3.14 Main Mode Choice Models.................................................................. 3- 35
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. v
List of Figures
Figure 1.1 California Urban Areas and HSR Station Locations .......................... 1- 2
Figure 1.2 Integrated Modeling Process................................................................. 1- 4
Figure 2.1 New Statewide Model Highway Network.......................................... 2- 9
Figure 2.2 New Statewide Model Transit Network............................................ 2- 13
Figure 2.3 Transit Network in Southern California ............................................ 2- 14
Figure 2.4 Statewide Area Types........................................................................... 2- 15
Figure 2.5 Northern California Area Types ......................................................... 2- 16
Figure 2.6 Southern California Area Types.......................................................... 2- 17
Figure 3.1 Interregional Model Structure............................................................... 3- 1
Figure 3.2 Market Segments in Each Model .......................................................... 3- 4
Figure 3.3 Model Component Linkages ................................................................. 3- 5
Figure 3.4 Net Effect of Distance on Trips in Destination Choice Models ...... 3- 20
Figure 3.5 Regions for Destination Choice Models............................................. 3- 22
Figure 3.6 Access/ Egress Nested Model Structure ............................................ 3- 27
Figure 3.7 Main Mode Choice Nested Model Structure..................................... 3- 32
Figure 3.8 Model Application Structure Outline................................................. 3- 37
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 1- 1
1.0 Introduction
1.1 PURPOSE OF THE REPORT
The focus of this report is on the development of the interregional models for the
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study.
These interregional models are estimated from a combination of existing and
new household and intercept traveler surveys collected in California. There is a
full set of new interregional models, including trip frequency, party size, and
destination and mode choice models. These models are segmented by trip
purpose, distance, and location of the interregional trip households.
This report does not include validation or forecasting using these interregional
travel models; that is the subject of the next phase of the project. It also does not
include the development, validation, or forecasting of the urban travel models,
which will determine high- speed rail ridership within the urban areas of
California. The urban models are derived from existing urban models, with
enhancements to include forecasting of the high- speed rail mode. These models
will be validated along with the interregional travel models to confirm their
reliability, and will be included in the forecasting activities. The urban model
documentation will, therefore, be included in the model validation and
forecasting reports, and will be presented to the peer review at the third peer
review panel meeting, along with the validation and forecasting of the
interregional travel models.
1.2 OVERALL MODEL DESIGN
The model design for the Bay Area/ California High- Speed Rail Ridership and
Revenue Forecasting Study includes the following components:
• Urban travel;
• Interregional travel;
• External travel; and
• Trip assignment.
Urban trips include all trips with both ends in one of the three urban areas with
more than one proposed high- speed rail station. These areas are the San
Francisco Bay Area, Greater Los Angeles, and San Diego regions. Sacramento is
also considered to ensure that this capability is available for future purposes.
The metropolitan planning organizations ( MPO) representing these areas are the
Metropolitan Transportation Commission ( MTC), the San Diego Association of
Governments ( SANDAG), the Southern California Association of Governments
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
1- 2 Cambridge Systematics, Inc.
( SCAG), and the Sacramento Area Council of Governments ( SACOG). These
urban areas are presented in Figure 1.1.
Figure 1.1 California Urban Areas and HSR Station Locations
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 1- 3
Interregional trips include all trips with both ends in California and whose
origin and destination are in different urban areas ( or different counties outside
the urban areas) having proposed high- speed rail stations.
External trips include trips with one end outside California and one end in an
urban area with a proposed high- speed rail station.
We recognize that some urban trips may be longer than some interregional trips
by this definition and vice- versa. However, these definitions do clearly fit in
with urban and statewide planning definitions, and do identify most
interregional trips as those that begin or end outside an urban area. One
example of an anomaly is a trip from Modesto to San Jose ( defined as an
interregional trip), which is similar in distance to a trip from Palmdale to Los
Angeles ( defined as an urban trip). Even taking these anomalies into
consideration, there was consensus that the definition of urban and interregional
trips fit well with most trips in the system, and that the models proposed for
each would adequately address the behavioral nature of each trip type. In
addition, as discussed below, we have segmented the interregional trips into
short trips ( less than 100 miles) and long trips ( longer than 100 miles) to help
address this issue.
During the design and data collection of interregional trips through intercept
surveys at air and rail stations, we decided to focus the resources of data
collection on travel within California. As a result, there are no data on external
travel that may access the high- speed rail system in California. We will
separately estimate external travel from Mexico into California through Tijuana,
especially on the Tijuana Trolley system.
Trip assignment includes the merging of the urban, interregional, and external
trips into modal trip tables that are assigned to highway, rail, and air networks.
These assignments will be validated in the base year and forecast year to
evaluate reasonableness and accuracy compared to observed data sources. The
model base year is 2005, but we will also prepare a year 2000 model run to
compare with data sources that are from that year. Sensitivity tests will also be
performed to ensure that the models capture behavioral changes to key
parameters, such as time and cost. As mentioned above, the interregional trips
are the focus of this report, while the urban, external, and assignment model
components will be reported in the next phase of the project.
The California interregional models will explicitly model peak and off- peak
travel for both urban and interregional trip movements. Consistent with most
urban and statewide models, this model will estimate average weekday riders
for the high- speed rail system. These average weekday riders will be converted
to average annual riders using annualization factors developed from available
high- speed rail systems around the world. To the extent possible, we will use
available data by trip purpose to develop annualization factors.
The integrated modeling process for the development of the statewide model is
presented in Figure 1.1. This process shows that the accessibility of the system
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
1- 4 Cambridge Systematics, Inc.
( represented by travel time) is included in the mode choice models and in the
interregional trip frequency and destination choice models. This feature allows
us to estimate the induced travel for the interregional travel market.
Figure 1.2 Integrated Modeling Process
Trip Generation Trip Frequency
Trip Distribution
Mode Choice
Destination Choice
Mode Choice
Urban Models Interregional Models
Travel Times
Trip Assignment
Travel Times
1.3 CONTENTS OF THE REPORT
There are three sections in this report: the introduction, a discussion of data
sources, and descriptions of each model component. Data sources include travel
surveys, highway and transit networks, and socioeconomic data. Model
components include trip frequency, party size, destination choice, access and
egress mode choice, and main mode choice models.
This report builds on several other reports developed in earlier stages of this
project:
• Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Model Design, Data Collection and Performance Measures, Cambridge
Systematics, Inc., with Mark Bradley Research & Consulting and Corey,
Canapary & Galanis Research, June 2005;
• High- Speed Rail Study Survey Documentation, prepared for Cambridge
Systematics, Inc., and the Metropolitan Transportation Commission by
Corey, Canapary & Galanis Research, December 2005; and
• Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Socioeconomic Data, Transportation Supply, and Base Year Travel Patterns Data,
Cambridge Systematics, Inc., December 2005.
These reports are available from MTC or the CHSRA.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 1
2.0 Data for Model Estimation
A variety of travel survey data sources, highway and transit networks, and
socioeconomic data were used for model estimation of the interregional travel
models. These sources are summarized below. Data sources developed for use
in model validation of the urban and interregional travel models will be reported
in the next phase of the project.
2.1 TRAVEL SURVEYS
Air, Rail, and Auto Passenger Surveys
A combination of intercept surveys and household surveys was conducted to
obtain the new data needed for the study. The survey data includes revealed-preference
( RP) and stated- preference ( SP) mode choice data from air, rail, and
auto trip passengers. These surveys were coordinated and conducted by Corey,
Canapary & Galanis Research ( CC& G) of San Francisco.
In total, 3,172 surveys were conducted on this project. This includes:
• 1,234 airline passenger intercept surveys;
• 430 rail passenger intercept and telephone surveys; and
• 1,508 auto trip telephone surveys.
Air Passenger Surveys
Airline passenger surveys were conducted at six key airports throughout
California. The surveys were conducted on the following dates:
• Sacramento Airport – Conducted August 17 to 18, 2005;
• San Jose Airport – Conducted August 24 to 25, 2005.
• San Francisco Airport – Conducted September 20 to 22, 2005;
• Fresno Airport – Conducted for October 13, 2005;
• Oakland Airport – Conducted November 1, 2005 ( outside the security area);
and
• San Diego Airport – Conducted November 9, 2005 ( outside the security area).
Surveying was conducted inside the terminals at boarding gates at Sacramento
( SMF), San Jose ( SJC), San Francisco ( SFO), and Fresno ( FAT) airports. Surveying
was conducted outside the security areas at Oakland ( OAK) and San Diego
( SAN) airports. In the airports where surveying was done at the boarding gates,
teams of surveyors were assigned to specific flights that were going to targeted
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 2 Cambridge Systematics, Inc.
destination airports in California. Potential respondents at Oakland and San
Diego were approached, and asked their travel destinations. California- bound
travelers were administered the survey.
Mailback envelopes with postage paid were offered to respondents who did not
complete the questionnaire in time to give it back to surveyor at the airport.
Most surveys completed at the SMF, SJC, SFO, and FAT airports were collected
at the airport from passengers who filled them out while waiting for their planes.
Nearly all of the surveys distributed at OAK and SAN were mailed back by
respondents. This is because passenger at these two airports did not have a
significant amount of time to complete the survey outside the security area.
Rail Passenger Surveys
The rail passenger survey was conducted using two methodologies: 1) as an on-board
self- administered survey similar to the air passenger survey; and 2) as a
telephone survey conducted among qualified users of existing rail services. On-board
surveys were conducted on two commuter rail systems on the following
dates:
• Altamont Commuter Express ( ACE) Trains – Conducted October 11, 2005;
and
• Metrolink Trains – Conducted November 10, 2005.
Telephone surveys were conducted using a rider database from Amtrak that
included riders from the following services:
• Capitol Corridor;
• Pacific Surfliner; and
• San Joaquins.
Rail passenger intercept ( on- board) surveys were conducted on- board the
Altamont Commuter Express ( ACE) and Metrolink trains. Teams of surveyors
were assigned to specific routes that were traveling across targeted regions
served by this system. For example, on the Metrolink trains, routes that traveled
between the San Diego and Los Angeles region were targeted. Mailback
envelopes with postage paid were offered to respondents who did not complete
the questionnaire in time to give it back to surveyor on the train.
Auto Passenger Surveys
To capture the mode choice decisions of interregional travelers who have chosen
to use autos, a Random Digit Dial ( RDD) sample of household surveys was
conducted among residents of the study area. A stratified sampling approach
was utilized. This entailed dividing the State into the relevant regions, and
setting a targeted number of completes for households within each region.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 3
The final target quotas for the retrieval surveys were:
• A minimum of 120 responses from 9 regions = 1,080 plus;
• 120 additional responses from some combination of the six smaller areas
( Bakersfield, Tulare/ Visalia, Fresno, Merced, Modesto/ Stockton,
Sacramento); plus
• 250 additional responses from some combination of the three larger areas
( San Diego, Los Angeles, San Francisco Bay).
The final retrievals by region are as follows:
• San Diego ( 158);
• Los Angeles ( 243);
• Bakersfield ( 144);
• Tulare County/ Visalia ( 98);
• Fresno ( 149);
• Merced ( 155);
• San Francisco Bay Area ( 283);
• Modesto/ Stockton ( 145); and
• Sacramento ( 133).
The actual number of retrieval surveys conducted was a total of 1,508.
Table 2.1 presents a summary of the air, rail, and auto passenger surveys
collected for this project. These are presented by trip purpose, mode, and
distance to demonstrate the contribution to each market segment used in the
interregional travel models.
Table 2.1 Air, Rail, and Auto Passenger Surveys by Mode, Distance, and
Purpose
Drive Air Rail Bus Other Total
Long Trips
Business 138 611 27 – – 776
Commute 4 15 8 – – 27
Recreation 805 228 80 – – 1113
Other 159 82 15 – – 256
Short Trips
Business 43 14 46 – – 103
Commute 6 0 159 – – 165
Recreation 146 2 27 – – 175
Other 54 1 8 – – 63
Total 1,355 953 370 – – 2,678
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 4 Cambridge Systematics, Inc.
Caltrans Household Travel Survey
The California Statewide Travel Survey was conducted in 2000- 2001 for weekday
travel1. This survey was an activity- based survey and included all in- home
activities and travel completed in accessing activity locations over a 24- hour
period. The survey of 17,040 households was conducted in each of the
58 counties throughout the State. The survey reported 8.6 total trips per
household.
The survey was conducted by NuStats Research and Consulting, who surveyed
randomly selected households using the telephone recruitment/ diary mail-out/
telephone trip retrieval method. These data were used in this study as
disaggregate data so expansion and adjustment factors developed for the survey
were not utilized. This includes adjustment factors developed from Global
Positioning System ( GPS) surveys conducted to identify trip under- reporting and
those developed to account for changes in travel behavior due to the
September 11, 2001, attacks on the World Trade Center and Pentagon, which
severely disrupted travel throughout the U. S. The survey was conducted in
waves, with the fall 2000 and spring 2001 waves completed before 9/ 11 and the
fall 2001 wave completed before and after 9/ 11.
Table 2.2 presents a summary of the California Department of Transportation
( Caltrans) household travel surveys filtered for interregional travel. These are
presented by trip purpose, mode, and distance to demonstrate the contribution
to each market segment used in the interregional travel models.
Table 2.2 Caltrans Travel Surveys of Interregional Trips by Mode,
Distance, and Purpose
Drive Air Rail Bus Other Total
Long Trips
Business 110 9 – – – 119
Commute 181 – 1 – 4 186
Recreation 175 – – 1 3 179
Other 122 3 1 5 7 138
Short Trips
Business 271 – 2 2 – 275
Commute 854 – 9 9 7 879
Recreation 550 – – 1 3 554
Other 465 – – 14 11 490
Total 2,728 12 13 32 35 2,820
1 State of California, Department of Transportation, Division of Transportation System
Information, Office of Travel Forecasting and Analysis, Statewide Travel Analysis
Branch, 2000- 2001 California Statewide Travel Survey Weekday Travel Report, June 2003.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 5
Urban Area Household Travel Surveys
There are three urban area household travel surveys that were used to
supplement the statewide travel survey for interregional travel:
• Southern California Association of Governments ( SCAG) 2;
• Bay Area Metropolitan Transportation Commission ( MTC) 3; and
• Sacramento Area Council of Governments ( SACOG) 4.
The SANDAG survey was obtained and reviewed but did not have sufficient
geocoding of interregional travel to retain these trips for use in this study.
The SCAG survey was a large- scale regional household travel survey conducted
in six counties in Southern California. The survey was conducted using Random
Digit Dial ( RDD) methods for six sample types ( base, Caltrans, Regional
Statistical Area Augment, Weekend, Mode User Augment, and a GPS sample).
Data collection was conducted during spring 2001, fall 2001, and spring 2002.
After data quality and cleaning, a total of 16,939 households completed the
survey.
Table 2.3 presents a summary of the SCAG household travel surveys filtered for
interregional travel. These are presented by trip purpose, mode, and distance to
demonstrate the contribution to each market segment used in the interregional
travel models.
The MTC survey conducted in 2000 is called the Bay Area Travel Survey 2000 or
BATS2000. This survey was conducted by Morpace International and collected
travel information from residents of the nine- county Bay Area for weekday and
weekend travel both inside and outside the region. For the purposes of this
study, weekend travel was not included and weighting and expansion factors
were not considered because only disaggregate data were used for model
estimation. There were 15,000 households in BATS2000, with an additional
sample of 3,000 BART- using households. BATS2000 was an activity- based travel
survey that collected information on in- home and out- of- home activities over a
two- day period.
2 NuStats Research and Consulting, Year 2000 Post- Census Regional Travel Survey Final
Report of Survey Methodology, prepared for the Southern California Association of
Governments, June 30, 2003.
3 Metropolitan Transportation Commission, San Francisco Bay Area Travel Survey 2000
Regional Travel Characteristics Report Volume I, August 2004.
4 NuStats Research and Consulting, 2000 Sacramento Area Household Travel Survey Final
Report, prepared for the Sacramento Area Council of Governments, November 2000.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 6 Cambridge Systematics, Inc.
Table 2.3 SCAG Travel Surveys of Interregional Trips by Mode, Distance,
and Purpose
Drive Air Rail Bus Other Total
Long Trips
Business – – – 16 5 21
Commute 21 – – – 1 22
Recreation 42 – – – 1 43
Other 15 – – – 2 17
Short Trips
Business 39 – – – – 39
Commute 120 – – – 2 122
Recreation 53 – – – 1 54
Other 25 – – – – 25
Total 315 – – 16 12 343
Table 2.4 presents a summary of the MTC/ BATS household travel surveys
filtered for interregional travel. These are presented by trip purpose, mode, and
distance to demonstrate the contribution to each market segment used in the
interregional travel models.
Table 2.4 MTC Travel Surveys of Interregional Trips by Mode, Distance,
and Purpose
Drive Air Rail Bus Other Total
Long Trips
Business 6 – – 1 3 10
Commute 24 – – 1 15 40
Recreation 55 – – 2 18 75
Other 38 – 1 1 10 50
Short Trips
Business 22 – – 1 15 38
Commute 156 – – – 99 255
Recreation 117 – 2 2 47 168
Other 44 – 2 9 32 87
Total 462 – 5 17 239 723
The SACOG survey was conducted in six counties in California ( Sacramento,
Yolo, Yuba, Sutter, Placer, and El Dorado) from February to June 2000. A total of
3,942 households completed the survey over a 24- hour period. The survey
collected data on randomly selected households using a telephone recruit, mail-out
and telephone retrieval method of collection.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 7
Table 2.5 presents a summary of the SACOG household travel surveys filtered
for interregional travel. These are presented by trip purpose, mode, and distance
to demonstrate the contribution to each market segment used in the interregional
travel models.
Table 2.5 SACOG Travel Surveys of Interregional Trips by Mode,
Distance, and Purpose
Drive Air Rail Bus Other Total
Long Trips
Business 60 – – 1 9 70
Commute 33 – – – 54 87
Recreation 37 – – – 1 38
Other 31 – – 2 72 105
Short Trips
Business 6 – – – – 6
Commute - – – – – –
Recreation 7 – – – 1 8
Other 3 – – – 1 4
Total 177 – – 3 138 318
A full summary of the combined surveys by mode and purpose is presented in
Table 2.6. There are 7,366 trip records of interregional travel in this combined
dataset that was used ( in part or in full) to estimate the interregional travel
models described in the next section.
Table 2.6 Total of All Survey Interregional Trips by Mode, Distance, and
Purpose
Drive Air Rail Bus Other Total
Long Trips
Business 314 620 27 18 17 996
Commute 263 15 9 1 74 362
Recreation 1114 228 80 3 23 1448
Other 365 85 17 8 91 566
Short Trips
Business 381 14 48 3 15 461
Commute 1136 0 168 9 108 1421
Recreation 873 2 29 3 52 959
Short Other 591 1 10 23 44 669
Total 5,037 965 388 68 424 6,882
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 8 Cambridge Systematics, Inc.
2.2 HIGHWAY AND TRANSIT NETWORKS
Highway Network
The representation of highway network supply is primarily determined by the
level of detail in the highway network and the attributes associated with the
roadway system, such as lanes, distances, speed, and capacity. The full
description of the development of these networks will be described in a separate
report on network coding ( Task 7). The highway network was constructed by
incorporating network detail from each of the urban model networks into the
California statewide model network. A brief summary of these networks is
provided here.
Beginning with the existing statewide highway network, detail was added using
the following regional models:
• In the Metropolitan Transportation Commission ( MTC) region, the entire
highway network was incorporated into the model;
• In the Southern California Association of Governments ( SCAG) region, the
entire highway network was incorporated into the model;
• In the San Diego Association of Governors ( SANDAG) region, highway
network was incorporated only within a five- mile radius of the three
proposed high- speed rail stations;
• In the Sacramento Area Council of Governors ( SACOG) region, highway
network was incorporated only within a five- mile radius of the proposed
high- speed rail station; and
• In the Kern County region, highway network was incorporated only within a
five- mile radius of the proposed high- speed rail station.
Figure 2.1 shows the highway network in CUBE software. The new highway
network includes 4,667 zones, 127,600 links, and 206,150 nodes.
Roadway and area type classifications from the various regional models have
been consolidated. Speed and capacity definitions by functional class and area
type are different for each regional model. These values are based on local
conditions in each region and modifications made during model validation. To
take advantage of the work done in each region, values from the individual
models were kept intact instead of developing a new lookup table based on area
type and functional class. Tables 2.7 and 2.8 show the range of values by area
type and roadway classification.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 9
Figure 2.1 New Statewide Model Highway Network
Table 2.7 Speeds ( Miles Per Hour) by Area Type and Functional
Classification
Area Type
No. Functional Class Urban Suburban Rural
1 Freeway 55- 65 60- 70 60- 70
2 Expressway 40- 60 45- 60 40- 65
3 Major Arterial 30- 50 35- 60 40- 60
4 Minor Arterial 20- 50 25- 50 25- 55
5 Collectors 20- 35 25- 45 25- 55
7 Ramps 20- 45 20- 45 35- 40
8 Freeway- Freeway Connector 40- 50 50- 55 50- 55
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 10 Cambridge Systematics, Inc.
Table 2.8 Capacities ( Per Lane Per Hour) by Area Type and Functional
Classification
Area Type
No. Functional Class Urban Suburban Rural
1 Freeway 1,750- 2,100 1,750- 2,100 1,950- 2,100
2 Expressway 900- 1,800 900- 1,900 900- 1,900
3 Major Arterial 800- 1,800 800- 1,900 800- 1,900
4 Minor Arterial 700- 1,800 700- 1,800 700- 1,800
5 Collectors 550- 1,600 700- 1,600 700- 1,600
7 Ramps 500- 1,600 600- 1,600 1,250- 1,600
8 Freeway- Freeway Connector 1,700- 2,000 1,800- 2,000 1,800- 2,000
Air Networks
The State of California has 28 airports that offer commercial airline passenger
service between California cities and elsewhere. Of these, 18 airports represent
more than 99 percent of the in- state demand, so these 18 airports were selected to
represent the air network for the statewide model. Table 2.8 lists these airports
and provides estimates of their numbers of annual passenger boardings in 2000
and 2005. Since the events of September 11, 2001, air demand in California ( and
elsewhere) has declined overall, but the biggest decline was in 2002 and 2003,
and since 2003, air demand has been increasing. The dramatic increase in
demand at Long Beach airport is due to the beginning of service by Jet Blue.
Table 2.9 California Airport Demand for In- State Travel
Airport
Code City Airport Name
2000 In-state
Boardings
2005 In-state
Boardings
Percent
Change
OAK Oakland Metropolitan Oakland
International 2,357,530 2,608,620 10.7%
LAX Los Angeles Los Angeles International 2,647,460 1,724,530 - 34.9%
SMF Sacramento Sacramento International 1,573,400 1,649,350 4.8%
SAN San Diego San Diego International 1,791,980 1,548,700 - 13.6%
SJC San Jose Norman Y. Mineta San Jose
International 1,930,520 1,502,460 - 22.2%
SNA Santa Ana John Wayne Airport- Orange
County 1,253,290 1,130,960 - 9.8%
BUR Burbank Bob Hope 1,219,680 1,038,020 - 14.9%
ONT Ontario Ontario International 962,780 884,530 - 8.1%
SFO San Francisco San Francisco International 1,961,320 812,670 - 58.6%
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 11
Table 2.9 California Airport Demand for In- State Travel ( continued)
Airport
Code City Airport Name
2000 In-state
Boardings
2005 In-state
Boardings
Percent
Change
LGB Long Beach Long Beach/ Daugherty Field 260 233,250 89611.5%
PSP Palm Springs Palm Springs International 89,190 88,910 - 0.3%
ACV Arcata/ Eureka Arcata 29,200 35,790 22.6%
FAT Fresno Fresno Yosemite International 26,390 22,340 - 15.3%
SBA Santa Barbara Santa Barbara Municipal 84,950 22,150 - 73.9%
MRY Monterey Monterey Peninsula 19,380 21,270 9.8%
MOD Modesto
Modesto City County- Harry
Sham Field 6,080 3,720 - 38.8%
BFL Bakersfield Meadows Field 5,940 3,130 - 47.3%
OXR Oxnard Oxnard 6,260 2,280 - 63.6%
All Total 15,965,610 13,332,680 - 16.5%
Source: Federal Aviation Administration Ten Percent Ticket Sample
Conventional Rail Networks
Year 2000 passenger rail services consist of a variety of intraregional and
interregional services. Passenger rail services are also subdivided by mode –
metro rail ( i. e., BART), conventional rail ( both intercity and commuter services),
and light rail:
• The San Diego Region has two rail operators – San Diego Trolley ( light rail)
and the Coaster ( conventional rail).
• The SCAG region has metro, conventional, and light- rail services. The Los
Angeles Metropolitan Transportation Authority ( MTA) operates metro and
light- rail services. The Southern California Regional Rail Authority ( SCCRA)
operates Metrolink conventional commuter rail services. The MTA Rail
system is comprised of the Metro Blue, Green, Red, and Gold Lines. The
Metro Red Line subway operates between Union Station, the Mid- Wilshire
area, Hollywood, and the San Fernando Valley. The remaining light- rail lines
are the Blue Line ( Long Beach to Los Angeles), the Green Line ( Norwalk to
Redondo Beach), and the Gold Line ( Los Angeles Union Station [ LAUS] to
Pasadena).
• Within the MTC region, metro, conventional, and light- rail services are
provided. Services include BART, Caltrain, Muni Metro, and Santa Clara
VTA light- rail systems. In 2000, the BART system consisted of 39 stations
serving four East Bay lines ( Fremont, Dublin/ Pleasanton, Pittsburg/ Bay
Point, and Richmond), as well as the Daly City/ Colma line through San
Francisco and the West Bay. In 2002, BART service was extended south of
Colma to San Francisco Airport and to Millbrae, and four new stations were
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 12 Cambridge Systematics, Inc.
added. San Francisco rail and cable car routes include the five light- rail
( metro) lines that operate in the Market Street subway, three cable car routes,
and the historic trolley line operating on Market Street. Santa Clara light- rail
lines have been extended to East San Jose ( Alum Rock) and to Winchester
( Vasona line) since 2000.
• Also in the MTC region, Caltrain currently operates 86 daily trains between
San Jose and San Francisco, including three daily peak periods, peak
direction round trips to Gilroy. Trains run to San Francisco an average of
every 12 minutes during peak periods, and 30 minutes during off- peak
periods. Since the year 2000, Baby Bullet trains have been introduced,
significantly reducing San Jose to San Francisco Express train travel times.
• The SACOG region’s rail services are limited to the Sacramento RT light- rail
system. Since 2000, two RT extensions have come on- line. In 2003, the South
Line extension was implemented. This new extension resulted in RT running
two lines for the first time. More recently, the Folsom extension became
operational. The Folsom Line is an extension of the existing line that operates
along the U. S. 50 corridor.
• Interregional rail services are all conventional rail systems. These include the
Capitol Corridor, Altamont Commuter Express ( ACE), Surfliner, and San
Joaquin systems.
Urban Area Transit Networks
The Statewide model intercity routes have been updated to include urban area
transit networks from the MTC, SACOG, SCAG, SANDAG, and Kern regional
systems. In addition, local transit services serving areas around high- speed rail
stations in Stanislaus, Merced, and San Joaquin Counties were added. Figure 2.2
shows the transit network detail for the intercity routes and the regional transit
in the MTC area. Figure 2.3 shows the transit routes for Southern California.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 13
Figure 2.2 New Statewide Model Transit Network
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 14 Cambridge Systematics, Inc.
Figure 2.3 Transit Network in Southern California
Area Type
The area type used in the HSR models was based on the Caltrans Statewide
Model ( STM) socioeconomic data, processed to represent a zonal population and
employment density for each zone. The area type is defined as follows:
• Rural – Less than 1,000 persons per square mile;
• Low suburban – 1,000 to 6,000 persons per square mile;
• High suburban – 6,000 to 10,000 persons per square mile;
• Urban – 10,000 to 20,000 persons per square mile; and
• Urban Core – More than 20,000 persons per square mile.
Persons per square mile are based on either the population or employment in a
zone, whichever is higher. These area types are presented in Figure 2.4.
Additional maps are provided for northern California ( in Figure 2.5) and
southern California ( in Figure 2.6) for a better representation of the more
urbanized areas.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 15
Figure 2.4 Statewide Area Types
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 16 Cambridge Systematics, Inc.
Figure 2.5 Northern California Area Types
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 17
Figure 2.6 Southern California Area Types
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
2- 18 Cambridge Systematics, Inc.
2.3 SOCIOECONOMIC DATA
For model application, socioeconomic data are being developed by combining
urban area socioeconomic data by traffic analysis zone from the urban models
with the Caltrans statewide model socioeconomic data and the U. S. Census
Bureau data. These data have slightly different household classifications and
categories, so some processing of these data is necessary. In addition, we
developed household classification models to forecast household classifications
that were not being developed by one of the existing sources. Table 2.10
describes the list of socioeconomic data that are being developed to support the
interregional and urban models for the base and forecast years. Summary totals
for these data in the year 2000 are shown in Table 2.10.
Table 2.10 Socioeconomic Data Classifications
Category
2000
California Total
Household
Size 1 person 2,704,585
2 persons 3,385,735
3 persons 1,831,480
4+ persons 3,590,220
Income group Low (<$ 35,000) 4,249,200
Medium
($ 35,000-$ 75,000)
3,948,834
High (>$ 75,000) 3,313,986
Number of workers 0 worker 2,901,170
1 worker 4,317,905
2+ workers 4,292,945
Car ownership and worker category 0 car 1,083,945
0 < cars < workers 873,700
cars >= workers 9,554,370
Total Households 11,512,020
Employment
Type Retail 2,293,524
Service 5,760,849
Other 7,214,346
Total Employment 15,268,719
Source: 2000 Census Transportation Planning Package for California.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 2- 19
The household classification model uses joint distributions of households in the
travel demand models and Census Public Use Microdata Sample ( PUMS) data to
stratify the marginal distributions of households provided by the statewide and
urban area models. Household income categories will be converted to 2005
dollars.
The traffic analysis zones used in this modeling system are derived from the
Caltrans Statewide Model and disaggregated within select urban areas to
provide more detail around high- speed rail stations. Table 2.11 presents a
comparison of the number of zones in the original Caltrans Statewide Model and
the new statewide modeling system developed for this study for each of
14 regions.
Table 2.11 Traffic Analysis Zones
Region Region Number
Number of
Caltrans Model
Zones
Number of HSR
Model Zones
AMBAG 1 49 49
Central Coast 2 26 26
Far North 3 111 111
Fresno/ Madera 4 123 123
Kern 5 89 166
South SJ Valley 6 128 128
Merced 7 42 42
SACOG 8 173 209
SANDAG 9 94 538
San Joaquin 10 97 97
Stanislaus 11 36 36
W. Sierra Nevada 12 24 24
MTC 13 291 1,454
SCAG 14 664 1,664
Total 1,947 4,667
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 1
3.0 Interregional Models
The interregional models are comprised of four sets of models: trip frequency,
destination choice, main mode choice, and access/ egress mode choice. The
structure and contents of the interregional modeling system is presented in
Figure 3.1.
Figure 3.1 Interregional Model Structure
Trio Frequency/ Day
• Household Characteristics
• Trip Purpose/ Distance Class
• Level of Service ( Logsum & Accessibility
• Region
• Party Size ( For Short Distance)
Destination Choice
• Level of Service ( Logsum & Accessibility
• Employment & Household Characteristics
• Region and Area Type
• Trip Purpose/ Distance Class
• Party Size ( For Long Distance)
Main Mode Choice
• Level of Service
• Household Characteristics
• Purpose/ Distance Class
• Party Size ( For Long Distance)
• Access & Egress ( Logsum)
Access Mode Choice
• Level of Service
• Household Characteristics
• Purpose/ Distance Class
• Party Size ( For Long Distance)
• Main Mode ( Rail/ HSR/ Air)
Egress Mode Choice
• Level of Service
• Household Characteristics
• Purpose/ Distance Class
• Party Size ( For Long Distance)
• Main Mode ( Rail/ HSR/ Air)
One Trip Two- Plus
No Trips Trips
Zone 1 Zone 2 Zone N- 1 Zone N
Car Rail HSR Air
Drive
and Park
Drop
Off
Rental
Car
Taxi Transit Walk Taxi Transit Walk
Unpark Picked Up Rental Car
and Drive
The trip frequency model component predicts the number of interregional trips
that individuals in a household will make based on the household’s
characteristics and location. The destination choice model component predicts
the destinations of the trips generated in the trip frequency component based on
zonal characteristics and travel impedances. The mode choice components
predict the modes that the travelers would choose based on the mode service
levels and characteristics of the travelers and trips. The mode choice models
include a main mode choice, where the primary interregional mode is selected,
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 2 Cambridge Systematics, Inc.
and access/ egress components, where the modes of access and egress for the air
and rail trips are selected.
Because of the way that the model components will be linked, model
development occurs in the reverse order of model application:
• Access and egress mode choice models – The choice of mode to and from
airports, conventional rail stations, and HSR stations. The available modes
include drive and park, picked- up/ dropped off, rental car, taxi, transit and
walk. This will be based on the actual and hypothetical access and egress
modes reported in the SP surveys – either 4 or 6 observations per respondent.
( Note: We are assuming that the path building process for the main modes
will do an adequate job assigning stations and airports, but, if not, this may
need to be a joint station and mode choice model.
• Main mode choice models – The choice of main mode, from among car, air,
conventional rail, and HSR. This is based on the 4 hypothetical SP responses
for each respondent in the SP surveys. This model uses information from the
access and egress mode choice component for each mode ( except car).
• Destination choice models – The choice of destination zone outside the
region. The model is segmented for destinations within and beyond
100 miles, and the alternatives are all TAZs applicable for the distance
segments. For the long- distance model, we use a 2- stage structure of
predicting “ macro- zone” and then TAZ, because that seems to be more
behaviorally realistic. The model input data are a mix of trips from the
statewide survey and the SP survey. The models use information from the
mode choice model components, calculated for each TAZ as the key measure
of impedance between zones.
• Trip frequency models – The choice of number of interregional trips to make
during a person- day ( 0, 1, or 2) for a given purpose/ distance segment. The
Statewide survey diary- days are the data source. The models use
information from the destination choice model component calculated across
all possible TAZs as a measure of zone accessibility.
The market segmentations used for the models are:
• Purpose:
– Business ( peak period);
– Commute ( peak period);
– Recreation ( off- peak period); and
– Other ( off- peak period).
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 3
• Distance range/ residence area type:
– Less than 100 miles, from large MPO regions;
– Less than 100 miles, from small MPO regions; and
– More than 100 miles.
• Household size – 1 person, 2 people, 3 people, more than 4 people.
• Household income range – Low, medium, or high.
• Household auto- ownership – 0, 1, 2+.
• Household number of workers – 1) no workers, 2) 1 worker, 3) 2+ workers.
• Party size: Traveling alone, traveling with others.
The distance ranges of less than or greater than 100 miles were determined by
reviewing the trip length distributions from the surveys and judgment about
behavior for short versus long trips. Party size is a segmentation variable
primarily for the Recreation and Other segments, because it has a large effect on
the travel cost of the car mode versus the other modes, and thus on the choices
throughout the model chain.
These market segments vary by model component to take advantage of
additional detail in some areas or aggregation of market segments in other areas.
The market segments in each model component are presented in Figure 3.2.
Model Component Linkages
The trip frequency, destination choice and mode choice models all use
accessibility or impedance measures as inputs to the logit choice equations. For
each model component, these measures are calculated from subsequent model
components and as a result, were not available during the initial model
estimation. So, for each model component, a substitute accessibility or
impedance measure was calculated to use for initial model estimation, then
replaced with the actual measure. These linkages are presented in Figure 3.3.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 4 Cambridge Systematics, Inc.
Figure 3.2 Market Segments in Each Model
Short Trips – All Regions
Long Trips
Business
Long Trips
Commute
Long Trips
Recreation
Long Trips
Other
Short Trips – All Regions Short Trips – All Regions Short Trips – All Regions
Business/
Commute
Traveling
Alone
Business/
Commute
Traveling
with Others
Recreation/
Other
Traveling
Alone
Recreation/
Other
Traveling
with Others
Business Commute Recreation Other
Business Commute Recreation Other
Business Commute Recreation Other
Business/
Commute
Traveling
Alone
Business/
Commute
Traveling
with Others
Recreation/
Other
Traveling
Alone
Recreation/
Other
Traveling
with Others
Business/
Commute
Traveling
Alone
Business/
Commute
Traveling
with Others
Recreation/
Other
Traveling
Alone
Recreation/
Other
Traveling
with Others
Business/
Commute
Traveling
Alone
Business/
Commute
Traveling
with Others
Recreation/
Other
Traveling
Alone
Recreation/
Other
Traveling
with Others
Short Trips
MPO
Regions
Business
Short Trips
Other
Regions
Short Trips
MPO
Regions
Commute
Short Trips
Other
Regions
Short Trips
MPO
Regions
Recreation
Short Trips
Other
Regions
Short Trips
MPO
Regions
Other
Short Trips
Other
Regions
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 5
Figure 3.3 Model Component Linkages
Trip Frequency
Destination Choice
Access/ Egress
Mode Choice
Trip Assignment
Initial Estimation
Final Estimation
and Application
Main Mode Choice
Access/ Egress
Logsums
Mode Choice
Logsums from STM
Accessibility
from STM
Access/ Egress
Logsums
Mode Choice
Logsums
Dest Choice Logsums
and Accessibility
Congested Modal
Skims
Congested Modal
Skims from STM
Accessibility Measures
In the development of the trip frequency models, accessibility measures were
estimated for all trips to approximate the destination choice logsum measure. In
the final models, accessibility measures were retained for intraregional trips
because the intraregional models maintained by the MPOs do not include
destination choice models, which are necessary to produce logsum measures.
Accessibility measures for interregional trips were replaced with logsum
measures from the destination choice models in the final models, as described
below. There were four accessibility measures calculated, as follows:
• Auto peak work trip accessibility
( )⎥⎦
⎤
⎢⎣
⎡
− + = Σd
Apeak _ auto LN 1 TotalEmpd * exp 2* Timepeak _ auto / Timepeak _ mean
• Auto off- peak non- work trip accessibility
( )⎥⎦
⎤
⎢⎣
⎡
− + + + = Σd
Aoffpeak _ auto LN 1 ( Households d Re tailEmp d ServiceEmp d ) * exp 2 * Time offpeak _ auto / Time offpeak _ mean
• Non- Auto peak work trip accessibility
( )⎥⎦
⎤
⎢⎣
⎡
− + = Σd
Apeak _ nonauto LN 1 TotalEmpd * exp 2 * Timepeak _ nonauto / Timepeak _ mean
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 6 Cambridge Systematics, Inc.
• Non- Auto off- peak non- work trip accessibility
( )⎥⎦
⎤
⎢⎣
⎡
− + + + = Σd
Aoffpeak _ nonauto LN 1 ( Households d Re tailEmp d ServiceEmp d ) * exp 2 * Time offpeak _ nonauto / Time offpeak _ mean
Where:
TotalEmpd = total employment at the destination zone;
Householdsd = total households at the destination zone;
RetailEmpd = retail employment at the destination zone;
ServiceEmpd = service employment at the destination zone;
Timepeak_ auto = highway travel time during the peak ( based on congested time)
from the origin zone to the destination zone;
Timepeak_ nonauto = transit travel time during the peak ( based on congested time)
from the origin zone to the destination zone;
Timeoffpeak_ auto = highway travel time during the off- peak ( based on free- flow
travel time) from the origin zone to the destination zone;
Timeoffpeak_ nonauto = transit travel time during the off- peak ( based on free- flow
travel time) from the origin zone to the destination zone;
Timepeak_ mean = average travel time from the origin zone to all possible
destination zones during the peak period, calculated from the average of
survey respondents travel time based on peak network times; and
Timeoffeak_ mean = average travel time from the origin zone to all possible
destination zones during the off- peak period, calculated from the average of
survey respondents travel time based on off- peak network times.
Logsum Measures
Logsum measures are a means to estimate a weighted average of travel time and
cost that can be fed back from one component to another. A summary of the
logsum measures for each model component is as follows:
• Trip frequency models use “ logsum” measures from the destination choice
models, which are intended to capture the fact that it is easier to make
relevant interregional trips from some zones than from other zones. For
initial model estimation, a synthesized network zone accessibility measure
was used.
• Destination choice models use logsum measures from the main mode choice
models that are intended to provide measures of the composite impedance
across all modes of travel between each of the zones. For initial model
estimation, a mode choice logsum calculate from the Caltrans statewide
model was used.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 7
• Main mode choice models use a logsum from the access/ egress mode choice
models. This was estimated prior to the main mode choice models, so a
substitute measure was not necessary.
This allows higher level model components to reflect accessibility measured
accurately from lower level models.
3.1 TRIP FREQUENCY MODELS
Model Structure
Although we maintained the multinomial logit model structure for these models,
over the course of trip frequency model estimation several decisions were made
about details of the model structure. These model structure decisions are
described below:
• Decision Unit – After exploring using a “ household- day” and a “ person-day,”
we decided to use “ person- day” as the decision unit. The aggregation
of people to households did not provide enough non- zero interregional trip
households to outweigh the cost of losing decision units ( since there are
fewer households than people in the surveys).
• Segmentation by Length – To differentiate between the type of trip that
could be undertaken on a daily basis and one that is more likely a special
trip, we decided to model short ( less than 100 miles) and long ( 100 miles or
greater) interregional trips separately. This 100 mile cutoff was determined
based on an evaluation of the trip length frequency distributions of
interregional trips for each trip purpose.
Although we had initially tested models with separate frequency choices of zero,
one, two, and three or more interregional trips per person day, the decision to
segment the trip frequency models both by length and purpose limited the
number of choices in the choice set to zero, one, or two or more interregional trips
per person- day. The frequency of trips in the survey for each of these 8 market
segments is provided in Table 3.1.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 8 Cambridge Systematics, Inc.
Table 3.1 Frequency of Trip Frequency in the Combined Surveys
Number of Interregional Person- Day Trips
Model 0 1 2+
Business 216,509 186 107
Long Trips 108,313 57 31
Short Trips 108,196 129 76
Recreation 216,159 494 149
Long Trips 108,233 134 34
Short Trips 107,926 360 115
Commute 215,910 697 195
Long Trips 108,273 100 28
Short Trips 107,637 597 167
Other 216,321 364 117
Long Trips 108,287 95 19
Short Trips 108,034 269 98
Model Specification
We estimated 12 models for trip frequency, based on 4 trip purposes ( business,
commute, recreation, and other) and 2 distance segments ( long trips and short
trips). The model specifications for these models are described below:
• Constraining Variable Coefficients – In preliminary model specifications,
we included 0, 1, 2, and 3 or more interregional trips per day as individual
choices, with unique variable coefficients for each. Because of smaller sample
sizes in each of our market segments ( trip purpose and trip type), we
constrained the final model specification to set variable coefficients on one-trip
and two- trip choices are set to be equal. This overcame some illogical
individual variable coefficients for each market segment, but allowed us to
keep all 12 market segments and retain separate choices for interregional
travel. In addition, the alternative- specific- constants are still estimated
individually. For instance, the effect of household size on the utility of
making one interregional trip in a day is constrained to equal the effect of
household size on the utility of making two interregional trips in a day, but
the overall utility of those two choices are different because the constants are
different.
• Variables Explored and Expected Signs – The variables that we explored in
the final model specifications were restricted to the types of variables that we
can forecast in the future. The most notable restriction is that all socio-demographic
data are at the household level. While we have explored
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 9
several person- level variables, discussion here will be limited to those
variables that are “ forecastable.” It is important to note that the effect of
certain variables on interregional travel is not necessarily the same as it is for
general travel.
• Alternative- Specific Constants – Alternative- specific constants ( ASC) for
each choice are included in each model specification. These represent the
combined effect of variables that are not included in the model ( those that
cannot be captured and/ or forecasted). Small alternative- specific constants
are desirable and can signify that the variables within the model are doing a
good job predicting the outcome. However, because interregional travel is
rare for most people, it is not surprising that constants on this type of travel
would be significantly negative.
Household characteristics were developed to support a series of additional
variables in the trip frequency models, as follows:
• Household Size/ Household Size is Greater than Two – These variables can
act as a proxy for having a family. Since traveling long distances with
children can be difficult, we expect these variables’ effects on interregional
travel to be negative – especially for long trips.
• Household Workers – As the number of workers in a household increases in
a household, it is more likely that one of them will make an interregional
work- related trip. We expect a positive effect of the number of workers in a
household on interregional commute and business trips. On the other side,
having more workers in a household limits the availability of time and
flexibility for discretionary- type interregional travel ( controlling for income).
Therefore, we expect the number of workers to have a negative effect on
recreation and other type interregional trips.
• Zero- Worker Household – This dummy variable serves as a proxy for
limited available discretionary spending for interregional travel ( no workers
can mean little or no income) and for retirees, who may have limited mobility
and vehicle- driving capabilities, and for other households with limited
available discretionary spending for interregional travel. We expect a
strongly negative sign on the effect of a zero- worker household on one’s
propensity to make a commute or business trip.
• Household Vehicles – We expect the effect of the number of vehicles in a
household to have a positive effect on all types of interregional trips, because
vehicles are probably indicative of overall household mobility.
• Number of Vehicles Less than Number of Workers – We expect this
measure of vehicle unavailability to have a negative effect on all types of
interregional trips.
• Zero- Vehicle Household – The general expected sign for this variable is
negative. After accounting for the number of vehicles per households,
though, this variable is insignificant in the models.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 10 Cambridge Systematics, Inc.
Household income is included in the models based on three income categories:
low- income households are less than $ 35,000; medium- income households are
from $ 35,000 to $ 75,000; and high- income households are more than $ 75,000. The
income variables are described as follows:
• Households by Income Group – As a general rule, we expect travel to
increase as income increases.
• Missing Income Households – We have also included a dummy variable for
an un- coded income in every model that we estimated because income is not
captured in every survey record. This dummy variable is used during model
estimation, but is not included in the final model specification for model
application. As this is the case, we would like the missing income dummy
parameter to be small in all cases.
Accessibility
As discussed above, the trip frequency models include measures that capture the
accessibility of all relevant travel opportunities from travelers’ home zones. For
each residence, we calculated three peak/ work and three off- peak/ non- work
accessibility measures for destinations in 1) their home region, 2) outside their
region, within 100 miles of home, and 3) over 100 miles from home. The final
model specifications rely on synthesized accessibility measures for the within
home region destinations and on logsums calculated from the destination choice
models for the remaining accessibility measures. The synthesized accessibility
measure is necessary within the home region since the urban area models are not
destination choice models ( they are gravity models) and are therefore not able to
produce logsums for the destination choices within the region. Logsums are a
means to produce a weighted average of all potential destinations.
We calculated the accessibility to jobs, goods, and services within one’s region of
residence (“ Regional Accessibility”). If there is a high accessibility level within a
region, it is less likely that one needs to travel outside of the region. Therefore,
we expect this variable to have a negative effect on all interregional travel. These
measures try to capture the potential substitution between trips of different
lengths.
We also calculated logsum measures for areas outside the region of residence.
Because our models estimate short and long trips separately, the logsum
measures are included only for the relevant distance class. For example, if we
are estimating destination choice for long trips, then the logsum measure is
measured only for trips over 100 miles. If there are more places outside your
region to travel, then you are more likely to travel outside the region and the
coefficient on this accessibility measure is positive.
Regional Dummy Variables
We have included regional dummy variables for MTC, SANDAG, SACOG, and
SCAG regions in many of the interregional trip frequency models. We expect
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 11
that, on balance, those living within the large metropolitan areas will be less
likely to leave and make an interregional trip because there is a lot to offer within
the region. However, the geographic locations of the regions vis- à- vis one
another and the interregional connectedness of certain regions will affect the size
and direction of these parameters.
Estimation Results
The ASCs for the one long distance interregional trip per day and the two long
distance interregional trips per day choices are large and negative compared to
the zero trips per day base for both business and commute trips. In all cases, the
two trip constants are more negative than the corresponding one trip constants,
as we would expect.
The household characteristics and location variables differ among the trip-purpose-
specific model specifications, as we selected what we judged to be the
best models for each purpose from the different estimation results. Of note, for
the long distance models:
• The commute and business models have a strongly negative no workers
variable as we would expect. These models also have an increasing
probability of travel as the fraction of workers in the household increase.
• Household size variables for 1- person and 3- person households are negative
and significant for the recreation and other long trip models.
• Income has a positive effect on long distance business, commute, and other
purpose travel, as expected, but not on recreation travel.
• The SACOG region dummy variable coefficients are positive and significant
for all purposes. This may mean that there are fewer opportunities for
intraregional travel for Sacramento residents, so there is a greater tendency to
make interregional trips.
• The SANDAG, SCAG, and MTC dummy location variables are negative for a
business and commute trip, which means that residents of these regions are
less likely to make interregional work trips than other residents. This is due
to increased business opportunities in these regions.
• SANDAG and MTC residents are more likely to make interregional travel for
recreation and other purposes than other residents.
• The MTC coefficient is positive for both long distance recreation and other
trip purpose trips. This is probably due to the tourist and other attractions in
the Bay Area.
• The long distance accessibility measures for long distance trips were all
positive in the initial models, as expected, but relatively small. In the final
model estimations using logsum measures, these were constrained to provide
a more reasonable estimate than was produced in the estimation stage.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 12 Cambridge Systematics, Inc.
• The within- region accessibility measures are all strongly negative, indicating
a strong trip length substitution effect.
• The outside- region short- distance accessibility measures are significant and
positive, but very small. The only exception is the recreation logsum
measure, which is the same as the long distance coefficient.
Table 3.2 presents the estimation results of the trip frequency models for long
trips in each of the four trip purposes: business, commute, recreation, and other.
For the most part, only those variables that are significant at the 95- percent level
are retained in the models, except in the case of the accessibility variable, where
all three variables were retained in all models due to its importance from a policy
perspective. The accessibility variable allows induced travel to be estimated
from the trip frequency models, which is an important component of the overall
ridership estimates.
Table 3.3 presents the estimation results of the trip frequency models for short
trips for each of the four trip purposes: business, commute, recreation, and
other. Of note in these models:
• The ASCs for the small region models are insignificant for all purposes.
• The household size coefficients are negative and significant for recreation
and commute models, and the small region business trip purpose model also
has a household- size- greater- than- two dummy variable coefficient that is
negative and significant.
• The worker coefficients behave as expected for the business and commute
models, with workers significantly positive for business and commute trips
and significantly negative for other trips from small regions.
• The income coefficients are all of the correct sign and relative magnitudes.
• The interregional accessibility ( logsum) measure coefficients are all positive
and business and recreation coefficients are significant. This indicates that
improved accessibility for interregional travel will increase the likelihood of
making an interregional trip. The intraregional accessibility measures are all
negative, so improved accessibility within a region will diminish the
likelihood of making an interregional trip.
The overall fit of the trip frequency models is strong for business trips but low
for the other purposes, as exhibited by the log- likelihood reductions compared to
the constants- only models. While this is obviously of concern, trip frequency
model levels of fit seem to have been generally low for previous similar
modeling efforts.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 13
Table 3.2 Trip Frequency Models for Long Trips
Business Commute Recreation Other
Observations 108,401 108,401 108,401 108,401
Final log- likelihood - 1,168.3 - 1,823.7 - 2,048.8 - 1,865.3
Rho- squared ( 0) 0.99 0.99 0.98 0.98
Rho- squared( const) 0.08 0.10 0.04 0.09
Variable Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat
Level of Service
Intraregion accessibility - 0.128 - 1.5 - 0.217 - 4 - 0.4 - 6 - 0.532 - 7.4
Mode/ destination choice logsum 0.466 1.5 0.123 0.6 0.656 2.8 0.159 0.6
Household Characteristics
Medium income 0.527 1.5 0.188 0.8
High income 1.139 3 0.291 1.1 - 0.246 - 1.3 0.393 2.1
Missing income1 0.955 2.3 0.34 1.1 0.282 1.3 0.158 0.7
Fewer cars than workers in HH - 0.412 - 1 - 0.457 - 1.6 - 0.922 - 2.4 - 0.915 - 2.2
No cars in HH
Fraction of HH who are workers 0.537 1.9 1.274 5.8
No workers in HH - 2.098 - 3.4 - 2.668 - 3.7 0.372 2.4
Household size
1 person household - 0.424 - 2
3+ person household - 0.482 - 3.9 - 0.379 - 2.8
Location Variables
SACOG resident 0.976 3.7 0.918 4.7 1.084 4.4 2.527 10.3
SANDAG resident - 0.704 - 1.1 - 0.419 - 1 1.344 3.5 0.92 1.8
SCAG resident - 1.176 - 3.6 - 1.644 - 6.3 - 0.031 - 0.1 0.259 0.8
MTC resident - 1.372 - 3.6 - 0.729 - 2.9 1.011 3.4 1.134 3.4
Constants2
1 trip - 15.67 - 2.3 - 6.48 - 1.4 - 3.416 - 3.1 - 0.493 - 0.4
2+ trips - 16.3 - 2.4 - 7.914 - 1.7 - 5.083 - 4.6 - 2.823 - 2.4
1Missing income not used in model application.
2Will be modified during model calibration.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 14 Cambridge Systematics, Inc.
Table 3.3 Trip Frequency Models for Short Trips
Business Commute Recreation Other
Observations 104,667 104,667 104,754 104,754
Final log- likelihood - 1,704.1 - 5,000.7 - 3,619.6 - 2,744.8
Rho- squared ( 0) 0.985 0.957 0.969 0.976
Rho- squared( const) 0.101 0.166 0.109 0.124
Variable Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat
Level of Service
Intraregion accessibility - 0.329 - 5.3 - 0.176 - 6 - 0.438 - 8.4 - 0.536 - 9.2
Mode/ destination choice logsum 0.205 4.4 0.262 11.8 0.262 7.5 0.22 6.3
Household Characteristics
Medium income 0.331 1.2 1.045 6 0.355 2.5
High income 0.835 3.1 1.523 8.6 0.432 2.8
Missing income1 0.446 1.4 0.696 3.4 0.137 0.8
Fewer cars than workers in HH - 0.947 - 2.4 - 0.225 - 1.6
No cars in HH - 1.27 - 2.5 - 0.736 - 1.6
Fraction of HH who are workers 1.153 5 1.57 13
No workers in HH - 0.863 - 2.5 - 2.163 - 5.9 0.493 4.8
Household size - 0.136 - 3.5
1 person household - 0.401 - 2.6
3+ person household
Location Variables
SACOG resident - 0.977 - 3.3 - 2.736 - 12.4 - 1.241 - 5.6 - 1.177 - 4.4
SANDAG resident - 0.88 - 2.2 - 1.446 - 5.5 - 1.802 - 3.9 - 0.66 - 1.7
SCAG resident - 1.969 - 8.6 - 1.524 - 10.9 - 1.16 - 5.3 - 1.265 - 4.8
MTC resident - 1.275 - 5.3 - 1.982 - 17.1 - 0.25 - 1.3 - 0.524 - 2.3
Constants2
1 trip - 4.946 - 6.7 - 8.242 - 15.2 - 2.881 - 4.3 - 0.845 - 1.4
2+ trips - 5.513 - 7.5 - 9.07 - 16.7 - 3.787 - 5.7 - 1.624 - 2.6
1Missing income not used in model application.
2Will be modified during model calibration.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 15
3.2 PARTY SIZE MODELS
Model Structure
The party size model was estimated as a simple binomial choice model between
traveling alone and traveling in a group. Two separate models were estimated,
one for business and commute trips and one for recreation and other trips. The
estimation dataset was the stated- preference ( SP) survey, which had
considerably more complete data on party size compared to the household travel
surveys. Table 3.4 shows the party size characteristics of the estimation dataset.
The overwhelming tendency for recreation/ other interregional trips to be with
another person compared to business/ commute demonstrates the need to model
the party size of these trip purpose categories separately.
Table 3.4 Party Size Estimation Dataset
Business/ Commute Recreation/ Other
Traveled alone 576 372
Traveled in a group 236 1,012
Model Specification
A variety of combinations of household variables available for model estimation
were tested in the party size models. In the end, we kept the model specification
with fewer variables because these were the most intuitive and did not sacrifice
the overall fit of the model.
Alternative- Specific Constants
Traveling alone was the base alternative in the model estimation for both models.
Alternative- specific constants should reflect any effect that is not captured within
the explanatory variables. The alternative- specific constants in the business/
commute model are negative, reflecting the general tendency to travel alone for
these trip purposes. The positive constant in the recreation/ other model reflects
a tendency to travel with a companion that is not captured in the explanatory
variables.
Household Income
Income was tested and is insignificant in all cases for the recreation/ other party
size model. However, high income had a positive effect on traveling alone in the
business/ commute model relative to the lower income classes. This makes
sense, as most people who carpool do so to save money.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 16 Cambridge Systematics, Inc.
Household Size
Having a one- person household has a very negative effect on making any type of
interregional trip with another person. If there is nobody in the household to
travel with, then it makes sense that a traveler would generally go alone.
Household size has a positive effect on traveling in a group for the recreation/
other purpose but no effect ( other than the negative one- person household effect)
on business and commute trip party sizes. It makes sense that recreation trips
are more of a family/ household event and therefore the size of your family/
household would have an effect on your travel party size for this trip purpose,
but not necessarily for business/ commute trips, where party- size may be
determined primarily by your workplace characteristics.
Number of Household Vehicles
If one’s use of a vehicle is constrained, then they will likely take a “ group” mode
of transportation including HOV. Therefore it is no surprise that having “ no
vehicles” makes an individual more likely to travel in a larger party – they have
less individual travel choices and relatively more group travel choices.
Trip Purpose
A trip purpose dummy was included in each model. In the business/ commute
model it indicated that people are more likely to travel in groups for business
trips relative to commute trips. If one has to make a long commute trip it may be
unlikely that they find a carpool buddy that lives in their area. It is also unlikely
that their spouse works in the same area. This makes business trips relatively
more likely to be group travel than commute trips. In the recreation/ other
model, the recreation dummy variable indicates that recreation trips are more
likely to encourage group travel than other trips. This sign makes sense because
recreation activities are often undertaken in a group, so they would likely travel
to the recreation- destination as a group as well.
Estimation Results
Table 3.5 presents the business/ commute party size model estimation results.
The variables in this model are described in the previous section. All variables
are significant at the 95- percent level, except the commute trip purpose variable,
which was kept in the model because it is intuitive and useful to separate
business from commute trips in this manner.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 17
Table 3.5 Business/ Commute Party Size Model
Variables ( For Party Size GT 1) Coefficients T- Stat
A. S. C. TWO - 0.3217 - 2.620
High Income - 0.7337 - 4.490
One Person HH - 1.2219 - 4.310
No Vehicles in HH - 1.097 - 2.150
Commute Trip - 0.6765 - 1.450
Model Statistics
Log- Likelihood Constants Only - 489
Log- Likelihood Model - 470
R- Squared ( with respect to constants) 0.0404
Table 3.6 presents the recreational/ other party size model results. The variables
in this model are also described in the previous section. All variables are
significantly different than zero. As expected, the one- person household variable
is the most significant variable.
Table 3.6 Recreation/ Other Party Size Model
Variables ( For Party Size GT 1) Coeff. T- stat
A. S. C. TWO 0.6804 3.060
Household Size 0.1402 2.340
One Person HH - 1.5459 - 7.850
Recreation Trip 0.3016 1.900
Model Statistics
Log- Likelihood Constants Only - 806
Log- Likelihood Model - 735
R- Squared ( with respect to constants) 0.1025
3.3 DESTINATION CHOICE MODELS
Model Structure
The destination choice models were estimated with a simple multinomial logit
model structure using ALOGIT software. The dataset used for the trip frequency
models ( comprised of interregional trips from the California Statewide survey,
the SCAG survey, the SACOG survey and the MTC/ BATS survey) was
combined with the stated- preference ( SP) survey ( used in the mode choice
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 18 Cambridge Systematics, Inc.
models) to produce a combined estimation dataset for the destination choice
estimation models. The addition of the SP dataset significantly increased the
number of “ long” ( more than 100 miles) trips in the dataset ( by nature, the
household surveys are generally better at capturing the more typical “ short”
trips). Table 3.7 shows the distribution of trips in the estimation data across trip
purposes, length, and survey. This table demonstrates the number of samples in
each market segment available for model estimation.
Table 3.7 Estimation Data by Purpose, Length, and Source
Caltrans BATS SACOG SCAG SP Total
Long Trips
Business 119 10 70 21 780 1,000
Commute 186 40 87 22 32 367
Recreation 179 75 38 43 1,125 1,460
Other 138 50 105 17 259 569
Short Trips
Business 275 38 6 39 268 626
Commute 879 255 - 122 198 1,454
Recreation 554 168 8 54 384 1,168
Other 490 87 4 25 116 722
Total 2,820 723 318 343 3,162 7,366
Segmentation by Length
We modeled interregional destination choice separately for “ short” ( less than
100 miles) and “ long” ( 100 miles or greater) trips. Since the trip frequency
models already differentiate between the two, we can use this information as a
valuable input to the destination choice models. This not only constrains an
individual’s choice set based on destinations being greater or less than 100 miles,
but it recognizes that an individual may value different trip characteristics for
different distance- categories of travel.
Trip Purposes
The short trip destination choice models used all four trip purposes modeled in
the trip frequency step: business, commute, recreation, and other. Due to
sample size considerations, only two aggregate trip purposes were estimated for
the long trip destination choice models: business/ commute and
recreation/ other.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 19
Model Specification
Travel Impedance
The models presented here use multimodal composite impedance from the
statewide model ( mode choice model logsum) broken up into four different
categories: home- based- work, home- based- recreation, home- based- other, and
work- based- other. We have included an appropriate impedance variable in
every specification and expect there to be a positive relationship between the
impedance that this variable represents, and destination choice.
This variable measures the combined utility of all available modal choices and
level of service characteristics. The coefficient turned out to be positive and
significant at the 95 percent confidence level in the destination choice models,
indicating that the destination zone is more attractive if it is better accessible.
Distance
In all of the destination choice models presented, we have used a distance power
series including distance, distance- squared, and distance- cubed. While common
sense would say that all distance coefficients should be negative, one cannot
analyze the distance coefficients individually, but as their collective impact.
Graphs illustrate the collective impact of all three distance coefficients on one’s
destination choice in Figure 3.4. Further caution should be used in interpretation
because a great deal of the impedance between origin- destination pairs is
captured within the travel impedance term and coefficient. Therefore it is not
wrong for the collective effect of distance to be either positive or negative. It
should be noted that since we are estimating separate models for “ short” and
“ long” trips, that the “ short” trips are automatically capped at 100 miles from the
origin. All short trip distance functions show a decreasing function up to
100 miles, which is consistent with our expectations. One example for short
recreation trips is shown in Figure 3.4. Both long trip distance functions show a
decreasing function from 100 miles to about 250 miles and then an increasing
function for trips greater than 250 miles.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 20 Cambridge Systematics, Inc.
Figure 3.4 Net Effect of Distance on Trips in Destination Choice Models
Long Business Trips
- 3.0000
- 2.5000
- 2.0000
- 1.5000
- 1.0000
- 0.5000
0.0000
100 200 300 400 500 600 700
Distance
Net Effect
Long Recreation/ Other Trips
- 4.0000
- 3.5000
- 3.0000
- 2.5000
- 2.0000
- 1.5000
- 1.0000
- 0.5000
0.0000
100 200 300 400 500 600 700
Distance
Net Effect
Short Business Trips
- 5.0000
- 4.0000
- 3.0000
- 2.0000
- 1.0000
0.0000
0
20
40
60
80
100
120
Distance
Net Effect
Short Commute Trips
- 8.0000
- 6.0000
- 4.0000
- 2.0000
0.0000
0 20 40 60 80 100 120
Distance
Net Effect
Short Recreation Trips
- 7.0000
- 6.0000
- 5.0000
- 4.0000
- 3.0000
- 2.0000
- 1.0000
0.0000
0 20 40 60 80 100 120
Distance
Net Effect
Short Other Trips
- 6.0000
- 5.0000
- 4.0000
- 3.0000
- 2.0000
- 1.0000
0.0000
0 20 40 60 80 100 120
Distance
Net Effect
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 21
Area Type
Each possible destination zone could have one of three basic area- types assigned
to it: Rural, Suburban, or Urban ( as defined in the California Statewide Model). In
the destination choice models we chose “ Suburban” to be the base. Additionally,
we created several interaction terms to capture whether travelers were starting
and ending in the same area type: Rural to Rural, Suburban to Suburban, Urban to
Urban. We expect that the sign on Urban to Urban to be positive, and the sign on
Rural to Rural and Suburban to Suburban to be negative or close to zero.
Location/ Region
A total of 25 variables indicating the general geographic region were assigned to
each zone. They were: AMBAG, Central Coast, Far North, Fresno/ Madera, Kern,
South SJ Valley, Merced, SACOG, SANDAG, San Joaquin, Stanislaus, W. Sierra
Nevada, Alameda, Contra Costa, Napa, Sonoma, Marin, San Francisco, San Mateo,
Santa Clara, Solano, Imperial, Los Angeles, Orange, Riverside, San Bernardino, and
Ventura. We chose two of these variables as base variables for the model:
SACOG in the north and Orange in the south. Since they are similar, Marin,
Napa, and Sonoma were combined together into one variable to simplify the
estimation process. In many of the models, there were no records to Imperial
County or San Bernardino County, so these two variables were also dropped
from estimation. Figure 3.6 shows these destination regions used in the location
type variable.
Location Interaction Variables
Similar to the area type interaction variables, the location type interaction
variables allow us to relate where you want to go, to where you currently are.
We tested four origin- destination location type interaction variables for all the
“ long” destination choice models: LA to/ from San Francisco, Sacramento to/ from
San Francisco, San Francisco to/ from San Diego, and Sacramento to/ from LA. Due to
the distances between many of these locations, we were only able to test
Sacramento to/ from San Francisco in the “ short” destination choice models. The
recreation/ other “ long” model did not have any records of people traveling to/
from Sacramento and LA so this variable was dropped from that specification.
Since all of these locations are major population centers and destinations in the
state we generally expect them to have a synergistic quality between them that
these variables represent, and thus have positive coefficients ( although it makes
sense that this may not occur for all trip purposes).
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 22 Cambridge Systematics, Inc.
Figure 3.5 Regions for Destination Choice Models
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 23
Size Functions
Size functions are used measure the amount of activity that occurs at each
destination zone and incorporate this into the utility of alternative variables.
This type of variable is frequently used in destination choice models to account
for differences in zone sizes and employment levels.
Four size variables are used in these models: retail employment, service
employment, other employment, and households. Other employment is used as
the base size variable for business and commute trips and is constrained to 1.0
while retail and service are further segmented by household income levels – low,
medium, high, and missing. Households are used as the base size variable for
recreation and other trips.
Income is used as a per person variable as an interaction between employment
and income to show that different income levels of the destination choices will
affect the attractiveness of the zone for particular travelers. For commute trips,
short and long, as income increases, retail employment has a bigger impact on
destination choice than service employment.
For example:
U( 1) = p70* dist( 1) + p80*( dist( 1)* dist( 1)) + p90*( dist( 1)* dist( 1)* dist( 1)) +
p91* log( dist( 1) + 0.001) + p23* wolsum( 1) + p35* urb( 1) + p37* rur( 1)
.... rest of utility functions
size( 1) = oth( 1) + p100* retlinc( 1)+ p101* retminc( 1)+ p102* rethinc( 1) +
p103* serlinc( 1)+ p104* serminc( 1)+ p105* serhinc( 1)
.... rest of size functions
This translates into the following utility for first alternative:
V( 1) = p70* dist( 1) + p80*( dist( 1)* dist( 1))+ p90*( dist( 1)* dist( 1)* dist( 1))+
p91* log( dist( 1)+ 0.001) + p23* wolsum( 1) + p35* urb( 1)+ p37* rur( 1) + L_ S_ M * log
{ oth( 1) + exp( p100)* retlinc( 1) + exp( p101)* retminc( 1) + exp( p102)* rethinc( 1) +
exp( p103)* serlinc( 1) + exp( p104)* serminc( 1) + exp( p105)* serhinc( 1)}
Where:
dist = Distance from a congested time path ( miles)
wolsum = Work- based other logsum
urb = Urban area type
rur = Rural area type
oth = Other Employment
retlinc = Retail Employment * Low Income
retminc = Retail Employment * Medium Income
rethinc = Retail Employment * High Income
serlinc = Service Employment * Low Income
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 24 Cambridge Systematics, Inc.
serminc = Service Employment * Medium Income
serhinc = Service Employment * High Income
L_ S_ M = Log Size Multiplier ( constrained to 1 by default).
Estimation Results
Table 3.8 presents the model estimation results of the destination choice models
for long trips by trip purpose: business/ commute, and recreation/ other. The
distance power series of coefficients for these models are both decreasing
functions as expected. All other variables have the sign and size we expect.
Many of the location and origin- destination interaction variables were dropped
in the final estimation because the coefficients were insignificant.
Table 3.9 presents the model estimation results of the destination choice models
for short trips by trip purpose: business, commute, recreation, and other. The
distance power series of coefficients for these models are all insignificant with
the inclusion of the mode choice logsum measure, but they are retained for
completeness in the final models. These show an increasing function for
commute and recreation trips, but a decreasing function for business and other
trips, changing to an increasing function above 50 miles. All of the origin-destination
interaction variables and some of the location variables were
dropped in the final model estimation because the coefficients were insignificant.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 25
Table 3.8 Destination Choice Models for Long Trips
Business Other
Observations 1,342 1,922
Initial log- likelihood - 12,102.6 - 17,029.0
Final log- likelihood - 11,475.4 - 16,219.3
Rho- squared 0.052 0.048
Coef T- stat Coef T- stat
Level of Service
Mode choice logsum1 0.107 5.1 0.103 6.7
Mode choice logsum2 0.107 constrained 0.103 constrained
Distance ( miles) - 0.024 - 8.5 - 0.031 - 11.7
Distance squared/ 100 0.0070 8.9 0.0087 10.8
Distance cubed/ 10000 - 0.0005 - 8.0 - 0.0007 - 9.5
Area type
Urban destination 0.724 6.7 0.810 9.5
Rural destination 0.222 2.0 0.607 6.8
Urban to urban - 0.010 - 0.1 - 0.096 - 0.8
Suburban to suburban - 0.185 - 1.5 - 0.029 - 0.3
Rural to rural - 0.112 - 0.7 - 0.036 - 0.3
Destination District
AMBAG 0.154 0.8 - 0.347 - 2.1
Central Coast - 1.357 - 3.9 - 1.316 - 5.1
Far North 0.190 1.0 - 0.295 - 2.0
Fresno 1.379 9.2 1.012 8.3
Kern 1.028 5.9 0.612 4.3
Merced 1.416 8.0 0.790 5.2
S. San Joaquin 0.882 3.2 0.408 1.7
SANDAG - 0.001 0.0 0.080 0.6
San Joaquin - 0.280 - 1.1 0.360 2.3
Stanislaus - 1.264 - 3.0 - 0.256 - 1.2
W. Sierra Nevada 1.114 4.4 - 0.406 - 1.4
Alameda - 1.277 - 6.1 - 0.983 - 6.0
Contra Costa - 0.276 - 1.4 - 0.415 - 2.5
Marin/ Sonoma/ Napa - 0.354 - 1.8 - 0.522 - 3.0
San Francisco - 1.350 - 6.2 - 1.433 - 7.2
San Mateo - 1.190 - 4.6 - 1.263 - 5.5
Santa Clara - 1.213 - 6.1 - 0.912 - 5.7
Solano 0.298 1.4 - 0.671 - 2.8
Los Angeles - 1.135 - 6.5 - 1.125 - 8.4
Orange - 1.624 - 7.2 - 2.433 - 10.4
Riverside - 2.606 - 5.5 - 2.001 - 7.8
San Bernardino - 2.020 - 6.0 - 1.898 - 8.1
Ventura - 1.191 - 3.4 - 1.638 - 5.3
Regional Interactions
MTC to SCAG 0.651 4.0 0.607 4.8
MTC to SANDAG 0.321 1.6 0.107 0.7
SACOG to SCAG 0.068 0.2 - 0.515 - 1.8
SACOG to SANDAG - 0.454 - 1.1 0.390 1.5
SCAG to MTC 0.256 1.6 0.153 1.0
SCAG to SACOG - 0.538 - 1.3 0.089 0.3
SANDAG to MTC 0.364 1.9 0.200 1.2
SANDAG to SACOG 0.208 0.7 - 0.285 - 1.0
Size variables ( exponentiated)
Other employment 1.000 constrained
Households 1.000 constrained
Retail employment- low income 2.889 2.1 0.960 - 0.1
Service employment - low income 1.728 1.5 0.287 - 3.6
Retail employment - med income 9.318 4.9 0.850 - 0.4
Service employment - med income 2.292 1.8 0.373 - 3.3
Retail employment - high income 7.338 5.6 1.385 0.8
Service employment - high income 2.525 2.8 0.393 - 2.4
Retail employment - missing income3 100.000 0.1 0.001 - 0.1
Service employment - missing income3 100.000 0.1 0.433 - 1.4
1Estimated without distance terms.
2Constrained in final model.
3Not used in application.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 26 Cambridge Systematics, Inc.
Table 3.9 Destination Choice Models for Short Trips
Observations Business Commute Recreation Other
Observations 397 1,153 865 556
Initial log- likelihood - 2,718.8 - 8,133.9 - 5,933.8 - 3,756.4
Final log- likelihood - 2,452.5 - 7,199.4 - 5,105.5 - 3,082.2
Rho- squared 0.098 0.115 0.140 0.179
Coef T- stat Coef T- stat Coef T- stat Coef T- stat
Level of Service
Mode choice logsum1 0.751 10.5 0.664 10.3 2.081 23.0 2.739 24.1
Mode choice logsum2 0.751 constrained 0.664 constrained 1.000 constrained 1.000 constrained
Distance ( miles) - 0.130 - 3.7 - 0.130 - 6.1 - 0.167 - 7.9 - 0.104 - 4.0
Distance squared/ 100 0.155 2.3 0.116 2.7 0.139 3.3 0.061 1.1
Distance cubed/ 10,000 - 0.067 - 1.7 - 0.045 - 1.8 - 0.030 - 1.2 - 0.011 - 0.3
Area type
Urban destination 0.760 3.8 0.872 7.4 0.502 3.8 0.419 2.3
Rural destination 0.036 0.2 0.126 1.1 0.081 0.6 0.190 1.1
Urban to urban - 0.499 - 1.6 - 0.019 - 0.1 - 0.142 - 0.7 0.457 1.9
Suburban to suburban 0.253 1.1 - 0.055 - 0.4 0.051 0.3 - 0.016 - 0.1
Rural to rural - 0.505 - 1.8 - 0.075 - 0.5 0.336 1.9 0.245 1.0
Destination District
AMBAG 0.878 3.3 0.425 2.3 0.396 2.2 0.617 2.7
Central Coast - 2.214 - 2.2 - 2.460 - 4.6 - 1.190 - 3.0 - 1.010 - 2.1
Far North 0.678 2.4 0.170 0.8 0.349 1.9 0.961 4.7
Fresno - 0.300 - 1.0 0.297 1.8 - 0.132 - 0.8 0.283 1.4
Kern 0.114 0.4 0.532 3.2 0.147 0.8 0.169 0.7
Merced 0.783 3.2 1.052 7.0 - 0.038 - 0.2 - 0.004 0.0
S. San Joaquin 1.317 4.0 1.017 4.3 0.346 1.4 0.311 1.0
SANDAG
San Joaquin 0.234 0.9 0.391 2.5 - 0.146 - 0.8 - 0.181 - 0.8
Stanislaus - 0.076 - 0.2 0.088 0.3 - 0.323 - 1.2 - 0.168 - 0.4
W. Sierra Nevada 1.744 5.5 1.153 5.1 0.257 0.8 0.531 1.4
Alameda - 1.159 - 3.9 - 0.524 - 3.4 - 1.551 - 7.3 - 0.646 - 2.6
Contra Costa - 0.619 - 2.3 - 0.086 - 0.6 - 0.858 - 4.8 - 0.509 - 2.3
Marin/ Sonoma/ Napa - 0.767 - 2.7 - 0.211 - 1.4 - 1.617 - 7.1 - 1.654 - 4.9
San Francisco - 0.993 - 3.3 - 0.893 - 5.1 - 2.274 - 7.2 - 1.680 - 4.4
San Mateo - 0.894 - 2.6 - 0.379 - 2.2 - 1.864 - 5.3 - 1.232 - 3.2
Santa Clara - 1.129 - 4.4 - 1.016 - 6.6 - 0.856 - 5.1 - 0.810 - 3.5
Solano - 1.102 - 1.8 0.113 0.5 - 1.627 - 3.5 - 0.422 - 0.9
Los Angeles - 1.391 - 5.5 - 1.717 - 9.7 - 1.335 - 8.8 - 1.885 - 8.2
Orange
Riverside - 1.538 - 1.5 - 0.646 - 1.5 - 1.827 - 3.1 - 2.094 - 2.1
San Bernardino - 0.991 - 2.5 - 0.113 - 0.3
Ventura - 0.846 - 0.8 - 0.131 - 0.3 - 0.602 - 1.2 - 0.001 0.0
Regional Interactions
MTC to SCAG
MTC to SANDAG
SACOG to SCAG
SACOG to SANDAG
SCAG to MTC
SCAG to SACOG
SANDAG to MTC
SANDAG to SACOG
Size variables ( exponentiated)
Other employment 1.000 constrained 1.000 constrained
Households 1.000 constrained 1.000 constrained
Retail employment- low income 1.039 0.0 9.826 3.7 1.160 0.3 0.000 0.0
Service employment - low income 3.414 2.1 3.022 1.7 0.069 - 1.0 0.228 - 2.4
Retail employment - med income 2.050 1.2 3.196 4.1 0.897 - 0.2 0.000 0.0
Service employment - med income 0.945 - 0.1 1.059 0.2 0.489 - 2.0 0.373 - 2.2
Retail employment - high income 23.243 3.1 10.257 6.1 0.855 - 0.2 2.737 1.8
Service employment - high income 2.724 0.9 3.047 2.9 0.169 - 1.4 0.367 - 0.8
Retail employment - missing income3 1.763 0.6 2.249 1.3 1.877 0.8 1.331 0.4
Service employment - missing income3 0.204 - 0.7 0.779 - 0.4 0.311 - 0.8 0.000 - 0.1
1Estimated without distance terms.
2Constrained in final model.
3Not used in application.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 27
3.4 ACCESS/ EGRESS MODE CHOICE MODELS
Model Structure
The access and egress models produce probabilities that each access and egress
mode will be chosen, for each origin- destination pair, given the specific
transportation and demographic characteristics of that traveler and trip. Several
nesting structures were tested in model estimation to derive the nesting structure
that provided the most logical and statistically sound nests. This nesting
structure is displayed in Figure 3.6 and demonstrates that all driving modes are
estimated at the upper nest, while non- driving modes are estimated at the lower
nest.
Figure 3.6 Access/ Egress Nested Model Structure
Drive/ Park Drop Off Rental Car
Taxi Transit Walk/ Bike
Main Mode
Didn’t Drive
Model Specification
Table 3.10 below shows the distribution of the survey data used in model
estimation. These access and egress choices reflect the survey data, but are not
used directly in producing access and egress choices for each trip. For access, the
majority or trips are drive and park or drop off. For egress, the shares vary more
by purpose and distance, with transit more popular for short trips, and rental car
and taxi more popular for long trips and business trips. The shares for short
commute trips are unusually high for unpark and drive ( indicating that someone
keeps a car at the destination station) and taxi but these will be modified by
observed values in the Census during model calibration.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 28 Cambridge Systematics, Inc.
Table 3.10 Access and Egress Mode Choice Shares
Long Short
Choice Shares Business Other Business Commute Other
Access
Get dropped off 21.8% 41.9% 10.0% 10.1% 26.6%
Drive and park 58.5% 44.3% 76.8% 82.9% 60.4%
Rental car 3.7% 0.6%
Taxi 10.7% 6.0% 1.9% 1.1% 4.3%
Transit 4.5% 6.5% 10.0% 5.0% 7.6%
Walk 0.8% 0.6% 1.4% 0.8% 1.2%
Egress
Get picked up 16.0% 44.4% 14.2% 6.7% 36.8%
Unpark and drive 9.4% 1.7% 13.7% 22.2% 1.2%
Rental car 34.5% 26.6% 10.9% 0.6% 8.7%
Taxi 31.7% 18.0% 36.6% 26.7% 27.6%
Transit 5.2% 7.7% 18.0% 40.0% 17.5%
Walk 3.2% 1.5% 6.6% 3.8% 8.2%
The access and egress mode choice models were based on actual reported and
hypothetical stated data. For people who were intercepted making actual air or
rail journeys, the access and egress mode choices are the actual reported ones.
For people whose actual journey was by car, the air and conventional rail
access/ egress mode choices are hypothetical. Obviously, the HSR access and
egress mode choices are hypothetical for all respondents. So, each respondent
provided up to 3 access choices and 3 egress choices, although most respondents
only provided 2 of each, because conventional rail and air were only included
together in the mode choice set for the LA- SD surveys.
For model estimation, the data were combined with network level of service
measures for auto and transit, and nested mode choice models were estimated.
The models also included a scale factor on the hypothetical choices relative to the
actual ones, to test the hypothesis that the residual error is less in the actual
choices.
Estimation Results
The access mode choice estimation results are shown in Table 3.11, and the
egress mode choice estimation results are in Table 3.12. Some important results
to note include the following:
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 29
Table 3.11 Access Mode Choice Models
Long Trip Short Trip
Business Other Business Commute Other
Observations 1,500 2,724 206 341 497
Final log- likelihood - 1,662.3 - 2,519.4 - 132.6 - 148.4 - 403.7
Rho- squared( 0) 0.276 0.365 0.486 0.639 0.316
Rho- squared( cons) 0.003 0.068 0.012 0.022 0.079
Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat
Level of Service
Cost ($) - 0.075 constrained - 0.120 constrained - 0.050 constrained - 0.100 constrained - 0.100 constrained
In- vehicle time ( min) - 0.060 constrained - 0.030 constrained - 0.040 constrained - 0.030 constrained - 0.025 constrained
Out of vehicle time ( min) - 0.147 - 6.4 - 0.083 - 2.5 - 0.100 - 2.9 - 0.060 constrained - 0.061 - 2.5
VOT IVT ($/ hour) $ 48.00 $ 15.00 $ 48.00 $ 18.00 $ 15.00
Ratio OVT/ IVT 2.45 2.76 2.51 2.00 2.43
Drive and park
Constant 4.503 4.6 1.319 2.6 3.705 3.8 6.947 1.9 1.618 5.2
Travel alone - 1.925 - 3.0
Fewer cars than persons - 1.547 - 2.2 - 1.903 - 2.8 - 3.775 - 1.9 - 1.166 - 3.2
Low income - 2.741 - 1.8 - 1.960 - 2.8 - 2.017 - 1.2 - 0.494 - 1.6
High income 0.709 1.6 0.339 1.4
Airport is LAX - 3.128 - 3.8 - 1.275 - 1.7
Airport is SFO - 4.082 - 4.4 - 3.036 - 2.6
Airport is SJC - 1.479 - 2.1
Airport is SAN - 1.410 - 2.3 - 1.370 - 2.3
Rental car
Constant - 7.010 - 4.3 - 8.801 - 3.2
To conventional rail - 5.000 constrained - 5.000 constrained
No cars in HH 5.110 3.2
High income 2.953 2.4
Get dropped off
In- vehicle time ( min) - 0.014 - 2.5 - 0.031 - 3.1 - 0.003 - 0.7
Household size 0.606 2.9 0.478 2.8 0.672 1.4 0.273 2.6
Taxi
Auto distance - 0.084 - 4.8 - 0.071 - 3.8 - 0.041 - 0.8 - 0.014 - 2.4
Constant 0.927 1.4 - 2.207 - 2.7 - 1.520 - 1.5 - 2.526 - 1.7 - 1.243 - 3.3
To conventional rail - 2.827 - 2.6 - 2.265 - 2.4
To high- speed rail - 1.092 - 2.1
Travel alone - 0.877 - 1.8
Low income - 3.010 - 1.9
High income 0.849 1.9
Transit
No walk egress - 4.836 - 4.6 - 1.807 - 1.9 - 1.469 - 1.1 - 3.345 - 3.6
Rail used in path 3.689 5.2 1.727 2.4 3.313 2.7 3.271 4.2
Constant 0.912 1.0 - 1.705 - 2.1 1.904 1.7 0.375 0.2 0.318 0.4
Travel alone 1.569 2.3
No cars in HH 1.439 1.7
Fewer cars than persons 1.480 2.1 1.985 2.6
Low income 0.846 1.0
Walk
Constant 3.142 2.9 0.901 0.8 3.778 2.1 1.983 0.9 2.497 2.3
To airport - 5.000 constrained - 2.634 - 1.0
Nesting and scaling
Nest- transit, walk, taxi 0.387 5.9 0.451 3.3 0.570 4.3 0.458 2.0 1.000 constrained
Scale on hypothetical choices 0.682 15.9 1.000 constrained 1.000 constrained 1.000 constrained 1.000 constrained
* Taxi not in the nest.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 30 Cambridge Systematics, Inc.
Table 3.12 Egress Mode Choice Models
Long Trip Short Trip
Egress Mode Choice Business Other Business Commute Other
Observations 1,466 2,668 171 300 444
Final log- likelihood - 2,121 - 3,066.6 - 267.5 - 390.7 - 515.2
Rho- squared( 0) 0.075 0.231 0.015 0.197 0.241
Rho- squared( cons) - 0.023 0.053 - 0.109 - 0.049 0.054
Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat
Level of Service
Cost ($) - 0.075 constrained - 0.120 constrained - 0.050 constrained - 0.100 constrained - 0.100 constrained
In- vehicle time ( min) - 0.060 constrained - 0.030 constrained - 0.040 constrained - 0.030 constrained - 0.025 constrained
Out of vehicle time ( min) - 0.140 - 6.2 - 0.060 constrained - 0.117 - 3.1 - 0.075 constrained - 0.050 constrained
VOT IVT ($/ hour) $ 48.00 $ 15.00 $ 48.00 $ 18.00 $ 15.00
Ratio OVT/ IVT 2.33 2.00 2.92 2.50 2.00
Unpark and drive
Constant 1.580 1.5 - 7.241 - 5.2 0.645 0.8 5.098 2.3 - 7.113 - 3.3
From conventional rail - 9.490 - 2.5
From high- speed rail - 2.251 - 1.8
Low income - 18.010 - 2.5 - 1.263 - 1.1
Rental car
Constant 6.345 4.8 - 0.280 - 1.3 - 1.282 - 1.2 - 14.520 - 2.1 - 3.074 - 3.3
From conventional rail - 3.522 - 2.4 - 1.176 - 3.1
From high- speed rail - 0.552 - 2.4
Travel alone - 2.588 - 4.7
Low income - 2.082 - 0.9 - 1.891 - 3.7
Get picked up
In- vehicle time ( min) - 0.015 - 3.9
Household size 0.974 2.8
Taxi
Auto distance - 0.126 - 7.9 - 0.052 - 6.6 - 0.230 - 3.1 - 0.096 - 3.5
Constant 7.705 5.5 - 0.749 - 3.3 4.962 3.0 6.179 2.1 0.048 0.1
From high- speed rail 2.507 3.6
Travel alone - 2.768 - 4.6
Low income - 3.002 - 2.3 - 1.038 - 2.3
High income 1.499 2.8
Transit
No walk egress - 5.118 - 4.3 - 4.466 - 6.2
Rail used in path 2.960 5.0 2.570 3.5
Constant 4.441 2.7 - 3.715 - 3.6 4.342 2.5 8.170 2.7 - 0.525 - 0.9
From conventional rail 3.580 5.2 1.830 2.8
From high- speed rail 0.592 0.7 1.032 1.9
Low income 1.216 1.9 1.948 2.2
High income - 0.581 - 1.1
Walk
Constant 10.330 5.7 - 0.815 - 1.3 5.607 2.8 4.825 1.6 1.942 3.7
From airport - 2.074 - 2.0
Nesting and scaling
Nest- transit, walk, taxi 0.280 6.9 0.470 5.3 0.649 2.9 0.487 2.6 0.758 4.1
Scale on hypothetical choices 0.516 9.8 1.000 constrained 0.412 3.0 0.334 5.2 0.610 4.8
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 31
• The in- vehicle time and cost parameters had to be constrained at reasonable
values for all segments, as the initial results were insignificant or the incorrect
sign in all cases. A reasonable value of time was asserted for each segment
based upon a review of other research. As the survey was not designed
primarily to estimate access and egress choice models, and the zone size is in
a statewide model is quite large for this type of local choice, the result is
perhaps not surprising. Also note that the costs of options such as taxi and
rental car and airport/ station parking are not readily obtained from network
data.
• The out- of- vehicle time coefficients were estimated for most segments, and
result in ratios of out- of- vehicle time to in- vehicle time that are in the range of
2.0 to 2.9.
• For the Long Segments, the Drop Off and Pick Up alternatives have an
additional negative in- vehicle time effect, capturing the disutility of the
driver that has to make the round trip to the airport.
• We did not include taxi cost explicitly, but did include an additional distance
coefficient for taxi, which is significant and negative for most segments,
typically with an equivalent value of over $ 1.00 per mile.
• For most segments, transit is less likely to be chosen if there is no reasonable
walk access to transit at the trip end, meaning that a drive to transit path was
included instead.
• For most segments, transit, which can include rail and/ or bus, is more likely
to be chosen if rail is included in the best transit path.
• Due to much smaller sample sizes for the short trip segments, we were not
able to estimate many other segmentation effects for those segments.
• For the long segments, taxi, parking, and rental cars are generally less
desirable to rail stations than to airports, while transit is more desirable from
rail stations. Walking is very rare to or from airports, capturing accessibility
affects that are not captured well in the zone system.
• Drive and park access is less likely at the busiest airports – SFO, LAX, and
SAN – and somewhat at SJC as well. This may capture both cost and
inconvenience effects at those airports.
• Those traveling alone in the Other Long segment are more likely to use
transit and less likely to use taxi and auto, relative to those traveling with
others.
• For most segments, those in larger households are more likely to be dropped
off.
• In general, high income favors rental car, taxi and drive and park, and low
income slightly favors transit in some segments.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 32 Cambridge Systematics, Inc.
• In 7 of the 10 models, there is a logsum coefficient less than 1.0 on a nest that
includes transit, walk, and taxi. Each of the other three alternatives is in its
own “ nest,” and scaled by the same logsum parameter to preserve equal
scaling at the elemental level. The logsum coefficient is typically near 0.5. In
2 of the models ( Business Short Access and Commute Short Access), the taxi
alternative is not included in the nest.
• For most of the Access mode segments, the scale ( the inverse of the residual
error variance) for the hypothetical choices was not significantly lower than
1.0. It was only so for the Business Long segment, which is mainly air
travelers. This result suggest that most people are fairly familiar with the
travel options near their home, but that business travelers may be more
familiar with the airport access situation than with possible access to rail
stations.
• In contrast, for most of the Egress model segments, the scale factor on
hypothetical choices is significantly less than 1.0. This result indicates that
many respondents have difficulty making accurate tradeoffs for mode choice
in less familiar surroundings at the non- home end of their trip, so that
hypothetical choices should be weighted less in estimation than actual ones.
3.5 MAIN MODE CHOICE MODELS
Model Structure
The main mode choice models produce probabilities that each trip will choose
one of the main modes ( auto, air, conventional rail, and high- speed rail). Several
nesting structures were tested for the main mode choice models and the final
nesting structure chosen is presented in Figure 3.7. This structure provided the
most logical and statistically sound nesting structure for the mode choice models.
Figure 3.7 Main Mode Choice Nested Model Structure
Auto
Air Conventional
Rail
High- Speed
Rail
Destination
Non- Auto
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 33
We tested a few model variables that did not impact the final model
specification, as follows:
• We tested “ inertia” effects related to the actual mode that people used
relative to their SP choices. This variable was significant in the models, but
produces illogical results for most of the other variables, so was left out of the
final models.
• We segmented the cost coefficients by income group, but these were not
significant in the models. The high income coefficients used by mode were a
more effective means to include income in the models.
We separated reliability and frequency variables for high- speed rail, but these
were not significant so were not included in the final models.
Model Specification
The main mode choice models are based on stated- preference choice data, with
each respondent making a choice for four separate scenarios. Three different
types of choice sets were used in the SP surveys:
• Within Southern California ( between the SCAG and SANDAG regions): All
four modes – car, air, conventional rail, and HSR.
• Within Northern/ Central California ( both trip ends north of the SCAG
region): Three modes – car, conventional rail, and HSR. Air not included.
• Between Southern and Northern/ Central California: Three modes – car, air,
and HSR. Conventional rail not included.
In general, most of the respondents in the Short trip segments less than 100 miles
were in the first two groups, while most of those in the Long trip segments were
in the third group.
The overall choice shares in the SP data are shown in Table 3.13 below by
segment. Conventional rail was rarely made available for Long trips, and Air
was very rarely made available for Short trips, which partly explain the low
shares for those modes in particular segments. In general, the share for HSR is
quite high, and is highest for business trips and long trips, giving a first
indication that HSR substitutes more closely with air than with car.
Table 3.13 Overall Choice Shares in SP Data
Long Trip Short Trip
Business Other Business Commute Other
Car 9.2% 34.7% 27.9% 11.2% 50.4%
Air 20.9% 6.2% 0.0% 0.0% 0.0%
Conventional rail 1.3% 3.0% 21.8% 33.5% 14.1%
High- speed rail 68.6% 56.2% 50.3% 55.3% 35.6%
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 34 Cambridge Systematics, Inc.
To prepare the data for estimation, the access and egress mode choice models
were first applied to calculate access and egress mode logsums for each
alternative. Then, a nested logit model was estimated across the four main
modes for each of the segments ( only three alternatives for the Short segments, as
air was not available for those segments).
Estimation Results
The estimation results are shown in Table 3.14. Some results of note include the
following:
• As in the access/ egress models, there are fewer cases for the Short segments,
and fewer significant coefficients as a result.
• The residual mode- specific constants for HSR are generally not very much
higher than for the other modes. This result indicates that the high choice
shares found for HSR are mainly due to the attractiveness of the time and
cost, by the mode, rather than to SP- related survey effects or biases.
• For the three largest segments, the cost and in- vehicle time parameters were
estimated non- constrained and give very reasonable values of time. For the
Short Business and Commute segments, the original in- vehicle time
coefficients were quite low, and so were constrained to give values of time
that seem more in line with other models. In general, VOT for the longer,
more expensive trips is higher than for the shorter, more frequent trips. This
is a typical result.
• The value of frequency ( headway) is significant for all segments, but is only
about 20 percent as large as the in- vehicle time coefficient. If wait time were
half the headway and valued twice as highly as in- vehicle time, then we
would expect the same coefficient on headway and in- vehicle time. For these
modes, and particularly air, headway is less related to wait time than it is to
scheduling convenience. Because none of the levels used in the SP had
headways higher than a few hours, the implications for scheduling may not
have been large enough to greatly influence mode choice.
• The value of reliability is fairly low for all segments, although with the
correct sign. It is very difficult to measure the effect of reliability in a large-scale
mailout SP survey, so we decided to use a somewhat higher effect of
reliability in application, based on any evidence from elsewhere. In addition,
we are redefining reliability based on percent within 60 minutes of schedule
time rather than the 15 minutes used in the survey to identify more
significant reliability problems.
• For the Long segments, those traveling with others are more likely to use car
and less likely to use air. This effect was also tested on the cost coefficients
and not found to be significant, so this relative mode preference appears to be
related to more than just cost – such as the fact that people can share driving
for long trips.
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
Cambridge Systematics, Inc. 3- 35
Table 3.14 Main Mode Choice Models
Long Trip Short Trip
Business Other Business Commute Other
Observations 2,918 8,075 326 564 852
Final log- likelihood - 1,969 - 3,933 - 295 - 445 - 744
Rho- squared( 0) 0.389 0.31 0.175 0.281 0.205
Rho- squared( cons) 0.163 0.155 0.123 0.159 0.117
Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat Coeff. T- stat
Main Mode Characteristics
Constants
Car ( base)
Air - 1.645 - 4.7 0.6898 2.8
Conventional rail - 0.387 - 0.9 0.6149 2.6 - 0.268 - 0.5 4.232 2.6 - 0.3847 - 1.4
High- speed rail - 0.3503 - 1.1 1.434 7 - 1.557 - 2.8 4.048 2.5 0.5041 1.7
Level of Service
Cost ($) - 0.01626 - 12.8 - 0.035 - 18.5 - 0.109 - 5.4 - 0.148 - 11.3 - 0.109 - 8.2
In- vehicle time ( min) - 0.016 - 11.1 - 0.011 - 14.2 - 0.5 constrained - 0.025 constrained - 0.014 - 5.2
Service headway ( min) - 0.003 - 3.7 - 0.003 - 3.5 - 0.006 - 2.5 - 0.0023 - 2.4 - 0.009 - 5.5
Reliability (% on time) 0.001 0.3 0.005 1.9 0.023 1.8 0.006 0.6 0.004 0.6
Implied Value of Time IVT ($/ hour) $ 57.71 $ 18.33 $ 27.60 $ 10.12 $ 7.93
Ratio Frequency/ IVT 0.21 0.24 0.12 0.1 0.66
Trip Characteristics
Travel in a Group
Car 0.8492 4.2 1.417 9.1
Air - 0.3375 - 2.7 - 0.5061 - 3.7
Household Characteristics
Household Size
Car 0.0704 0.9 0.225 4.9 0.655 2
Income
High – car - 1.211 - 2.3 - 1.247 - 1.8
High – air 1.018 4.5
High – conventional rail 0.5237 1.2
High – high- speed rail 0.9807 4.8
Fewer Cars than Workers
Car - 0.7696 - 2.4 - 0.4354 - 2.8 - 0.7873 - 0.8 - 2 - 1.5
Nesting and scaling
Nest – air, rail, high- speed rail 0.8514 8.8 0.7426 13 0.5159 2.7 0.5892 3.4 0.6855 6.1
Access mode choice logsum 0.115 3.1 0.2134 3.8 0.4628 1.9 0.33 1.5 0.3148 3.5
Egress mode choice logsum 0.1561 3.8 0.3974 7.1 0.4628 constrained 0.33 constrained 0.3148 3.5
Bay Area/ California High- Speed Rail Ridership and Revenue Forecasting Study
3- 36 Cambridge Systematics, Inc.
• People in larger households are more likely to use car. Even though we
already have the group/ alone segmentation, people in larger households are
likely to be in larger groups.
• Higher income generally favors air and high- speed rail versus auto.
• Low auto availability within the household is somewhat related to less
chance of choosing the auto.
• A nest with air, rail, and HSR, ( with car in its own “ nest”) produced a logsum
coefficient below 1.0 for all segments, indicating that this was a reasonable
nesting structure for interregional trips.
• The access mode choice logsums were estimated with positive coefficients in
the range of 0.11 to 0.46 for all segments. The egress mode choice logsums
gave negative values ( which are illogical) for the business and commute short
trips, so these were constrained to be the same as the access logsum.
• The access and egress logsums are somewhat lower for the long trips than for
the short trips, which may reflect the fact that the access and egress legs are a
smaller percentage of the total travel time for the long trips.
• For the long trips, the egress mode accessibility seems to have somewhat
more influence on mode choice than does the access mode. Travelers may be
less concerned about the home end, where they know the options and can
use their own auto, and then they are about the destination end.
The main mode choice models are likely to be the key determinants of the
sensitivity of the model system as a whole – particularly the models for the Long
trip segments where HSR is likely to be most attractive.
3.6 MODEL APPLICATION
The interregional models will be applied with customized software within the
Cube software framework. This applica