Acrp

THE NATIONAL ACADEMIES PRESS
This PDF is available at http://nap.edu/23106 SHARE   
Airport Ground Access Mode Choice Models
DETAILS
104 pages | | PAPERBACK

ISBN 978-0-309-42034-1 | DOI 10.17226/23106
AUTHORS
BUY THIS BOOK
FIND RELATED TITLES

Visit the National Academies Press at NAP.edu and login or register to get:
– Access to free PDF downloads of thousands of scientiﬁc reports

– 10% off the price of print titles
– Email or social media notiﬁcations of new titles related to your interests
– Special offers and discounts

Distribution, posting, or copying of this PDF is strictly prohibited without written permission of the National Academies Press.
(Request Permission) Unless otherwise indicated, all materials in this PDF are copyrighted by the National Academy of Sciences.
Copyright © National Academy of Sciences. All rights reserved.

AIRPORT COOPERATIVE RESEARCH PROGRAM
ACRP SYNTHESIS 5
Airport Ground Access
Mode Choice Models
A Synthesis of Airport Practice
CONSULTANT
GEOFFREY D. GOSLING
Aviation System Consulting, LLC
Berkeley, California
S UBJECT A REAS
Aviation and Planning
Research Sponsored by the Federal Aviation Administration
TRANSPORTATION RESEARCH BOARD

WASHINGTON, D.C.
2008
www.TRB.org
Copyright National Academy of Sciences. All rights reserved.

AIRPORT COOPERATIVE RESEARCH PROGRAM ACRP SYNTHESIS 5
Airports are vital national resources. They serve a key role in Project 11-03, Topic 03-02
transportation of people and goods and in regional, national, and ISSN 1935-9187
international commerce. They are where the nation’s aviation sys- ISBN 978-0-309-09798-7
tem connects with other modes of transportation and where federal Library of Congress Control Number 2007910441
responsibility for managing and regulating air traffic operations
intersects with the role of state and local governments that own and © 2008 Transportation Research Board
operate most airports. Research is necessary to solve common oper-
ating problems, to adapt appropriate new technologies from other
industries, and to introduce innovations into the airport industry. COPYRIGHT PERMISSION
The Airport Cooperative Research Program (ACRP) serves as one Authors herein are responsible for the authenticity of their materials and for
of the principal means by which the airport industry can develop obtaining written permissions from publishers or persons who own the
innovative near-term solutions to meet demands placed on it. copyright to any previously published or copyrighted material used herein.
The need for ACRP was identified in TRB Special Report 272: Cooperative Research Programs (CRP) grants permission to reproduce
Airport Research Needs: Cooperative Solutions in 2003, based on material in this publication for classroom and not-for-profit purposes.
a study sponsored by the Federal Aviation Administration (FAA). Permission is given with the understanding that none of the material will
The ACRP carries out applied research on problems that are shared be used to imply TRB or FAA endorsement of a particular product, method,
by airport operating agencies and are not being adequately or practice. It is expected that those reproducing the material in this
addressed by existing federal research programs. It is modeled after document for educational and not-for-profit uses will give appropriate
the successful National Cooperative Highway Research Program acknowledgment of the source of any reprinted or reproduced material. For
and Transit Cooperative Research Program. The ACRP undertakes other uses of the material, request permission from CRP.
research and other technical activities in a variety of airport subject
areas, including design, construction, maintenance, operations,
safety, security, policy, planning, human resources, and adminis- NOTICE
tration. The ACRP provides a forum where airport operators can
cooperatively address common operational problems. The project that is the subject of this report was a part of the Airport
The ACRP was authorized in December 2003 as part of the Cooperative Research Program conducted by the Transportation Research
Vision 100-Century of Aviation Reauthorization Act. The primary Board with the approval of the Governing Board of the National Research
participants in the ACRP are (1) an independent governing board, Council. Such approval reflects the Governing Board’s judgment that the
the ACRP Oversight Committee (AOC), appointed by the Secretary project concerned is appropriate with respect to both the purposes and
of the U.S. Department of Transportation with representation from resources of the National Research Council.
airport operating agencies, other stakeholders, and relevant indus- The members of the technical advisory panel selected to monitor this
try organizations such as the Airports Council International-North project and to review this report were chosen for recognized scholarly
America (ACI-NA), the American Association of Airport Execu- competence and with due consideration for the balance of disciplines
tives (AAAE), the National Association of State Aviation Officials appropriate to the project. The opinions and conclusions expressed or
implied are those of the research agency that performed the research, and
(NASAO), and the Air Transport Association (ATA) as vital links
while they have been accepted as appropriate by the technical panel, they
to the airport community; (2) the TRB as program manager and sec-
are not necessarily those of the Transportation Research Board, the National
retariat for the governing board; and (3) the FAA as program spon-
Research Council, or the Federal Aviation Administration of the U.S.
sor. In October 2005, the FAA executed a contract with the National Department of Transportation.
Academies formally initiating the program. Each report is reviewed and accepted for publication by the technical
The ACRP benefits from the cooperation and participation of air- panel according to procedures established and monitored by the
port professionals, air carriers, shippers, state and local government Transportation Research Board Executive Committee and the Governing
officials, equipment and service suppliers, other airport users, and Board of the National Research Council.
research organizations. Each of these participants has different
interests and responsibilities, and each is an integral part of this
cooperative research effort.
Research problem statements for the ACRP are solicited period- The Transportation Research Board of the National Academies, the National
ically but may be submitted to the TRB by anyone at any time. It is Research Council, and the Federal Aviation Administration (sponsor of
the responsibility of the AOC to formulate the research program by the Airport Cooperative Research Program) do not endorse products or
identifying the highest priority projects and defining funding levels manufacturers. Trade or manufacturers’ names appear herein solely because
and expected products. they are considered essential to the clarity and completeness of the project
Once selected, each ACRP project is assigned to an expert panel, reporting.
appointed by the TRB. Panels include experienced practitioners and
research specialists; heavy emphasis is placed on including airport
professionals, the intended users of the research products. The panels
prepare project statements (requests for proposals), select contractors,
and provide technical guidance and counsel throughout the life of the Published reports of the
project. The process for developing research problem statements and
selecting research agencies has been used by TRB in managing coop- AIRPORT COOPERATIVE RESEARCH PROGRAM
erative research programs since 1962. As in other TRB activities, are available from:
ACRP project panels serve voluntarily without compensation.
Primary emphasis is placed on disseminating ACRP results to the Transportation Research Board
intended end-users of the research: airport operating agencies, service Business Office
500 Fifth Street, NW
providers, and suppliers. The ACRP produces a series of research Washington, DC 20001
reports for use by airport operators, local agencies, the FAA, and other
interested parties, and industry associations may arrange for work- and can be ordered through the Internet at
http://www.national-academies.org/trb/bookstore
shops, training aids, field visits, and other activities to ensure that
results are implemented by airport-industry practitioners. Printed in the United States of America

The National Academy of Sciences is a private, nonprofit, self-perpetuating society of distinguished schol-
ars engaged in scientific and engineering research, dedicated to the furtherance of science and technology
and to their use for the general welfare. On the authority of the charter granted to it by the Congress in
1863, the Academy has a mandate that requires it to advise the federal government on scientific and techni-
cal matters. Dr. Ralph J. Cicerone is president of the National Academy of Sciences.
The National Academy of Engineering was established in 1964, under the charter of the National Acad-
emy of Sciences, as a parallel organization of outstanding engineers. It is autonomous in its administration
and in the selection of its members, sharing with the National Academy of Sciences the responsibility for
advising the federal government. The National Academy of Engineering also sponsors engineering programs
aimed at meeting national needs, encourages education and research, and recognizes the superior achieve-
ments of engineers. Dr. Charles M. Vest is president of the National Academy of Engineering.
The Institute of Medicine was established in 1970 by the National Academy of Sciences to secure the
services of eminent members of appropriate professions in the examination of policy matters pertaining
to the health of the public. The Institute acts under the responsibility given to the National Academy of
Sciences by its congressional charter to be an adviser to the federal government and, on its own initiative,
to identify issues of medical care, research, and education. Dr. Harvey V. Fineberg is president of the
Institute of Medicine.
The National Research Council was organized by the National Academy of Sciences in 1916 to associate
the broad community of science and technology with the Academyís p urposes of furthering knowledge and
advising the federal government. Functioning in accordance with general policies determined by the Acad-
emy, the Council has become the principal operating agency of both the National Academy of Sciences
and the National Academy of Engineering in providing services to the government, the public, and the scien-
tific and engineering communities. The Council is administered jointly by both the Academies and the Insti-
tute of Medicine. Dr. Ralph J. Cicerone and Dr. Charles M. Vest are chair and vice chair, respectively,
of the National Research Council.
The Transportation Research Board is one of six major divisions of the National Research Council. The
mission of the Transportation Research Board is to provide leadership in transportation innovation and
progress through research and information exchange, conducted within a setting that is objective, interdisci-
plinary, and multimodal. The Board’s varied activities annually engage about 7,000 engineers, scientists, and
other transportation researchers and practitioners from the public and private sectors and academia, all of
whom contribute their expertise in the public interest. The program is supported by state transportation depart-
ments, federal agencies including the component administrations of the U.S. Department of Transportation,
and other organizations and individuals interested in the development of transportation. www.TRB.org
www.national-academies.org

ACRP COMMITTEE FOR PROJECT 11-03 COOPERATIVE RESEARCH PROGRAMS STAFF

CHRISTOPHER W. JENKS, Director, Cooperative Research Programs
CHAIR CRAWFORD F. JENCKS, Deputy Director, Cooperative Research
BURR STEWART Programs
Port of Seattle ROBERT E. DAVID, Senior Program Officer
EILEEN DELANEY, Director of Publications
MEMBERS
GARY C. CATHEY ACRP SYNTHESIS STAFF
California Department of Transportation STEPHEN R. GODWIN, Director for Studies and Special Programs
KEVIN C. DOLLIOLE JON M. WILLIAMS, Associate Director, IDEA and Synthesis Studies
Lambert–St. Louis International Airport GAIL STABA, Senior Program Officer
BERTA FERNANDEZ DON TIPPMAN, Editor
Landrum & Brown CHERYL Y. KEITH, Senior Program Assistant
JULIE KENFIELD
Jacobs Carter Burgess TOPIC PANEL
CAROLYN MOTZ DAN BURKE, Port of Seattle
Hagerstown Regional Airport FRED CUMMINGS, Philadelphia International Airport
KIMBERLY FISHER, Transportation Research Board
FAA LIAISON HANI S. MAHMASSANI, University of Maryland
LORI PAGNANELLI LLOYD A. McCOOMB, Greater Toronto Airports Authority
GUY ROUSSEAU, Atlanta Regional Commission
ACI–NORTH AMERICA LIAISON PHILLIP SHAPIRO, Vanesse Hangen Brustlin, Silver Spring, MD
RICHARD MARCHI KEVIN F. TIERNEY, Cambridge Systematics, Inc.
PATRICK SULLIVAN, Federal Aviation Administration (Liaison)
TRB LIAISON
CHRISTINE GERENCHER
ACKNOWLEDGMENTS
The preparation of this report benefited greatly from the extensive of the Port of Portland; Stacey Falzarano of Resource Systems Group,
comments, suggestions, and help by the members of the Project Panel, Inc.; Ricardo Fernandez of Earth Tech; Cary Greene of San José Inter-
as well as the input and assistance from a large number of organizations national Airport; Ian Harrington of the Central Transportation Planning
and individuals involved in airport ground access studies and planning. Staff, Boston; Richard Hazlett of the city of Chicago; Craig Leiner of the
The respondents to the survey, who provided the information described in Massachusetts Port Authority; Lloyd McCoomb of the Greater Toronto
the report, although too numerous to mention individually, are gratefully Airport Authority; Guy Rousseau of the Atlanta Regional Commission;
acknowledged. The case studies documented in the report would not Peter Smith of Halcrow Group, Ltd., London; and Richard Walker of
have been possible without the help of many individuals in obtaining Portland Metro. Finally, the help and support of Gail Staba, the Senior
copies of reports or other documents. Particular appreciation is extended Program Officer for Airport Synthesis Studies, proved invaluable in
to Gary Donn of the Florida Department of Transportation; Scott Drumm bringing the study to completion.

FOREWORD Airport administrators, engineers, and researchers often face problems for which infor-
By Staff mation already exists, either in documented form or as undocumented experience and prac-
Transportation tice. This information may be fragmented, scattered, and unevaluated. As a consequence,
Research Board full knowledge of what has been learned about a problem may not be brought to bear on its
solution. Costly research findings may go unused, valuable experience may be overlooked,
and due consideration may not be given to recommended practices for solving or alleviat-
ing the problem.
There is information on nearly every subject of concern to the airport industry. Much of
it derives from research or from the work of practitioners faced with problems in their day-
to-day work. To provide a systematic means for assembling and evaluating such useful
information and to make it available to the entire airport community, the Airport Coopera-
tive Research Program authorized the Transportation Research Board to undertake a con-
tinuing project. This project, ACRP Project 11-03, “Synthesis of Information Related to
Airport Practices,” searches out and synthesizes useful knowledge from all available
sources and prepares concise, documented reports on specific topics. Reports from this
endeavor constitute an ACRP report series, Synthesis of Airport Practice.
This synthesis series reports on current knowledge and practice, in a compact format,
without the detailed directions usually found in handbooks or design manuals. Each report
in the series provides a compendium of the best knowledge available on those measures
found to be the most successful in resolving specific problems.
PREFACE This synthesis extends previous efforts to document the state of practice for airport
ground access mode choice models. It examines the characteristics of existing models and
discusses the issues involved in the development and use of such models to improve the
understanding and acceptance of their role in airport planning and management. Informa-
tion presented in this report may be of interest to a range of airport managers, airport and
regional transportation planners, consultants and transportation modeling specialists, and
researchers interested in issues involving airport ground access mode choice.
For this synthesis, a comprehensive review of the relevant literature was undertaken. To
document the extent of the recent use of airport ground access mode choice models and to
identify sources of technical documentation on existing models, this literature review was
supplemented by a survey of airport authorities, metropolitan planning organizations, con-
sulting firms and research organizations, and other government agencies and industry orga-
nizations. Follow-up communications by telephone and e-mail were made where necessary.
Geoffrey D. Gosling, Aviation System Consulting, LLC, Berkeley, California, collected
and synthesized the information and wrote the report. The members of the topic panel are
acknowledged on the preceding page. This synthesis is an immediately useful document
that records the practices that were acceptable within the limitations of the knowledge avail-
able at the time of its preparation. As progress in research and practice continues, new
knowledge will be added to that now at hand.

CONTENTS
1 SUMMARY
11 CHAPTER ONE INTRODUCTION

Background, 11
Methodology, 11
Potential Uses of Airport Access Mode Choice Models, 12
Report Audience, 12
Report Organization, 12
14 CHAPTER TWO AIRPORT GROUND ACCESS MODE CHOICE

MODELING PROCESS
Motivation for Modeling Airport Ground Access Mode Choice, 14
Overall Mode Choice Model Development Process, 16
Mathematical Form of Typical Mode Choice Models, 22
Model Estimation Considerations, 25
Model Application, 28
31 CHAPTER THREE REVIEW OF THE LITERATURE

Air Passenger Mode Choice Models, 31
Airport Employee Travel Choice, 38
39 CHAPTER FOUR USE OF AIRPORT GROUND ACCESS MODELS

IN AIRPORT PLANNING
Surveys of Airports, Planning Agencies, and Other Organizations, 39
Survey Findings, 40
43 CHAPTER FIVE STATE OF PRACTICE OF AIR PASSENGER

MODE CHOICE MODELS
Technical Approach, 43
Modes Included in Model, 46
Explanatory Variables, 48
Market Segmentation, 53
Model Performance, 54
55 CHAPTER SIX AIRPORT EMPLOYEE MODE CHOICE

Technical Approach, 57
Modes Included in Model, 63
Explanatory Variables, 64
Modeling Considerations, 65
Model Performance, 65

67 CHAPTER SEVEN TRANSFERABILITY OF AIRPORT GROUND ACCESS

MODE CHOICE MODELS
Issues in Model Transferability, 68
Variability in Model Specification, 68
Comparative Analysis of Model Parameters, 69
Data Considerations, 69
Summary, 70
72 CHAPTER EIGHT INTEGRATION OF AIRPORT GROUND ACCESS MODELS

IN REGIONAL PLANNING PROCESS
Technical Considerations, 72
Case Studies, 73
76 CHAPTER NINE CONCLUSIONS AND FURTHER RESEARCH
77 REFERENCES
81 BIBLIOGRAPHY
82 GLOSSARY
83 APPENDIX A SURVEY QUESTIONNAIRES
96 APPENDIX B SUMMARY OF SURVEY RESULTS
102 APPENDIX C LIST OF PARTICIPATING AGENCIES

AIRPORT GROUND ACCESS MODE CHOICE MODELS
SUMMARY The prediction of air passenger and airport employee mode choice decisions for travel to and
from the airport forms a key analytical component of airport landside planning, as well as air-
port system planning. However, there is currently no generally accepted and validated
approach to modeling how airport users will change their access or egress mode in response
to changes in the airport ground transportation system. The factors affecting airport travel are
recognized as being significantly different from those affecting typical trips accounted for in
regional transportation models. Therefore, it is difficult to determine the economic feasibil-
ity of proposed projects to improve airport ground transportation or effectively manage the
existing airport ground transportation system using traditional regional models.
Travel forecasting models, particularly those developed to address airport ground access
and egress trips, are highly specialized and not well understood by most airport managers and
planners. With increasing emphasis on intermodal connections, there is a pressing need for
more widely accepted and accessible reference material and information on such models.
This report has been undertaken to update and extend previous efforts to document the state
of practice for airport ground access mode choice models. It examines the characteristics of
existing models and discusses the issues involved in the development and use of such mod-
els, as well as research and development efforts that are needed to improve the state of the art
of modeling airport ground access mode choice and address technical issues that are not cur-
rently well handled.
The synthesis project has undertaken a comprehensive review of the relevant literature in
the field. This was supplemented by an extensive survey of airport authorities, metropolitan
planning organizations (MPOs), consulting firms and research organizations, and other rele-
vant government agencies to document the extent of the recent use of airport ground access
mode choice models and to identify sources of technical information on existing models.
Based on the findings of the literature review and survey, the report summarizes the current
state of practice of both air passenger and airport employee ground access mode choice mod-
els and discusses the type of planning issues to which the models have been applied, the
technical approach adopted, the ground access modes included in the models, and the ex-
planatory variables and market segmentation used to account for air passenger ground access
behavior. The report also addresses the extent to which air passenger ground access mode
choice models may be transferable to situations different from the one for which they have
been originally developed and issues that arise in integrating air passenger and airport
employee ground access mode choice models into more general regional transportation plan-
ning models.
The development of air passenger ground access mode choice models has been the sub-
ject of ongoing research for more than 30 years, during which time the state of practice has
evolved from relatively simple multinomial logit (MNL) models to more complex nested
logit (NL) models involving several levels of nesting and four or more market segments.
However, no clear consensus has yet emerged as to what explanatory variables should be
included or how the various modes and sub-modes should be nested and a number of prob-
lematic issues have not yet been addressed in a meaningful way, including how to treat rental

car use by non-residents of a region and how best to account for the role of traveler income
in the mode choice process. In addition, there has been almost no attention given to how
reliably existing models predict air passenger access mode use when circumstances change
from those from which the model was developed.
In contrast to air passenger mode choice models, there has been very little effort directed
at developing mode choice models for airport-based employees, including airline personnel
and employees of the many other organizations that are involved in airport operations. The
majority of MPOs model airport employee trips the same way as they model any other
journey-to-work trips, although the characteristics of airport employee trips are often very
different from regular journey-to-work trips because major airports operate 24 hours a day,
7 days a week. Thus, many airport-based employees work shift patterns outside the usual
commute periods and airline flight and cabin crews may be away from their crew base for
several days at a time.
Airport access mode choice decisions by air passengers and airport employees affect a wide
range of airport planning and operational management decisions, including the development
of landside facilities, airport revenue from parking and other ground transportation services,
and programs to reduce the growth in vehicle trips generated by the airport and the associated
emissions. Potential uses of models that can predict the effect on access mode use of proposed
changes to the system include sizing new planned facilities, evaluating the financial implica-
tions of proposed changes in parking rates or other ground transportation fees, determining the
expected air quality impacts of planned new facilities or proposed mitigation measures, and
assessing the feasibility of proposed projects to improve airport access. Airport accessibility
is also a significant factor in airport choice in multi-airport regions or situations where air ser-
vice competition exists between local and more distant airports. Airport access mode choice
is often embedded within models of airport choice and improved representation of airport
access mode choice behavior should benefit those applications as well.
This report has been prepared to meet the needs of a wide range of airport managers and
planners, transportation planning professionals, researchers, and others interested in model-
ing airport ground access mode choice, some of whom are primarily interested in a brief
overview of the topic, whereas others are more interested in technical details of specific mod-
els. This executive summary provides a nontechnical introduction to the issues involved in
airport access mode choice modeling and summarizes the current state of practice that is
described in more detail in the remainder of the report. Airport planners involved in landside
planning and regional transportation planners involved in travel demand modeling who are
interested in the airport access mode choice modeling process in general, the details of spe-
cific models, or the way in which airport access travel can be integrated into the regional
travel demand model will find a more thorough discussion in the relevant chapters of the
report. Consultants and transportation modeling specialists may be more interested in the
detailed case studies of a number of recent airport access mode choice models that are con-
tained in an appendix to the report. Finally, members of the aviation or transportation
research community who are interested in airport access mode use modeling may be inter-
ested in the potential areas for future research suggested in the report.
• Airport Ground Access Mode Choice Modeling Process
In broad terms, the general approach to developing a mode choice model is no different
from the development of any other mathematical model of a physical or behavioral process.
A set of data is assembled that describes the process being modeled. Next, a suitable func-
tional form for the mathematical model is defined that expresses the value of the variable that
the model is intended to predict in terms of some number of explanatory variables and model
coefficients (sometimes referred to as parameters), the values of which are to be determined
from the data. Statistical model estimation software is then used to estimate the values of

model coefficients that best explain the observed values of the variable that the model is
intended to predict, termed the dependent variable, using the observed values of the ex-
planatory variables, termed the independent variables. The extent to which the model is able
to reproduce the observed values of the dependent variable for any given set of values of the
independent variables is referred to as the goodness-of-fit of the model and is an important
measure of its usefulness.
In the case of airport access mode choice models, there are two types of data that are
required to estimate a model. The first consists of the access mode choices made by a repre-
sentative sample of airport users, together with explanatory data about their characteristics
(such as their household income) or the characteristics of their trip (such as where they com-
menced their journey to the airport or the purpose of their air trip). These data are typically
obtained from surveys of air passengers or airport employees. The second type of data con-
sists of the transportation characteristics (such as travel time and cost) of the various access
modes between which the mode choice decisions were made.
Because of the wide range of factors that affect airport access mode choice and the num-
ber of alternative modes that are typically available to a decision maker, airport access mode
choice models are usually disaggregate models that attempt to predict how an individual de-
cision maker with a given set of characteristics will behave. These models attempt to predict
the probability that a given airport user will choose a particular mode, because two airport
users with apparently identical characteristics may well choose different access modes.
Because a disaggregate mode choice model predicts the probability of a decision maker
choosing a given mode from among a defined set of alternatives, these models are also
referred to as discrete choice models. If the probability of choosing each access mode is
estimated for each airport user in a given sample of users, then the percentage of any group
of such users choosing a given mode can be calculated.
There are two different approaches to assembling the necessary data on airport access
mode choice behavior to develop mode choice models: revealed preference and stated pref-
erence surveys. Revealed preference surveys identify the travel choices actually made by air-
port travelers as well as collect information on other traveler characteristics and details of the
trip that are believed to influence the choice. The model estimation process attempts to
develop a model that explains the mode choice decisions in terms of the traveler characteris-
tics and the service characteristics of the different airport ground access modes available to
the traveler. Stated preference surveys follow a similar process, except that the respondent is
presented with a set of hypothetical choices and asked to select from them. For realism, the
stated preference experiment is usually structured so that the choices presented to the
respondent correspond to their current trip or a recent actual trip, but change the characteris-
tics of the ground transportation options available. This allows the model to incorporate
ground access options that do not currently exist or to explore the effect of changing factors
that do not exhibit much variation in the real world.
Once an estimation dataset has been assembled, model specification involves selecting an
appropriate functional form and market segmentation for the model and defining relevant
explanatory variables. Model estimation software is then used to obtain the estimated values
of the model coefficients. The statistical significance of these estimated values and the over-
all goodness-of-fit of the model are examined and the model specification revised as neces-
sary to address any problems with the model coefficients or statistical fit. Once a satisfactory
model has been estimated, model calibration involves making any necessary adjustments to
the model so that the model predictions agree with the observed pattern of mode use. Model
validation is the final step in model development and involves comparing the predictions of
the model with actual values of the phenomena being modeled, ground access mode use in
this case, under different conditions from those under which the model was developed, usu-
ally after some change has occurred in the system being modeled.

The overall process of developing an airport access mode choice model is summarized in
Figure 1 and discussed in more detail in chapter two of this report.
The basic concept underlying most disaggregate discrete choice analysis is that each alter-
native in the choice set provides the decision maker with some utility that can be expressed in
terms of measurable or observable characteristics of both the decision maker and the alterna-
tive (e.g., the travel time involved or the income level of the decision maker). The larger the
difference in the utility between two alternatives, the more likely the decision maker is to
choose the alternative with the higher utility. Because the probability of choosing a particular
alternative cannot be greater than one or less than zero, this results in an S-shaped relationship
between the difference in utility between two alternatives and the probability of choosing the
alternative with the greater utility for that decision maker. A common mathematical form for
this relationship is the logistic function, which for more than two alternatives results in the
MNL model that has been widely used for airport access mode choice studies. This model can
be expressed as:
eUi
P(i) =
∑ eU j
j ∈J
where P(i) is the probability of a decision maker choosing alternative i, Ui and Uj are the util-
ities of alternatives i and j, and J is the number of alternatives. The utility function for a given
alternative is assumed to comprise a deterministic part that consists of a function of measured
and observed variables and an error term that accounts for unobserved characteristics and
variability in the perceived utility of a given set of characteristics across different individu-
als. In logit choice models, the error term is assumed to be a random variable and the vari-
ance of the error term reflects the goodness-of-fit of the model. The deterministic part of the
utility function typically consists of a linear combination of explanatory variables with their
associated model coefficients, the values of which are determined in the model estimation
process. Therefore, a utility function can be expressed as:
Ui = ai + b1x1 + b2x2 + . . . + bnxn + ε
where ai and the b’s are the model coefficients, the x’s are the values of the explanatory vari-
ables such as travel time and cost, and ε is the error term. In general, the utility function for
each alternative will have a constant term ai, known as the alternative-specific constant,
which reflects attributes of the alternative that are not accounted for by the other variables.
Therefore, a fairly simple utility function for a mode choice model might comprise:
Vi = ai + b1(travel time) + b2(waiting time) + b3(walk distance) + b4(cost/income)
where Vi is the deterministic part of the utility function (it is common to omit the error term
in presenting the components of a utility function). In this example, travel cost is divided by
income in the fourth explanatory variable so that the choice process becomes less sensitive
to cost for higher-income travelers.
Although the MNL model has been widely used, it is vulnerable to problems that arise
from a property of the model termed the Independence from Irrelevant Alternatives. This
states that including a new alternative in the choice set (or changing the perceived value of
one of the alternatives) should not affect the relative probabilities of choosing any of the other
alternatives. However, in many situations in airport access mode choice it is quite unlikely
that changing the characteristics of one mode or sub-mode will leave the relative probabili-
ties of choosing all the other modes and sub-modes unchanged. For example, changes in one
public transportation service are likely to affect the use of other public transportation services
to a greater extent than the use of private vehicles. These limitations can be addressed by

grouping similar modes or sub-modes into separate groups or nests in a choice structure
referred to as a NL model, as illustrated by Figure 4 in chapter three.
This figure (from a ridership study for a proposed airport express train in Chicago) shows
that private transport modes have been grouped together in one nest, whereas public trans-
port modes have been grouped in a different nest. It also shows another feature of NL mod-
els, that it is possible to define lower-level nests that contain sub-modes of a particular mode,
in this case the access mode by which travelers reach the airport express train. The grouping
of modes in the NL model requires some changes to the mathematical form of the model,
which are not discussed here but are described in chapter two.
Once a calibrated or validated model is available, the process of applying the model is
technically fairly straightforward, although there are a number of aspects that need to be care-
fully considered in developing the required input data and interpreting the results. One is that
although a relatively small survey sample size (a few thousand respondents in the case of air
passengers and perhaps even fewer for airport employees) may be adequate to estimate an
airport access mode choice model, a much larger sample may be required for a given appli-
cation of the model, depending on the issues of interest and the desired level of geographic
resolution of the results. Two other considerations involve how to adjust the model to be able
to predict behavior in future years, as is typically needed for planning studies. The first of
these is how to adjust travel times and, particularly, costs to correspond to future conditions.
These adjustments will need to consider expected changes in highway congestion as well as
anticipated changes in real costs (in constant dollars) over time. The second consideration
involves adjustments for the effect of changes in the levels of real household incomes over
time. Implicit in the calibrated coefficients of a mode choice model are assumptions about
how travelers trade off time and cost. If real incomes change, these tradeoffs can be expected
to change as well. These issues are discussed in more detail in chapter two.
• Review of Literature
Given the importance of understanding air passenger airport ground access mode use it is not
surprising that there have been a number of studies over the years that have developed air pas-
senger ground access mode choice models. One of the earliest efforts to develop a formal model
of air passenger airport ground access mode choice was undertaken by Ellis et al. in the early
1970s. This study used a MNL model, as did several other studies that developed air passenger
ground access mode choice models over the next ten years. However, by the mid-1980s, it was
becoming recognized that some of the limitations of the MNL model could be addressed through
the use of NL models. One of the first applications of NL models to airport ground access mode
choice was undertaken as part of a study of surface access to London Heathrow Airport, followed
shortly thereafter by another study that used a NL structure to develop an integrated model of
airport choice and ground access mode choice for the San Francisco Bay Area. Subsequent air
passenger ground access mode choice models developed for Boston, Massachusetts; Portland,
Oregon; and airports in the southeast and east of England used a nested structure, whereas other
studies continued to use MNL models to represent air passenger ground access mode choice. In
addition to models that have exclusively addressed airport access mode choice, a number of
recent studies have used NL models to represent air passenger airport choice, with airport ground
access mode choice as a lower level nest. However, these models generally only include a single-
level nest for the airport ground access mode choice process and thus are equivalent to MNL
models from the perspective of ground access mode choice.
In addition to papers in the open literature, the synthesis project identified several studies
that had developed airport access mode choice models, the details of which had not been
widely reported, in several cases because the models had been documented in technical
reports that had restricted distribution or did not obviously involve airport access mode
choice. These included the regional travel demand model for the Atlanta region, a ridership

forecasting study for a proposed airport express train serving the two Chicago airports, a
travel demand forecast study for the planned Miami Intermodal Center, a ridership analysis
of a planned automated people mover connection between Oakland International Airport and
the nearby Coliseum station of the Bay Area Rapid Transit system, and a revenue and rider-
ship forecasting study for a proposed Air Rail Link between Toronto Union Station and
Lester B. Pearson International Airport.
• Use of Airport Ground Access Models in Airport Planning
To better understand the current state of practice with airport access mode choice models,
as well as to identify models that may have been developed for specific studies but not
reported in the published literature, a web-based survey was undertaken of airport authori-
ties, regional and state planning agencies, federal agencies involved in airport or surface
transportation planning, airport consulting firms, selected universities and other research
organizations, and relevant industry associations. The survey inquired about recent airport
ground transportation studies undertaken by the responding organization and whether these
involved the use of formal models of airport ground access mode choice. The survey also
inquired about respondents’ perceptions of the usability of such a model, as well as their
awareness of other organizations that have experience with the use of these models.
Survey responses were obtained from 105 different organizations. These responses identified
85 specific studies completed in the past ten years that had included some analysis for airport
access mode choice, of which 52 had involved the use of mode choice models. However, only
four of these studies were available on the organization’s website. The survey also asked about
prior experience with airport access mode choice models, and from these responses it does not
appear that there has been a significant increase in the use of analytical models in recent years.
Respondents who reported the use of analytical models of airport access mode choice were asked
to characterize the current state of practice with these models. Approximately 55% indicated that
current models were adequate for their needs, 35% reported that current models are not reliable
enough, 30% noted that they are too costly to use, and 10% indicated that they are too complex
to use. However, it is worth noting that only 5 of the 13 consulting firms reporting involvement
in studies using such models indicated that current models are adequate, whereas 7 of the 8 air-
port authorities and all 5 of the MPOs reporting the use of such models indicated that current
models are adequate for their needs. Because in many cases the actual modeling is done by con-
sultants rather than by airport authority or MPO staff, the limitations of the current models may
not be fully appreciated by the organization sponsoring the studies.
The survey also explored how airport trips were modeled in the regional travel modeling
process. Of the 23 MPOs responding to the survey, 15 (65%) reported using a special-generator
sub-model for air passenger trips, whereas about 50% reported using a special-generator sub-
model for airport employee trips. The other MPOs either treated airport trips the same way as
other regional trips or did not consider airport trips at all.
• Air Passenger Mode Choice Models
Although the details of the different air passenger airport access mode choice models iden-
tified in the literature review vary widely, it is clear that a standard of best practice has
evolved, although by no means is it always followed. This standard of best practice uses NL
choice models with separate coefficients (and possibly including different variables) for at
least four market segments:
• Resident business trips,

• Resident non-business trips,
• Non-resident business trips, and
• Non-resident non-business trips.

The modes available for resident and non-resident trips will generally be different
because non-residents do not have the option of parking a private vehicle at the airport
(indeed this would make no sense because their visit to the region begins at the airport and
they return to the airport at the end of their stay). On the other hand, many non-residents
rent a car on their arrival at the airport to provide local transportation during their stay in
the region.
Although no generally accepted practice has yet emerged for how to structure the nests of
a NL model, this should largely be determined by the characteristics of the different modes
because the primary purpose of using a NL model is to allow higher rates of substitution
between modes that have similar characteristics. Therefore, it would appear logical to group
private vehicle modes in one nest, with different parking options as a second-level nest, group
exclusive ride on-demand modes (taxi and limousine) together in a second first-level nest,
and group shared-ride scheduled modes (public transit and scheduled airport bus) together in
a third first-level nest, possibly with different transit options (e.g., rail and bus) as a second-
level nest. It is not clear where door-to-door shared-ride van should best fit in this structure,
as a separate mode at the top level, in the on-demand nest with taxi and limousine, or in the
shared-ride nest with the scheduled modes. This may be an issue to resolve empirically by
exploring which option gives the best fit to the data. Alternative access modes to scheduled
services can also be included as lower-level nests to each mode.
Rental car and hotel shuttle use by non-residents is best modeled outside this choice
process, because use of both modes is determined by factors that are largely independent of
the service levels of other modes. Rental car use is often determined by local travel needs
other than the airport egress and access trips. Therefore, visitors to the region may rent a car
even if they are staying at a nearby hotel that has a free shuttle to and from the airport.
The form of the utility functions for each choice alternative will generally be a linear com-
bination of explanatory variables with their associated coefficients. However, some variables
are best entered in the utility function as an inverse or ratio. For example, the service head-
way of scheduled modes, which is a direct measure of average waiting time, is the inverse of
the service frequency. The effect of household income may best be entered in the utility func-
tion by expressing direct travel costs as a ratio of the cost to some function of the per capita
or total household income. Thus, higher-income travelers will be less influenced by cost than
lower-income travelers.
• Airport Employee Mode Choice
Airport employee ground access and egress mode choice has received much less attention
in the literature than that of air passengers, and only three studies were identified that describe
an airport access mode choice model developed to account for airport employee access mode
choice behavior. These three models each adapted other journey-to-work mode choice mod-
els to predict airport employee mode use rather than developing an entirely new model from
airport employee travel data.
A special-purpose airport employee mode choice model was developed for the Greater
London region as part of the U.K. South East and South of England Regional Air Service
(SERAS) Study. This model was a fairly simple binary (two-mode) logit model that predicted
the percentage use of private vehicle and public transport and was based directly on one
developed for the South and West London Transport Conference for a study covering the area
to the south and west of London Heathrow Airport that was felt to provide a good basis for
the SERAS work. A study of the potential ridership on a proposed automated people mover
link between Oakland International Airport and a nearby rail transit station and a second
study of a similar link between San Jose International Airport and a nearby light-rail stop)
both used a similar approach of adopting model coefficients from regional travel demand

models for home-based work trips and then estimating alternative specific constants to cali-
brate the model predictions to survey data on airport employee travel.
The most common way to model employee travel to and from airports is to treat the air-
port in exactly the same way as any other transportation analysis zone in the regional travel
demand model and use the trip generation, trip distribution, and mode choice sub-models for
home-based work trips to generate the number of person and vehicle trips associated with air-
port employee travel. Those MPOs using special-generator models for airport employee
travel tailor this process to better fit the number of airport employee trips, typically through
the use of airport employment data and surveys of airport employee travel.
• Transferability of Airport Ground Access Mode Choice Models
Given the considerable cost and effort required to develop models of airport ground access
mode choice, it is natural to ask whether separate models need to be developed for every air-
port or whether it would be possible to adapt or apply a model developed for one airport for
use at another. Indeed, several of the existing models described in this report did just that. In
general, experience in applying models of transportation behavior in situations that are dif-
ferent from the one for which they were developed has not been very encouraging. However,
this experience has largely focused on models of general urban travel behavior, and airport
ground access travel behavior may be more consistent.
In principle, one would expect that air travelers would behave similarly when faced with
a similar choice situation, controlling for differences in ground access service characteristics
(e.g., fares or travel times) and differences in traveler characteristics (e.g., trip purpose and
duration, household composition, and income). Therefore, to the extent that a model accu-
rately reflects the effect of these variables, it should explain the behavior of air parties in other
geographical regions. However, this is a significant caveat, because many models are heav-
ily dependent on alternative specific constants that at best reflect a range of local factors that
are not explicitly included in the model and at worst correct for problems in the model spec-
ification. In particular, there may be regional differences in attitudes toward the service char-
acteristics of different modes as well as differences in the nature of the services offered.
One important consideration is the way in which different modes are included in NL mod-
els. Because the nesting structure of the limited number of models that have used a NL struc-
ture is very dependent on the particular modes that exist in the region being modeled, it is
difficult to generalize about how the various models have grouped the modes. It is unclear
whether differences in the nesting structure of different modes reflects fundamental differences
in choice behavior across different regions is a consequence of the modes included in the mod-
els or the explanatory variables used, or merely reflects different modeling philosophies by
different developers. The way in which household income is included (or not included) in the
models will also affect how well they can be expected to explain behavior in other regions
where the distribution of household incomes is different.
Given the current lack of consensus over model specification and typical coefficient values
between different airport access mode choice models, it can be assumed that the transferabil-
ity of these models is highly suspect. Although it seems plausible that the underlying airport
traveler behavior may not differ that much from region to region, after taking into account dif-
ferences in air passenger or airport employee characteristics and transportation system service
levels, it appears unlikely that current airport access mode choice models do this in a way that
is transferable to other regions, based on the significant differences between the different mod-
els. There is an urgent need to better understand how well current models reflect the factors
influencing the underlying travel behavior and how they can be improved to better reflect this
behavior, both because of the obvious value of being able to apply airport access mode choice
models in different situations from those for which they were originally developed, as well as

the concerns about the reliability of even applying them to different situations at the airports
for which they were developed.
• Airport Ground Access Models in the Regional Planning Process
Although modeling airport access mode choice by air passengers and, to a lesser extent,
airport employees, has largely been restricted to specialized studies addressing airport land-
side and system planning issues, there is a growing interest in explicitly modeling such trips
in the regional transportation planning process. A number of MPOs have begun to address air
passenger trips using a special-purpose mode choice model or special-generator sub-model
and a somewhat smaller number of MPOs have begun to do the same for airport employee
trips. However, the majority continues to model trips to and from airports as regular regional
travel using a standard trip classification such as home-based non-work trips. Because these
standard trip classifications encompass a very wide range of activities, most of which have
very little in common with airport travel, it would be surprising if the model components did
a very good job of predicting airport access mode choice.
This is compounded by the concept that air passengers in particular typically have access
to a much larger number of alternative modes for airport access and egress trips than are usu-
ally modeled in regional travel demand models, including taxi and limousine services,
shared-ride door-to-door van services, and scheduled airport express bus services. Further-
more, a significant fraction of all airport access and egress trips is made by visitors to the
region. Most current regional travel models are only designed to model travel by residents of
the region and largely ignore travel by visitors. Therefore, modes such as rental car and hotel
courtesy shuttle are typically not included in the models. Finally, the largest single air pas-
senger access or egress mode at many airports, for both visitors and residents, is being
dropped off or picked up by private vehicle. This option is also typically not explicitly mod-
eled in regional travel models. Models predicting private vehicle use that are based on the
assumption that the vehicle will be parked at the destination until the return trip will thus
underestimate the vehicle-miles of travel involved by a factor of two.
These concerns may not be particularly important in terms of total regional travel, because
airport trips comprise a fairly small fraction of all regional trips. However, these issues
become of much greater concern when the regional travel models are used to predict trips on
parts of the transportation network in the vicinity of the airport or are used for airport access
and egress studies, including predictions of airport access and egress trips for use in envi-
ronmental impact studies. Therefore, a fairly strong case can be made that airport access and
egress trips need to be modeled separately from general regional travel patterns (or at least as
a special-generator sub-model within the overall modeling framework) and then integrated
with other trips in the traffic assignment process.
This synthesis project examined some of the technical issues involved in modeling air-
port trips within the context of regional travel demand models and identified a range of
approaches that has been followed by different MPOs that have explicitly modeled airport
trips in their regional travel modeling process. These approaches vary from a special-purpose
sub-model within the travel modeling process of the Atlanta Regional Commission,
through the external generation of airport trip tables that are combined with the trip tables
generated by the regular travel modeling process of the Metropolitan Washington Council
of Governments, to two examples of the use of external airport access mode choice mod-
els. In the case of the Boston Central Transportation Planning Staff, an air passenger mode
choice model was developed in-house in cooperation with the airport authority. In contrast,
the Southern California Association of Governments uses a proprietary air passenger mode
choice model, the output of which is then used as input to the regular regional travel mod-
eling process. Further information on these four approaches is provided in chapter eight of
this report.

10
• Conclusions and Further Research
Airport ground access and egress mode choice models play a critical role in airport land-
side planning studies and modeling traffic on the regional transportation system in the vicin-
ity of airports. The ability to predict how air passenger and airport employee access and
egress mode use will change in response to changes in the airport landside access system or
other anticipated changes in the regional transportation system is essential to the proper eval-
uation of proposed measures and projects. However, these decisions are influenced by very
different factors from those affecting general regional travel patterns and the transportation
options available to airport travelers are often quite different from those for other types of
regional trip. Therefore, there is a need for specialized models that can represent these mode
choice decisions as well as the means to integrate these models or their output into the
regional traveling modeling process.
The development of air passenger ground access mode choice models has been the sub-
ject of ongoing research for more than 30 years. Over this period, the state of practice has
slowly evolved from relatively simple MNL models to more complex NL models involving
several levels of nesting and four or more market segments. However, no clear consensus has
yet emerged as to what explanatory variables should be included or how the various modes
and sub-modes should be nested. In addition, even the most recent models have still not
addressed a number of problematic issues in a meaningful way. These include how to treat
rental car use by non-residents of a region and how best to account for the role of traveler
income in the mode choice process. Aside from these technical considerations, there has been
almost no attention given to how reliably existing models predict air passenger access mode
use when they are used to predict mode use under very different conditions from those pre-
vailing when they were developed, including changes in the physical infrastructure, ground
transportation services, and household income levels. There is an urgent need for more
research into these specific aspects, as well as continuing research directed at improving the
current state of practice.
In contrast to air passenger mode choice models, there has been very little effort directed
at developing airport employee mode choice models. The majority of MPOs model airport
employee trips the same way as they model any other journey-to-work trips. The develop-
ment of better airport employee access models is a promising research opportunity.
Finally, many existing regional travel models do not explicitly model airport trips, but
treat them as general regional travel. Because of the unique characteristics of airport travel
and the range of transportation options typically available at airports, this is likely to give
fairly poor predictions of airport mode use and the resulting vehicle trips. Further research is
needed to explore how well existing regional travel models account for airport trips and to
provide guidance on how best to implement explicit modeling of airport access mode choice
in the regional travel modeling process.

11
CHAPTER ONE
INTRODUCTION
BACKGROUND funding. As a result, no clearly recognized body or practice

has emerged and relatively few people have worked in this
The prediction of air passenger and airport employee mode area long enough to develop significant expertise. Typically,
choice decisions for their travel to and from the airport forms airport access mode choice models have been developed by
a key analytical component of airport landside planning, as transportation modeling specialists, airport planners, or acad-
well as airport system planning. There is currently however emic researchers who are mainly concerned with other issues.
no well-accepted and validated process for modeling how Part of the reason for this is that opportunities to develop such
airport users will change their access or egress mode in models occur relatively infrequently. At the same time, mod-
response to changes in the airport ground transportation sys- eling airport ground access mode choice is arguably more
tem (such as changes in fares, rates, or service levels) or the complicated and challenging than general urban travel. First,
introduction of new modes (such as the extension of a light airport ground access usually involves more transportation
rail system to the airport). The factors affecting airport travel modes than are typically considered in general urban travel
are recognized as being significantly different from those demand models, which are commonly restricted to single-
affecting typical trips accounted for in regional transporta- occupant automobiles, carpools and vanpools with varying
tion models, as a result of both the characteristics of the travel occupancy levels, and conventional transit services. The air-
party and the timing and duration of the trip, as well as the port ground access system includes all of these, as well as
differences in available modes and services. Therefore, it is other modes that are rarely considered in general urban travel
difficult to determine the economic feasibility of proposed models, such as rental car, taxi, shared-ride van, dedicated
projects to improve airport ground transportation or effec- express bus services, and hotel courtesy shuttles, all of which
tively manage the existing airport ground transportation sys- can have a significant mode share at large airports. Indeed, at
tem using traditional regional models. many airports the rental car is the largest single mode after
private vehicles, and the other shared-ride transportation
Travel forecasting models, particularly those developed to modes typically carry more passengers than transit.
address airport ground access and egress trips, are highly
specialized and not well understood by most airport man- Second, the circumstances surrounding airport access
agers and planners. With increasing emphasis on intermodal trips by air passengers are very different from other types of
connections, there is a pressing need for more widely urban travel. A high proportion of travel is by visitors to the
accepted and accessible reference material and information region; passengers travel in groups more often than for other
on such models. This synthesis project was undertaken to urban trips, they make airport access trips far less frequently
update and extend previous efforts to document the state of than other urban trips, they commonly have luggage, and
practice for airport ground access mode choice models. It they are often away for several days or even several weeks,
also examines the characteristics of existing models and dis- which affects the cost and feasibility of parking a car at the
cusses the issues involved in the development and use of airport while they are away. Picking up and dropping off air
such models to improve the understanding and acceptance of passengers at the airport with a private vehicle is one of the
their role in airport planning and management. It is hoped most common access and egress modes. This generates twice
that the report will also serve to focus research and develop- as many vehicle trips to and from the airport as air party trips
ment efforts on improving the state of the art of modeling air- owing to the round trip involved in either dropping off or
port ground access mode choice by assembling a detailed picking up air travelers. These factors greatly complicate the
database of comparative information on recent model devel- issues that need to be considered in developing airport access
opment efforts and identifying technical issues that are not mode choice models.
currently well handled.
In contrast to the current state of practice of urban trans- METHODOLOGY

portation modeling, which has been the subject of study and
improvement by a large body of practitioners and researchers A comprehensive review of the relevant literature in the
for many decades, airport ground access mode choice model- field was undertaken. This was supplemented by an exten-
ing has received only limited attention and relatively little sive survey of airport authorities, metropolitan planning

12
organizations (MPOs), consulting firms and research orga- REPORT AUDIENCE

nizations, and other government agencies and industry
organizations to document the extent of the recent use of The information presented in this report may be of interest to
airport ground access mode choice models and to identify a range of airport managers and planners, transportation
sources of technical information on existing models. Based planning professionals, researchers, and others interested
on the responses to the survey, technical documentation on in modeling airport ground access mode choice. Airport
existing models has been assembled and reviewed, supple- managers may be interested in obtaining a fairly high-level
mented with follow-up communication by telephone and understanding of the current state of practice and an appreci-
e-mail where necessary. The findings of this review were ation of where and how such models can be used, as well as
then assembled into a project database that was used to what is involved in using them. Such readers may find the
prepare summary descriptions of each model in a standard executive summary sufficient or they may prefer to go into
format. more detail in selected sections of the report, particularly the
description of the mode choice modeling process in chapter
The survey of airport authorities and other organizations two. Airport planners and regional transportation planners
was undertaken to complement the review of published liter- who are more directly involved in airport landside planning
ature as well as to obtain information on the extent to which or regional travel demand modeling but do not have exten-
each organization makes use of formal models of airport sive experience with airport access mode choice modeling
access travel and perceives the ease of use or reliability of may find the description of the mode choice modeling
such models. The survey was undertaken using Internet sur- process in chapter two of interest, as well as the technical
vey software hosted on the TRB website. E-mails were sent details of existing models in the following chapters. Regional
to survey recipients requesting their participation and pro- transportation planners may find the discussion of the role of
viding a link to the appropriate survey. Four somewhat airport ground access mode choice modeling in the regional
different survey questionnaires were developed reflecting the travel demand modeling process in chapter eight of particu-
different concerns and needs of the various types of organi- lar interest. Consultants and transportation modeling spe-
zations included in the survey. Survey recipients were given cialists who have prior experience with airport access mode
the option of responding using a form that could be com- choice models may find the comparative discussion of the
pleted off-line and returned by e-mail, fax, or mail. various existing models of interest, together with the review
of their strengths and weaknesses contained in the detailed
case studies in Appendix D (which is found in the online ver-
sion only). Finally, researchers interested in issues involving
POTENTIAL USES OF AIRPORT ACCESS
MODE CHOICE MODELS
airport ground access mode choice may be interested in the
discussion of the aspects of the current state of practice that
Airport access mode choice decisions by airport travelers, are not well understood and the associated suggestions for
including both air passengers and airport employees, impact future research in this topic that appear throughout the report.
a wide range of airport planning and operational manage-
ment decisions, including the development of landside
facilities, airport revenue from parking and other ground REPORT ORGANIZATION
transportation services, and programs to reduce the growth
in vehicle trips generated by the airport and the associated The remainder of this report consists of eight chapters and
emissions. Potential uses of models that can predict the four appendixes. Chapter two describes the airport access
effect on access mode use of proposed changes to the system mode choice modeling process and discusses a range of
include sizing new planned facilities, evaluating the finan- issues that arise in modeling airport access mode choice.
cial implications of proposed changes in parking rates or Chapter three reviews the published literature on air passen-
other ground transportation fees, determining the expected ger airport access mode choice models and addresses airport
air quality impacts of either planned new facilities or pro- employee mode choice for journey-to-work trips. Chapter
posed mitigation measures, and assessing the feasibility of four reviews the use of airport ground access mode choice
proposed projects to improve airport access. Airport acces- models in airport planning based on the findings of the
sibility is also a major determinant of airport choice in multi- survey of airport authorities, regional and state planning
airport regions or situations where air service competition agencies, and other organizations. The chapter examines the
exists between local and more distant airports. Airport extent to which formal models of air passenger or airport
access mode choice is often embedded within models of air- employee ground access mode choice have been used in air-
port choice, because the availability and use of different port planning studies, as well as how airport access trips are
access modes affects the perceived accessibility of different addressed in regional transportation planning models.
airports, and improved representation of airport access mode
choice behavior should benefit those applications as well. Chapter five presents the current state of practice of air
Several of these potential applications are discussed in more passenger ground access mode choice models and forms the
detail in chapter two. core of this synthesis review. The chapter discusses the type

13
of planning issues to which the models have been applied, the Chapter seven addresses the extent to which air passenger
technical approach adopted to represent air passenger choice ground access mode choice models may be transferable to sit-
behavior, the ground access modes included in the models, uations different from the one for which they have been origi-
the explanatory variables used to model the transportation nally developed, or whether it is necessary to custom build
system and air passenger characteristics, and the market seg- ground access mode choice models for every airport. Chapter
mentation that has been used to improve the ability of the eight discusses the issues that arise in integrating air passenger
models to account for air passenger ground access behavior. and airport employee ground access mode choice models into
It also discusses a number of considerations that arise in more general regional transportation planning models.
developing and applying these models, as well as issues of
model performance. Finally, chapter nine presents the conclusions of the syn-
thesis of current practice and recommendations for further
Chapter six reviews the corresponding state of practice research. These address needed improvements in the current
with the analysis of airport employee journey-to-work travel. state of practice for air passenger ground access mode choice
In contrast to air passenger ground access, the literature models as well as the development of appropriate models for
review and survey of airport authorities and other organiza- airport employee mode choice for journey-to-work trips.
tions identified relatively few formal mode choice models
explicitly developed to represent airport employee journey- The four appendixes provide more detailed information
to-work travel. Rather, analysis of airport employee travel in as background to the discussion in the body of the report.
airport planning studies tends to be based on extrapolation Appendix A presents the questionnaires used in the surveys
from survey data of existing travel patterns, although of airport authorities, planning agencies, and other organiza-
regional planning studies typically use models designed for tions; Appendix B contains a detailed summary of the survey
general urban travel. findings; Appendix C lists the organizations that responded
to the survey; and Appendix D (provided only online) con-
The next two chapters explore two issues that arise from the tains detailed technical summaries of nine of the mode choice
review of current practice described in the previous chapters. models identified in the project.

14
CHAPTER TWO
AIRPORT GROUND ACCESS MODE CHOICE MODELING PROCESS
This chapter provides a short introduction to the motivation In some cases, an airport may want to understand the
and methodology for modeling airport ground access mode consequence of decisions that are largely outside its con-
choice. It is primarily intended to give airport management trol, such as changes in the price or service pattern of
and planning staff some background on how these models ground transportation services operated by other entities or
can be used in the planning and decision-making process, changing congestion levels on the regional transportation
and provide an overview of the technical issues involved in system. In other cases, it may want to understand how deci-
developing and applying these models so that they can inter- sions that it is considering will affect the airport ground
act with more technical specialists in an informed way and transportation system or be affected by it. Examples of such
properly supervise contracts for the development and use of decisions could include proposed changes in parking rates
such models. It is also hoped that it will be helpful to airport for airport-operated parking lots, the introduction of a shut-
planning consultants and other planning specialists, includ- tle bus link to a nearby rail station, or the construction of a
ing transportation planners working for MPOs and other consolidated rental car facility some distance from the air-
planning agencies, who do not have particular expertise in port terminal. Each of these decisions is likely to result in
airport ground access modeling, but become involved in air- changes in the relative attractiveness of the different access
port ground access modeling issues. and egress modes (including the different on-airport and
off-airport parking facilities) and resulting shifts in the pro-
As will be apparent from the detailed discussion in this portion of airport travelers using each mode, as well as the
report, airport ground access mode choice modeling is a highly effect of this on the revenue generated by each facility or
specialized field, with many complex aspects that make this a service. Finally, the requirement to predict the environ-
particularly challenging problem. This has two implications. mental impacts of decisions that change or affect the airport
The first is that the development of these models typically ground transportation system, particularly the need to pre-
requires the use of specialists with prior experience in model- dict air quality impacts, demands an ability to assess how
ing airport ground access mode choice. The second is that it is those decisions will affect the use of the various modes
therefore helpful if other planners and managers involved in (Gosling 2005). This can be a particularly critical issue for
the larger planning process have some idea of what is involved airports in regions that are not in attainment of the National
in developing and using these models, so that they have Ambient Air Quality Standards, where it may be necessary
appropriate expectations of both the resources required and to be able to demonstrate that a proposed project will not
what it is reasonable to be able to accomplish with a given increase total emissions or that appropriate mitigation mea-
level of resources. sures will offset any increase in emissions from the project.
Indeed, it may be necessary to be able to demonstrate a
MOTIVATION FOR MODELING AIRPORT
reduction in total emissions.
GROUND ACCESS MODE CHOICE
Prediction of how changes in the airport ground trans-
Airport travelers make use of a wide variety of different portation system will influence the mode choice decisions of
modes for their ground access trips to and from the airport, airport travelers is complicated because those decisions
including private vehicles, rental vehicles, taxis, and multi- depend not only on the price and level of service of the alter-
ple private and public transportation services. Large airports native modes but also on the characteristics of the individual
in particular are served by a large number of different ground travelers. In the case of air passengers, these characteristics
transportation modes and services. Planning landside and air- include their trip purpose, whether they are residents of the
port ground access facilities, as well as accounting for the region or visitors to it, how long residents of the region will
environmental impacts of airport ground transportation be away from home on their trip, and whether visitors to the
activities, requires the ability to predict how airport users will region will need a rental car for local travel during their visit.
change their ground access and egress decisions in response The distribution of these characteristics across the population
to changes in the array of options that they face. It would be of airport travelers not only varies seasonally but also in
impractical to imagine that the proportion of airport users response to external influences, such as currency exchange
choosing a particular mode will remain constant when the rates and the state of the regional economy, and changes in
factors influencing their choices are continuously changing. the air services offered at the airport.

15
Given the many different factors influencing the propor- much-needed terminal modernization and expansion pro-
tion of airport travelers using each mode, it is unrealistic to gram. The airport planning staff has suggested that as part of
expect planners and decision makers to be able to make the terminal modernization program the airport construct an
quantitative estimates of the effect on mode use of any given automated people mover (APM) link to the nearest station
change in the system without the use of formal analytical on the regional light-rail system, located approximately one
tools that model how airport users respond to changes in the mile from the airport, to replace the existing infrequent tran-
available airport ground transportation services. Thus, air- sit bus service between the airport and the light rail system.
port ground access mode choice models provide the basic However, there are concerns about the capital and operating
input to other analysis tools that are used to support airport cost of the proposed people mover link and the likely use it
landside and ground access planning and decision making, will attract. As part of the landside planning for the modern-
such as traffic flow models, simulation models, or financial ization program, the airport decides to undertake a feasibil-
planning tools. ity study of the APM link in cooperation with the regional
transit authority. The scope of work for the feasibility study
includes the development of a ground access mode choice
Representative Applications model that can be used to evaluate the likely increase in the
use of the light rail system for airport trips if the APM is
To illustrate the type of decision that can benefit from the constructed, as well as explore alternatives such as an
availability of an airport access mode choice model, this sec- enhanced shuttle bus connection to the light rail system or an
tion describes three representative situations where such a express bus service to several off-airport terminals with
model could be applied. remote parking.
Airport A is experiencing steady growth in air traffic and The MPO for the region served by Airport C has recog-
is planning to construct a new passenger terminal alongside nized for some time that its regional travel demand model
the existing terminal. The existing surface parking lot in front significantly underestimates vehicle trips to and from the air-
of the current terminal is already reaching capacity at peak port when compared with the vehicle traffic volumes pro-
periods and there is no space between the terminal roadways jected in landside planning studies undertaken by the airport
and the site of the new terminal to significantly expand the authority and environmental documentation prepared for air-
parking lot. The terminal redevelopment plan envisages con- port projects. A detailed review of the travel demand model
structing a multi-level parking structure on the site of the and discussions with airport planning staff reveal that the
existing surface lot. To pay for the parking structure, the air- underestimate is the result of two factors. The first is that
port authority is considering raising the daily parking rate the standard trip generation and attraction relationships in the
once the new structure is in operation. However, it is con- model substantially underestimate the number of air passen-
cerned that this will reduce the parking demand for the new ger trips to and from the airport owing to the absence of
structure and divert air passengers to the surface long-term relevant variables in the model and that the modeling process
parking lot some distance away, privately operated off- does not explicitly consider visitor trips to the region. Sec-
airport parking facilities, or even convince some air passen- ond, the mode choice model in the overall modeling process
gers who might otherwise park at the airport to be dropped underestimates vehicle trips and produces poor estimates of
off and picked up instead, or make use of other modes such the traffic composition by not accounting for several key
as a shared-ride van. It recognizes that the extent of any such characteristics of airport ground access and egress travel.
diversion will depend on the new rate charged for the park- These issues include not accounting for the two-way trips by
ing structure, any change in the rate for the use of the long- private vehicles picking up or dropping off air passengers,
term lot, and whether the off-airport parking lot operators underestimating the proportion of shared-ride automobile
also adjust their rates. To study the effect of different rate trips by ignoring that many air travel parties have more than
structures on the demand for parking at both the new parking one person, and those dropped off or picked up by private
structure and the long-term lot, and the implications for the vehicles necessarily involve a shared-ride trip in one direc-
financial feasibility of the new parking structure, the airport tion, not considering taxi and limousine trips, and treating
retains a consultant to develop an airport access mode choice vehicle trips by public modes such as a shared-ride van as
model that can be used to analyze the impact of different rate part of regular public transit use.
structures on parking demand and identify the optimal pric-
ing strategy and resulting parking demand in both facilities. To improve the ability of the regional travel demand
model to estimate airport trips the MPO decides to develop a
Airport B is in a non-attainment region for several ambi- special-generator sub-model for airport access and egress
ent air quality standards and is under pressure from the local trips. This sub-model combines a trip generation module and
Air Resources Board to reduce the emissions from airport a mode choice module and generates vehicle trip tables for
ground access and egress travel. The airport authority rec- each category of vehicle trips included in the regional model
ognizes that unless it can show significant progress at that are combined with the trip tables for general regional
reducing emissions, it is unlikely to be able to undertake a travel before the traffic assignment stage of the regional

16
model. The trip generation module converts airport passenger functional form is modified and new coefficients are esti-
traffic forecasts into estimates of person-trips between the mated, hopefully improving the fit of the model. The process
airport and each regional travel analysis zone, whereas the continues until a satisfactory model is obtained. Changes to
mode choice module converts these estimated person-trips the functional form of the model that are typically explored
into the corresponding vehicle trips for use in the regional include adding or dropping independent variables, changing
travel demand model. the way that an independent variable is defined or appears in
the model, or segmenting the data so that different coefficient
values are obtained for different subsets of the data, or dif-
Airport Choice ferent independent variables or model functional forms are
used for different subsets of the data (e.g., developing sepa-
The need to model air traveler choice of airport in multi-
rate models for business and non-business travel).
airport regions or situations where air service competition
exists between local and more distant airports is becoming
In the case of airport access mode choice models, there are
increasingly prevalent in airport system planning studies, par-
two types of data that are required to estimate a model. The
ticularly as low-fare carriers introduce service at secondary
first type of data consists of the access mode choices made by
airports often some distance from congested major hubs. It is
a representative sample of airport users, together with ex-
recognized that airport accessibility is a major determinant of
planatory data about their characteristics (such as their
airport choice, because the availability and use of different ac-
household income) or the characteristics of their trip (such as
cess modes affects the perceived accessibility of different air-
where they commenced their journey to the airport or the
ports. In consequence, the airport access mode choice process
purpose of their air trip). These data are typically obtained
is often embedded within models of airport choice and im-
from surveys of air passengers or airport employees, as the
proved representation of airport access mode choice behavior
case may be. The second type of data consists of the trans-
should benefit those applications as well.
portation characteristics (such as travel time and cost) of the
various access modes between which the mode choice deci-
OVERALL MODE CHOICE MODEL sion was made. Because the mode choice decision depends
DEVELOPMENT PROCESS not only on the transportation characteristics of the mode
actually chosen, but also on the characteristics of the modes
This section presents an overview of the process of develop- that were not chosen, it is generally necessary to obtain the
ing an airport access mode choice model. The application of transportation characteristics for all the modes included in
such a model to support planning or decision-making activi- the model corresponding to each airport user included in the
ties is discussed in a subsequent section. In broad terms, the estimation dataset. This can be a significant amount of work
general approach to developing a mode choice model is no and is discussed further later.
different from the development of any other mathematical
model of a physical or behavioral process. A set of data is The overall process of developing an airport access mode
assembled that describes the process being modeled. Then, a choice model is summarized in Figure 1 and discussed in
suitable functional form for the mathematical model is more detail in the following sections.
defined that expresses the value of the variable that the model
is intended to predict in terms of some other explanatory
variables and model coefficients (sometimes referred to as Aggregate Versus Disaggregate Models
parameters), the values of which are to be determined from
the data. Statistical model estimation software is then used to There are two broad types of behavioral models such as a
estimate the values of model coefficients that best explain the transportation mode choice model. An aggregate model
observed values of the variable that the model is intended to attempts to predict the value of an attribute of a group of
predict, termed the dependent variable, using the observed decision makers, such as the percentage of air passengers
values of the explanatory variables, termed the independent from a given origin zone choosing a particular mode. In con-
variables. The extent to which the model is able to reproduce trast, a disaggregate model attempts to predict how an indi-
the observed values of the dependent variable for any given vidual decision maker with a given set of characteristics will
set of values of the independent variables is referred to as the behave. In the context of airport access mode choice, such a
goodness-of-fit of the model and is an important measure of model typically attempts to predict the probability that a
the usefulness of the model. given airport user will choose a particular mode, because two
airport users with apparently identical characteristics may
As with most model development efforts, the process is well choose different access modes. If the probability of
usually iterative. An initial functional form for the model is choosing each access mode is estimated for each airport user
proposed. The values of the model coefficients are estimated in a given sample of users, then the percentage of any group
and the fit of the model to the data examined. Then, based on of such users choosing a given mode can be calculated.
how well the model appears to fit the estimation dataset and Because a disaggregate mode choice model predicts the
the reasonableness of the estimated coefficient values, the probability of a decision maker choosing a given mode from

17
Assemble
Airport-User
Transportation
Survey
Service Data
Airport-User Transportation
Survey Data Service Data
Model Specification
Model
Estimation Model Estimation
Dataset
Review
Goodness of Fit
Assemble Model
Calibration Calibration Model Calibration
Data Dataset
Assemble Model
Validation Validation Model Validation
Data Dataset
Model Application
FIGURE 1 Mode choice model development process.
among a defined set of alternatives, these models are also the estimation data give the access mode actually chosen, the
referred to as discrete choice models. statistical approach to estimating the model coefficients typ-
ically uses a maximum likelihood method rather than the
Because the airport access mode choice decision depends more common statistical regression methods. The maximum
on the characteristics of each individual travel party as well likelihood method determines the value of the model coeffi-
as the transportation characteristics of the different modes cients that maximize the likelihood that the model will pre-
faced by that travel party, which in turn vary with the travel dict the mode actually chosen by each decision maker in the
party characteristics (e.g., the travel costs typically vary with sample, where this likelihood is defined as the overall proba-
the access trip origin and the travel party size), most airport bility that the model predicts the actual mode choices (this is
access mode choice models take a disaggregate approach. simply the product of the predicted probabilities of choosing
Indeed, it arguably impossible to develop a reasonable the mode actually chosen across all the decision makers in
aggregate airport access mode choice model that adequately the sample). Readers interested in more background on the
reflects all the relevant variables. mathematical details of estimating disaggregate choice mod-
els are referred to standard texts on the subject, such as
However, the decision to develop a disaggregate model Ben-Akiva and Lerman (1985) or Train (2003).
imposes a number of constraints on the form of the model
and the approach used to estimate the model coefficients. A It should be noted that each data point (or record) in a
mathematical form must be chosen for the model that pre- dataset used to estimate a disaggregate choice model repre-
dicts the probability of a given air travel party or airport sents a single choice decision. In the case of airport access
employee choosing a given access mode. Suitable mathe- mode choice, air passengers often travel as a party of more
matical forms for the model are discussed in the next section. than one person and the data point is the travel party, whether
Because the model predicts the probability of a given deci- composed of one or more air passengers. Each travel party is
sion maker choosing each of the available access modes but considered to make the mode choice decision as a single unit,

18
whether or not the decision is made jointly by the members on the day of the survey. In addition, the survey will obtain
of the party or by one individual within the party. In contrast, information on respondent characteristics that may be used as
airport employees are usually assumed to make airport ac- independent variables in the mode choice model or to segment
cess mode choice decisions individually, even if they decide the data for model estimation.
to travel to the airport in a group (e.g., a car pool).
The survey will also need to identify where the access trip
A further complication with airport ground access travel began to determine the transportation service characteristics
by air passengers that is commonly ignored in developing of the different modes that were faced by the respondent. The
mode choice models is that the ground access travel party more accurately the trip origin location can be determined,
and the air travel party may be different. For example, two the more precisely the corresponding transportation service
colleagues making a business trip together may travel sepa- levels for the various modes can be estimated. Because the
rately to the airport and then meet up at the airport and fly data on transportation service levels are typically assembled
together to their destination. Conversely, situations may arise on a zonal basis, the trip origin locations need to be expressed
in which the members of a ground access travel party take in terms of the system of analysis zones used for the trans-
different flights after they reach the airport, such as attendees portation service data. The issues involved in selecting an
at a conference who decide to share a taxi to the airport. In appropriate system of analysis zones are discussed further
these cases the unit of decision is the ground travel party, not later. However, with the possible exception of U.S. Postal
the air travel party. Therefore, the access mode choice model Service zip code areas, most practical systems of analysis
should strictly predict the mode choice decision of each zones will not be such that respondents can be expected to
ground travel party, rather than each air travel party. Shared- know which zone their trip began in. Some respondents may
ride modes such as a door-to-door van will typically combine even have difficulty with zip codes. Therefore, the usual
more than one ground travel party in a single vehicle, al- approach is to attempt to obtain the street address of the trip
though each ground travel party is considered to make a sep- origin or the name of a specific origin location, such as a
arate decision, because its members could have chosen a hotel, for which the address can be obtained later. For pri-
different access mode without affecting the decisions of the vacy reasons, the exact street address is not required and
others on the vehicle. In practice, any error that would be typically the block number or a nearby street intersection is
introduced in an airport access mode choice model by ignor- considered adequate. These addresses can then be geocoded
ing the distinction between the ground access travel party and and later assigned to the appropriate analysis zone.
the air travel party is likely to be very small, because the dis-
tinction is only relevant for a small proportion of air travel- The respondent characteristics that are used as indepen-
ers. However, it is a point that should be borne in mind when dent variables or to segment the data will depend on the func-
designing air passenger surveys that will collect data for use tional form of the model. Because this is typically not known
in developing airport access mode choice models, because it at the time that the survey is conducted, but evolves during
can affect the way that questions are worded. the model estimation process, the survey should attempt to
collect information on those respondent characteristics that
are believed to influence the mode choice process, even if
Model Estimation Data some end up not being used in the model. Although there is
a small cost to collecting data that are not used, if the infor-
As noted earlier, model estimation data are obtained in two
mation is not collected, one will never know how important
different ways:
it might have been in the model. The following is list of air
passenger or airport employee characteristics that have either
1. Surveys of air passengers or airport employees, as the
been shown to influence airport access mode choice or might
case may be; and
reasonably be expected to do so.
2. Assembly of transportation service data for each mode
from operator and agency records and data files, pub-
• Air Passengers
lished information, and other sources.
– Essential
䉭 Trip purpose (business vs. personal)
Survey Data 䉭 Resident of region or visitor

䉭 Primary ground access mode
The air passenger or airport employee surveys obtain informa- 䉭 How accessed primary mode (where relevant)
tion on the access mode used for a specific trip, typically the 䉭 Where parked (if relevant)
most recent one. In the case of air passengers who are surveyed 䉭 Trip origin location (address, hotel name, etc.)
at the airport, this will usually be the access trip that they have 䉭 Trip origin type (residence, hotel, etc.)
just completed. In the case of airport employees, the survey 䉭 Number of air passengers in travel party
may ask about access trips over a recent period, such as the 䉭 Air trip duration (nights away on trip)
previous week, to identify variation in access mode use from 䉭 Household income
day to day or to reduce the effect of some special circumstance 䉭 Household size.

19
– Potentially useful (used in some models) an ACRP study currently underway (Project ACRP 03-04) is
䉭 Amount of checked baggage developing a “Guidebook for Airport-User Surveys,” which
䉭 Number of air trips from airport in past year will provide detailed guidance on these and other related
䉭 Whether trip costs paid by employer or client issues.
䉭 Time arrived at airport
䉭 Gender of respondent.
• Airport Employees Stratified Sampling
– Essential
Because the use of different transportation modes for airport
䉭 Primary ground access mode
access varies widely, and in particular at many airports pub-
䉭 How accessed primary mode (where relevant)
lic transportation has a fairly small market share, surveying a
䉭 Where parked (if relevant)
random sample of air travelers (e.g., in airport terminal
䉭 Monthly parking cost (if any)
departure lounges) will result in relatively few respondents
䉭 Whether any travel costs paid by employer
Δ Trip origin location (home address)
who used the less common modes. This will in turn adversely
affect the ability of a model estimated on that data to explain
䉭 Work location on airport
the choice behavior of the respondents using those modes.
䉭 Times shift starts and ends
One approach to overcoming this problem is to perform
䉭 Variability of shift times
stratified sampling, in which the survey is done in a way that
䉭 Household income
obtains a greater number of responses from a particular sub-
䉭 Household size.
set of travelers than would be expected from a truly random
– Potentially useful
sample. For example, surveys could be performed where pas-
䉭 Job type/classification
sengers alight from transit vehicles. Alternatively, the survey
䉭 Employer
could be done in a location where respondents selected at
䉭 Number of automobiles owned by household.
random will intercept all types of travelers, but a screening
question asked at the start of the survey will identify respon-
Some of the characteristics may appear directly in the
dents in the subset of interest and the survey of those respon-
mode choice model, whereas others are needed to deter-
dents will be performed in more depth.
mine the appropriate travel costs of the various modes. In
the case of air passengers, the number of air passengers in
Estimating mode choice models on the basis of data ob-
the ground access travel party affects the relative cost of
tained from a stratified sample requires some adjustments to
different modes, whereas the duration of the air trip affects
the standard model estimation techniques to weight the
the cost of parking for residents of the area. For those
response data in the estimation process; however, these ad-
modes (such as rail transit or express airport bus) where a
justments are discussed in most standard textbooks on dis-
secondary access mode is needed to reach the primary
crete choice modeling. Similarly, in presenting the results of
mode, the access mode used will affect the cost and time of
a stratified sample, it is important to weight the responses
the trip. Some mode choice models may even model this
appropriately to reflect the correct distribution of respondent
access mode choice.
characteristics across the population of airport users as a
whole. Developing appropriate response weights for both
Accounting for income effects in mode choice models is
model estimation and presenting survey results requires data
problematic, as discussed elsewhere in this report, and many
on the frequency of occurrence of the different categories of
past models have ignored the issue entirely. However, clearly
respondent. These weights can be obtained from actual traf-
income must have some effect on airport access mode
fic counts of the different respondent categories where these
choice. Household income may be a more appropriate mea-
are available or by comparing the results of the stratified
sure than individual income, because this reflects the contri-
sample with that of a random sample of the larger population
bution of other members of the household in covering basic
of air passengers or airport employees as the case may be.
household costs; however, at the same time the discretionary
income for a given level of household income will also
depend on the size of the household. Transportation Service Data
As important as deciding what information should be col- To calculate the travel times and costs faced by each travel
lected in the survey is deciding how to word the questions. party in using the different airport ground transportation ser-
Poorly worded questions will produce unreliable data, be- vices, it is necessary to assemble the transportation service data
cause respondents may misunderstand the question and give for each mode on the basis of a defined system of analysis zones
an incorrect answer. In the case of self-completed surveys, so that the appropriate value of any particular service charac-
this applies not only to the questions themselves, but also to teristic for a given travel party can be determined from the trip
any predefined response options that are provided. The de- origin zone for that party. Some data, such as the highway
sign of survey questionnaires and question wording is a topic travel time to the airport, will vary across the zones. Other data,
in its own right and beyond the scope of this report. However, such as daily airport parking rates, are independent of the trip

20
origin zone and thus constant for every zone, although the cost However, a large metropolitan region may easily have
for parking will depend on how long the vehicle is parked. more than 1,000 TAZs, and some larger regions have signif-
Some costs, such as fares for some public transportation ser- icantly more than that. The current regional travel demand
vices will depend on both the origin zone and the number of model for the San Francisco Bay Area utilizes 1,454 TAZs,
people in the travel party. whereas that for the Washington metropolitan area utilizes
2,191. For those travel variables that can be obtained directly
Highway travel times, highway distances, and transit from the regional transportation modeling datasets, such as
travel times and fares are typically available from the re- highway times or transit fares, this is not a particular prob-
gional transportation planning agency. However, modes such lem. However, for other modes, such as shared-ride van,
as taxi, shared-ride van, and scheduled airport bus, which are where fares are not typically available from the regional
not usually included in the regional travel demand model, datasets, or are usually expressed in terms of the TAZs,
will require some work to assemble the necessary data. The obtaining the relevant data and converting it to the TAZs can
ground transportation information pages of the airport web- be a major task.
site may have current information on many of the ground
transportation services and parking rates. Otherwise, it may It might appear that the data management problem can be
be necessary to contact the individual operators to obtain fare reduced somewhat by using larger zones, such as U.S. Postal
and schedule information. In the case of scheduled services Service zip code areas. However, apart from the loss of pre-
with defined stops or stations, the schedule will generally cision involved in using larger zones, most transportation
provide headways and travel times from each stop or station. service data are not available on a zip code basis anyway, so
It will be necessary to determine the analysis zone for each the work involved in converting the data to a zip code-based
stop, as well as the analysis zones served by that stop and the system may not be significantly less than using TAZs. Where
transportation service characteristics (travel times and costs) data are available by zip code (e.g., some shared-ride van
for the secondary access trip to reach the stop from each zone operators base their fares on zip codes), it is fairly easy to
served by the stop. Because there may be several alternative develop a mapping from zip code areas to the corresponding
secondary access modes (such as walk, drop off by private TAZs and convert the data to a TAZ basis. Most geographi-
vehicle, taxi, or public transit), the transportation services cal information systems have functions that can do this auto-
characteristics will have to be determined for each mode and matically provided that the TAZ boundaries are available as
each analysis zone. a geographical information systems file.
The result of this process is a large data table with a row

(or record) for each analysis zone and the transportation ser- Revealed Versus Stated Preference
vice variables for each mode and sub-mode forming the
columns (or position in the record). The appropriate trans- There are two different approaches to assembling the neces-
portation service values can then be assigned to each travel sary data on airport access mode choice behavior to develop
party in the model estimation dataset by looking up the rele- mode choice models: revealed preference and stated prefer-
vant data for the trip origin analysis zone for that party and ence surveys. Revealed preference surveys identify the travel
computing the values where necessary to account for the choices actually made by airport travelers, as well as collect
travel party size, air trip duration, or other travel party char- information on other traveler characteristics and details of
acteristics that affect the value of the transportation service the trip that are believed to influence the choice. The model
variable for that party. estimation process attempts to develop a model that explains
the mode choice decisions in terms of the traveler character-
istics and the service characteristics of the different airport
Analysis Zones ground access modes available to the traveler.
Selection of a suitable system of analysis zones involves a Stated preference surveys follow a similar process, except
tradeoff between the precision of the transportation service data that the respondent is presented with a set of hypothetical
used for each travel party in the sample and the work involved choices and asked to select from them. For realism, the stated
in assembling the necessary data. Some of the data may already preference experiment is usually structured so that the choices
be available in a particular system of analysis zones. For exam- presented to the respondent correspond to their current trip or
ple, regional transportation planning agencies will generally a recent actual trip but change the characteristics of the
have computer files with highway travel times between each ground transportation options available, such as different
transportation analysis zone (TAZ) used in their regional travel prices or travel times or the introduction of a new service or
demand model and the TAZ containing the airport. Indeed it mode. Estimating a mode choice model on such data allows
may be desirable or even required that an airport access mode the model to incorporate ground access options that do not
choice model is based on the regional travel demand model currently exist or to explore the effect of changing factors that
TAZ system so that the results of the modeling can be inte- do not exhibit much variation in the real world. Although this
grated with the regional travel demand modeling process. is a powerful capability, there are concerns about how well

21
travelers’ expressed choices between hypothetical situations However, for this to happen it will be necessary to incorporate
that they have not actually encountered correspond to how traveler information more explicitly in the models.
they would really behave if faced with those situations in
practice. To attempt to address this concern, stated preference Model Specification, Estimation, Calibration,
studies are often combined with analysis of revealed prefer- and Validation
ence to at least ensure that the stated choice behavior is
consistent with the actual behavior when applied to situations The terms model estimation and model calibration are often
that have actually been experienced. loosely used interchangeably. However, strictly speaking
they are two different steps in developing an airport access
Even so, there remains the concern that survey respon- mode choice model (or indeed any model). Model estimation
dents may overstate their willingness to use new alterna- refers to the use of statistical procedures to determine the val-
tives that do not yet exist, whether out of a desire to appear ues of the model coefficients that best fit the data from which
responsive to the survey or because they misinterpret how the model is being developed. Typically, this will be derived
attractive the new service will be relative to their current from a sample of air passenger or airport employee trips
choices. There is a limit to how detailed the description of obtained from a survey.
a new service can be in the context of a stated preference
survey that has to be completed in a fairly short time period, Once an estimation dataset has been assembled, the first
and respondents may not fully consider all the factors that step in model estimation is model specification. This involves
would arise in using the new service, such as how they selecting an appropriate functional form and market seg-
would get to it at the time that they need to travel. A related mentation for the model, and defining relevant explanatory
factor is that a stated preference survey necessarily informs variables. Model estimation software is then used to obtain es-
the respondents about the options that are hypothetically timated values of the model coefficients. The statistical sig-
available so that they can make a choice. However, in prac- nificance of these estimated values and the overall goodness-
tice many travelers may not be aware of the existence of the of-fit of the model is examined and the model specification
service or may have a misperception of the nature of the revised as necessary to address problems with the resulting
service offered. model coefficients or statistical fit. Model development typi-
cally proceeds iteratively. A fairly simple functional form with
relatively few explanatory variables is initially estimated.
Habit and Information Then the model is improved progressively by adding variables
or modifying the functional form, such as changing the struc-
An important aspect of air passenger airport access mode ture of the way that modes are grouped within the model to im-
choice is the role of travel habits and the awareness of travel prove the statistical fit of the model to the estimation data.
alternatives. In contrast to most urban trips, such as the journey- However, statistical fit is not everything. A model must also
to-work or shopping trips, most air passengers do not make make sound behavioral sense. A model that reflects a plausible
air trips that often, and many visitors to a region may be vis- structure of behavioral causality is generally preferred to one
iting for the first time. It is self-evident that travelers will not that contains counterintuitive features, even if the latter has a
choose to use modes that they do not know exist, or even if better statistical fit.
they have significant misperceptions about the nature of the
service offered, such as the travel time involved or the cost. Once a satisfactory model has been estimated, model cal-
However, most air passenger surveys ignore questions of ibration should be undertaken to make any necessary adjust-
travelers’ awareness of difference services and their percep- ments to the model so that the model predictions agree with
tions of the service offered by modes that they did not use. As observations. If the model estimation has been done cor-
a result, the estimation process of most mode choice models rectly, the model predictions will agree with the observed
has implicitly assumed that travelers have full information data in the model estimation dataset. However, because these
about each mode. The issue of how to address habit and in- data may not be a truly representative sample of the larger
formation in airport access mode choice models (or any mod- population being modeled, the model may need to be ad-
els of travel behavior for that matter) is not well understood justed to produce satisfactory predictions. In the case of an
and is one that could benefit from further research. airport access mode choice model, where the composition of
the air travel market will vary seasonally or even from day to
This is an important consideration because it directly ad- day and assumptions will need to be made about average
dresses the role of marketing in the provision of airport access vehicle occupancy of some modes to convert the number of
services. Airport authorities or transportation operators can person-trips using those modes to the equivalent number of
choose to spend resources improving services or more inten- vehicle trips, the total volume of vehicle trips by each mode
sively marketing the services that they already provide. It predicted by the model using the estimated coefficients may
would be extremely helpful if airport access mode choice differ from the observed volume of vehicle trips for any
models could help shed some light on how best to allocate given period. This is particularly the case when a model has
resources between service improvements and marketing. been estimated on survey data collected at one point in time,

22
typically a few weeks or less, but is being used to predict Because of the effort and resources required to assemble
mode use for a different time period, such as a year. the necessary data to perform a proper validation of an air-
port ground access mode choice model, such validations are
Even if the survey data used for model estimation is an ac- rarely if ever done and it is simply assumed that a calibrated
curate representation of the larger population, the conversion model that has been developed from data on mode use pat-
of model predictions from travel party trips to vehicle trips terns at one point in time will remain valid when used to
generally will require some model calibration. Because access model airport access mode use at other points in time or
mode choice models are estimated on data for a sample of air under different conditions. Also, because model validation
passenger travel parties or airport employees, they generate involves comparing model predictions with observed condi-
predictions of travel party or employee trips using each mode, tions under different conditions from those from which it has
which typically need to be converted to vehicle trips for use in been calibrated, this often cannot be done until some time
planning studies or other applications. For most private vehi- after the initial model estimation effort, by which time it may
cle modes the conversion is straightforward, because each already have been used for the application for which it was
travel party generates one vehicle trip in each direction (or two developed. However, airport access mode choice model de-
in the case of drop off or pick up by private vehicle). However, velopment should not be viewed as a one-time effort, any
for most other modes the ratio of vehicle trips to travel party or more than any other aspect of travel demand analysis. Rather,
employee trips depends on assumptions about average vehicle opportunities to validate a model should be sought and pur-
occupancy or (in the case of modes such as taxi or limousine) sued following its initial development, and the model refined
the proportion of deadhead trips to revenue trips. For sched- and improved over time. In this way the model will be avail-
uled modes, the number of vehicle trips is determined by the able for subsequent applications with a growing level of con-
schedule rather than the mode use. Where observed counts of fidence in its predictive reliability.
vehicle trips are available, it will generally be necessary to
adjust the assumptions about average vehicle occupancy or MATHEMATICAL FORM OF TYPICAL MODE
proportion of deadhead trips to match the predicted values of CHOICE MODELS
vehicle trips to the observed values.
The challenge of developing mathematical models of dis-
Model validation is the final step in model development crete choice behavior has attracted the interest of statisti-
and involves comparing the predictions of the model under cians, economists, social scientists, and transportation plan-
different conditions from those under which it was estimated ners over a long period of time and thus not surprisingly there
and calibrated, usually after some change has occurred in the is an extensive literature on the subject that is beyond the
system being modeled, with the actual values of the phe- scope of this report to summarize. Some of the earliest appli-
nomena being modeled. In the case of an airport access mode cations of discrete choice behavior models to transportation
choice model, this could involve comparing projections of travel demand were undertaken by Daniel McFadden and
mode use in subsequent years or after changes have occurred colleagues (described in Domencich and McFadden 1996),
in the ground transportation services available at the airport as part of work for which McFadden was awarded the Nobel
with the observed mode use under those different conditions. Prize in economics. Readers interested in the theoretical
Although limited model validation can be done using partial background and evolution of the current state of practice of
data on the use of certain modes, such as comparing the num- transportation mode choice models can refer to standard texts
ber of private vehicles parked at the airport with the number such as Ben-Akiva and Lerman (1985) or Hensher et al.
projected by the model, a more thorough model validation (2005). However, to help readers who have limited or no
will require an extensive effort to collect comprehensive data prior familiarity with these techniques understand the gen-
on mode use for the validation period. eral approach; this section will attempt to provide a simpli-
fied introduction to the current state of practice. Those with
Because the pattern of access mode use at an airport will de- some familiarity with transportation mode choice modeling
pend on the composition of the air travel market or of the may choose to skip this discussion.
employee workforce, as well as the transportation services
available, a true validation of an access mode choice model The basic concept underlying most disaggregate discrete
should include a new air passenger or airport employee survey choice analysis is that each alternative in the choice set pro-
to ensure that the market composition assumptions being used vides the decision maker with some utility that can be
in the modeling are correct. Otherwise, it is unclear whether expressed in terms of measurable or observable characteris-
differences between the predicted mode use and the observed tics of both the decision maker and the alternative (e.g., the
mode use are the result of problems with the model or invalid travel time involved or the income level of the decision
assumptions about the market composition. Similarly, it is maker). The larger the difference in the utility between two
important that the transportation service assumptions for the alternatives, the more likely the decision maker is to choose
various modes that are used in the modeling are updated to the alternative with the higher utility. This can be illustrated
reflect changes in costs and service levels since the model was by the relationship shown in Figure 2, which shows the prob-
originally developed. ability of choosing alternative 1, P(1), as a function of the

23
1.0
Probability of Choosing Alternative 1

0.5
0.0
-4 0 4
Difference in Utilities (U1-U2)
FIGURE 2 Illustrative binary choice relationship.
difference between the utility of alternative 1, U1, and that of where Ui and Uj are the utilities of alternatives i and j,
some other alternative 2, U2, (termed a binary choice be- respectively, and P(i) is the probability of choosing alterna-
cause it involves only two alternatives). As this difference tive i. This in turn can be reexpressed as:
becomes large and positive (i.e., the utility of alternative 1 is
much greater than that of alternative 2), the probability of P(i) = eUi ( eUi + eU j )
choosing alternative 1 approaches one (a probability by def-
inition must lie between zero and one). Conversely, if the which became known as the logit model (strictly the bino-
difference in utility is large and negative (i.e., the utility of mial logit model). It can be shown fairly easily that with
alternative 2 is much greater than that of alternative 1), the more than two alternatives the model can be extended as
probability of choosing alternative 1 approaches zero. If follows:
the difference in the utilities is zero, the two alternative are
equally attractive and the probability of choosing either is eUi
P(i) =
equal to 0.5 (50%). ∑ eU j
j ∈J
Multinomial Logit Model

where J is the number of alternatives. This is termed the
As shown in Figure 2, the functional relationship between multinomial logit (MNL) model and has been widely used
the difference in utility and the probability of choosing a for airport access mode choice models, because most air-
particular alternative will be an S-curve, because it must be port access mode choice situations involve more than two
asymptotic to one and zero as the utility difference becomes alternatives.
very large in either the positive or negative direction. Early
work on choice modeling identified the logistic function, Although the logit model was initially simply a convenient
defined as: way of generating the required S-shaped relationship, later
work showed that under certain assumptions regarding the
f ( x ) = e x (1 + e x ) = 1 (1 + e − x ) form of the utility terms the logit model can be derived from
theoretical principles of utility maximization, whereby the
decision maker chooses the alternative that offers the highest
as providing a suitable S-shaped curve, where ex is the expo-
utility, and the probability of choosing a given alternative is
nential function of x. It can be seen that as x becomes very
thus given by the probability that that alternative offers the
large and positive, f(x) approaches one and as x becomes very
highest utility of all the alternatives.
large and negative, f(x) approaches zero.
In the context of a choice model between two alternatives Utility Function

(termed a binary or binomial choice) the logistic function can
be restated as: In developing discrete choice models, the utility of a given
alternative is assumed to comprise two parts; a deterministic
P(i) = 1 (1 + e − (Ui −U j ) ) part that consists of a function of measured and observed

24
variables and an error term that accounts for unobserved char- the estimated values of the coefficients have no direct inter-
acteristics and variability in the perceived utility of a given set pretation. However, the ratio of the coefficients for two
of characteristics across different individuals, therefore: variables (or the ratio of the ASC to a coefficient of an ex-
planatory variable) is another matter. This ratio expresses
U i = Vi + ε how an increase in one variable (or the ASC) will offset a
decrease in the other variable and thus can be expressed as
where Vi is the deterministic part of the utility and ε is the implied values. The ratio of a given coefficient or ASC to the
error term. In logit choice models, the error term is assumed coefficient for the cost term gives the implied value of that
to be a random variable with values that are independent and variable or constant in the units of the cost term. With appro-
identically distributed with a Gumbel (double exponential) priate adjustments for the units of two variables, this can give
distribution with a zero mean. (This assumption allows the implied values of time in dollars per hour. Where the travel
logit model to be derived from utility maximization theory.) cost variable incorporates some function of household or
The variance of the error term reflects the goodness-of-fit of other income the resulting implied values will be expressed
the model. in terms of this income measure.
The deterministic part of the utility function typically con- This has an important implication for the specification of
sists of a linear combination of explanatory variables with the utility function. Although different components of
their associated model coefficients, the values of which are travel time can (and often should) be expressed with sepa-
determined in the model estimation process, therefore: rate variables, cost terms should be combined into a single
variable. This will avoid problems with the model giving
Vi = ai + b1x1 + b2x2 + . . . + bnxn different implied values of a given variable depending on
which cost term coefficient is used. This is also conceptu-
where ai and the b’s are the model coefficients and the x’s are ally sound. Although travelers may (and usually do)
the values of the explanatory variables, such as travel time perceive different components of travel time as having a
and cost. In general, the utility function for each alternative different disutility per unit time (i.e., different implied val-
will have a constant term ai, known as the alternative-specific ues), it would be surprising if they view a dollar spent on
constant (ASC), which reflects attributes of the alternative one aspect of the airport access journey any differently
that are not accounted for by the other variables. Therefore, from a dollar spent on another aspect, because the money
a fairly simple utility function might comprise: involved is completely interchangeable. (The one exception
to this principle would be if some travel costs are reim-
Vi = ai + b1(travel time) + b2(waiting time) bursable and others not. In this case, it would make sense to
+ b3 (walk distance) + b4(cost/income) use separate variables for the two types of cost so that the
ratio of the two cost coefficients reflect the relative impor-
Note that in this example travel cost is divided by house- tance of reimbursable to nonreimbursable costs. This would
hold income in the fourth explanatory variable so that the still allow values of time to be expressed consistently in
choice process becomes less sensitive to cost for higher terms of nonreimbursable costs.)
income travelers. This is included as an illustration that all
explanatory variables do not have to enter the utility function
as separate terms and may not be the best way to reflect the Limitations of Multinominal Logit Model
effect of household income.
Although the MNL model has been widely used, it is vulner-
It should also be noted that changing the utility functions able to problems that arise from a property of the model
of each alternative in a logit choice model by the same termed the “Independence from Irrelevant Alternatives.”
amount will not affect the resulting probabilities (because the This states that including a new alternative in the choice set
change will factor the numerator and denominator of the logit (or changing the perceived value of one of the alternatives)
expression for each alternative up or down by a constant should not affect the relative probabilities of choosing any of
amount that will cancel out). Therefore, it is usual practice to the other alternatives. It can be seen from the previous equa-
set the ASC for one of the alternatives to zero, so that the tion for the MNL model that the ratio of the probability of
ASCs for the other alternatives then reflect the differences in choosing any two alternatives is determined only by the per-
the constant part of the utility for those alternatives relative ceived utilities of those alternatives.
to the alternative without an ASC.
However, in many situations in airport access mode
The estimated coefficients of the utility function can be choice it is quite unlikely that changing the characteristics
thought of as weighting factors that convert the units of the of one mode or sub-mode will leave the relative probabili-
explanatory variable (e.g., minutes of travel time) to a mea- ties of choosing all the other modes and sub-modes
sure of perceived utility. Because perceived utility is an unchanged. For example, increasing the parking rates in the
abstract concept that has no intrinsic units of measurement, short-term parking lot is likely to have a greater effect on

25
the probability of an air party choosing to park in the long- where P(i|m) is the probability of choosing mode i in nest m
term parking lot than on the probability of choosing to use from among the set Nm of modes in nest m, given that nest m is
a shared-ride van, because those who would have parked in chosen (strictly that a mode in nest m is chosen) and P(m) is the
the short-term lot at the former rates are much more likely probability of choosing nest m from the set S of nests at the same
to choose to park in the long-term lot instead than to decide level as nest m. If one branch of a nest consists of a discrete
to use a shared-ride van. Similarly, changes in one public mode m* rather than a lower-level nest, the value for the scaling
transportation service are likely to affect the use of other parameter for that mode μm* = 1. Therefore, if there is only one
public transportation services to a greater extent than the nest, the previous equations reduce to the MNL model.
use of private vehicles.
Estimation of NL models involves estimating values
This is not usually a significant problem with a model that for the scaling parameters as well as the utility function
only includes a limited number of well-differentiated modes, coefficients.
but becomes increasing problematical with models that at-
tempt to distinguish between the use of similar modes (such
as taxi and limousine) or to account for sub-modes (such as Other Model Specifications
different parking facilities).
Although the NL model overcomes some of the inherent
limitations of the MNL model, there remain a number of
Nested Logit Model other limitations to the use of NL models for modeling air-
port access mode choice. Perhaps the most significant of
The limitations of the MNL model can be addressed by these is the assumption that the variance of the error term in
grouping similar modes or sub-modes into separate groups or the utility function is the same for all air parties and all
nests in a choice structure referred to as a nested logit (NL) alternatives. Another limitation can arise where the same
model, as illustrated by Figure 3. In the nested model shown alternative appears in different nests; for example, if several
in the figure, alternative b consists of a second-level nest of public transportation alternatives have station or stop access
two sub-alternatives, b1 and b2, the second of which consists sub-mode nests, because these will typically involve the
of a third-level nest of two further sub-alternatives, b21 and same sub-modes. Efforts to explore alternative model
b22. For example, alternative b might represent the use of a formulations to NL models for transportation mode choice
private vehicle, with alternative b1 representing the air party applications have taken two approaches. One is to use more
being dropped off at the airport and b2 the use a private advanced logit model formulations that address some of
vehicle that is parked at the airport for the duration of the air the limitations in the standard model. The other is to use an
trip, where b21 represents the use of the short-term parking entirely different conceptual approach to representing the
lot and b22 the use of the long-term parking lot. mode choice process.
The general form of the NL model is similar to the MNL However, to date neither of these approaches has been
model, with the addition of a scaling parameter μm for each applied to stand-alone airport ground access mode choice mod-
nest m, as follows: els, although some work has been done on modeling airport
choice in which ground access mode choice forms part of the
1
P(i m) =
(eU (i ) )μ m
choice process (Hess and Polak 2005, 2006). Although these
alternative approaches may be a promising area for future
1
∑ (eU ( j ) )μ m research, given the limited experience applying approaches
j ∈N m other than MNL and NL models to airport ground access, the
μ details of these approaches are not discussed further here. Read-
⎛ 1 ⎞ m
⎜⎝ ∑( eU ( j ) μ)m
⎟⎠ ers interested in more information will find a discussion of
j ∈N m the potential application of alternative mode choice model
P (m) = μl
⎛ 1 ⎞ approaches to airport access travel in Lu et al. (2006).
∑ ∑( ⎜⎝ )
eU ( k ) μl ⎟⎠
l ∈S k ∈Nl
MODEL ESTIMATION CONSIDERATIONS
Apart from questions of selecting an appropriate functional

a form and market segmentation for the model and defining
relevant explanatory variables, careful consideration needs
a b c d b1
b21 b22
to be given to developing the necessary estimation dataset.
alternatives
alternatives
The reliability of the resulting model is critically dependent
Multinomial Model
on the accuracy of the data on which it is estimated. If the
Nested Logit Model
data contains errors or biased values of explanatory vari-
FIGURE 3 Multinomial and nested logit models. ables, the model will attempt to explain the traveler behavior

26
in terms of those values rather than the real values. This may management features are often fairly basic, and user support
not be a problem in obtaining a good fit to the data, but will may be limited or nonexistent. On the other hand, the soft-
produce biased predictions when the model is later applied to ware may contain features or provide capabilities that are
other datasets that do not have the errors or biases. not available in commercial packages. A good example of
this class of software is BIOGEME, developed by Michel
Model Estimation Software Bierlaire at the Ecole Polytechnique Fédéral de Lausanne,
Switzerland (Bierlaire 2003), and available on the Internet at
The process of estimating airport access mode choice mod- http://biogeme.epfl.ch.
els, such as mode choice models for other transportation
applications, requires specialized software that is designed Unfortunately, obtaining comparative information on dif-
to perform maximum likelihood estimation with fairly com- ferent software options is not as easy as might be thought.
plex model specifications. This software falls into three With some persistence, Internet searches can generally locate
broad categories: information on the principal software packages. However,
some creativity is required in defining the search expressions
1. Special-purpose commercial software that has been to avoid being swamped by links to articles about choice
developed specifically to estimate logit-type choice model estimation methodology rather than the software in-
models or the mode choice component of transporta- volved or econometric software in general.
tion planning models.
2. General purpose commercial statistical packages that
include modules for performing maximum likelihood Market Segmentation
estimation of discrete choice models.
The air passenger market is not homogeneous and different
3. Software developed by academic researchers to esti-
market segments have different airport access needs and
mate a fairly broad class of econometric models that
available options. The most obvious distinction is between
includes discrete choice models.
residents of the local area and visitors. Residents typically
have access to a private vehicle and often someone who can
As might be expected, there is typically a tradeoff be-
take them to the airport or pick them up. Visitors on the other
tween the ease of use and cost of the different categories of
hand may need to rent a car to meet their transportation needs
software. Special-purpose commercial software for estimat-
while in the area or may be staying at a hotel that does not
ing logit choice models or transportation mode choice mod-
provide a courtesy shuttle service to the airport. Another
els is often the easiest to use, because considerable attention
important distinction is between those traveling on business
has been given to the design of the user interface and the soft-
trips, whose travel costs may be reimbursed by their em-
ware has been specifically developed for this purpose. How-
ployer or client, and those making trips for non-business pur-
ever, the software is often more expensive than the other
poses. These distinctions are typically addressed by defining
options, although the cost of acquiring software is usually
different air passenger market segments and estimating a
small compared with the overall cost of the model develop-
separate sub-model for each segment. The market segment
ment process.
sub-models may include different modes, may use different
explanatory variables, and will generally have different esti-
A number of general purpose statistical software pack-
mated coefficients for a given variable.
ages provide the capability to estimate discrete choice mod-
els using maximum likelihood techniques. Sometimes this
Although a common market segmentation approach, dis-
requires the purchase of an additional software module.
cussed in more detail in subsequent chapters, is a four-way
Although the cost of this additional capability (if required)
division into resident business trips, resident non-business
is usually less than acquiring special-purpose software, esti-
trips, visitor business trips, and visitor non-business trips,
mating a model with this software may require somewhat
other market segments may be worth considering. One is to
more effort than using special-purpose software, although
differentiate visitors staying in a hotel from those staying with
the interface between the model estimation function and the
relatives or friends, because the latter may have access to pri-
data management capabilities of the package is usually
vate vehicles owned by the people they are staying with as
straightforward. This may be an attractive option for organi-
well as people who can pick them up or drop them off at the
zations that already are using one of these software packages
airport.
for general statistical analysis.
Software developed by academic researchers to estimate Changes in the Airline Industry

various types of econometric models, often in support of
their research or teaching activities, is often available at little The terrorist attacks of September 11, 2001 (9/11) have led
or no cost. Depending on the scope of the software, this may to dramatic changes in the airline industry, in particular pas-
require some customization to use to estimate a particular senger security processing at airports. No longer can greeters
model specification. Documentation, user interface, and data and well-wishers enter the secure side of airport terminals,

27
and passengers need to arrive at the airport earlier than before inexpensive matter. This raises the questions of how large an
to ensure being able to clear security in time to make their airport needs to be to justify the effort and what such an effort
flight. In addition, airports rigorously enforce the prohibition might cost, not easy questions to answer. For most large air-
on leaving vehicles unattended at the terminal curb front or ports, with complex ground access systems to plan and man-
even waiting when not actively loading or unloading passen- age, it is very difficult to conceive how this can be done well
gers. The combined effect of these measures has been a without such a model (although many airports try to make do
significant change in airport ground access mode use. Fewer without one). For smaller airports, perhaps with traffic levels
well-wishers accompanying passengers to the airport (or driv- in the range of 5 to 10 million annual passengers, the need
ers dropping passengers off at the airport) park for a short may depend on the type of planning issue being faced.
time to accompany the passengers into the terminal, because
the passengers are usually anxious to get through security as The resources needed to develop an airport access mode
quickly as possible and the well-wishers cannot accompany choice model depend in part on the availability of air passen-
them. Similarly, greeters can no longer meet arriving pas- ger survey data (or airport employee survey data in the case
sengers at the gate, and the increased use in cell phones over of employee access mode choice models). As discussed ear-
the past five years has simplified the process of picking up lier, good air passenger survey data are critical to model
arriving passengers at the terminal curb front. As a result, development. Experience suggests that for adequate model
many airports have introduced cell phone lots where drivers development such a survey should have at least 3,000 re-
picking up air passengers can wait until the passengers are sponses, although models have been developed with smaller
ready to be picked up. sample sizes. Stated preference surveys have been conducted
with fewer respondents, approximately 800 and 1,100 in two
Another significant change in the airline industry since recent cases; however, each respondent typically answers
9/11 is a reduction in the percentage of short-haul trips. This several choice experiments, each of which provides a data
is widely believed to be at least in part a result of the need to point, and such surveys cost more to perform per respondent
arrive at the airport for a flight earlier than before, which than a revealed preference survey. Although the cost per
increases the time required to make an air trip and in turn completed survey of a revealed preference survey can vary
makes driving or other surface transportation modes rela- significantly with local circumstances as well as the number
tively more attractive. There may also be a heightened con- of questions asked, typical costs identified in an on-going
cern over aviation security that leads some who do have a ACRP study of airport-user survey methodology (Project
surface transportation alternative to take this option even if it ACRP 03-04: “Guidebook for Airport-User Survey Method-
involves more time than flying. A third consideration is the ology”) are in the range of approximately $30 to $50 per
increasing market share of low-fare airlines and the compet- completed response.
itive response of the network carriers. This has led to a much
higher proportion of air travelers using low-fare or heavily This suggests that developing the necessary survey data
discounted tickets. It would seem likely that passengers might cost somewhere in the range of $90,000 to $150,000.
attracted to air travel by cheaper fares would also be more Assembling the corresponding transportation service data
cost-sensitive when making their airport access decisions. It is also not trivial, although the local MPO may be able to
is also likely that a higher proportion of air travelers are fly- provide some of these data in electronic format from the
ing for personal rather than business reasons, although that regional travel demand model network data. The amount of
has also been affected by cyclical changes in the economy. work involved in assembling the remaining data required
will depend in part on how much information the airport au-
Although these changes in the airline industry do not fun- thority already has available on the ground transportation
damentally affect the basic approach to modeling airport pages of its website or in other files. Overall, this task might
access mode use, they do affect the relative attractiveness of require between one and two person-months of effort and
the different transportation services represented in the model cost somewhere between $20,000 and $50,000 at typical
and hence the estimated model coefficients, as well as the mar- consultant rates. Once the necessary data have been assem-
ket composition. Any airport access mode choice model bled, estimating, calibrating, and validating the model might
estimated on data from before 9/11 is likely to produce biased require two to three person-months of effort, or perhaps be-
predictions of current or future air passenger access behavior. tween $40,000 and $75,000. Therefore, a complete study,
Similarly, current data on air passenger market composition, including an air passenger survey, might cost between
particularly the split between business and non-business $150,000 and $275,000. However, the largest part of this
travelers and the distribution of trip durations, is likely to have cost is performing the air passenger survey, which has other
changed significantly from data collected before 9/11. value for airport planning purposes and should be done
periodically anyway.
Level of Effort Required
Whether such an investment is worthwhile depends on
It is clear from the foregoing discussion that developing an what the resulting model would be used for and the likely cost
airport ground access mode choice model is not a simple or of making a bad decision. Certainly, in the case of evaluating

28
the feasibility of a major infrastructure investment such as an of scenarios it may be helpful to develop utility routines to
airport rail link or even an automated people-mover link to a perform the various steps.
nearby rail station that might cost several hundred million
dollars or more, the cost of having a good modeling capabil-
ity is trivial and would easily be justified by avoiding a poor Model Application Considerations
decision that results in an increase in the cost of the project by
even 1%. Indeed, having such a modeling capability is prob- Although the process of applying an airport access mode
ably a necessary requirement for obtaining environmental choice model is technically fairly straightforward, there are a
approval and funding. In the case of a smaller airport where number of aspects that need to be carefully considered in de-
the issues being addressed involve less money, the justifica- veloping the required input data and interpreting the results.
tion for such a modeling capability is less clear. Even so, an
airport handing 5 million annual passengers, 20% of whom The first is that whereas a relatively small survey sample
park at the airport for their trip duration, could be generating size (a few thousand respondents in the case of air passengers
as much as $10 million per year in parking revenue. An and perhaps even fewer for airport employees) may be ade-
increase in revenue of only 3% as a result of better pricing quate to estimate an airport access mode choice model, a much
decisions could pay for the cost of developing the modeling larger sample may be required for a given application of the
capability in less than year. model, depending on the issues of interest and the desired level
of geographic resolution of the results. For example, a study to
estimate the likely use of a proposed off-airport terminal in a
MODEL APPLICATION particular location in the region served by the airport needs to
have enough data points in the application dataset in the vicin-
Once a mode choice model has been developed, to apply it to ity of the proposed terminal that the resulting estimate of the
support planning and decision making it is necessary to have likely use of the facility is sufficiently accurate. To create a
the ability to use it to analyze specific scenarios. This typi- large enough application dataset, it may be necessary to de-
cally involves a significant amount of data management and velop a synthetic sample using Monte Carlo simulation meth-
model configuration to define the scenarios to be analyzed. ods based on the distribution of travel party characteristics in
One approach is to use the model in conjunction with stan- the original survey. This is more accurate than simply dupli-
dard transportation planning software or proprietary airport cating the survey records, because that cannot create a record
landside modeling software. These software tools are de- with different characteristics from those that appear in the sur-
signed to facilitate the data management involved in model- vey. Therefore, record duplication will simply create multiple
ing transportation network flows and typically provide users records with identical characteristics, and combinations of
with the flexibility to define the structure and coefficients of traveler characteristics that do not appear in the original survey
the mode choice model incorporated in the analysis. If such sample will never appear in the expanded sample no matter
models are not available, or the models that are available do how large it is, whereas a synthetic sample will create records
not have the necessary capabilities, it will then be necessary with characteristics that do not appear in the original sample.
to develop custom software to apply the model to analyze This will ensure that analysis zones that were thinly populated
any given scenario. with survey respondents in the original sample will have a
much more representative mix of traveler characteristics in the
Many discrete choice model estimation tools also provide expanded sample and not be biased toward the characteristics
the capability to apply a defined choice model to any suitably of those travelers that happened to appear in the original sur-
configured dataset of decision-maker characteristics and as- vey sample.
sociated properties of the choice alternatives to estimate the
resulting choice probabilities. The result of this process is Two other considerations relate to the application of a
typically a table of the probability of choosing each alterna- model estimated on data at one point in time to predict
tive for each decision maker (air passenger travel party or behavior in future years, as is typically done in planning
airport employee) in the dataset. Converting this table to a studies. The first consideration is how to adjust travel times
projection of passenger trips or vehicle trips by mode is then and costs for the various modes to correspond to future con-
a matter of factoring up the results to correspond to the total ditions. Highway travel times for future years should reflect
airport activity for the period in question and applying the ap- any anticipated changes in highway congestion. Future costs
propriate ratios of vehicle trips to travel party or employee are more problematical, because their effect on traveler
trips. These estimated vehicle trips can in turn be allocated to decision making will depend on their value relative to the
the transportation network by segmenting the results by the overall cost of living as well as changes in the income level
trip origin zone and creating a zonal trip table. Although con- of travelers over time. Typical practice is to consider that the
ceptually this is not difficult and the required calculations can cost and income variables in the model are expressed in real
generally be performed fairly easily using spreadsheet or dollars (i.e., do not need to be increased to account for infla-
database management software, because the process typi- tion). However, that is not to say that the cost of different
cally has to be repeated multiple times to analyze a number modes will not change over time in real dollars. If general

29
income levels rise or fuel prices increase in real terms, then equations for a NL model is somewhat more involved, owing
transportation operators will need to raise their prices to to the existence of the scaling parameters and the more com-
cover their higher costs of doing business and the cost of plex structural form of the model, but the same approach can
operating private vehicles will rise. Conversely, if trans- be applied.
portation operators are able to achieve productivity gains or
rising levels of air travel increase the traffic that they carry, The advantages of this approach are two-fold. The first is
allowing them to be more efficient, then they may be able to that it eliminates any error in the predictions resulting from
reduce their prices in real terms. Although these effects may differences between the mode use predicted by the model at
tend to offset each other, the net effect is likely to vary by current values of the transportation service variables and the
mode. Therefore, some thought should be given to assump- actual mode use. The second is that it only requires values for
tions about future travel costs and not simply assume that the model coefficients of the variables that change in value
they will remain unchanged in real terms. and not for the other variables or the ASCs. However, this is
also a potential weakness of the approach. If the ASCs reflect
Historically, household incomes have increased in real some of the contribution of the transportation service vari-
terms over time, although over the past 30 years the distribu- ables to the predicted mode shares (perhaps owing to speci-
tion of household incomes has also changed, with the incomes fication errors in the model or errors in measuring the trans-
of higher-income households increasing faster than that of portation service levels), then those effects will be ignored in
lower-income households (U.S. Census Bureau 2006, Table the analysis. Therefore, an alternative approach would use
A-3). Whether the effect of household income is explicitly the full model, but apply the predicted change in mode use to
included in the mode choice model or implicitly included in the existing observed mode use.
the coefficients of cost terms, some thought should be given
to how to incorporate future growth in household income in
the application of the model for future years. This is not a triv- Demand Elasticity
ial matter. From 1995 to 2005 the average household income
in the United States increased by almost 11% in real terms The concept of demand elasticity refers to the percent
(U.S. Census Bureau 2006, Table A-1). This is equivalent to change in demand for a 1% change in some variable. There-
a reduction in the relative costs of different modes by at least fore, the demand elasticity with respect to the price for taxi
that amount, which could easily have a greater effect on mode use at an airport would express the percent change in taxi
use than the type of transportation system enhancement that use for each percent change in fare levels. In the case of a
mode choice models are used to evaluate. Because higher MNL model with a linear utility function, it can be shown
income households use air travel more than lower income that the elasticity of the probability of choosing a particular
households, the real increase in the average household income mode with respect to a given variable in the utility function
of air travelers and the corresponding effect on airport access is given by:
mode choice is probably even greater.
εik = αki • Xki • (1 − Pi)
Pivot Point Analysis where εik is the elasticity of the probability of choosing
mode i with respect to changes in the value of the explana-
Rather than use a disaggregate mode choice model directly to tory variable Xk for mode i (Xki), αki is the coefficient of Xk
predict the change in mode shares as a result of some change in the utility function for mode i, and Pi is the current prob-
in transportation service levels, it is possible to use the relevant ability of choosing mode i. In simple terms, this is saying
model coefficients and the existing mode shares to predict the that the elasticity is given by the product of the coefficient
change in mode share for a given change in the explanatory of the variable, the current value of the variable, and the
variables. This has come to be known as the incremental logit probability of not choosing the mode. As the probability of
model or pivot point analysis (the predicted mode shares are choosing a particular mode increases, the elasticity of the
considered to pivot about the existing mode shares) (Kumar probability of choosing that mode with respect to any vari-
1980). Another way to think about this type of analysis is to able becomes less.
use the model coefficients and existing mode shares to deter-
mine the slope of the demand curve at the current values of the However, this is of limited use because it is clear from the
transportation service variables and mode use, and then calcu- previous equation that the elasticity of demand varies with
late the change in mode share for a given movement along the the value of the variable in question for every air party and
demand curve corresponding to the change in the value of the with the probability of each air party selecting the mode in
transportation service variable (Meyer and Miller 1984). question at current values of all the explanatory variables.
Therefore, elasticity is not a constant property of a given
In the case of MNL models (or binary logit models), the mode or a given situation, but varies with the values of the
incremental logit equations are fairly straightforward and are transportation service variables and with the market shares of
given by Kumar (1980). The derivation of the corresponding the different modes. Nonetheless, for any given situation the

30
elasticity of demand for a given mode with respect to a given value is only valid for the particular situation for which it has
service variable can be calculated numerically, and this may been calculated and will change as the situation changes. In
be a useful thing to do to give planners and managers an eas- particular, if changes occur in other modes the elasticity for
ily understood tool to make quick assessments of the likely the mode in question will change because changes in other
effect of any proposed change. The important caveat to en- modes will change the probability of choosing the mode for
sure this is clearly understood is that any given elasticity which the elasticity has been calculated.

31
CHAPTER THREE
REVIEW OF THE LITERATURE
AIR PASSENGER MODE CHOICE MODELS Therefore, given the importance of understanding air pas-
senger airport ground access mode use it is not surprising that
The ability to predict how airport users will respond to there have been a number of studies over the years that have
changes in the service characteristics of the ground access developed air passenger ground access mode choice models.
modes serving the airport or the addition of new ground One of the earliest efforts to develop a formal model of air
access services is clearly an essential part of any effective passenger airport ground access mode choice was under-
analysis of proposals to enhance the airport ground trans- taken in the early 1970s (Ellis et al. 1974). This study used a
portation system. In addition, airport ground access models MNL model, as did several other studies that developed air
play an important role in studies addressing how future air passenger ground access mode choice models over the next
travel demand will be distributed among airports in a multi- ten years (Leake and Underwood 1977; Sobieniak et al.
airport region. The relative accessibility of airports serving a 1979; Gosling 1984; Spear 1984; Harvey 1986). However,
region is recognized as one of the key determinants of air by the mid-1980s it was becoming recognized that some
passenger airport choice (in addition to the air service offered of the limitations of the MNL model could be addressed
at the airports). Whereas early airport choice models simply through the use of NL models (Ben-Akiva and Lerman
used highway travel times as a measure of airport accessibil- 1985). One of the first applications of NL models to airport
ity, later models have recognized that appropriate measures ground access mode choice was undertaken as part of a study
of airport accessibility need to account for the range of of surface access to London Heathrow Airport (Howard
ground transportation services available and the proportion Humphreys and Partners 1987), followed shortly thereafter
of airport users who choose different ground transportation by a study by Harvey (1988) that used a NL structure to
modes for their travel to and from the airport. develop an integrated model of airport choice and ground
access mode choice for the San Francisco Bay Area. Subse-
Because airport ground transportation involves trips both quent air passenger ground access mode choice models
to and from the airport, ideally what are required are airport developed for Boston, Massachusetts (Harrington et al.
ground access and egress models. However, in practice, 1996; Harrington 2003); Portland, Oregon (PDX Ground
most modeling efforts to date have only addressed airport Access . . . 1998); and airports in the southeast and east of
access, and it has been assumed (often implicitly) that the England (Halcrow Group Ltd. 2002b) used a nested structure,
reverse trip reflects a symmetrical pattern of mode use. whereas other studies continued to use MNL models to rep-
Personal experience and a growing body of evidence sug- resent air passenger ground access mode choice (Tambi and
gests that this is not the case, at least on the basis of the Falcocchio 1991; Dowling Associates, Inc. 2002; Psaraki
behavior of individual air parties, but this report focuses pri- and Abacoumkin 2002). In addition to models that have
marily on airport ground access mode choice models, exclusively addressed airport access mode choice, a number
because those are the models that are generally available. of recent studies have used NL models to represent air pas-
The review of the literature undertaken as part of this study senger airport choice, with airport ground access mode
found no examples of prior studies that explicitly modeled choice as a lower-level nest (Bondzio 1996; Monteiro and
airport ground egress travel behavior. The focus on ground Hansen 1996; Mandel 1999; Pels et al. 2003). However,
access trips no doubt results primarily from the use of air these models generally only include a single-level nest for
passenger surveys as the source of data on which to base the the airport ground access mode choice process and thus are
development of airport ground travel mode choice models. equivalent to MNL models from the perspective of ground
Because it is much easier to survey departing passengers access mode choice.
rather than arriving passengers, air passenger surveys have
generally addressed the ground access trip that the respon- The technical details of many of these models have been re-
dent has just completed, rather than the egress trip that a viewed by Lunsford and Gosling (1994) and later by Gosling
respondent beginning their air trip will make on their return et al. (2003). However, the level of detail reported in the liter-
or the egress trip that a visiting air traveler made on their ature for each of the models varies, with some authors only
arrival in the area some time before. The issue of the differ- providing partial information on estimated parameter values or
ence between airport ground access and egress travel even on the independent variables included in the model. It is
behavior is discussed later in this report. common to estimate separate sets of model parameters, or

32
even different model specifications, for different market seg- vate vehicle and rental car in their choice set, whether or not
ments, such as residents of the area versus visitors, or air they are staying with residents of the region and thus have
travelers on business trips versus those on leisure trips. Some someone who could drop them off at the airport or the need
published articles describing these models only present the for a rental car during their visit. The explanatory variables
estimated values of the model coefficients for some of the mar- consist of the travel times and costs for each mode. Separate
ket segments. This makes comparison of the different models variables are defined for in-vehicle, walk, and wait times.
difficult. However, detailed results are available for a number The transit in-vehicle times use the total in-vehicle time for
of recent models. The principal features of these models are the trip from the origin zone to the airport zone, transit walk
briefly described in the following sections. More detailed doc- times combine access, egress, and sidewalk times from the
umentation of each of these models is provided in Appendix D transit network, and the transit wait times combine the initial
included in the web version only. wait with any transfer wait times.
Boston Logan International Airport Model

Atlanta Regional Commission Model
This model was developed by the Central Transportation Plan-

The Atlanta Regional Commission (ARC), the MPO for the
ning Staff (CTPS) in Boston using a 1993 air passenger survey
Atlanta region, has developed an Airport Passenger Model
done at Boston Logan International Airport (Harrington et al.
(APM) to model air passenger trips to and from Hartsfield–
1996; Harrington 2003). Separate sub-models were developed
Jackson Atlanta International Airport (H-JAIA) as a compo-
for resident business trips, resident non-business trips, non-
nent of the overall regional travel demand modeling process
resident business trips, and non-resident non-business trips.
(Model Documentation . . . 2005). The model was originally
The two resident sub-models consist of a NL model, with sep-
developed in 2003 using data from the H-JAIA Peak Week Air
arate nests for door-to-door modes (taxi and limousine) and
Passenger Survey performed in 2000 and was updated in 2006
automobile modes (drop off, short-term parking, long-term
for new income groups for 2000 (Travel Demand . . . 2006).
parking, and off-airport parking). There are four shared-ride
The ARC Airport Passenger Model (ARC/APM) consists of
public modes at the top level (regular transit, scheduled airport
two components: a trip generation/distribution model that as-
bus, the Logan Express service to off-airport terminals in
signs the total originating air passenger traffic at H-JAIA to
the region, and the Water Shuttle between the airport and the
regional TAZs and a mode choice model that predicts the
downtown Boston waterfront). The visitor sub-models are
ground access mode use of those air passenger trips. A sub-
MNL models and omit the long-term parking alternatives but
sequent step converts air passenger trips to vehicle trips for
add a hotel shuttle mode.
inclusion on the traffic assignment step of the regional trans-
portation demand model. The model is one of the few exam-
This model is particularly relevant to studies involving
ples (and the only one documented in this study) of a special-
improved public transportation access to airports because it
generator airport access mode choice model fully integrated
includes both a rail access mode, the Massachusetts Bay
into a regional travel demand modeling process.
Transportation Authority (MBTA) regional rail transit sys-
tem, and off-airport terminals, the Logan Express service
The ARC/APM predicts air passenger and vehicle trips
operated by the Massachusetts Port Authority (Massport),
using four market segments: resident business, resident
the airport authority for Logan Airport. The MBTA Airport
non-business, non-resident business, and non-resident non-
Station is adjacent to the airport and linked to the passenger
business. The mode choice model considers five modes: air
terminals by a free shuttle bus service operated by Massport.
passengers dropped off by private vehicle, private vehicle
Unlike many other airport access mode choice models, the
parked at the airport for the duration of the air trip (termed
CTPS model is also interesting in that it treats rental car
drive self), rental car, transit, and taxi. Although not
use as an independent decision and excludes it from the mode
explicitly stated in the model documentation, it appears
choice decision process.
from the documentation that the transit mode includes the
Metropolitan Atlanta Rapid Transit Authority rail and bus
services, commercial shared-ride shuttle van services, and Chicago Airport Express Ridership
other high-occupancy shared-ride modes such as charter Forecasting Study
bus, whereas the taxi mode includes exclusive ride limou-
sine services and hotel and motel courtesy vehicles, as well In 2003, the Chicago Department of Transportation retained
as conventional taxi use. Resource Systems Group, Inc. and Wilbur Smith Associates
to undertake a ridership and revenue forecasting study of a
The basic form of the mode choice model is a NL model proposed Airport Express train service between downtown
with a separate structure for trips by residents of the region Chicago and O’Hare International and Midway Airports
from that for non-residents. The model does not take the type (Wilbur Smith Associates 2004). To identify the pattern of
of ground access trip origin into account. Therefore, all visi- air passenger trip ends, travel party characteristics, and exist-
tors to the region are considered to include drop off by pri- ing access mode use, and to understand how airport travelers

33
might change their access mode choice if the Airport Express private mode, Airport Express train, Airport Bus, Chicago
train were available, two surveys of air travelers were under- Transit Authority train, and other public mode. The “other
taken by Resource Systems Group, Inc. at O’Hare and Mid- private mode” included limousine, hotel/motel courtesy
way Airports, an origin–destination survey and a stated pref- shuttle, and shared-ride airport van service, whereas the
erence survey (Resource Systems Group Inc. 2004). The “other public mode” included local bus, region train and bus
origin–destination survey results were used to develop a pro- service, and charter bus. The mode choice model used a NL
file of existing originating air passenger characteristics, in- form with a somewhat different nest structure for travelers on
cluding trip purpose, trip origin, and ground access mode business and non-business trips, as shown in Figure 4.
use, as well as current and future trip tables that predicted the
number of trips by market segment originating in each of 145 The utility functions for each airport access mode included
TAZs within the study area. two continuous variables, total travel time and travel cost, in
addition to alternative specific constants. The estimated coef-
The stated preference survey interviewed 1,110 air travel- ficients for each of the two continuous variables within a
ers in the two airports, who were asked about details of their given market segment were constrained to have the same
trip to the airport, and then asked to complete eight stated value for each mode. Traveler income was not explicitly in-
preference choice experiments in which they were presented cluded in the model, but travelers were divided into two
with a choice between three trip modes, including the mode income categories on the basis of household income and sep-
they had just used and the Airport Express train. The results arate travel cost coefficients estimated for each category.
of the stated preference survey were used to estimate a mode Although a single travel time variable was used in each utility
choice model that defined nine airport access modes: private function, weights were applied to various components of the
vehicle parked at the airport for the duration of the air trip, total travel time that was used in the model estimation to
drop off at airport by private vehicle, rental car, taxi, other account for different disutility of access, transfer, and waiting
Business Mode Choice Model Structure

Mode Choice
Public Transport Public Transport

Mode Airport Express Train
Mode
Express Train
Access Mode
Drove & Dropped Rental Other Airport CTA Other

Taxi Private Public
Park Off Car Bus Train
Mode Mode
Free Drive & CTA

Walk Taxi
Van Park Service
Non-Business Mode Choice Model Structure

Mode Choice
Public Transport
Mode Public Transport Mode
Drove & Dropped Rental Other Other

Taxi Private Airport Express Train Airport CTA
Park Off Car Bus Train Public
Mode Mode
Express Train
Access Mode
Free Drive & CTA

Walk Taxi
Van Park Service
FIGURE 4 Chicago Airport Express mode choice model nesting structure (Source: Wilbur Smith
Associates 2004).

34
time (as applicable) for each mode. The weights were deter- private vehicle parked for the air trip duration, rental car,
mined by iteratively adjusting their value to obtain the best taxi, limousine, premium transit, local transit, shared-ride
overall model estimation result. In addition to the two contin- van, and hotel courtesy shuttle. The model used a NL struc-
uous variables, a dummy variable for the availability of bag- ture with the first five modes grouped into a nest called Non-
gage check-in at the downtown terminal was included in the Group modes and the other four modes grouped into a nest
Airport Express and Airport Bus modes, and a second dummy called Group modes. The model defined four market seg-
variable was included in the Airport Express mode for those ments: resident business trips, resident non-business trips,
passengers boarding at an intermediate station. non-resident business trips, and non-resident non-business
trips.
Separate model coefficients were estimated for business
and non-business travelers. The same model coefficients The model has only three explanatory variables, apart
were used for residents and non-residents of the Chicago from the ASCs: in-vehicle travel time, out-of-vehicle travel
region, although the available modes for these two market time (including waiting time and terminal time, such as walk-
segments were different. After the model coefficients were ing from the parking lot to the airport terminal or returning a
estimated using the stated preference data, the model was rental car), and travel cost. There is no consideration given to
calibrated to correspond to the mode shares for the study area the effect of income differences in the model. The values for
obtained from the origin–destination survey data by adjust- the coefficients for the continuous variables were not esti-
ing several of the ASCs until the model predicted the ob- mated from the air passenger survey data, but rather adopted
served mode shares from the study area. from the values estimated for Newark International Airport
in the original model with the same coefficient values being
The calibrated mode choice model was applied to forecast used for each market segment. Values for ASCs for each
ridership and revenue for the Airport Express train for vari- mode were estimated from air passenger survey data to fit the
ous service scenarios for 2009 and 2020, including two dif- model predictions of mode use to the observed data.
ferent fare levels, several different travel time assumptions
for the service to O’Hare Airport, two different growth rates Bay Area Rapid Transit–Oakland International
for off-peak highway travel time, and whether or not a free Airport Connector Study
downtown shuttle or baggage check at the downtown termi-
nal would be provided (Wilbur Smith Associates 2004). Since the 1970s, a number of studies have been undertaken
by the Port of Oakland (the operator of Oakland International
Miami Intermodal Center Travel Demand Airport), Bay Area Rapid Transit (BART), and other agen-
Forecast Study cies to explore the feasibility of developing an APM connec-
tion to replace the current AirBART shuttle bus link between
The Miami Intermodal Center (MIC) is being planned as a the airport and the Coliseum BART station located approxi-
major transportation interchange facility located immedi- mately 2.5 miles from the airport (BART–Oakland . . . 2002,
ately to the east of the Miami International Airport (MIA) to Executive Summary). As part of on-going efforts to imple-
provide an integrated terminal for several intercity trans- ment the Oakland Airport Connector, a Final Environmental
portation services, including Amtrak, Tri-Rail commuter Impact Report/Environmental Impact Statement (EIR/EIS)
rail, and Greyhound buses, as well as Metrobus and Metro- was completed and approved in March 2002. The analysis
rail transit services (Miami Intermodal Center . . . 1997). An for the EIR/EIS included the development of an airport
APM (the MIC/MIA Connector) will link the MIC to the air- access mode choice model by CCS Planning and Engineer-
port. Provision has also been made in the planning for future ing, Inc., that was applied to generate ridership projections
High Speed Rail and East–West Corridor rail lines as well as for the Connector (BART–Oakland . . . 2002, Appendix B:
an Airport/Seaport Connector rail link. The MIC will also Transit Ridership Procedures and Inputs). The mode choice
help accommodate growth of MIA by providing expanded model addressed both air passenger trips and airport em-
airport landside facilities, including rental car and long-term ployee trips, with the employee trips treated as a separate
parking facilities. As part of the planning for the MIC, a market segment. The general form of the model is a MNL
travel demand forecast was prepared by ICF Kaiser Engi- logit model with air passenger trips divided into four market
neers (1995) that incorporated an airport access mode choice segments: resident business trips, resident personal trips, vis-
model that was used to forecast ridership on the MIC/MIA itor business trips, and visitor personal trips.
Connector. The mode choice model was based on a model
originally developed by KPMG Peat Marwick for a study for Data on the air party characteristics for each market seg-
Newark International Airport in New Jersey. ment were obtained from the 1995 Air Passenger Survey un-
dertaken for the Metropolitan Transportation Commission
Because of the large number of modes available at MIA (MTC) at the three Bay Area airports (including Oakland
and the complexity of the choices available as a result of the International Airport), supplemented by surveys of AirBART
MIC project, the mode choice model was expanded to in- passengers performed by CCS Planning and Engineering,
clude the following nine modes: drop off by private vehicle, Inc., in December 1999 and May 2000 as part of the study.

35
The model assigns airport trips among the following eight modes that were not then available, namely light rail, express
modes: private vehicle, rental car, scheduled airport bus, pub- bus, and shared-ride transit (it is unclear from the documen-
lic transit, shared-ride van, hotel courtesy shuttle, taxi or lim- tation how this was defined).
ousine, and other. Public transit included both the use of
BART by means of the AirBART shuttle (or the Connector in An initial model estimation was done by Cambridge Sys-
the future), as well as local transit bus service directly to the tematics (Bowman 1997; Cambridge Systematics 1998) that
airport. The transit alternative for travelers with trip origins in jointly estimated MNL models using both the revealed pref-
zones near the airport was assumed to be local bus, whereas erence and stated preference data for the same four market
the transit alternative for those from more distant zones was segments as the Boston Logan model. Separate ASCs were
assumed to be BART. The use of “other” modes was not estimated for each mode for trips originating within the Port-
explicitly modeled, but rather the use of those modes was land metropolitan area (termed internal trips) and those orig-
assumed to remain constant from the mode share observed in inating outside the metropolitan area (termed external trips).
the 1995 Air Passenger Survey. Two different sets of model parameters were estimated for
each market segment, reflecting different assumptions for the
The model utility functions included the following six ASCs for the light rail and express bus modes. The models
variables: highway travel time, travel time by rail transit, were subsequently revised by Metro staff to combine some
travel time by bus transit, walking distances, waiting times, of the choice alternatives and recalibrate the models by ad-
and travel costs, although not all variables applied to each justing the ASCs (PDX Ground Access . . . 1998).
mode. The distinction between rail transit travel time and bus
transit travel time allowed the analysis to consider the effect In addition to the development of a mode choice model, a
of replacing the AirBART shuttle bus with the planned APM, review of the experiences of other U.S. airports with a range
as well as the different level of service between BART and of airport ground access strategies was undertaken as part of
local bus. Household income was included in the model by the overall study of alternative airport access modes (Coogan
dividing the costs for personal trips by the household income 1997). This report included statistics on the ground access
in thousands of dollars raised to the power 1.5. This adjust- mode shares of various airports that had implemented ground
ment was not applied to business trips as it was considered transportation services similar to those being considered for
that business travel decisions are unaffected by income Portland, as well as a discussion of the operational experi-
because business travelers are usually reimbursed for travel ence of those airports with the ground transportation services
expenses. The model coefficients for the continuous vari- and the lessons that might be applicable to the Portland situ-
ables were adopted directly from an earlier airport ground ation. Although there was no explicit comparison of the
access mode choice model for the Bay Area developed by results of the mode choice analysis with the experience at
Harvey (1988). The values of the ASCs were then estimated other airports, this study put the results of the mode choice
to fit the model to the mode use data from the 1995 MTC Air modeling into a larger context and served to provide some as-
Passenger Survey. surance of the likely validity of the modeling results.
The mode choice model market segment for airport em- San José International Airport Model
ployees included only two modes, private vehicle and public
transit, and is discussed further in the following section on This model was developed by Dowling Associates (2002)
airport employee travel choice. and was designed to estimate the ridership on a planned APM
to connect the airport to a nearby Santa Clara Valley Trans-
Portland International Airport portation Authority light rail line. The model was estimated
Alternative Mode Study using data from an air passenger survey performed at the air-
port for the Bay Area Metropolitan Transportation Commis-
Soon after the Boston Logan model was developed, a similar sion in 1995 and supplemented with the results of stated pref-
modeling effort was undertaken in Portland, Oregon, as part erence surveys that were conducted as part of the study to
of a ground access study for Portland International Airport, determine how air passenger mode choice might be influ-
jointly undertaken by the Port of Portland and Metro, the enced by the availability of the APM and to compensate for
regional MPO, with the assistance of Cambridge Systemat- the limited number of users of the light rail line in the 1995
ics, Inc. (Bowman 1997; Cambridge Systematics 1998; PDX survey sample. The model used a MNL form with separate
Ground Access . . . 1998). The primary purpose of the model coefficients for the same four market segments used in the
was to forecast the potential ridership on potential ground Oakland International Airport–BART Connector model, as
access enhancements, including a planned extension of the well as an airport employee segment. Each air passenger
Portland MAX (Metropolitan Area Express) light rail system market segment included six modes: private vehicle, rental
to the airport. An air passenger survey was done at the airport car, scheduled airport bus, shared-ride door-to-door van,
that combined a revealed preference survey that examined air taxi, and public transit. In addition, the visitor market seg-
passengers’ actual mode use and a stated preference survey ments included hotel shuttle. The model implementation
that was designed to determine travelers’ preferences for allowed for up to four different public transit routes from any

36
given analysis zone, and the model used separate coefficients largely have begun their trip from a private residence,
for bus travel time and rail travel time. In addition, a separate whereas those using taxi or airport bus would be more likely
ASC was used for those routes that included the use of the to have begun their trip from a hotel or place of business. The
APM to reflect the greater attractiveness of this link as deter- model utility functions also do not consider the household
mined from the stated preference survey. The model is very income of the air travelers.
similar in structure and form to the model used for the Oak-
land International Airport–BART Connector analysis and ASCs were initially included in the model utility func-
used the same approach of adopting the model coefficients tions; however, these were found to be not statistically sig-
from Harvey (1988) and estimating ASCs. nificant and were dropped from the model. This is surprising
given the relatively simple form of the utility functions and
The mode choice model market segment for airport em- the absence from the model of such factors as household
ployees included only two modes, private vehicle and public income and the access time involved in reaching the rail link
transit, and is discussed further in the following section on station. Separate model coefficients were estimated for four
airport employee travel choice. market segments: resident business, resident non-business,
non-resident business, and non-resident non-business.
Toronto Air Rail Link Revenue
and Ridership Study In addition to estimating a formal mode choice model, the
study included a benchmark comparison analysis that exam-
In May 2003, Transport Canada issued a Request for Business ined the mode share of existing airport rail links in 24 cities
Case for a public–private partnership to develop an Air Rail in the United States, Europe, and Australia. This analysis
Link between the Toronto Lester B. Pearson International developed cross-sectional regression relationships that ex-
Airport and Toronto Union Station (Request for Business pressed the rail mode share in terms of a series of market and
Case . . . 2003). In preparation for the Request for Business geographical characteristics, such as the percentage of air
Case, a revenue and ridership forecasting study was under- passengers with central city origins, the distance of the air-
taken in 2002 (Halcrow Group 2002a). The study included the port from the central city, and the ratio of rail travel time to
conduct of a stated preference air passenger survey and the taxi travel time from the city center. These relationships were
development of a mode choice model to predict the diversion then used to predict the corresponding rail mode share for
of airport access trips from existing modes to the proposed Toronto using the same regional characteristics. The result-
new rail link. The stated preference survey was carried out in ing range of rail mode shares (which varied with the charac-
February 2002 in the terminal departure lounges and collected teristic chosen) was compared with the results of the formal
data on the air party characteristics, ground trip origin, and mode choice modeling process, to provide a reality check on
ground access mode for the current trip. Some 807 respon- the modeling analysis.
dents were identified as potential air rail link users and com-
pleted a stated preference questionnaire, the results of which
United Kingdom South East and East of England
were then used to estimate a set of binomial logit mode choice Regional Air Service Study Air Passenger
models, each of which models the choice between an existing Surface Access Model
mode and the planned rail link.
As part of the South East and East of England Regional Air
Air travelers using hotel bus or rental car to access the air- Service (SERAS) study undertaken for the U.K. Department
port were excluded from the diversion analysis, because of Transport, Local Government and the Regions, a set of
users of a hotel bus were assumed to have a door-to-door ser- surface access models was developed that included an air
vice that was effectively free, whereas those using a rental passenger mode choice model, an airport employee trip dis-
car were assumed to require the car for other purposes during tribution model, and an airport employee mode choice model
their visit and thus not consider the use of other modes. The (Halcrow Group 2002b). The air passenger mode choice
mode choice model relationships use three continuous vari- model is a NL model that covers 12 defined ground access
ables: the in-vehicle travel time on each mode, the service modes and has separate coefficients and model structures for
headway for the mode, and the travel cost involved in using the following six market segments:
the mode. In addition, the utility function for the rail link
includes two dummy variables: one that indicated whether • U.K. business passengers on domestic trips,
the air traveler was accompanied (whether by other members • U.K. business passengers on international trips,
of the air travel party or by well-wishers) or traveling alone • U.K. leisure passengers on domestic trips,
and one indicating whether the air traveler(s) intended to • U.K. leisure passengers on international trips,
check any bags. The model structure does not directly con- • Non-U.K. passengers on business trips, and
sider the type of trip origin. However, the approach of devel- • Non-U.K. passengers on leisure trips.
oping separate diversion models for each existing access
mode indirectly addresses some of these effects, because air The 12 ground access modes consist of several different
passengers being dropped off by private vehicle would types of rail links, including a dedicated express rail service

37
(such as the Heathrow Express service from Central London model that incorporates an airport trip special-generator
to Heathrow Airport), London Underground, and coach con- model (Dallas–Fort Worth Regional . . . 2007). This model
nections to nearby mainline rail stations, as well as private generates home-based non-work and non-home-based trips
automobile (both drop off and park), rental car, taxi, local corresponding to the forecast air passenger traffic at the air-
bus, and charter and intercity coach. The mode choice struc- port. These trips are then added to other regional trips in each
ture uses a multi-level tree to account for the complex pattern category to determine the overall distribution of regional
of public modes and alternative rail services, with the modes trips before mode choice analysis and trip assignment to the
grouped in a different tree structure for each market segment. regional transportation network. Airport employment is han-
A representative choice structure for one of the market seg- dled through the regular home-based work trip generation
ments is shown in Figure 5, and details of the choice struc- process. The mode choice models do not treat airport trips
ture for the other market segments are included in Appendix differently from other trips in each trip category (Cambridge
D, which can be found in the web version of this report. The Systematics 2005).
utility functions for each mode use a generalized cost ap-
proach that considers travel time, out-of-pocket costs, and Following the 9-11 attacks on the World Trade Center in
time penalties for interchanges, with all costs converted to New York, the Lower Manhattan Development Corporation
equivalent minutes of travel time. was established to rebuild Lower Manhattan and create a
sense of place that would revitalize the economy of the area.
The airport employee mode choice model is discussed To improve the transportation links to Lower Manhattan, the
further in the following section on airport employee travel Lower Manhattan Development Corporation, in cooperation
choice. with the Metropolitan Transportation Authority, the Port
Authority of New York & New Jersey, and the New York
City Economic Development Corporation, undertook a fea-
Other Recent Studies sibility study of alternative ways to provide a new rail link
between Lower Manhattan and suburban commuter markets
In addition to the foregoing nine models, the survey of airport
in Long Island and John F. Kennedy International Airport
authorities, MPOs, and consultants discussed in the next
(Parsons/SYSTRA Engineering 2004). Following an identi-
chapter identified several other recent airport access model-
fication and screening process of potential project alterna-
ing studies that are not discussed in any detail in this report
tives, two alternatives were selected for further analysis: a
owing to limited technical documentation.
new tunnel under the East River between Lower Manhattan
and Brooklyn and the use of the existing Montague Street
In 2002, the North Central Texas Council of Governments
tunnel and realignment of existing subway services. Both
(NCTCOG), together with the Dallas/Fort Worth Interna-
alternatives have a number of potential variants. As part of
tional Airport, Dallas Area Rapid Transit, and Fort Worth
the on-going environmental analysis of these alternatives, a
Transportation Authority completed a Major Investment
more detailed ridership analysis is being undertaken to eval-
Study to evaluate potential rail links to the Dallas/Fort Worth
uate the likely use of the proposed service by airport travel-
International Airport (DMJM Aviation, Inc. 2002). As part
ers under the various project alternatives being considered.
of the study, ridership estimates were prepared for a range of
project alternatives that included shuttle bus links to existing
In 2005, the Aéroports de Montréal undertook a feasibil-
commuter rail lines near the airport as well as direct com-
ity study for establishing a rail shuttle between the city cen-
muter rail service to the airport and extension of a regional
ter of Montréal and Montréal–Trudeau International Airport
light rail system to the airport. The demand modeling was un-
(Guilbault et Associés 2005). The rail link would operate
dertaken by the NCTCOG using its regional travel demand
largely over existing tracks of the Canadian National Rail-
way from the Central Station in Montréal to the vicinity of
0.13338
the airport. The study involved a stated preference survey of
1,000 air passengers at Montréal–Trudeau International
Park & Fly Airport and a further 200 air passengers at Montréal–Mirabel
International Airport, the results of which were used to de-
0.41689
0.18127 velop a mode choice model to predict the likely ridership on
the rail shuttle at varying train frequencies and fare levels.
Coach Taxi Kiss & Fly
0.40598 In July 2000, the Sacramento Area Council of Govern-
ments published the findings of a study of transit access to the
Sacramento International Airport (Sacramento International
Rail Routes . . . 2000). This study reviewed the findings of prior studies ad-
FIGURE 5 SERAS model—Representative mode choice dressing the feasibility of extending the Sacramento light rail
structure: U.K. Business Passengers on International Trips system to the airport through a developing area to the north of
(Source: Halcrow Group 2002b). the city known as the Natomas Basin. The study recommended

38
that the Sacramento Regional Transit District pursue more de- were using either before or after the improvement in service
tailed analysis of the feasibility of an extension of the light rail and did not develop a formal model of airport employee
system to North Natomas and the airport and to preserve the access mode choice.
right-of-way identified in prior studies. In October 2001, the
Regional Transit District initiated the Downtown/Natomas/ A subsequent paper by Riccard (1995) described an air-
Airport Transit Alternatives Analysis/Draft Environmental port employee commute program at Boston Logan Interna-
Impact Statement and Report project. This included the devel- tional Airport that was instituted to encourage employees to
opment of a travel demand forecasting methodology (DKS use alternatives to single-occupant vehicles for their journey
Associates 2002), the conduct of a combined revealed prefer- to work. The paper contains an extensive discussion on fac-
ence and stated preference survey, and the development of an tors that affect airport employee decisions on how to get to
air passenger mode choice model (DKS Associates 2004). and from work as well as a large amount of data on airport
employee commute patterns at the airport, but no formal
modeling of the mode choice process.
AIRPORT EMPLOYEE TRAVEL CHOICE
In general, it appears that most studies addressing airport
In contrast to air passenger ground access mode choice, there employee mode choice have simply used the journey-to-
has been very little attention paid to airport employee ground work component of the regional travel demand model for the
access mode use in the literature. In one of the earliest stud- area, considering the airport as no different from any other
ies to explicitly address airport employee access trip patterns, employment center. However, three specific models of air-
Dunlay (1978) developed a model of airport employee vehi- port employee travel have been identified in the literature. A
cle trips based on information on employee shifts from a sur- special-purpose airport employee mode choice model was
vey of airport employees at Dallas/Fort Worth International developed for the Greater London region as part of the U.K.
Airport. However, the model did not address mode choice SERAS study (Halcrow Group 2002b) and airport employee
and assumed that the mode split found in the survey (which market segments were included in two mode choice models
was almost exclusively private vehicle use anyway) would that were developed to analyze planned APM connections,
remain constant, or at least could be adjusted from exoge- one between the Oakland International Airport and the Coli-
nous data if the model were to be applied in another situation. seum station of the BART system (BART–Oakland . . . ,
Instead, the objective of the model was to predict access and Appendix B: Transit Ridership Procedures and Inputs 2002)
egress vehicle traffic flows by time of day, reflecting that air- and one between the San José International Airport and a
port employees do not enter or leave the airport exactly at the nearby light rail station of the Santa Clara Valley Trans-
start and end of their shift. portation Authority (Dowling Associates 2002). However,
none of these models was estimated directly on airport em-
A later study by Boyle and Gawkowski (1992) examined ployee travel data. The SERAS model was based on one
the changes in ridership on the Q3 bus in Queens, New York, developed for a different study covering the area to the south
when the bus route was extended to better serve employment and west of London Heathrow Airport that included all
areas at John F. Kennedy International Airport, and service journey-to-work travel, not just airport employees. The Oak-
was subsequently improved to provide higher frequency and land International Airport and San José International Airport
earlier and later hours. Because a large part of the increase in models adapted regional travel demand models for home-
ridership was comprised of employees at the airport, it fol- based work trips, but calibrated these to airport employee
lows that this improvement in service resulted in a change of mode use data. All three models are discussed in more detail
journey-to-work mode split. However, the study did not at- in chapter six and the relevant sections of Appendix D (web
tempt to collect data on what other modes airport employees version only).

39
CHAPTER FOUR
USE OF AIRPORT GROUND ACCESS MODELS IN AIRPORT PLANNING
SURVEYS OF AIRPORTS, PLANNING AGENCIES, operators, airport consultants, and airport ground transporta-
AND OTHER ORGANIZATIONS tion operators.
To better characterize the current state of practice with air- In addition to the survey recipients on the mailing list,
port access mode choice models, as well as to identify mod- an e-mail request was sent to the chairs of several TRB
els that may been developed for specific studies but not committees with a scope relevant to airport ground trans-
reported in the published literature, a survey was undertaken portation or transportation modeling issues. The e-mail de-
of airport authorities, regional and state planning agencies, scribed the scope of the study and asked them to forward
federal agencies involved in airport or surface transportation the information to the members and friends of their com-
planning, airport consulting firms, selected universities and mittees and invite any with relevant experience to partici-
other research organizations, and relevant industry associa- pate in the survey. This generated a number of additional
tions. The survey enquired about recent airport ground trans- survey responses.
portation studies undertaken by the responding organization
and whether these involved the use of formal models of air- Four slightly different survey questionnaires were devel-
port ground access mode choice. The survey also enquired oped reflecting the different concerns and needs of the various
about respondents’ perceptions of the usability of such mod- types of organizations included in the survey. In addition to the
els, as well as their awareness of other organizations that online versions of the survey, the four questionnaires were pre-
have experience with the use of these models. pared as Microsoft Word documents that could be downloaded
from the TRB website and completed off-line. Survey recipi-
A list of potential survey recipients was developed and ents were given the option of responding to the survey by
reviewed with the Project Panel. This list was comprised of downloading the appropriate version of the questionnaire and
all U.S. large hub and medium hub airports in 2005 (defined returning it by e-mail, fax, or mail. The four questionnaires are
by the FAA as those enplaning more than 1% and 0.25% of found in Appendix A.
total U.S. enplaned passengers, respectively), as well as the
eight largest Canadian airports, together with a sample of 47 A memorandum from the ACRP staff officer responsible
MPOs and the aeronautics or aviation agencies for each state. for the project was sent by e-mail to each survey recipient
Three of the state aviation agencies (Hawaii, Maryland, and requesting their assistance with the project by participating in
Rhode Island) are also responsible for operating large or the survey and providing a link to the online survey and the
medium hub airports included in the sample and were not Word version of the questionnaire. The memorandum
surveyed separately as state agencies. The sample of MPOs requested survey recipients to respond within approximately
comprised all of those responsible for regions with a Metro- 3 weeks of the initial requests. A reminder e-mail message
politan Statistical Area population in 2003 of one million or was sent at the end of this period to those survey recipients
more, because those were felt to be the most likely to become who had not responded and several follow-up messages were
involved in airport ground transportation planning. Although sent over the subsequent weeks.
the hub size definition used by the FAA is based on enplaned
passengers rather than originating passengers (who use the
ground access system), those airports with a high proportion Survey Response
of connecting passengers are generally the larger airports that
also generate a large amount of ground access trips. As of July 2, 2007, 107 responses representing 105 different
organizations were received. The composition of the re-
In addition, the survey recipients included 42 consulting sponses is shown in Table 1.
firms involved in airport planning studies and 17 academic
and other researchers who were known to have an interest in The overall response rate to the survey was approximately
airport ground transportation studies, as well as selected key 43%. The response rate varied across the different types of
staff in four federal agencies involved in airport ground organizations, from a response rate for large hub airport au-
transportation issues: the FAA, FHWA, FTA, and Volpe thorities, MPOs, and airport planning consultants of around
National Transportation Systems Center. The survey was 50% to a response rate for research organizations of approx-
also sent to three industry organizations representing airport imately 30%. The response rate for state aviation agencies

40
TABLE 1
SURVEY RESPONSE
Sample Sample Other Total
Type of Organization Size Responses Responses Responses
Airport Authorities 26 13
Large hubs 35 15
Medium hubs 8 3 1
Other (Canadian)
Subtotal 69 31 1 32
Metropolitan Planning Organizations 47 23 2 25
State Aviation Agencies 47 16 16
Federal Transportation Agencies 7 3 3
Airport Planning Consultants 42 21 3 24
Research Organizations 17 5 2 7
Industry Organizations 3 0 0
Total 232 99 8 107
was approximately 35%, whereas that for medium hub air- in studies over the past ten years that included some analysis
ports and federal agencies was approximately 43%. One of airport ground access mode choice. The proportion of dif-
MPO forwarded the survey request to its state DOT, which ferent types of organizations that reported some involvement
does the transportation modeling for the MPO. The resulting in ground access studies varied considerably. Perhaps not
response has been included with the MPO responses. surprisingly, most large hub and equivalent Canadian air-
ports (13 of the 14 responding), airport consulting firms
The overall response to the survey was believed to be rea- (20 of 23 responding), and researchers (6 of the 7 respond-
sonable, given the specialized nature of the topic and the ing) reported involvement in studies of this nature. In the
diverse composition of the sample. It is likely that those or- case of airport consulting firms and researchers, this may
ganizations with the most experience with airport ground reflect that those organizations with experience of this type
access modeling would have been more inclined to respond, of study would be more likely to respond to the survey.
and indeed many respondents provided valuable informa- MPOs and state aviation agencies reported a much lower
tion. However, the relatively small number of responses in level of involvement in such studies, with only 8 of the 24
any category limits the ability to draw any statistically robust MPOs that responded to the survey and only one of the 16
conclusions from the survey results for different types of state aviation agencies reporting any involvement.
organizations. Nevertheless, it is believed that the survey
provides a reasonable profile of the extent to which airport The survey responses identified 103 specific studies, to-
ground access mode choice models are used in airport land- gether with some references to other studies that were too
side planning and some indication of how this varies across vague to identify the specific studies in question. Some
different types and sizes of organizations. responses mentioned air passenger surveys. However, these
are not considered analytical studies in the sense intended by
the question and are not included in the count. Several of the
SURVEY FINDINGS respondents appeared to be referring to the same study and
after eliminating duplicate references the responses identi-
In view of the limited number of survey responses for each fied 85 separate studies, ranging from airport master planning
type of organization, the survey findings are presented for all studies through feasibility studies for improved airport
respondents combined, although the role of each type of ground transportation services to academic studies of air pas-
organization is somewhat different. Where there appear to be senger ground access mode choice.
significant differences between the responses to a given
question across the different types of organizations this is A significant number of these studies have not previously
discussed in terms of the actual count responses. been reported in the literature, at least in a way that would
allow a typical literature search to identify them as being rel-
Number of Airport Ground Access evant to airport ground access mode choice analysis. There-
and Airport Choice Studies fore, the survey has been very valuable in identifying a
broader range of airport ground access studies than is gener-
Of the 105 organizations responding to the survey, 57 (53%) ally reported in existing bibliographic reviews of airport
reported that they had sponsored, undertaken, or participated ground access issues.

41
The survey also asked respondents whether their organi- organization’s website. Only five respondents (not the same
zation has undertaken or participated in studies over the past five) indicated that this was the case. Thus, it appears that the
ten years that have included some analysis of air passenger majority of the studies identified in the survey are not readily
airport choice. Of the 105 organizations responding to the available from the Internet and direct follow-up with the
survey, 37 (35%) reported some involvement in such studies. sponsoring organizations would generally be necessary to
Respondents identified 32 separate studies. Although it obtain documentation on these studies.
appears from the titles of the studies that several may not be
airport choice studies in the sense intended by the question, To determine if the use of analytical mode choice models
the responses identified a number of studies that have not may have increased in recent years, as well as to identify ex-
been previously reported in the literature. perience that may not have been reported because it occurred
before the ten-year period mentioned in the earlier questions,
respondents from airport authorities and MPOs were also
Use of Analytical Models asked if their organization had undertaken or commissioned
studies prior to the last ten years that made use of formal
Respondents were asked to state whether any of the studies analytical models of airport ground access mode choice. Of
involving some analysis of airport ground access mode the 55 organizations responding to this question, 9 (16%)
choice made use of formal analytical mode choice models. indicated that they had and 29 (51%) indicated that they had
Of the 57 respondents reporting such studies, 34 (60%) indi- not. A further 17 were not sure whether they had or not.
cated that the studies involved such models and specifically Therefore, of the 38 respondents who knew whether such
identified 52 of the 85 separate studies previously reported. studies had been done, only approximately 24% reported the
Thus, it appears that the use of formal mode choice models in use of such models in earlier studies. This is not significantly
airport ground access studies is quite widespread. different from the percentage reporting the use of such mod-
els in studies undertaken during the past ten years (13 re-
The survey also asked whether any of the studies involv- spondents out of 56 organizations, or approximately 23%),
ing analysis of airport choice made use of formal analytical particularly because those who were not sure or did not
models of airport ground access mode choice as a factor in answer the question may well be less likely to have under-
airport choice decisions. Of the 37 respondents reporting taken such studies than those who knew either way. There-
such studies, 15 (41%) indicated that the studies included the fore, there does not appear to have been a significant increase
use of such models and specifically identified 16 separate in the use of analytical models in recent years.
studies. However, three of these studies were updates of a
regional transportation planning process by the same organi-
zation using the same model. Satisfaction with Existing Models
Respondents who indicated that airport ground access The survey asked those respondents who reported the use of
studies included the use of formal analytical models of mode formal analytical models of airport access mode choice how
choice were asked about the source of air passenger survey they would characterize the current state of practice with
data used in the development of the models. Of the 34 re- such models based on their organization’s experience. Of
spondents reporting studies involving the use of such mod- the 34 organizations that reported experience with the use
els, 23 (68%) indicated that the models utilized data from of such models, 19 (55%) indicated that current models
surveys performed specifically for model development and were adequate for their needs, 12 (35%) indicated that cur-
an equal number indicated that the models used data from rent models are not reliable enough, 3 (10%) indicated that
surveys performed for other purposes. Multiple responses current models are too complex to use, and 10 (30%) indi-
were possible from a given respondent, because their re- cated that they are too costly to use (multiple responses
sponses could cover more than one study or a given study were allowed).
could use data from more than one survey.
Although on its face this suggests a generally high level of
The survey asked those who reported involvement in air- satisfaction with the current state of practice, it is noteworthy
port ground access studies that made use of formal analytical that seven of the eight airport authorities and all five of the
models of mode choice whether reports from any of these MPOs reporting the use of such models indicated that current
studies are available on their organization’s website. Only models are adequate for their needs, whereas only 5 of the 13
five respondents indicated that this was the case, one airport consulting firms reporting involvement in studies using such
authority, two MPOs, and two universities. Further investi- models indicated that current models are adequate. Because
gation established that the airport authority website was an in many cases the studies using such models are performed
intranet site that is not accessible to the general public. The by consultants rather than the airport authority or MPO staff,
survey also asked whether reports on the air passenger sur- and the consultants may well be reluctant to share any con-
vey data used in the development of the mode choice models cerns about the adequacy of the analysis with their clients,
are available on the responding organization’s or another this may suggest that the apparent confidence in the current

42
models by the airport authorities and MPOs could be mis- passenger trips using a special-generator sub-model, 5 (20%)
placed. It is also possible that the definition of what is con- reported that they treated air passenger trips the same way as
sidered adequate varies among the different organizations. other home-based non-work trips, and another 5 reported that
they treated air passenger trips the same way as other non-
Those respondents that reported undertaking or being in- home-based non-work trips. Six MPOs (25%) reported that
volved in airport ground transportation studies that included they modeled airport employee trips using a special-generator
some analysis of airport ground access mode choice but did sub-model, whereas 12 (50%) reported that they treated air-
not make use of formal analytical models of mode choice port employee trips the same way as other journey-to-work
were asked for the reasons that the studies did not include the trips. Three MPOs reported that they modeled air passenger
use of such modeling. Of the 22 organizations in this cate- trips in the regional travel modeling process but not airport
gory that answered the question, 15 (70%) indicated that the employee trips, whereas four other MPOs reported that they
scope of the studies did not require it, 4 (20%) indicated that modeled airport employee trips but not air passenger trips.
there is inadequate information and guidance on the use of
such models, and 3 (15%) indicated that available models are One MPO reported that airport automobile driver trips
not reliable enough or too difficult to use. An additional five were modeled using exogenous trip tables that are developed
respondents indicated that the decision on which analytical from air passenger survey data. Future year trip tables are
techniques to use either was made by those performing the developed using a Fratar-type extrapolation technique
study or they did not know the reason. (Fratar et al. 1954; Papacostas 1987) based on future year
household and employment forecasts. In summary, this tech-
Modeling Airport Trips in Regional nique extrapolates a base year trip table to a future year by
Transportation Planning Process initially applying growth factors to each zonal interchange
flow based on the forecast change in the characteristics of
Of the 23 MPOs that responded to the question about how each zone and then iteratively calculating adjustment factors
they modeled airport trips in their general regional travel and applying them to the zonal interchange flows so that the
modeling process, 15 (65%) reported that they modeled air flows into and out of each zone balance.

43
CHAPTER FIVE
STATE OF PRACTICE OF AIR PASSENGER MODE CHOICE MODELS
Although the details of the different air passenger airport ac- result of a specific effort to model these trips, but rather are
cess mode choice models discussed in the literature review in simply the application of models that have been developed to
chapter three vary widely, it is clear that a standard of best predict urban travel behavior in general to account for airport
practice has evolved, although by no means is it always fol- trips. Given the differences between the characteristics of
lowed. This chapter summarizes that standard and identifies airport trips and other types of regional travel, it can be ex-
aspects where further improvement is needed. pected that the application of general regional travel models
to predict airport trips will not produce very good results, and
indeed there is a growing interest in developing airport-
MODEL APPLICATION specific sub-models to account for these trips within the over-
all regional transportation modeling framework.
The airport access mode choice models reported in the liter-
ature fall into four broad categories. The first category con-
sists of academic studies where the primary objective is to TECHNICAL APPROACH
develop a model that explains air passenger access travel
Current best practice uses NL choice models with separate
behavior. These studies address such issues as the appropri-
coefficients (and possibly including different variables) for at
ate functional form of the model to use, variables to include,
least four market segments:
and market segmentation issues. Some recent studies have
also explored alternative approaches to the traditional MNL
• Resident business trips,
and NL choice models.
• Resident non-business trips,
• Non-resident business trips, and
The second category consists of models that have been
• Non-resident non-business trips.
developed in support of specific airport ground access plan-
ning studies, such as an evaluation of the feasibility of con-
The modes available for resident and non-resident trips
structing an APM link between the airport and a nearby rail
will generally be different, because non-residents do not have
station or of extending an urban or intercity rail system to an
the option of parking a private vehicle at the airport (indeed
airport. These models are often less detailed than the first cat-
this would make no sense because their visit to the region
egory of models, possibly because the primary focus of the
begins at the airport and they return to the airport at the end of
study is not on model development, and the time and
their stay). On the other hand, many non-residents rent a car
resources available for model development are limited.
on their arrival at the airport to provide local transportation
during their stay in the region, whereas most residents already
The third category consists of airport access mode choice
have a car that they can use for airport trips and do not con-
components of models developed to explain or predict air
sider renting a car for their airport access or egress trip.
passenger airport choice. These too often tend to be fairly
simple in terms of the number of modes included in the
model, the variables used, and the structure of the choice Revealed or Stated Preference
process, in part because the primary focus of the study is on
airport choice rather than ground access mode choice, and As discussed in chapter two, there are two different ap-
many of the more detailed aspects affecting ground access proaches to assembling the necessary data to develop airport
mode choice (such as differences in fare and travel time for ground access mode choice models: revealed preference and
different public transportation services) have only a second- stated preference surveys. The majority of past airport access
order effect on airport choice. mode choice models have been developed using revealed
preference techniques. However, in a number of recent stud-
Finally, the fourth category of airport access mode choice ies, including ridership studies for proposed airport rail links
models consists of the components of regional transportation in Chicago and Toronto, demand analysis for the planned
planning models that are used to generate estimates of vehi- Airport MAX Red Line extension to Portland International
cle trips to and from the airports in the region. In many cases Airport, and a feasibility evaluation of a proposed APM link
these are not true airport access models, in that they are the to a light rail stop at San José International Airport, stated

44
preference studies have been done to develop model coeffi- car even if they are staying at a nearby hotel that has a free
cients for the planned modes or services that do not yet exist, shuttle to and from the airport.
and thus are not reflected in surveys of actual mode choice.
The form of the utility functions for each choice alternative
will generally be a linear combination of explanatory vari-
Functional Form of the Model ables with their associated coefficients. However, some vari-
ables are best entered in the utility function as an inverse or
Although the MNL model continues to be used in airport ratio. For example, the service headway of scheduled modes,
access mode choice models, it is becoming accepted that which is a direct measure of average waiting time, is the
the NL model is generally a more appropriate formulation. inverse of the service frequency. The effect of household
The airport ground transportation system consists of a large income may best be entered in the utility function by express-
number of different modes and sub-modes, which are likely ing direct travel costs as a ratio of the cost to some function of
to be viewed by a traveler as having very different substi- the per capita or total household income. Therefore, higher-
tutability. Therefore, a traveler trying to decide between income travelers will be less influenced by cost than lower-
parking in a close-in parking lot or a remote parking lot with income travelers.
a shuttle bus connection to the airport terminal is not likely to
care very much about differences between bus fares and taxi Some model developers have attempted to reflect nonlin-
rates. Conversely, a traveler trying to decide between the earity in the effect of some variables (such as service
lower cost but longer travel time of taking transit to the air- frequency or income) through the use of the logarithm of the
port and the higher cost but greater convenience of a taxi or variable value in the utility function. However, there are the-
shared-ride van is unlikely to care very much about differ- oretical problems with this approach and great care is needed
ences in parking rates between on-airport and off-airport lots. when introducing logarithms into the specification of terms
It is precisely the effect of these different substitution rates in the utility function. Although a logarithmic transformation
that the NL model is designed to reflect. Because the private may approximate some other function (such as the inverse)
vehicle generally accounts for the majority of airport ground over part of the data range (with appropriate coefficients), it
access trips at most airports, at a minimum an air passenger can differ significantly at lower and higher values. However,
mode choice model should nest the private vehicle choice it can be precisely at these values that the effect of the vari-
alternatives for residents of the region (drop off and park for able on traveler behavior is most important. Therefore, it is
the duration of the trip). critical to consider whether the logarithmic transformation of
the variable best reflects the nonlinear effect that is desired
Although no generally accepted practice has yet emerged compared with some other function.
for how to structure the nests of a NL model, this is largely
determined by the characteristics of the different modes be-
cause the primary purpose of using a NL model is to allow Survey Data Used in Model Development
higher rates of substitution between modes that have similar
characteristics. Therefore, it would appear logical to group Although air passenger or airport employee surveys are some-
private vehicle modes in one nest, with different parking op- times performed primarily (or entirely) to support develop-
tions as a second-level nest, group exclusive ride on-demand ment of airport access mode choice models, more commonly
modes (taxi and limousine) together in a second first-level model development makes use of surveys conducted for other
nest, and group shared-ride scheduled modes (public transit, purposes. However, because such models, particularly air
scheduled airport bus) together in a third first-level nest, pos- passenger models, require detailed information on air traveler
sibly with different transit options (e.g., rail and bus) as a characteristics and factors likely to influence their choice of
second-level nest. It is not clear where the door-to-door access mode, it is highly desirable that if a survey will be (or
shared-ride van would best fit in this structure, as a separate even might be) used later to support development of an airport
mode at the top level, in the on-demand nest with taxi and access mode choice model, the survey questions are reviewed
limousine, or in the shared-ride nest with the scheduled by modeling specialists before the survey is performed to
modes. This may be an issue to resolve empirically by ex- ensure that key questions are not omitted that would later
ploring which option gives the best fit to the data. Alternative compromise the ability to develop a reasonable model.
access modes to scheduled services can also be included as
lower-level nests to each mode. Among the factors that are often omitted from air passen-
ger and employee surveys, but are critical for model devel-
Rental car and hotel shuttle use by non-residents is best opment, is information on household income and household
modeled outside this choice process, because use of both composition. It is self-evident that traveler decisions between
modes is determined by factors that are largely independent alternative access modes that offer a trade-off between travel
of the service levels of other modes. Rental car use is often time and cost will be influenced by their perceived value of
determined by local travel needs other than the airport egress time (even if they do not think of the decision in those terms).
and access trip. Therefore, visitors to the region may rent a Because one cannot meaningfully ask survey respondents to

45
state their value of time, the functional form of the model estimation, it is implicitly assumed in the modeling that the
needs to include some measure of household income so that estimated values of the model coefficients convert the values
the values of time implied by the model coefficients will vary of the service levels used in the modeling to the values per-
by income. However, as noted elsewhere in this report, will- ceived by the travelers. Therefore, the model coefficients
ingness to pay to save time is influenced not only by the total serve two purposes: to adjust the values of the service levels
household income, but also by how many people that income used in the modeling to the values perceived by the travelers
has to support. Although these questions are often omitted as and to express those perceived values in terms of their contri-
intrusive or even of little relevance for other purposes to bution to the overall perceived utility of each alternative.
which the survey data will be put, experience has indicated Because it is highly likely that many airport travelers have
that reasonable responses can be obtained if the questions are only limited information about many of the alternatives in-
asked in the right way. cluded in the modeling, and the perception of the transporta-
tion service levels relative to the values used in the modeling
Another factor that affects airport access travel decisions is may vary widely across the respondents to surveys used to
the time that a traveler begins a trip to or from the airport, both develop the models, this is likely to have a significant impact
in terms of the time of day and (in the case of departing air pas- on the overall ability of the model to explain the observed
sengers) the time remaining before their scheduled flight choices. Unfortunately, the question of how well airport trav-
departure. This information can also be useful in the case of eler perceptions of the service levels of the different modes
shared-ride van services to determine how much lead time the conform to the values used in modeling is one that has been
operator required to schedule other pickups, because this largely ignored in the literature on modeling practice.
can affect the overall access time experienced by the traveler.
Similarly, with scheduled modes, this information can provide Because it is necessary to determine the transportation
an indication of how much margin of error travelers allow in service levels faced by each travel party included in the mod-
their travel plans to avoid missing scheduled services. eling process, it is normal practice to define a system of
analysis zones and assign the ground access trip origin of
each respondent in an airport traveler survey to the appropri-
Transportation Service Data ate analysis zone. To provide a reasonable level of precision
for the ground access service levels, in particular travel
In addition to data on the characteristics of the airport travel- times, these analysis zones cannot be too large. A fairly com-
ers, mode choice model development and application mon practice is to use the TAZs defined for the regional
requires data on the transportation service levels of the dif- travel demand model developed for general transportation
ferent modes. These will include travel times, fares or costs, planning in the region.
service frequencies of scheduled modes, any walking dis-
tances involved, and network connectivity issues such as the The relevant regional transportation planning agency will
number and type of anticipated transfers required between generally be able to provide the highway travel times and
different services. The extent to which different explanatory transit service levels (travel times and fares) from each TAZ
variables have been included in various existing models is to the airport from the regional transportation network model.
discussed later in this chapter. The extent to which these travel times can be expressed in
terms of their components (e.g., walking time, waiting time,
Generally, these data are obtained from transportation net- and in-vehicle time) or different values can be provided for
work models or from published information by transportation different times of day or different days of the week is likely to
providers, rather than by asking survey respondents. There are vary from agency to agency. In general, regional travel
two reasons for this. The first is that transportation service- demand modeling tends to be based on average weekday con-
level data are required for all the modes included in the model ditions and distinguish between peak and off-peak travel
and it would be too cumbersome to ask each survey respon- times. However, it is not uncommon to only model one peak
dent for their perception of the value of these service levels for period (typically the morning peak).
every mode, and indeed they may have no idea of the details
of those modes that they did not use. The second reason is that To develop the appropriate transportation service data for
for any application of the models, different values of at least the other airport access modes, such as shared-ride van ser-
some of the transportation service levels are typically vices, taxis, and any scheduled airport bus services, it will be
assumed, whether to reflect expected future values or to necessary to obtain the relevant service information (travel
model the effect of changing the ground access system in times, fares, and schedules) from the transportation pro-
some way, and therefore the values used in the model appli- viders and convert these to the analysis zone system being
cation can be established in a consistent way to those used in used. Given the number of analysis zones typically used,
the model development. this can be a significant task. This task can become more
complicated if some time has passed since the airport trav-
To the extent that the traveler perception of the service lev- eler survey that is being used to develop the model was per-
els of the various modes differs from the values used in model formed, because transportation providers may have changed

46
their service levels in the interval and do not always main- van service. Public transit service is likely to be very location-
tain good records of past service levels. One potentially use- specific. Representation of the use of private vehicles needs
ful source of information on transportation provider service to distinguish between air passengers who are dropped off
levels may be the ground transportation section of the air- at the airport and those who park the vehicle at or near the
port website, where this provides the relevant level of de- airport for the duration of the air trip, because there are sig-
tail, or websites of the transportation providers. Ideally, this nificantly different costs involved and being dropped off at
information would be archived whenever an airport trav- the airport generates additional vehicle travel for the return
eler survey is done, so that it can be referred to later if the trip. Whether or not those dropping off or picking up air
data changes; however, unfortunately this is not often done. passengers park the vehicle for a time at the airport is less
An historical archive of some website content is available important and generally not addressed in mode choice mod-
through the Internet Archive, a nonprofit organization that els. In any event, the decision of whether or not to park is
maintains an online web archive called the Wayback Ma- likely to be determined by factors not typically included in
chine (http://www.archive.org). However, this only provides ground access mode choice models, such as whether the ve-
the content of web pages and cannot recover information in hicle driver arrives at the airport earlier than intended.
underlying databases that may have been accessed by web
pages. Whether or not to model the choice of parking location by
those parking for the duration of their air trip is a more diffi-
Travel times by services such as taxi and limousine will cult question. Most airports offer a choice of parking facili-
generally be the same as other highway modes. Taxi fares ties, with less expensive options often involving a shuttle bus
that are distance-based can be calculated from the meter rates ride to reach the terminal. In many cases there are also pri-
and highway distances obtained from the regional travel vately operated off-airport parking lots that provide a shuttle
demand model network data. Shared-ride van travel times bus service to the airport. These various facilities typically
will typically depend on the number of other parties that have charge different daily rates. Thus, the choice of which facil-
to be picked up after picking up the party in question and the ity to use affects the cost of the ground access trip as well as
additional travel involved. Because this will vary from air the travel time involved in the access trip. Where these
party to air party, it will be necessary to make some assump- choices are not explicitly included in the model, the travel
tions about the average travel times experienced from a given time and cost assumed for parking for the duration of the air
analysis zone. In the case of fixed-route services, such as trip needs to reflect the proportional use of the different park-
scheduled airport bus or rail modes, consideration needs to ing options, which will vary with the air trip duration.
be given to how airport travelers will access the relevant stop
or station where they will board the service. In the case of The degree to which it is necessary to identify specific
nested mode choice models, the access mode to the fixed- public transportation services depends to an extent on the
route service may be treated as part of the choice process and purpose of the model. Because in general public transit tends
will require separate travel times and costs for the different to attract a fairly small mode share at most U.S. airports,
access alternatives (e.g., walk, drop off by private vehicle, except at those with an extensive fixed-rail system serving
taxi, and public transit). Where this secondary access choice the airport (e.g., the Washington Metro at Ronald Reagan
process is not explicitly modeled, assumptions will need to National Airport), it may be sufficient to consider this a sin-
be made about the access time and cost to be included in the gle mode and assume that travelers will use the best path
utility function of the primary mode. through the network. However, where the model will be used
to evaluate specific services or analyze the market for the
introduction of a new mode, it will then be necessary to iden-
MODES INCLUDED IN MODEL tify the alternative services in more detail so that ridership on
specific services can be calculated. The model developed for
The modes to be included in an airport access model are Boston Logan International Airport discussed earlier in-
largely determined by the modes available at the airport, cluded rail transit, the Logan Express scheduled airport bus
although not all the modes available at an airport may be in- service from off-airport terminals in the region, and a Water
cluded in a given model. In particular, some modes may be Shuttle ferry that ran between the airport and downtown
combined in the model or excluded from the analysis with Boston. However, the model does not distinguish between
users assumed to be captive to the mode. The modes included the different Logan Express terminals. In contrast, the model
in nine recent airport access mode choice models described developed for the SERAS study in the United Kingdom in-
in chapter three are shown in Table 2, together with the type cludes an explicit representation of several different types of
of model used and the nature of the data from which the rail services to be able to model how changes in specific rail
model was estimated. services affect not only the overall mode share of rail but how
those using rail would choose between the different services.
Rental car and taxi are generally available at all airports.
Most of the larger airports will also have some form of lim- Two important issues to be addressed are how to incorpo-
ousine (black car) service and some door-to-door shared-ride rate rental car use and the use of a courtesy shuttle bus service

47
TABLE 2
MODEL TYPE AND MODES INCLUDED IN RECENT AIRPORT ACCESS MODE
CHOICE MODELS
Airport or Study
ATL BOS CHI MIA OAK PDX SJC YYZ UK
Year Model Developed 2002 1996 2004 1995 2001 1997 2002 2002 2002
Model Structure
Binomial logit
Multinomial logit
Nested logit
Estimation Data RP RP RP/SP RP RP RP/SP RP/SP RP/SP RP
Modes in Model
Private vehicle—drop off
Private vehicle—park
Private vehicle—park ST
Private vehicle—park LT
Private vehicle—park OA
Private vehicle—combined
Rental car
Taxi
Limousine
Shared-ride van
Scheduled airport bus
Express bus from OAT
Transit (all services)
Transit (rail)
Transit (bus)
Intercity coach
Airport express train
Intercity rail/coach links
Ferry (water shuttle)
Hotel shuttle bus/van
Charter coach
Other private modes
Other public modes
Model: ATL = Hartsfield–Jackson Atlanta International Airport (Travel Demand . . . 2005).

BOS = Boston Logan International Airport (Harrington 2003).
CHI = Chicago O’Hare International Airport, Chicago Midway Airport (Resource Systems Group 2004).
MIA = Miami International Airport (ICF Kaiser Engineers 1995).
OAK = Oakland International Airport (BART–Oakland . . . 2002).
PDX = Portland International Airport (PDX Ground Access . . . 1998).
SJC = San José International Airport (Dowling Associates 2002).
YYZ = Toronto Lester B. Pearson International Airport (Halcrow Group 2002a).
UK = United Kingdom SERAS study (Halcrow Group 2002b).
Notes: RP = revealed preference data; SP = stated preference data; ST = short-term (i.e., associated with air passenger
drop off); LT = long-term (i.e., park on airport for duration of air trip); OA = off-airport; OAT = off-airport
terminal (e.g., Logan Express service at Boston)—includes inter-airport transfer coach (London).

48
from nearby hotels in the model. Visitors to the region may number of transfers in both models can be thought of as
rent a car for local travel quite independently of their airport measuring the perceived disutility of a transfer in terms of
egress and access travel decisions. Those staying at hotels the equivalent additional riding time on a non-stop service.
with courtesy shuttle bus service are likely to use that service, The SERAS model distinguished between different types of
because it is usually free, unless they rent a car for other rea- transfer, such as cross-platform or those involving a level
sons. Therefore, it would appear appropriate to model rental change.
car use by visitors first and then assign visitors staying at
hotels with courtesy bus service to that mode if they are not In contrast, the model developed for the Portland ground
assigned to a rental car. access study (Bowman 1997; Cambridge Systematics 1998;
PDX Ground Access . . . 1998) combined the different com-
In some cases residents of the region may also rent cars ponents of travel time into a single variable for total travel
for their airport access trip, particularly if they live some dis- time without weighting any of the components. The model
tance from the airport and will be away for some time on developed by the Atlanta Regional Commission included sep-
their air trip. Local rental car agencies may not charge a drop- arate coefficients for highway travel time, transit in-vehicle
off penalty for returning the car to the airport, and two one- time, waiting time, and walk time, although the values of
way rentals may be much less expensive than parking their these coefficients were adopted from other models and not es-
own car at the airport for the duration of their trip or using timated for the Atlanta region. Similarly, the model developed
other modes. Therefore, a rental car can be included in the for the Miami Intermodal Connector project distinguished
regular choice set for resident travelers. In general, it will between in-vehicle time on all modes and the combination of
have a very low probability of being chosen owing to the waiting and terminal time (the time required to access the
high fixed cost, but will become more attractive for the type zone centroid in the transportation network), although this
of trip described earlier. model also adopted coefficients from another model for a dif-
ferent region.
EXPLANATORY VARIABLES
Cost and Income
Existing models differ widely in terms of the explanatory
variables used in the utility functions, as well as how those Not unreasonably, most models combine costs into a single
variables are defined. The variables included in the nine air- variable, although the Boston Logan model defined different
port access mode choice models discussed in the previous cost variables depending on the income of the traveler and
section are shown in Table 3. whether the employer was paying the cost of a business trip.
However, this was simply a way to obtain different cost
All existing models include some measures of travel coefficients for each case. Several models have divided the
time and cost. However, the extent to which travel time is sep- cost by a function of the household income, although there is
arated into its components (walk time, waiting time, in-vehicle no consistency in what function was used.
travel time, and time involved in private vehicle access to pub-
lic transportation services) varies. The model developed for The Boston Logan model (Harrington et al. 1996; Harring-
Boston Logan International Airport estimated separate coeffi- ton 2003) divided respondents who paid their own travel
cients for in-vehicle time, walk and wait time (combined as expenses into two income categories (low and high) and esti-
out-of-vehicle travel time), and automobile access time to pub- mated separate travel cost coefficients for each income cate-
lic transportation. The models developed for Oakland Interna- gory as well as for a third category of respondents whose travel
tional Airport and San José International Airport to evaluate expenses were paid by their employer. The Portland ground
proposed APM links (BART–Oakland . . . 2002; Dowling access study model (Bowman 1997; Cambridge Systematics
Associates 2002) used separate coefficients for walk distance, 1998; PDX Ground Access . . . 1998) divided the travel costs
waiting time, and in-vehicle time in private vehicles, buses, incurred by air passengers by the natural logarithm of the
and light rail, although these were adopted directly from household income, although this was not done for the automo-
Harvey (1988). The use of separate coefficients for in-vehicle bile operating cost for drop-off trips.
time in the different modes was intended to capture differences
in the perceived comfort and convenience of rail compared The Oakland Airport Connector model (BART–Oakland
with bus, as well as between private vehicle and transit. The . . . 2002) and the San José International Airport Model
models developed to estimate ridership on the proposed (Dowling Associates 2002) divided the travel costs for per-
Chicago Airport Express train (Resource Systems Group sonal trips by the household income raised to the power 1.5,
2004) used a weighted travel time, with the weight for access but did not do this for business trips.
time, waiting time, and transfer time adjusted incrementally to
obtain the best model fit. Both the Boston Logan model and the Although each of these approaches will ensure that
U.K. SERAS model included fixed penalties for transfers on higher-income travelers (at least those paying their own
the public transit system. The coefficient of the variable for the expenses in the case of the Boston Logan model and those

49
TABLE 3
EXPLANATORY VARIABLES INCLUDED IN RECENT AIRPORT ACCESS MODE
CHOICE MODELS
Airport or Study
ATL BOS CHI MIA OAK PDX SJC YYZ UK
Continuous Variables
Off-peak highway time
Travel time (private vehicle)
In-vehicle time (all modes)
In-vehicle time (transit)
In-vehicle time (rail transit)
In-vehicle time (bus transit)
In-vehicle time + walk time
Out-of-vehicle timea
Total travel time (all modes)
Weighted travel timeb
Wait time + terminal timec
Wait time/headway
Walk time
Walk distance
Auto access time to transit
Number of transfers
Interchange penalties
Drop-off driver time
Driving distance
Travel cost (all modes)
Private vehicle cost
Drop off vehicle operating
cost
Transit fare
Taxi fare
Household income
Dummy variables
Air party size
Baggage check-in
Employer pays cost
Flights/year
Household income
Luggage/checked bags
Non-residence trip origin
One-person travel party
Use of intermediate station
CHI = Chicago O’Hare International Airport, Chicago Midway Airport (Resource Systems Group 2004).
UK = United Kingdom SERAS study (Halcrow Group 2002b).
a
Out-of-vehicle time on Boston model combined waiting time and walking time (access, transfer, and egress) as
appropriate for the mode.
b
Weighted travel time in Chicago model combined in-vehicle time and egress time with a weighted sum of access time,
transfer time, and waiting time.
c
Terminal time in Miami model used the regional travel demand model zonal terminal times that represent the
additional travel time required to reach the TAZ centroid from the actual trip origin.

50
making personal trips in the case of the San José Interna- One issue that arises with scheduled modes is how to han-
tional Airport model) will be less sensitive to travel cost than dle the waiting time involved. There is an extensive literature
lower-income travelers, clearly the effect on travelers of a on how travelers value the disutility of waiting time relative
given income level varies significantly. It is most unlikely to travel time, particularly in the context of transfers between
that they all correctly reflect sensitivity to income, and those different transportation modes or services (e.g., Mohring
that simply classify travelers into two income categories et al. 1987; Moreau 1992; Small 1992; Chang and Hsu 2001;
obviously can only approximate the effect. Indeed, common Hensher 2001; Lam and Small 2001). The general consensus
sense suggests that total household income is only part of the is that waiting time has a higher disutility (higher perceived
story and the effect of household income must depend in part value of time savings) than in-vehicle travel time, although
on the size of the household. There is clearly a huge differ- the ratio between the two appears to vary with circumstances,
ence between a single person making $100,000 per year and typically in the range between 1.5 and 2.5.
a family of six making the same annual income.
Some authors have argued that because the schedule is
A related issue that is generally ignored is the difference published and people will arrive a certain amount of time be-
of purchasing power of a given income level between resi- fore departure, waiting time is effectively independent of ser-
dents of the area in question and visitors. Perceived values of vice headway. Others have used half the service headway as
time are likely to depend on the level of discretionary income the average waiting time. This is commonly referred to as
of individual travelers, which in turn is affected by the cost schedule delay, and it is argued that although the traveler
of living in their home region rather than that of the region may only wait at the bus stop or station for few minutes, they
where the airport is located. Therefore, it can be expected have a time that they would prefer to have departed and the
that visitors to a region from another region where the cost of difference between that time and the time that they actually
living is significantly different are likely to have different depart is part of the disutility of travel, even if they are actu-
perceived values of time for a given level of per capita house- ally doing something else. In the case of airport access and
hold income than residents of the region where the airport is egress trips, this argument is quite persuasive. Airport trav-
located. This issue has important implications for the trans- elers will generally take an earlier bus or train if one is not
ferability of mode choice models from one region to another, leaving at the time they would prefer to have departed, be-
as discussed further in chapter seven. cause taking a later service may cause them to miss their
flight or, in the case of airport employees, be late for work.
In view of the lack of any well-established practice re- However, arriving early at the airport has relatively little
garding how best to incorporate traveler income into airport value, because they are likely to just spend longer waiting in
ground access mode choice models, this would seem to be a the gate lounge or for their shift to start. Similarly, in the re-
useful topic for future research. verse direction, any time spent waiting at the airport for the
next departure from the airport is generally not usable for
other activities apart perhaps from reading.
Travel Time Components
The changes in the amount of time that air travelers have
As discussed earlier, it is widely recognized that the different to allow for airport security screening since the events of
components of travel time are likely to have a different dis- 9/11 adds a further complication to how passengers may per-
utility per minute. This is particularly true for waiting time and ceive the disutility of arriving at the airport earlier than they
walking time (although this may be measured by distance would prefer. On the one hand, any additional time provides
rather than time). There are two approaches to addressing something of a buffer against unexpected delays in clearing
this. The first is to weight the various components of travel security. On the other hand, if passengers are already con-
time differently before combining them into a single travel cerned about the amount of time that they have to spend at
time variable for the purpose of model estimation. The sec- the airport because of the need to allow enough time for
ond is to estimate separate coefficients for each component. security delays, they may be even more intolerant of having
The latter is preferable, because assumptions for the relative to arrive at the airport earlier than they would prefer owing to
weights of different travel time components based on the the waiting time while traveling to the airport.
experience of other models may not be appropriate for the
particular situation being modeled. Some models estimate A related issue arises from the schedule dependability
separate coefficients for travel time on different modes. This when a connection is involved. If the trip to the airport
has the advantage over the use of a single travel time variable involves connecting between two (or more) scheduled ser-
for all modes that if the relative perceived disutility of differ- vices, then the traveler needs to consider the possibility that
ent modes varies by length of trip, this will be reflected in the if one service runs late, they may miss their planned connec-
estimated values of the coefficients. Only relying on ASCs to tion and have to wait for the next departure on the subsequent
reflect differences in the perceived disutility of different segments of the trip. Therefore, the perceived schedule relia-
modes implicitly assumes that any differences are indepen- bility may play an even more important role than the actual
dent of the length of the trip, which is highly unlikely. waiting time involved according to the schedule. Ironically,

51
the less the waiting time involved in making connections gage that an air party has. However, some thought needs to
according to the published schedule, the higher the probabil- be given to the appropriate specification of a baggage vari-
ity that the traveler may miss the connection, unless the out- able in the post-9/11 environment, in which there are strict
bound services at the connecting point wait for the inbound limits on carry-on bags. As a result, passengers may well
services to arrive. This issue is not unique to airport ground check a bag, even though it is not large enough to present any
access trips and arises with all scheduled transportation limitation to their use of public transportation. The wide-
services. spread adoption of wheeled bags over the past decade has
also reduced the difficulty of carrying bags on public trans-
The issue of travel time dependability is not restricted to portation services.
waiting time. One advantage of rail access modes is that the
travel times involved are not subject to the variability that can Not surprisingly, the issue of whether any of the travel
affect highway travel times resulting from accidents or unex- costs will be paid by someone other than the traveler was
pected congestion. Thus, it might be expected that the per- found to be a significant factor in mode choice in the
ceived disutility of each minute of travel time on a rail mode Boston Logan model (Harrington et al. 1996; Harrington
might be different from that of each minute of travel time on 2003), the only model where this has been included.
highway modes. Because the disutility of time spent on dif- Although this is not a factor that can be independently fore-
ferent modes is also affected by the level of comfort offered cast and simply applies to a particular subset of respondents
by the mode and other factors, such as perceptions of personal in a given survey sample (generally those on business
safety, it is not obvious whether on balance the higher relia- trips), its inclusion in a model improves the model fit to the
bility of rail modes could be offset by other factors. The use data and helps reduce any bias in the coefficients of other
of separate coefficients for travel time on each mode allows variables that might result from not considering this issue.
the model estimation process to determine the relative per- Similarly, the Boston Logan model found that familiarity
ceived disutility. However, the more model coefficients that with the airport and available ground transportation ser-
have to be estimated, the larger the dataset needs to be to ob- vices, as measured by the number of flights from the airport
tain statistically significant estimates of the coefficients. in the past year by the survey respondents, was a significant
factor in explaining mode choice of residents of the region.
Although frequency of use of an airport may be a useful
Other Variables indicator of familiarity with alternative ground transporta-
tion services, it is likely to be a fairly inaccurate one. Some
In addition to travel time and cost, several models have at- frequent travelers may make no effort to learn about the al-
tempted to include other variables believed to influence air ternative transportation options, whereas some infrequent
passenger ground access mode choice, including the amount (or even first-time) users of the airport may have researched
of luggage that an air party has, familiarity with the airport, their different options or sought advice from more frequent
and whether the traveler or someone else is paying for users. Therefore, ideally, air passenger surveys would
the trip. Some models, such as the Boston Logan model attempt to obtain a better indication of the familiarity of
(Harrington et al. 1996; Harrington 2003), have also included respondents with alternative ground transportation ser-
air party size as a separate variable. However, to the extent vices. However, it is far from clear how to do this effec-
that some costs, such as transit fares, vary with the number of tively and this is an important aspect for further research,
people in the travel party, it is generally better to address the because it is self-evident that travelers will not choose ser-
effect of air party size through the calculation of the travel vices that they do not know exist.
costs of the different modes, rather than try to handle this
through a separate variable, which simply adds a fixed
amount to the disutility of the affected modes irrespective of Implied Values of Time
the actual value of the travel costs involved.
The ratio of the travel time coefficient to the cost coefficient
Those models that have included variables reflecting the can be interpreted as the implied value of time (strictly of
amount of checked baggage, such as the Boston Logan travel time savings). These values for eight of the mode
model (Harrington et al. 1996; Harrington 2003), the Toronto choice models described in Tables 2 and 3 and discussed in
Air Rail Link model (Halcrow Group 2002a), and the earlier more detail in Appendix D (web only version) are summa-
work by Harvey (1988) on which the models for Oakland rized in Table 4. The implied travel time values for the U.K.
International Airport and San José International Airport were SERAS model are not included in the table because they are
based, have generally found this to be a significant factor in expressed in pounds and were derived using a different
explaining model choice. It would therefore seem desirable methodology; therefore, they are not directly comparable to
to include this in any airport access mode choice model, par- the U.S. experience. Because separate coefficients were esti-
ticularly if the model is to be used to study proposed services mated for the two Chicago airports in the Chicago Airport
that include off-airport baggage check-in, where the attrac- Express study, Table 4 presents the implied values of time for
tiveness of that capability will depend on the amount of bag- each airport separately.

52
TABLE 4
IMPLIED VALUES OF TIME FROM RECENT AIRPORT ACCESS MODE CHOICE MODELS
Airport or Study
ATL BOS ORD MDW MIA OAK PDX SJC YYZ
Year of Cost Data a 1993 2003 2003 b c 1996 c 2002
Travel Times ($/hour) d e f
Highway time g h h i
Resident business trips 15 11 33 63 78 15 19 15 53
Resident non-business trips 13 17 25 22 78 16 29 10 29
Non-resident business trips 16 40 33 63 78 15 19 15 71
Non-resident non-business 12 13 25 22 78 16 30 10 34
trips
Transit in-vehicle time j k k
Resident business trips 11 26 33 63 78 11 19 11 53
Resident non-business trips 9 7 25 22 78 12 29 7 29
Non-resident business trips 12 15 33 63 78 11 19 11 71
Non-resident non-business 9 9 25 22 78 12 30 7 34
trips
Travel time (other cases) l m m n o n
Resident business trips 22 92 82 20 24 20
Resident non-business trips 38 55 57 19 37 12
Non-resident business trips 40 92 82 20 24 19
Non-resident non-business 13 55 57 19 39 11
trips
ORD = Chicago O’Hare International Airport (Resource Systems Group 2004).
MDW = Chicago Midway Airport (Resource Systems Group 2004).
a
Coefficients for time and cost adopted from other models (date unspecified).
b
Coefficients for time and cost adopted from another model (date unspecified).
c
Coefficients for time and cost adopted from Harvey (1988), 1995 cost data.
d
For household income (only applies to non-business trips) of $75,000/year.
e
For household income (only applies to non-business trips) of $55,000/year.
f
Canadian dollars.
g
Private vehicle drop off, costs not reimbursed, low income (undefined).
h
Weighted travel time, low household income (less than $100,000/year).
i
For household income of $50,000/year.
j
Costs not reimbursed.
k
Rail transit.
l
Private vehicle drop off, costs not reimbursed, high income (undefined).
m
Weighted travel time, all modes, high household income ($100,000/year or more).
n
Bus transit.
o
For household income of $150,000/year, all modes.

53
In addition to the implied values of time, Table 4 shows the recent models developed in the United States have used
the year for which the cost data were obtained where this is this approach, although the model developed for the United
known. Because the implied value of time is obtained from Kingdom the SERAS Study (Halcrow Group 2002b) to ana-
the ratio of the travel time coefficient to the cost coefficient, lyze surface access to airports in the London region included
it is expressed in dollars of the year for which the cost data two additional market segments to distinguish between do-
were assembled from which the model was estimated. In mestic and international trips by U.K. residents. Although this
cases where the travel time and cost coefficients were model distinguished between U.K. residents and non-U.K.
adopted directly from other models, the value of time is residents, rather than residents of the London region and vis-
expressed in dollars of the year for which the cost data used itors to the region, the distinction is less critical than it might
in estimating those models were assembled. at first appear. Distances in the United Kingdom are such that
many U.K. residents will travel to the London airports by sur-
For those models where household income was included face modes for international trips, whereas visitors to the
in the cost term of the utility function or different cost coef- United Kingdom will fly in to the London airports, even if
ficients were estimated for different income categories, the their final destination is elsewhere in the country. The inclu-
implied value of travel time is a function of income (or varies sion of a U.K. domestic market segment accounts for travel
by income category). In those cases where the values of time between London and the rest of the United Kingdom. Even
vary with income the values shown in Table 4 have been cal- so, this segmentation is problematical on a number of counts,
culated for specific household income levels. and if the distinction between domestic and international
travel is considered to be important (as it might well be in the
Because the different models were estimated using cost United States as well as the United Kingdom), then it would
data for different years, the values of time are not directly probably be desirable to distinguish between international and
comparable. Furthermore, for those models where the implied domestic visitors to a region as well as international and do-
value of time is a function of household income, the values mestic travel by residents of the region, giving eight market
shown in Table 4 have not been calculated for comparable segments rather than the six used in the SERAS study.
household income levels. Even so, it is clear from Table 4 that
the variation in implied value of time across the different Although the distinction between business and personal trip
models is much greater than can be explained by either the purposes is widely followed, air travel forecasting practice
different year for which the cost data were obtained or differ- often also makes a distinction between what is usually termed
ences in the household income used to calculate the implied vacation travel and visiting friends and relatives travel. The
value of time. Given the differences in model structure and factors that influence the choice of where to stay (a hotel ver-
explanatory variables included in the model, this is hardly sur- sus a local residence) as well as the airport egress and access
prising. Including other terms than travel time and cost in the travel options are quite different for visitors in the two cate-
utility functions, or changing the way that travel time is de- gories. This distinction is less important for residents of the
fined in the model, will change the estimated values of the region, because their airport access trip is generally unaffected
travel time and cost coefficients and hence the implied values by what they are going to do at the other end of their trip.
of time. However, this has important implications for the abil-
ity of the models to predict how airport travelers will respond The distinction between business and personal trips is
to changes in travel time and cost of airport access modes. If largely a reflection that travelers on business are typically not
other terms in the model utility functions (such as dummy paying their own travel costs. However, it does not always fol-
variables or ASCs) are accounting for part of the perceived low that business travelers are less concerned about cost than
disutility of travel time or cost for a particular mode, the con- non-business travelers. The airport access mode choice model
tribution of those terms to the overall disutility of the mode developed for Boston Logan International Airport by the
will not change when the values of travel time or cost are CTPS (Harrington et al. 1996; Harrington 2003) distinguished
changed in a given application of the model, resulting in an in- between business travelers paying their own out-of-pocket
correct prediction of the impact of the change on perceived costs from those whose travel costs were paid by their em-
disutility and hence on predicted mode use. ployer. This was handled by including a separate variable in
the model utility function, but it could equally well (perhaps
better) be handled by defining a separate market segment.
MARKET SEGMENTATION
One important consideration in choosing the number of
As noted earlier, although some early models considered market segments to include in a model is the implications for
more limited market segmentation, it is now widely recog- the number of survey responses required to obtain reliable es-
nized that at a minimum separate airport access mode choice timates of the model coefficients. Somewhat oversimplified,
models should be developed for at least four market segments the more market segments that are defined the larger the sam-
corresponding to residents of the region being modeled ver- ple size needed to estimate the model. This is not always true,
sus visitors to the region, with each segment further split into because a poorly specified model may need a larger data
business and non-business (or personal) trips. Almost all of sample size to achieve a desired level of confidence for the

54
model coefficients; therefore, adding another market seg- Flyvberg et al. 2005). This approach involves placing the
ment may actually improve the model fit with a given data project in a statistical distribution of outcomes from a class
sample size if indeed that better reflects differences in the un- of similar projects. Although the concerns that have moti-
derlying behavior of the travelers. However, consideration vated the development of reference-class forecasting are
also has to be given to the number of travelers in each mar- broader than simply the performance of travel demand mod-
ket segment. A market segment with very few travelers will els themselves and have as much to do with how those
be difficult to model with any degree of confidence without models are used, including the input assumptions on which
a very large data sample, unless a stratified sample is some- forecasts are based, the approach is directly applicable to air-
how obtained that ensures an adequate number of responses port ground access mode choice modeling, particularly
in the data sample for each defined market segment. where the motivation for the development of these models is
to forecast ridership on proposed new airport access links or
services. The benchmark comparison analysis that was un-
MODEL PERFORMANCE dertaken as part of the modeling for the Toronto Air Rail
Link described in chapter three represents an example of this
The issue of model performance is one that has largely been approach.
ignored in the literature, apart from the simplistic approach
of reporting the level of significance (t-statistics) of the This suggests that monitoring the performance of the air-
model coefficients and the overall goodness-of-fit of the port ground access system after the introduction of new ser-
model. The latter is usually expressed by the likelihood ratio, vices or transportation links should be a part of the project
a statistical measure of overall fit of the model to the data that funding, so that adequate resources are available to do the
is quite meaningless to most users of the models. However, necessary ongoing data collection and analysis after the proj-
as suggested by Gosling (2006), what really matters is not ect is complete, and indeed that the requirement to undertake
how well the model fits the data from which it has been esti- this monitoring should be a condition of the project funding.
mated, because the nature of the model estimation process Only by collecting data on the use of all of the ground access
largely forces this to be fairly good (that after all is the whole modes on an ongoing basis after a new service or link is
point of model estimation). Rather, what matters is how well opened can the changes in mode use be compared with the
the model predicts the airport ground access mode choice be- predictions of the mode choice models during the project
havior when conditions change, such as the transportation planning stage and the performance of the models assessed.
services levels are improved (or degraded) or a new service
is introduced. Even the stability of the model coefficients However, although improved models are clearly desirable,
over time is an important concern if the model is to be used as is a better understanding of the likely reliability of existing
for forecasting (as they almost always are). models; even an inaccurate model may be better than no model
at all. The range of transportation alternatives that comprise
Unfortunately, there is very little discussion of these as- the airport ground access system at most airports and the myr-
pects in either the general literature on airport access mode iad traveler decisions that result in the observed pattern of
choice modeling or the technical documentation of existing access mode use are sufficiently complex that it is highly
models. Some academic studies have selected a holdout sub- unrealistic to attempt to determine the likely effect of any sig-
set of the data that they can then use to test how well the es- nificant change in the system by intuition alone. A well-
timated model explains this subset. However, although this is constructed model is better than a guess. Even so, it is impor-
better than no test at all, it still does not address many of the tant to give appropriate consideration to likely sources of
concerns identified previously, such as the effect of changing forecast error in any model application and undertake sensitive
the transportation system or the stability of the model coeffi- analysis to explore the possible effect of potential error in both
cients over time. the modeling and forecast assumptions on the model predic-
tions. Potential sources of error include omission or incorrect
This issue is not unique to airport access mode choice representation of important explanatory variables, incorrect
models and there is very little literature on how well most representation of the mode choice process implicit in the
travel demand models perform over time or when conditions model structure, and optimism bias in selecting future values
change. There is a growing interest in this issue by the FTA of explanatory variables. One useful approach to assessing the
in the context of the modeling used to support the Federal likely reliability of a model is to undertake a back-casting
New Starts capital grant program (U.S. Federal Transit analysis that applies the model to explain observed behavior in
Administration 2006) and the term “reference-class forecast- the period before that for which the model was developed. This
ing” has started to come into use to describe a forecasting is particularly useful if significant changes occurred in the air-
method that is based on the development of a so-called “out- port ground transportation system during that period, but may
side view” of a particular project on the basis of information also shed light on the ability of the model to reflect the effect
from a class of similar projects (Flyvberg 2005, 2007; of changes in real costs and income levels over time.

55
CHAPTER SIX
AIRPORT EMPLOYEE MODE CHOICE
As discussed in the literature review in chapter two, the topic of the values for major airline hub airports in the range from 0.3
airport employee ground access and egress mode choice has re- to 0.6; whereas airports with predominantly origin and desti-
ceived much less attention in the literature than that of air pas- nation traffic generally fall in the range from 0.2 to 0.3. The
sengers, and only one study was identified that described an air- high value for Oakland International is most likely the result
port access mode choice model developed to account for airport of the presence of an on-site major sorting facility for Federal
employee access mode choice behavior. In the case of airport Express, which had relatively low passenger traffic com-
employees, airport access trips are more commonly referred to pared with the other airports in the sample. The very low
as journey-to-work trips. However, these trips will be referred to value for San Diego International could be the result of part
as airport access trips in the following discussion for consistency of the very constrained site, which limits the opportunities
with the treatment of airport access travel by air passengers. for ancillary activities at the airport. The higher values for the
major airline hub airports reflect the additional airline activ-
As with air passengers, discussion of airport employee ac- ities that occur at those airports, including maintenance ac-
cess mode choice behavior should really address both access tivities and crew bases.
and egress travel. However, unlike air passengers, airport
employee egress mode use is likely to be the same as their ac- The more interesting statistics from the perspective of
cess mode use. If they drive to work by private vehicle, then airport ground access are the ratios of average daily em-
that constrains their choice mode for the return trip. Simi- ployees to average daily originating passengers. Apart from
larly, if they use public transportation for the journey to San Diego International, these range from about 0.25 in the
work, they do not generally have a private vehicle available case of predominantly origin and destination airports to a
for the return trip. Although situations may arise on an indi- high of 1.55 for Dallas/Fort Worth International. Most of
vidual basis from time to time in which airport employees the major airline hub airports have values between 0.7 and
travel to work one way and return home another (e.g., they 1.0, with Lambert–St. Louis International (which at the
may come to work by public transportation but get a ride time was the main hub for TWA) having a ratio of 1.27.
home from a co-worker), it can be expected that the effects Although person-trips do not translate directly to vehicle
of these irregular commute patterns will generally cancel trips owing to differing use of high-occupancy vehicles
each other out in the overall mode split. (HOV) and the additional vehicle trips resulting from air
passenger drop-off and pick-up trips by private vehicle, this
Airport employee trips can account for a large share of the suggests that daily employee vehicle trips at major airline
ground access trips generated by an airport, particularly for hub airports are of the same order of magnitude as vehicle
large airports that have a wide range of support activities, in- trips generated by air passengers, and could be significantly
cluding air cargo handling, aircraft maintenance, and airline higher.
crew bases. An earlier TCRP study (Leigh Fisher Associates
2000) assembled data for a sample of airports on air pas- The statistics given in Table 5 also provide a perspective
senger traffic and the estimated average number of daily on the relative importance of airport employee trips in the
employees working at the airport. These data are shown in overall pattern of regional travel. A flow of 10,000 to
Table 5, together with the ratios of average daily employees 40,000 trips per day each way clearly has a significant im-
to average daily enplaned passengers and average daily orig- pact on the regional transportation system in the vicinity of
inating passengers. The first ratio indicates the relationship the airport, even though it may be only about 1% of the total
between the number of airport employees and the overall regional journey-to-work travel (the number of average
level of airport activity, whereas the second ratio indicates daily employee trips shown in Table 5 averages approxi-
the relationship between the number of airport employee mately 1.1% of the corresponding metropolitan area aver-
airport access person-trips per day and the number of air pas- age weekday journey-to-work trips, although the ratios
senger ground access person-trips per day, assuming that vary from 0.2% to 2.1% across the different metropolitan
each employee makes one access trip per day. areas).
The data in Table 5 show that the ratio of daily employees Airport employee travel to and from work differs in a
to enplaned passengers varies widely, from 0.13 to 0.83, with number of ways from typical journey-to-work travel. Major

56
TABLE 5
AIRPORT EMPLOYEE ACCESS TRIP GENERATION
Enplaning Originating Average Avg. Daily Employees

Passengers Passengers Daily to Avg. Pax per Day
1998 1998 Employees
Airport Enplaned Originating
Los Angeles International 30,826,859 18,313,990 40,000 0.47 0.80

Chicago O’Hare International 35,841,551 16,127,120 40,000 0.41 0.91
San Francisco International 19,658,626 12,531,590 31,000 0.58 0.90
Dallas/Fort Worth International 30,121,523 11,279,510 48,000 0.58 1.55
Las Vegas McCarran International 15,132,220 11,114,050 7,500 0.18 0.25
Boston Logan International 13,250,754 10,364,870 14,500 0.40 0.51
Phoenix Sky Harbor International 15,984,620 10,323,330 23,655 0.54 0.84
Seattle–Tacoma International 12,867,830 9,592,000 11,375 0.32 0.43
Denver International 18,444,540 8,956,900 17,400 0.34 0.71
San Diego International 7,453,186 6,980,960 2,600 0.13 0.14
Houston Bush Intercontinental 15,492,252 6,356,870 14,406 0.34 0.83
Tampa International 6,955,805 6,289,530 8,219 0.43 0.48
Lambert–St. Louis International 14,334,844 5,442,800 19,000 0.48 1.27
Portland International 6,487,226 5,353,170 5,000 0.28 0.34
San Jose International 5,202,502 4,861,650 3,500 0.25 0.26
Salt Lake City International 10,097,036 4,762,760 13,026 0.47 1.00
Metropolitan Oakland International 4,612,614 4,183,300 10,500 0.83 0.92
Sacramento International 3,593,647 3,521,200 2,300 0.23 0.24
Source: Leigh Fisher Associates (2000, Table 1-1). Average daily employees estimated for 1998 based on data
reported by airport operators. Passenger ratios calculated by author.
Notes: Avg. = average; Pax = passengers.
airports operate 24 hours a day, 7 days a week, and as a result models, to give an indication of the typical structure of
many employees work shifts that are significantly different such models.
from the conventional workday. In addition, airline flight and
cabin crews often have multi-day duty cycles during which MODEL APPLICATION
they are away from their crew base. It is unclear how these
characteristics cause employee access mode use patterns to The application of airport employee mode choice models,
be different from more general journey-to-work trips repre- whether special-purpose models, adapted special-generator
sented in regional travel demand models. Although there sub-models, or general-purpose journey-to-work models,
may be other trip generators in a region that have similar falls into two broad categories. The first is to predict airport
work shift patterns to those at the airport, these trips are prob- employee travel patterns for airport planning studies,
ably not well handled by standard regional travel demand whether specifically oriented to employee travel or more
models either. general airport ground transportation issues. Airport em-
ployee travel accounts for a significant proportion of the
The survey of MPOs described in chapter three exam- vehicle trips generated by airports, particularly large air-
ined how airport employee access travel is handled in reports, as well as of the ridership on transit services to and
gional travel demand models. Although no MPOs reported from airports. Therefore, modeling these trips forms an im-
the use of a special-purpose airport employee mode choice portant part of feasibility studies for improved airport
model, six reported the use of a special-generator sub- ground transportation services or infrastructure, as well as
model for airport employee trips, with the majority treating studies directed at reducing the environmental impacts of
airport employees the same way as other regional journey- airport activities. In particular, environmental impact docu-
to-work trips. This chapter therefore examines the use of mentation for airport projects needs to address airport em-
special-generator sub-models for airport travel, as well as ployee travel as part of considering transportation impacts.
a selection of general urban journey-to-work mode choice Although this can be (and often is) addressed by simply

57
using existing mode split data from employee surveys or SERAS Model
similar sources, to the extent that proposed projects will
change the factors affecting employee choices or that pro- A special-purpose airport employee mode choice model was
posed mitigation measures have the goal of modifying developed for the Greater London region as part of the U.K.
employee access mode use, it becomes necessary to have SERAS study (Halcrow Group 2002b). This model was a
tools that can predict the resulting changes in airport em- fairly simple binary (two-mode) logit model that predicted the
ployee access mode behavior. percentage use of private vehicle and public transport. The
model used an incremental logit formulation, sometimes
The second category of model application is the represen- described as a pivot-point analysis. Rather than directly pre-
tation of airport employee travel in general urban or regional dicting the percentage of employees using private vehicles, it
travel modeling. Where airport employee travel is simply predicted the change in the percentage use in terms of the
treated the same as any other journey-to-work trip then the change in the explanatory variables. This change was then
only question becomes the number of employee trip ends that applied to the observed mode split in the base case. Mathe-
are assigned to the airport TAZs in the trip generation matically there is no difference between this approach and the
process. However, where special-purpose airport employee use of a regular logit choice model as long as the utility func-
mode choice models or adapted special-generator sub-models tions in the regular logit model have ASCs to ensure that the
for airport employee trips are used, then in addition to the mode splits in the base case correspond to the observed data.
models themselves consideration needs to be given to how
the resulting airport employee trip patterns are integrated into The SERAS employee mode choice model was based di-
the overall trip assignment process. rectly on one developed for the South and West London
Transport Conference for a study covering the area to the
TECHNICAL APPROACH
south and west of London Heathrow Airport. It was there-
fore felt that this provided a good basis for the SERAS work.
In general, the airport ground access modes available to air- However, the Conference study covered all journey-to-work
port employees are the same as those available to air passen- travel, not just airport employees, and thus the model does
gers, although some modes, such as rental car and drop off by not include factors specific to airport employee travel (such
private vehicle, are not likely to be used by airport employ- as irregular shift times or multi-day duty cycles). The param-
ees on a regular basis, whereas other modes that may be eters used in the SERAS model were derived from an earlier
available to airport employees, such as car pool and van pool, revealed preference study of multi-modal travel, the details
are not applicable to air passenger travel. This suggests that of which were not reported. It is unclear whether this earlier
a similar modeling approach would be appropriate. Given the study comprised (or even included) airport employee travel.
number of alternative modes that would need to be included
in the model, as will be discussed later in this chapter, it Because the SERAS model included only two modes, pri-
would seem likely that a NL model would be the most vate vehicle and public transport, no account was taken of
appropriate structural form (at least at the current state of factors influencing shared-ride behavior. However, the costs
practice of airport access mode choice modeling). This also and travel times involved in using public transport were ob-
corresponds to the current state of best practice for general tained from a public transport network model that took into
regional travel mode choice models. Because different fac- account interchanges between different services (such as
tors influence air passenger and airport employee mode mainline rail, London Underground, and local bus). There-
choice decisions, the explanatory variables included in the fore, the use of different public transport modes is included
modal utility functions and the market segmentation (if any) in the model by implication. The model used a generalized
used in airport employee mode choice models will be differ- cost approach with separate travel time coefficients for
ent from that used in models of air passenger mode choice. access and egress time, waiting time, and in-vehicle time. In
Pivot-point analysis, as discussed in chapter two, may be an addition, public transport travel times included a constant for
appropriate approach for modeling employee travel where travel on each mode, termed a boarding penalty, which
data on existing employee mode use are available. reflected relative preferences for different modes. Each bus
journey incurred a boarding penalty of 7.5 min, whereas each
Airport Employee Access Mode Choice Models journey by rail, light rail transit, or guided bus incurred a
boarding penalty of 2.5 min. The boarding penalties were
The literature review only identified three studies in which added together if a trip involved a change of mode. However,
mode choice models were specifically developed to predict there were no penalties for an interchange within a mode,
airport employee access mode use. All three studies adapted apart from any waiting time involved. Costs were combined
other journey-to-work mode choice models to predict airport into a single variable. Costs and times were converted to a
employee mode use rather than developing an entirely new generalized cost by multiplying travel times by an assumed
model from airport employee travel data. average value of time. The same value was used for all

58
employees, because the model was applied at the level of an later for a study of the potential ridership on a planned APM
analysis zone rather than individual employee. that would connect San José International Airport with a
nearby station of the Santa Clara Valley Transportation
Authority light rail system (Dowling Associates 2002). As
Oakland Airport Connector Model with the Oakland Airport Connector model, coefficients for
travel time and cost were adopted from those for home-based
As part of the analysis for the preparation of an Environmen-
work trips in a regional travel demand model, in this case the
tal Impact Report/Environmental Impact Statement for a
planned APM connection between the Oakland International Santa Clara countywide transportation model, and the ASC
Airport and the Coliseum station of the BART system, for transit was estimated to fit the model predictions of mode
located approximately 2.5 miles from the airport, an airport use to airport employee survey data. Although details of the
access mode choice model was developed and applied to Santa Clara countywide model were not reviewed, it appears
generate ridership projections for the APM connection, from the values of the coefficients that the inconsistency in
termed the Oakland Airport Connector (BART–Oakland . . . the years in which the cost data was expressed in the two
2002, Appendix B: Transit Ridership Procedures and models may be even greater than for the Oakland Airport
Inputs). The mode choice model comprised five market seg- Connector model.
ments covering four air passenger market segments and an
airport employee segment. The application of the model allowed for up to four differ-
ent connecting transit routes between any given analysis zone
The airport employee segment of the mode choice model and the airport. For those routes involving the use of the APM,
considered only two modes, private vehicle and transit, and the ASC for transit was adjusted on the basis of the results of a
used the same coefficient value for both highway and transit stated preference survey performed as part of the study to
travel time, including rail and bus. In addition to travel time, reflect the greater reported likelihood of using transit if the
the utility functions included walking distance, waiting time, APM link was available. However, the adjustment was not
and cost. The coefficient values were not estimated from em- particularly large and the difference would be offset by a dif-
ployee survey data, but were adapted from the regional travel ference in fare of only 20 cents. A more detailed discussion of
demand model developed by the MTC, using the mode this model is provided in Appendix D (web version only).
choice model coefficients for home-based work trips. The
ASC for transit does not correspond to the MTC model coef-
ficients and appears to have been estimated to make the tran- Special-Generator Models
sit mode share match observed data for airport employees.
Special-generator models form part of the regional travel
However, the MTC model includes seven modes and sev- demand model and make use of customized techniques to
eral other explanatory variables, including household income predict travel to zones such as a sports stadium or airport
and vehicle ownership. It could be expected that excluding that is not likely to be properly represented by the regular
these variables from the model would change the appropriate modeling process. Six of the MPOs responding to the survey
values of the other coefficients, because the behavioral described in chapter three indicated that they used a special-
explanation that they provide in the MTC model would now generator model for airport employee trips, although only
have to be accounted for by the remaining variables. The one of these, the Houston–Galveston Area Council, provides
likely impact of this on the performance of the model is any relevant information on its website. The Council com-
unclear. Furthermore, the MTC mode choice model was missioned a travel survey at Houston Intercontinental Airport
estimated using constant 1990 dollars for travel costs. Apply- in November 1995 to support the development of a special-
ing these model coefficients to costs in 1999 dollars, as generator model. This survey assembled data on air passen-
appears to have been done in the analysis, without making any ger ground access travel, airport employee travel, and com-
adjustments for inflation would overstate the effect of cost in mercial vehicle trips to and from the airport. In addition to
mode choice decisions. surveys of air passengers, airport employees, and commercial
vehicle drivers, traffic counts were undertaken on airport ac-
A more detailed discussion of this model is provided in cess roadways to develop expansion factors to convert the
Appendix D (web version only) and additional information survey results to daily trips.
on the MTC home-based work mode choice model is given
later in this chapter. The airport employee survey obtained data on household
and travel characteristics for 193 employees, including the
San José International Airport Model number of people and workers in the household, the number
and age of vehicles available, and the household income. The
A very similar approach to that used to model airport em- survey respondents reported an average of 2.4 trips to and
ployee travel in the Oakland Airport Connector study was from the airport on the travel survey day. The majority were
followed in a similar model that was developed about a year by private vehicle, with 88% auto driver, 10% auto passen-

59
ger, and 1% each transit and other. These trips were ex- employee residences that can be compared with the distribu-
pressed in terms of the trip types used in the regional travel tion of employee trip origins predicted by the regional model
demand model and factored up on the basis of total airport trip distribution process. If the differences are significant,
employment to give average weekday trips by trip type. and this is important for a particular application such as pre-
dicting the likely demand for a new airport access service,
then further adjustments to the employee travel pattern pre-
General Regional Travel Models dicted by the regional model may be necessary.
The most common way to model employee travel to and from A survey of MPOs recently completed by Vanasse Hangen
airports is to treat the airport in exactly the same way as any Brustlin for TRB explored the state of practice in metropoli-
other TAZ in a regional travel demand model and use the trip tan area travel forecasting (Vanasse Hangen Brustlin 2007).
generation, trip distribution, and mode choice sub-models to The survey found that the great majority of the 228 MPOs re-
generate the number of person and vehicle trips associated sponding use a trip-based four-step or similar travel forecast-
with airport employee travel. Typically, an airport will be rep- ing process. However, although approximately 95% of the
resented as single TAZ within the regional travel analysis large MPOs reported that they used a mode choice model for
zone system. Most regional travel demand models include HBW trips, overall slightly less than 50% of the MPOs mod-
separate sub-models for different types of trips, such as travel eled HBW mode choice. About three-quarters of the large
to and from work or for shopping or recreation. Most models MPOs use a NL models for HBW trips. The study presented
consider journey-to-work, commonly termed home-based data on the frequency of use of different modes in the mode
work (HBW), trips and assume that the reverse trip is sym- choice models, but did not indicate which of these were used
metrical (although at a different time of day). for HBW trips or provide any information on the explanatory
variables used or how the modes were nested.
Because the focus of this report is on airport access mode
choice, this section provides a brief review of current prac-
tice with HBW mode choice models in regional travel Home-Based Work Mode Choice
demand modeling. However, it should be noted in passing
that the total number of employee trips and the resulting To illustrate the variety of model structures and explanatory
travel patterns on the regional transportation network depend variables included in typical HBW mode choice models, this
on the trip generation and trip distribution components of the section presents some technical details on five representative
regional models as much as on the mode choice component. models for the following regions: Atlanta, Dallas/Fort
The extent to which the trip generation relationships incor- Worth, the San Francisco Bay Area, Seattle, and Metropoli-
porated in the regional model are representative of the tan Washington (the Baltimore–Washington region). Each of
employment at an airport is an issue that should be consid- these models differed in a number of ways, including the
ered in interpreting the predictions of airport employee travel modes represented, the structural form of the model, and
produced by a regional model. There are two aspects that the explanatory variables. Because of the complexity of the
may need to be addressed. The first is how well the base year models, their need to reflect the different transportation
employment levels implicit in the model correspond to ac- facilities and geographical aspects of each region, and differ-
tual employment at the airport, whereas the second is how ences in the underlying transportation system data used to
well the model predicts the expected growth in employment apply the models, this discussion does not cover all the de-
at the airport in the future. The first aspect is primarily a ques- tails of the models, for which the interested reader is referred
tion of how well the regional model is calibrated to the air- to the technical documentation referenced in the discussion,
port employment data at the TAZ level. The second aspect but rather attempts to provide a comparative overview to give
relates to how well the way that the regional model predicts a broad sense of how these models function and differ.
future growth in employment corresponds to the expected
growth in airport activity and employment. Where there are The principal characteristics of the five models are sum-
exogenous forecasts of future airport employment levels, marized in Table 6, which shows the model structure, the
such as forecasts produced by the airport authority, it may modes included in the model, and the explanatory variables
be possible to improve the predictions of future airport em- used in each of the models. Four of the models were true NL
ployee travel produced by the regional travel model by fac- models with two or more levels, whereas the fifth consisted
toring the airport employee trips predicted by the regional of two linked MNL models that were estimated sequentially.
model for future years so that the number of employee trip The terminology varied in the documentation for the differ-
ends in the airport TAZ corresponds to the exogenous fore- ent models (e.g., some models refer to shared-ride auto trips,
cast of airport employment. whereas others use the term group ride), but consistent
terminology has been used in the following discussion for
Although trip distribution issues will generally be of clarity. Therefore, the terminology in the discussion of a spe-
lesser concern than trip generation and mode choice, the air- cific model may differ slightly from that used in the model
port authority will typically have data on the distribution of documentation.

60
TABLE 6
CHARACTERISTICS OF SELECTED HOME-BASED WORK MODE CHOICE MODELS
Regional Model
ARC MTC MWCOG NCTCOG PSRC
Model Structure
Sequential multinomial logit
Nested logit
Modes Included
Drive alone
Shared ride 2
Shared ride 3
Shared ride 3+
Shared ride 4+
Transit auto access
Transit walk access
Bus transit walk access
Rail transit walk access
Bicycle
Walk
Nests
Motorized (auto, transit)
Drive alone/shared ride
Drive alone/drive to transit
Shared ride a
Transit access
Bus vs. rail transit
Variables
In-vehicle travel time
Waiting time b
Walking time b
Bicycle time
Trip cost b
Persons/household
Workers/household
Vehicles/household
Vehicles/worker
Households w/ vehicles < people
Bus miles in peak hour
Household income
Employment density
Land use mix index
Sub-area dummy variables
Model: ARC = Atlanta Regional Commission (Travel Demand . . . 2005).

MTC = Metropolitan Transportation Commission (Purvis 1997).
MWCOG = Metropolitan Washington Council of Governments (COG/TPB Travel . . . 2007).
NCTCOG = North Central Texas Council of Governments (Dallas–Fort Worth Regional Travel Model . . .
2007).
PSRC = Puget Sound Regional Council (Cambridge Systematics 2003).
a
Estimated as separate multinomial logit model.
b
Coefficients adopted from other models.

61
FIGURE 6 Home-based work mode choice structure—Atlanta region. (Source: Travel Demand . . . 2005).
All five models distinguish between single-occupant transit service (termed premium transit) trips, as shown in
auto trips (drive alone) and at least two categories of Figure 6. This structure implicitly assumes that the disutil-
shared-ride auto trips. Three models have three vehicle oc- ity involved in choosing to walk to transit depends on a
cupancy categories (drive alone, two people, and three or lower-level choice of whether to walk to a local bus stop or
more people), whereas two of the models add a fourth cat- walk directly to a rail station. Although some travelers in-
egory and distinguish between three people and four or deed face this choice, most do not and have a trip end far
more. The representation of transit in the models is more enough from the nearest rail station that their options are to
varied. All the models distinguish between walk access and drive to the station or walk to a local bus stop. Those who
automobile access to transit in some way; however, the de- choose to walk to a local bus stop may subsequently use the
tails differ. Some models represent the access mode as a rail rapid transit system depending on their route through
secondary choice, whereas others treat walk access to tran- the transit network.
sit and auto access to transit as separate modes. The Metro-
politan Washington model (FY-2003 Models . . . 2007) as- In contrast, the HBW mode choice model for the Seattle
sumes either walk access or automobile access to transit region (Cambridge Systematics 2003) has five modes and
based on the distance from the trip origin to the nearest bus one composite alternative at the top level: auto, two-person
stop or rail station. Transit trips beyond the assumed maxi- shared-ride, three or more person shared-ride, walk access to
mum walking distance from the nearest rail station but transit, walk to destination, and bicycle, as shown in Figure 7.
within walking distance of a bus stop are assumed to walk The auto composite alternative is split into drive alone and
to the bus stop. Use of rail or bus then depends on the quick- auto access to transit at the second level. This structure im-
est path through the transit network. Two of the models in- plicitly assumes that the disutility involved in choosing to
cluded bicycle and walk the entire trip as alternatives in the make an auto trip depends on a lower-level choice of
choice set for some trips. whether to drive alone to the destination or to an interme-
diate transit stop or station. In the case of automobile ac-
Not only did the models differ in which modes they in- cess to transit, the perceived disutility of the alternative
cluded but the way in which the modes were nested dif- will include the effect of the travel time and cost involved
fered across every model. The Atlanta HBW regional in riding transit.
model (Travel Demand . . . 2005) divides person-trips into
highway trips and transit trips at the top level and then at The HBW mode choice model for the San Francisco Bay
the second level splits highway trips into drive alone and Area (Purvis 1997) lies somewhere between the Atlanta and
shared-ride and transit trips into walk access and automo- Seattle models, as shown in Figure 8. At the top level there is
bile access. Shared-ride auto trips are split into three vehi- a choice between motorized modes (both automobile and
cle occupancy categories at the third level, whereas walk transit), bicycle, and walking to the destination. The motor-
access transit trips are split into local bus and rail rapid ized modes are then split into three auto modes (drive alone,

62
HBW
Person Trips
Shared Shared Transit-Walk

Auto Walk Bicycle
Ride 2 Ride 3+ Access
Drive Auto Access

Alone to Transit
FIGURE 7 Home-based work mode choice structure—Seattle region. (Source: Cambridge

Systematics 2003.)
two-person shared-ride, and three or more person shared- son shared-ride. The split into three carpool/vanpool modes
ride) and transit. At the third level, transit trips are split into is done with a separate carpool occupancy model that is esti-
automobile access and walk access. There is no explicit dis- mated independently from the top-level model. The resulting
tinction between the use of local bus or regional rail services vehicle occupancy information is then used in the main
(BART or commuter rail); however, the use of rail depends model to determine the average costs for the shared-ride
on the path taken through the transit network. mode for each person.
The HBW models for the Dallas/Fort Worth region (Cam-

bridge Systematics n.d.; Dallas–Fort Worth . . . 2007) and for Variables Used in Models
the Metropolitan Washington region (COG/TPB . . . 2007)
both have a somewhat simpler structure than the other three The range of explanatory variables included in the utility
models, with only three alternatives at the top level, one of functions of the HBW models varies from region to region,
which is split into three modes at the second level. However, as indicated by Table 6. Furthermore, not all variables used
the modes at each level are different. The NCTCOG model in a given model are used in the utility function for each
has an auto composite alternative at the top level, together mode because many of the variables, such as waiting time or
with walk access to transit and auto access to transit as sepa- bicycle travel time, are only relevant for some modes. Other
rate modes. The auto composite alternative is then split into variables that are not mode-specific, such as measures of ve-
three modes at the second level: drive alone, two-person hicle ownership, have been found to provide a statistically
shared-ride, and three or more person shared-ride. The Met- significant contribution to the explanatory power of the
ropolitan Washington Council of Governments (MWCOG) model when included in the utility function for some modes
model has the following three alternatives at the top level: but not others.
transit, drive alone, and shared-ride auto (termed group ride).
As described previously, access to transit is not explicitly All five models include measures of in-vehicle travel
modeled as a mode choice, but transit trips are assumed to in- time, waiting time, walking time (or walking distance), and
volve either walk access or automobile access depending on travel cost, although the coefficients for travel time and cost
the distance to the nearest bus stop or station. The shared-ride in the Atlanta model were adopted from other regional mod-
alternative is then split into three carpool/vanpool modes at els rather than being estimated from local data. Some models
the second level for different vehicle occupancy: two-person used the same travel time or travel cost coefficients for each
shared-ride, three-person shared ride, and four or more per- mode, whereas others estimated different coefficient values
Drive Shared Shared Transit Transit

Bicycle Walk
Alone Ride 2 Ride 3+ Auto Acc Walk Acc
FIGURE 8 Home-based work mode choice structure—San Francisco Bay

area. (Source: Purvis 1997.)

63
for different modes, typically distinguishing between transit parable. The work required to reexpress this information
and automobile use. Similarly, some models estimated sepa- on a consistent basis is beyond the scope of the current
rate model coefficients for different travel time components, study.
such as waiting or walking, whereas others assumed that
these coefficients were a fixed multiple of the in-vehicle
travel time coefficient and only estimated a single coefficient Summary
for the weighted travel time. If the assumed weighting factor
Although the current state of practice of regional travel
for the out-of-vehicle time components is incorrect, this will
demand modeling is fairly consistent in broad terms among
bias the estimated values of the coefficients for each of the
large metropolitan planning agencies, the details of how the
travel time components. components of this process are implemented vary widely
from region to region. Therefore, the use of these models to
Only three of the regional models included household in- study airport employee mode choice in a specific region
come in the model in some form, and the approach varied should be approached with caution. There is a need for sig-
across the models. The Bay Area model used a continuous nificantly more work to determine how the differences in the
variable for household income up to a maximum value of models could affect their ability to predict journey-to-work
$25,000 in 1989 dollars. The NCTCOG model used dummy travel behavior of specific sub-groups of the larger regional
variables for low- and high-income households, defined as population, such as airport employees.
those with incomes of less than $30,000 or more than
$75,000 in 1996 (when the household travel survey used to
estimate the model was performed). Separate values of the MODES INCLUDED IN MODEL
dummy variable coefficients were estimated for each mode.
The Atlanta model stratified the model estimation sample Generally, the modes included in airport employee mode
into four household income groups, with household incomes choice models are more limited than those for air passenger
in 2000 dollars of less than $20,000, $20,000 to $49,999, mode choice models, because many of the modes used by air
$50,000 to $99,999, or $100,000 or more. Separate ASCs passengers are not relevant to airport employee journey-to-
were estimated for each income group for each mode apart work trips. Airport employees will not usually consider rent-
from local bus and rail rapid transit. Estimating separate ing a car to get to work (except perhaps in the rare case where
ASCs is effectively the same as using a dummy variable for their own car is not available), whereas modes such as taxi
those income groups. and limousine are too expensive for use on a regular basis.
Although modes such as scheduled airport bus or shared-ride
Two of the models considered employment density in door-to-door van will often be too expensive to be a viable
the destination zone, whereas a third model used an index option, the cost of using these modes may be affordable if
of land-use mix in both the origin and destination zone to discounted fares are available to airport employees.
explain differences in mode use behavior in different areas
of the region. Four of the models used dummy variables for On the other hand, airport employees may have access ser-
certain sub-areas within the region, although the way that vices available to them that are not available to air passengers,
these were defined and how extensively they were applied such as dedicated employee buses or van pools. Unlike air
varied widely. The use of sub-area dummy variables pro- passengers, who generally cannot arrange to share a ride to
vides a way to adjust the model predictions to better match the airport with another air party other than by the use of com-
observed mode shares in specific areas. Although this im- mercial shared-ride services, airports or airport employers
proves the fit of the model to the estimation data, it also im- often establish ride-matching services to arrange car or van
plies that the other variables in the model are not adequately pools, or airport employees may be able to make use of local
explaining the observed choice behavior for those areas. ride-sharing organizations.
This could in turn lead to forecasting errors when the model
is used to predict travel patterns in future years when the Given the interest of many regional transportation plan-
conditions or influences reflected by the other variables are ning agencies in modeling shared-ride trips, as well as that in
expected to change. many regions the number of vehicle occupants affects the
ability to use HOV lanes or avoid paying tolls, it is desirable
Because of the differences in the way that the variables that airport employee mode choice models identify the num-
are defined in each model and of the differences in func- ber of occupants in car or van pools. This is necessary any-
tional form, the specific values of the coefficients are not way to be able to convert from employee trips to vehicle
directly comparable. Similarly, differences in the years trips. However, where the number of occupants that are re-
when the data were collected from which the models were quired to use HOV lanes varies for different facilities in the
estimated, as well as differences in the way that household region, then determining an average vehicle occupancy for
income was (or was not) included in the models, means shared-ride trips is not sufficient and it will be necessary to
that implied values of travel time are also not directly com- distinguish between the occupancy categories that allow the

64
use of each facility (typically shared-ride occupancy criteria denced by the widely different model structures used in dif-
are either two or more people or three or more). Although ferent regional travel demand models, which is discussed
there may be no reason to distinguish between formal van later in this chapter.
pools and car pools with three or more people from the per-
spective of HOV lane use, it may be desirable for other Because of the timing of shift start and end times, air-
reasons, such as evaluating the effectiveness of ride-sharing port employee travel may well occur at times when some
programs or because van pool vehicles are provided by the modes, particularly public transportation services, are not
airport or employer. available. This will need to be reflected in the mode choice
model through constraints on mode availability. Shift
Although airports are typically located sufficiently far times will also constrain the ability of employees to find a
from residential areas that walking is not usually a feasible car or van pool at the appropriate time, and this needs to be
access mode for the entire journey to work (as distinct from reflected in how these modes are represented in the model.
the access or egress trip from other modes), some employees
may cycle to work, particularly if the airport provides facili-
ties to encourage this. Indeed, the provision of such facilities EXPLANATORY VARIABLES
may form part of airport ground access trip reduction or mit-
igation programs. Therefore, a fairly comprehensive airport As with air passenger airport access mode choice models,
employee mode choice model might include the following airport employee mode choice models will need to include
modes: travel time and cost variables. Although costs can be aggre-
gated into a single variable, the different components of
• Single-occupant private vehicle travel time (walking, waiting, and in-vehicle time) should be
• Shared-ride private vehicle—two occupants expressed as separate variables to allow for the possibility of
• Shared-ride private vehicle—three or more occupants different perceived disutility of each component.
• Van pool
• Charter bus or van An important consideration in determining the values of
• Public transit explanatory variables is how to represent travel constraints
• Scheduled airport bus service arising from the timing of shifts. Where highway travel times
• Bicycle. or transit travel times, service frequency, or fares vary sig-
nificantly by time of day, it will be necessary to ensure that
Depending on the application, it may also be desirable (or the appropriate values are used for each employee. It may be
even necessary to estimate the model coefficients ade- desirable to have separate travel time variables for use of
quately) to include sub-modes such as different parking lo- public transportation late at night to reflect differences in per-
cations, different transit services, and different access or ceived disutility owing to concerns for personal security, in
egress modes to or from the stations or stops used for public addition to greater travel time as a result of infrequent ser-
transit or other fixed-route services, as with air passenger vice. However, the majority of regional travel demand mod-
mode choice models. Although mode choice models devel- els provide limited resolution for differences in travel times
oped for general urban travel are increasingly incorporating by time of day. It is not uncommon to only model the morn-
a wider range of modes and many explicitly considering ing commute peak and average off-peak conditions, which
how transit users will access transit services, very few cur- will not distinguish between mid-day and late night condi-
rent models include the range of modes that are typically tions. In this situation, it will not be possible to obtain the
available to employees at a large airport. Even so, as a wider appropriate travel times directly from the regional travel
range of modes is included in the more sophisticated re- demand model data files and significant adjustments may be
gional travel demand models, including different types of needed to appropriately represent late night travel times on
shared-ride trip and nonmotorized modes such as cycling, public transportation.
these may have an adequate degree of resolution to model
airport employee trips. Where employees pay for parking (or a transit pass)
through a payroll deduction, it may be desirable to use a
As the number of different modes and sub-modes in- separate variable for this cost component to allow for the
cluded in the model increases, so the structure of the model possibility that employees may perceive this cost differ-
becomes an important issue. The MNL model is predicated ently from a regular out-of-pocket expense. In the case of
on the assumption that the utility of each mode is indepen- a transit pass, there is also the benefit that the pass can usu-
dent of that of other modes, which is clearly not the case with ally be used for other transit trips in addition to the journey
many of the modes discussed earlier. This difficulty can be to work.
overcome with the use of a NL model. However, this raises
the question of how best to group the modes into different Employee income level is likely to be a significant factor in
nests. This is an issue that deserves further research, as evi- mode choice and an appropriate variable should be included in

65
the model. Employee wage rate is likely to be a more relevant airport undertaking an employee survey for use in develop-
measure than household income, because the latter may be ing such a model should attempt to obtain as large a sample
unduly influenced by the income of other workers in the as possible, in the range of 1,000 to 3,000 responses.
household.
Although not strictly an airport employee mode choice
modeling issue, the application of an airport employee mode
MODELING CONSIDERATIONS choice model requires an airport employee trip generation
model that predicts the number of airport employees living in
Development of airport employee access mode choice mod- each TAZ and generates a representative sample of airport
els requires both survey data on existing employee mode use employee trips with their associated characteristics. Unless
patterns and service data on the costs and travel times in- the employee survey responses are a very large sample of
volved in using the alternative modes available. Much of the total employment, simply factoring up the employee charac-
modal service data, such as highway travel times and transit teristics from the survey by the ratio of the total airport em-
travel times and frequencies, are the same as that required for ployment to the survey responses will result in a trip genera-
developing air passenger airport access mode choice models. tion pattern that has a large number of employees with
However, additional data will have to be collected on identical characteristics from some TAZs and no employee
employee-specific costs, such as parking fees, transit passes, trips from other zones. This will tend to overstate the poten-
and employee discounts on scheduled airport bus services, as tial for ridesharing, as well as bias the predicted mode use
well as information on employer-provided services such as from any given zone.
free parking or employee charter buses.
This issue is not that different from regional travel de-
As with air passenger airport access mode choice, the mand models, which typically include a trip generation
range of factors that influence the mode choice decisions of model component. However, the journey-to-work trip gener-
a given employee are sufficiently individual-specific that any ation models in regional travel demand models cover all
reasonable model will have to be disaggregate in nature and types of employment and typically do not generate individ-
predict the probability of a given employee choosing a par- ual traveler characteristics, much less those appropriate to
ticular mode. This implies that such a model can only be airport employees (such as shift start and end times). There-
developed from fairly detailed survey data of a large enough fore, it may be desirable to generate a synthetic sample of
sample of employees. In addition to information on the trips using Monte Carlo simulation methods based on the
existing mode use, the survey needs to collect data on all the distribution of airport employee characteristics obtained
factors that might reasonably influence that choice, including from an employee travel survey.
the respondent’s shift patterns, type of work, residence loca-
tion, wage or salary level, and eligibility for any special It is common practice in regional travel demand modeling
benefits, such as free parking or transit passes, even if the to treat the airport as a single TAZ. This is reasonable within
respondent is not actually using that benefit. the context of the overall regional travel demand, but may not
be adequate for airport planning studies, particularly where
It will also be necessary to determine the respondent’s different areas of the airport may be served by different ac-
employer (or at least the type of employer), because this is cess roads. Therefore, it may be desirable to subdivide the
important in expanding the model to predict airport access airport into different zones and identify the level of airport
mode use patterns for all airport employees, not just the re- employment in each zone as well as use a more refined rep-
spondents to the survey. Because the survey response rate is resentation of the transportation network in the vicinity of the
likely to vary across employers or even departments within airport. Although this will not significantly change travel
an organization, reflecting different managerial support for times to trip origin TAZs at some distance from the airport, it
performing the survey or even willingness to participate, it may affect travel times for those employees who live closer
will be necessary to develop expansion factors for each to the airport and will be necessary to properly assign to trips
respondent based on the total employment within each orga- to the different access routes.
nization or unit. Fortunately, the requirement for most airport
employees to have security badges means that the airport
authority usually has good data on the number of employees MODEL PERFORMANCE
in each category.
Given the limited experience with special-purpose airport
Given the very limited experience developing employee employee mode choice models, there is no basis for assess-
access mode choice models, it is unclear how large a sample ing the likely performance of such models. Assessments of
size would be necessary to estimate a reasonably accurate the overall performance of general regional travel models
model. This is yet another aspect that would benefit from fur- shed little light on their performance at predicting airport
ther research. Until this issue is better understood, based on employee mode choice, because airport employees form
experience with air passenger access mode choice models, an such a small percentage of the total regional journey-to-work

66
travel. It would of course be possible to compare the regional As with air passenger airport access mode choice models,
travel model predictions of journey-to-work mode use for the assessment of the performance of existing models is critical
airport TAZs with data from airport employee surveys. How- to understanding how much reliance can be placed on their
ever, no such studies have been identified. This would appear predictions and identifying the need for further improvement
to be a promising area for further research, because it is fairly in these models. However, before the performance of special-
simple to do and would provide valuable information on the purpose airport employee mode choice models can be as-
extent to which existing regional travel models can be used sessed, it is first necessary to develop the models to assess.
to predict airport employee mode use. This is another area that could benefit from further research.

67
CHAPTER SEVEN
TRANSFERABILITY OF AIRPORT GROUND ACCESS

MODE CHOICE MODELS
Given the considerable cost and effort required to develop where adequate data are not available to estimate a model on
models of airport ground access mode choice, it is natural to local conditions, the results should be used with caution. They
ask whether separate models need to be developed for every also stress that the complete model should always be used and
airport or whether it would be possible to adapt or apply a not just selected coefficients, because variables may be corre-
model developed for one airport at another. This would lated with each other.
depend on how transferable such models are. In general,
experience applying models of transportation behavior in sit- For readers interested in more details on past work on
uations that are different from the one for which they were de- model transferability, there is a good summary of the litera-
veloped has been mixed. An early study of the transferability ture on the transferability of regional travel demand models
of work-trip mode choice models (Atherton and Ben-Akiva by Karasmaa (2003).
1976) suggested that these models might transfer fairly well.
However, further work that examined model transferability Past studies of travel demand model transferability have fo-
issues in more detail (McFadden et al. 1977) found that cused on models of general urban travel behavior, which
whereas the evidence for transferability of work-trip mode includes a wide range of trip purposes that are likely to be heav-
choice model coefficients between different market segments ily influenced by the local characteristics of the transportation
within a region was mixed, their transferability between system, and airport ground access travel behavior may be more
regions did not appear very encouraging. Subsequent work consistent. In principle, one would expect that air travelers
(Pas and Koppelman 1984; Badoe and Miller 1995) showed would behave similarly when faced with a similar choice situ-
that although mode choice models appeared reasonably trans- ation, controlling for differences in ground access service char-
ferable within the same metropolitan region, the model acteristics (e.g., fares or travel times) and differences in traveler
parameters were not temporally stable. McCoomb (1986) es- characteristics (e.g., trip purpose and duration, household com-
timated journey-to-work models for the ten largest Canadian position, and income). Airport access mode choice models at-
cities using the same mode choice model specification and tempt to account for the effect of these variables on the choices
consistent travel data that was collected for each city using the made by a given air party. Therefore, to the extent that a model
same survey for the same day. Although the model coeffi- accurately reflects the effect of these variables, it should explain
cients were found to differ across the cities, some models the behavior of air parties in other geographic regions. How-
were similar enough to lead the author to conclude that they ever, this is a significant caveat that may well not hold for rea-
could be transferred when the cities are reasonably similar in sons discussed here.
size, urban structure, and transportation system. The impor-
tance of consistent data as a requirement for model transfer- In particular, there may be regional differences in atti-
ability has also been noted by Galbraith and Hensher (1982). tudes toward the service characteristics of different modes, as
well as differences in the nature of the services offered. This
More recently, Rossi and Outwater (1999) undertook a is likely to be a significant consideration with the use of pub-
comparative analysis of mode choice model parameters for lic transit services. Air travelers living in large metropolitan
home-based work, home-based non-work, and non-home- areas with well-developed transit systems are more likely to
based segments of regional travel demand models for 11 U.S. consider using public transportation for a trip to the airport
metropolitan areas. The values of the parameters for a given than those from areas with less extensive transit services and
variable were found to vary quite widely, particularly for less use of transit, partly owing to familiarity and comfort
home-based non-work and non-home-based trips. The au- with using public transportation and partly the result of dif-
thors undertook a simple experiment of applying the param- ferences in automobile availability. Whereas differences in
eters for each of the regional models to a hypothetical origin- service characteristics of different modes should in principle
destination zone pair in some other region after obtaining be accounted for by the explanatory variables in the model,
ASCs for transit for the other region by fitting each model to in practice this is not necessarily the case. Misspecification of
an assumed transit share in a different origin-destination zone utility functions or inaccuracies in the values of service char-
pair. The results showed that the estimated transit share in the acteristics used for model estimation can result in biased
test market varied from 10.2% to 19.8%, or by a factor of al- values of the ASCs or coefficients of other variables that will
most two. The authors concluded that whereas transferring result in biased estimates of the utilities of each mode when
mode choice model parameters may sometimes be necessary these coefficients are applied in different situations.
68
ISSUES IN MODEL TRANSFERABILITY structure. There are sound theoretical reasons why the NL
structure should be superior to the MNL, particularly if the
Because the mix of air passenger or airport employee char- model attempts to include modes that are likely to be per-
acteristics and the details of the airport ground transporta- ceived as having a stronger substitutability with some alter-
tion services that are available differ widely from airport to native modes than others or to distinguish between different
airport, for an airport access mode choice model to be trans- secondary modes that are used to access a primary mode,
ferable it must correctly reflect the influence of the different such as parking at an off-airport terminal compared with
factors that determine airport traveler ground access mode being dropped off by private vehicle.
choice. If model coefficients are partly accounting for fac-
tors that are not directly associated with the variable in ques- Because the nesting structure of the limited number of
tion (or are inherent to the mode in the case of the ASCs), models that have used a NL structure is very dependent on
then differences in these factors when the model is applied the particular modes that exist in the region being modeled,
in other situations will result in biased estimates of the it is difficult to generalize about how the various models have
modal utilities and errors in the predicted probabilities of grouped the modes. The model for Boston Logan Interna-
using each mode. tional Airport developed by the Central Transportation Plan-
ning Staff (Harrington et al. 1996; Harrington 2003) adopted
Although this applies to the transferability of a model a hybrid structure, in which the resident models were nested
from one region to another, it also applies when a model is but the nonresident models were not. The resident models
used to analyze a significant change in the ground trans- included an automobile nest (drop off, short-term parking,
portation system serving the airport for which it was devel- and long-term parking on and off airport) and a door-to-door
oped, such as the addition of a new mode or service, as well nest (taxi and limousine). The public transport modes
as the application of a model to forecast future mode use, as (MBTA transit, scheduled bus/limousine, Logan Express,
discussed by Gosling (2006). Therefore, ensuring that mod- and Water Shuttle) were not nested, although that would
els correctly reflect the factors that influence mode choice have been redundant because there were no other modes at
decisions not only improves their transferability to other the top level of the tree. In contrast, the nesting structure of
regions, but also their use in planning studies at the airport or the SERAS Air Passenger Surface Access Model (Halcrow
airports for which they were developed. This is a lot easier Group 2002b) involves up to six levels of nest, including a
said than done. three-level nest of different rail options. The decision of
which modes are nested together appears to have been made
In particular, model transferability is likely to be influ- on the basis of what gave the model the best fit. For some
enced by the following factors: market segments the use of taxi and drop off by private vehi-
cle (termed kiss and fly) are grouped in the same second-
• Incorrect or incomplete model specification, level nest, with private vehicle parked for the trip duration
• Missing explanatory variables, (termed park and fly) at the top level, whereas for other mar-
• Incorrect market segmentation, and ket segments all three modes are at the top level or are
• Problems with the data used to estimate the model grouped differently at lower levels.
coefficients.
However, the discussion of household income brings up
an interesting issue with respect to model transferability,
VARIABILITY IN MODEL SPECIFICATION
namely how to reflect differences in the cost of living in
different regions. Although including household income in
Although there are a fairly limited number of different airport
an airport access mode choice model in some way is clearly
access mode choice models that have been documented in
necessary if the model is to correctly reflect traveler behav-
the review of recent literature and professional practice, it is
ior, a given income level may provide more disposable in-
relevant to ask how consistent these models are in terms of
come in one region than in another. The extent to which this
their functional specification and use of explanatory vari-
will influence traveler sensitivity to time and cost tradeoffs in
ables. Wide variation in model specification would suggest
airport access trips is unclear.
that either the underlying behavior that they are attempting to
explain varies widely from region to region or the models do
not correctly reflect that behavior in their technical specifica-
tion. Either way, this would suggest that they are not likely to Explanatory Variables
be reliably transferable to other situations.
Differences in the explanatory variables included in different
airport access mode choice models, although not necessarily
Functional Form a problem for model transferability, do raise questions about
the extent to which the model coefficients may be explaining
Although a number of recent models have continued to use a factors not directly measured by the variables in question.
MNL structure, the more advanced models have used a NL The absence of studies that have taken a model estimated in

69
one region and applied in another to compare the model pre- whereas those staying in a hotel or visiting a business typi-
dictions with the observed mode choices in the second region cally do not. Similarly, travelers on business who can charge
makes it difficult to assess how sensitive the model predic- their travel expenses to an employer or client may behave
tions might be in such a situation to the choice of variables very differently from those who have to meet their travel
included in the model. In particular, the absence of a house- expenses out of their own pocket or out of a limited travel
hold income variable (or equivalent measure of the perceived budget. Thus, there may be significantly different behavior
value of time) in many models means that such models by someone traveling to a professional conference at their
would most likely not work well at all if applied in a situation own (or their organization’s) expense compared with some-
where the distribution of household income is very different. one attending a meeting the costs of which they can charge to
a client. Although these issues have typically been handled
An ideal model would not have ASCs in the utility function (if at all) through the specification of the utility functions, this
because these imply that the mode will have some probability imposes limitations on the ability of the model to reflect the
of being selected even if the values of all the continuous vari- relevant behavior and it may be better to address these issues
ables are zero. Rather, the intrinsic attributes of the mode through more detailed market segmentation.
would be expressed through appropriate continuous variables
that could be adjusted for different situations. However, in
practice the ASCs account for missing variables and other data COMPARATIVE ANALYSIS OF MODEL
PARAMETERS
problems and frequently take values that are quite large com-
pared with the effect of the continuous variables. This is likely Although the numerical value of the model coefficients in a
to be a significant problem for the transferability of these mod- nested or MNL model have no direct significance because
els, because the factors represented (at least in part) by the they depend in part on the variance in the dataset from which
values of the ASCs are not likely to occur in the same way in they have been estimated, the ratio of two coefficients does
another situation. have an interpretable meaning that can be compared across
different models. The ratio of the travel time coefficient to
the cost coefficient represents the implied value of travel
Market Segmentation time. Similarly, the ratio of coefficients of other variables to
the travel time coefficient can be interpreted as a factor that
This is the aspect in which there is the most consistency in re- converts the values of the other variable to equivalent min-
cent airport access mode choice models. It has become general utes of travel time.
practice to develop separate sub-models for four market seg-
ments (although terminology may vary slightly) comprising: Even if the implied value of travel and other continuous
variables is reasonably consistent between two models, the
• Resident business trips, values of the ASCs are likely to be problematic from the per-
• Resident non-business trips, spective of the transferability of a given model. If indeed
• Non-resident business trips, and these coefficients represent intrinsic characteristics of the
• Non-resident non-business trips. mode that will be similar in other situations, then using them
in another situation should produce reasonable results. How-
This market segmentation recognizes that residents of the ever, in practice they are likely to account for missing
area have different airport access mode options than visitors variables, model specification errors, inaccuracies in the
to the region (non-residents) and that travelers on business estimation dataset, and other issues. To the extent that
trips may value their time differently from those on personal these are likely to be different in different situations, the use
trips. There is also the consideration that many business trav- of the ASCs in the second situation would most likely gener-
elers are able to charge their travel expenses to their employer ate incorrect results.
or client. The need of some visitors to rent a car for their local
travel while in the area is another important difference be-
tween residents and visitors. Once a decision is made to rent DATA CONSIDERATIONS
a car for other reasons, then that typically predetermines the
airport access mode choice. It is a truism that a model is only as good as the data from
which it was developed and with which it is applied. This is
However, examination of recent airport access mode particularly relevant to airport access mode choice models
choice models suggests that this four-way market segmenta- owing to the extensive amount of data that are needed to de-
tion may be too simplistic to fully reflect the factors that velop and apply them and because much of these data are not
shape airport access mode choice behavior. Visitors staying readily available. Errors in the data used for model estimation
with friends or relatives often have access to a private vehi- will result in biased model coefficients, as those coefficients
cle for local travel (indeed, the local travel may be under- attempt to explain behavior in terms of data that are incorrect.
taken together with those they are visiting) as well as people Errors in the estimation data will not typically be detected by
who can pick them up from the airport and drop them off, the usual tests of model goodness-of-fit. Similarly, even if the

70
model is estimated on valid data and the model coefficients mation systems, in practice this is often problematic and
are not biased by data problems, applying the model with prone to error. Highway travel times vary by time of day and
incorrect data will produce incorrect results. day of the week. Although regional transportation planning
agencies are increasingly developing datasets that better re-
Potential data problems arise in both of the two broad cat- flect this variation, it is still common for planning agencies to
egories of data required for their development: survey data of divide the day into a limited number of time periods (e.g., the
air passenger or airport employee characteristics and data on a.m. peak period, p.m. peak period, and off-peak period) and
the transportation service levels in the airport ground access determine average weekday travel times for each period.
system. However, airport travelers will base their ground access de-
cisions on how long they think their trip will take at a partic-
Air passenger and airport employee surveys are rarely truly ular time of day, which may be quite different from the travel
random samples of the underlying population, although this is times given by the regional transportation network datasets.
often overlooked in their analysis. In particular, the logistics of Airports, particularly large airports, are typically served by a
performing air passenger surveys necessarily restricts the sam- large number of private transportation providers, such as
ple to a subset of the total population. Surveys are typically shared-ride van or limousine (black car) services, that may
performed for a limited time period and often for only part of charge a wide range of fares that vary by the location of the
each day on which the survey takes place, owing to staffing trip origin as well as from provider to provider. The travel
considerations. Air passenger surveys are often done in airline time involved in taking a shared-ride van is also influenced
gate lounges, because that is a location in which passengers are by how many other travel parties have to be picked up or
most likely to be willing to take the time to answer a survey. dropped off. Basing the mode choice model estimation on
However, this limits the sample to passengers on specific travel time and cost assumptions for each mode that is dif-
flights traveling to a specific destination. Although an effort is ferent from what the travelers perceived those values to be
typically made to survey a reasonably representative sample of when they made their decision will inevitably bias the model.
flights, budget limitations on the number of flights that can be
surveyed will usually result in a biased sample. This can be
partly corrected by appropriate weighting techniques; how- SUMMARY
ever, careful thought needs to be given to how this is done.
Weighting responses to correct for one issue such as the distri- Given the current lack of consensus over model specification
bution of responses by flight destination can exacerbate any and typical coefficient values between different airport ac-
bias in the responses with respect to a different issue, such as cess mode choice models, it can be assumed that the trans-
the distribution of responses by time of day. ferability of these models is highly suspect. Although it
seems plausible that the underlying airport traveler behavior
The second set of problems that can arise with air passen- may not differ that much from region to region, after taking
ger and airport employee surveys relates to question wording into account differences in air passenger or airport employee
or which questions are asked. Omitting a topic in the survey, characteristics and transportation system service levels, it
such as the household income of the respondents, necessar- appears unlikely that current airport access mode choice
ily precludes including that factor in the specification of a models do this correctly, based on the significant differences
mode choice model estimated from the data. Poorly worded between the different models.
questions can also introduce errors in the data. Air passen-
gers in particular may not use the same terms to describe their Both because of the obvious value of being able to apply
trip characteristics or ground access modes used as the de- airport access mode choice models in different situations
signers of the survey expected. For example, an air passenger from those for which they were originally developed, as well
may refer to a shared-ride van as a limousine (indeed some as the concerns about the reliability of even applying them to
shared-ride van operators use the word limousine in the name different situations at the airports for which they were devel-
of the service). Although these errors in individual responses oped, there is a pressing need to better understand how well
may not be too serious in presenting the aggregate results of current models reflect the factors influencing the underlying
the survey, particularly where there may be offsetting errors travel behavior and how they can be improved to better
by other respondents that reduce the overall error in the reflect this behavior. It is not sufficient to simply say that
results, they can have a significant impact on the use of sur- existing models are unlikely to be transferable and that a new
vey results to estimate disaggregate mode choice models that model must be developed for every application. This is to
attempt to explain why individual survey respondents chose admit that the existing models do not properly reflect the
the access mode that they did. causal structure of the underlying traveler behavior and by
implication calls into question the reliability of the predic-
Whereas in principle it should be fairly easy to determine tions generated by these models.
the travel times and costs of using different access modes
from existing data sources, such as the regional transporta- In the event that a study sponsor does not have the re-
tion planning process or airport ground transportation infor- sources or time to develop a new mode choice model for a

71
particular planning study where an existing model is not ables are properly reflected in the model predictions. If the
available and wishes to try using an existing model that was prediction errors are reasonably consistent for each mode
developed for another situation elsewhere, at a minimum it across all subsets of the data (e.g., by distance from the airport
would be prudent to conduct validation tests of a number of or across different income levels or air trip duration), then it is
candidate models by examining how well they predict recent more likely that the model is reflecting the underlying behav-
travel behavior in the new location. Although the predictions ior. On the other hand, even if the model correctly predicts the
from these models can usually be improved by adjusting the total number of trips using each mode but predictions for sub-
ASCs, it should be recognized that this may well be creating sets of the data are wildly off, then it is very unlikely that the
a false sense of confidence. What is more important is to model will produce reasonable predictions of the likely effect
ensure that the effects of differences in the continuous vari- of changes in the system.

72
CHAPTER EIGHT
INTEGRATION OF AIRPORT GROUND ACCESS MODELS IN

REGIONAL PLANNING PROCESS
Although modeling airport access mode choice by air pas- amount of tourist activity, and can be expected to attract
sengers and, to a lesser extent airport employees, has greater interest in regional travel demand modeling.
largely been restricted to specialized studies addressing air-
port landside and system planning issues, there is a growing Finally, the largest single air passenger access or egress
interest in explicitly modeling such trips in the regional mode at many airports, for both visitors and residents, is drop
transportation planning process. As indicated in the survey off and pick up by private vehicle. This option is also typi-
responses discussed in chapter three, a number of MPOs cally not explicitly modeled in regional travel models.
have begun to address air passenger trips using a special- Models predicting private vehicle use that are based on the
purpose mode choice model or special-generator sub-model assumption that the vehicle will be parked at the destination
and a somewhat smaller number of MPOs have begun to do until the return trip will thus underestimate the vehicle-miles
the same for airport employee trips. However, a significant of travel involved by a factor of two.
number continue to model trips to and from airports as reg-
ular regional travel using a standard trip classification, such These concerns may not be particularly important in terms
as home-based non-work trips. Because these standard trip of the total regional travel, because airport trips comprise a
classifications encompass a very wide range of activities, fairly small fraction of all regional trips, although it is unclear
most of which have very little in common with airport to what extent the errors tend to cancel each other out or are
travel, it would be surprising if the model components did all in one direction. However, these issues become of much
a very good job of predicting airport access mode choice. greater concern when the regional travel models are used to
This is likely to be compounded because airport access and predict trips on parts of the transportation network in the
egress trips, by both air passengers and airport employees, vicinity of the airport or are used for airport access and egress
typically have access to a much larger number of alternative studies, including predictions of airport access and egress
modes than are usually modeled in regional travel demand trips for use in environmental documents.
models, such as shared-ride door-to-door van services and
in some cases scheduled airport express bus services. Even Therefore, a fairly strong case can be made that airport ac-
taxi and limousine services, which account for a significant cess and egress trips need to be modeled separately from gen-
proportion of ground access trips at many airports, are eral regional travel patterns (or at least as a special-generator
typically not included as a separate mode in most regional sub-model within the overall modeling framework) and then
travel models. integrated with other trips in the traffic assignment process.
This chapter discusses some of the issues that arise in inte-
Furthermore, a significant fraction of all airport access grating specialized airport access mode choice models into
and egress trips (as much as half or even greater at some the general modeling framework and describes how these
tourist destinations) is made by visitors to the region. Most have been addressed by selected MPOs that have incorpo-
current regional travel models are only designed to model rated special-purpose airport access mode choice models or
travel by residents of the region and largely ignore travel special-generator sub-models into their regional travel mod-
by visitors. Therefore, modes such as rental car and hotel eling process.
courtesy shuttle are typically not included in the models.
Although visitor trips have been considered in some activity-
based microsimulation models (Bradley et al. 2001; Waddell TECHNICAL CONSIDERATIONS
et al. 2002) as well as in travel demand models for areas with
a significant number of visitor trips, such as the Visitor Intra- Most regional travel demand models are based on the well-
CBD Model developed for the Central Area Loop Study for established four-step approach of trip generation, trip distri-
Cincinnati (Parsons Brinckerhoff 2001), the focus tends to be bution, mode choice, and traffic assignment, although not all
on the travel of visitors within the region rather than their regional models include all four components and implemen-
travel to and from the airport. However, to the extent that in- tation details differ widely (Vanasse Hangen Brustlin 2006).
tercity travel is increasing faster than intraregional travel, However, the regional travel demand models for most large
visitor trips will become a larger component of regional metropolitan areas include all four components. Most models
travel patterns, particularly in those areas that attract a large distinguish between home-based trips and non-home-based

73
trips and between work trips and non-work trips. For exam- This suggests that airport access trips (at least air passen-
ple, the regional travel demand model for the San Francisco ger trips) need to be modeled outside the general travel mod-
Bay Area (Baycast-90 Users Guide . . . 2004) considers the eling process and the resulting vehicle trip tables between the
following seven types of trip for intraregional personal travel: access trip origin or the egress trip destination zones and the
airport zones added in to the trip tables for other types of trips
• Home-based work before the traffic assignment step. This will allow an appro-
• Home-based shop/other priate mode choice model to be used for air passenger trips.
• Home-based social/recreational Although air passenger survey data can provide the distribu-
• Home-based school (grade school) tion of air passenger trip ends across the region, the fairly
• Home-based school (high school) small sample sizes typically used in such surveys will pro-
• Home-based school (college school) duce a relatively “lumpy” distribution of air passenger trip
• Non-home-based. ends and air party characteristics, with some analysis zones
having no trip ends and others having air party trips with cer-
In addition, the model considers interregional personal tain combinations of characteristics but not others. There-
trips and commercial (truck) trips, although projections of fore, it may be desirable to generate a much larger synthetic
each of these two types of trip are generated by a separate sample using Monte Carlo simulation techniques based on
analysis from that used to model intraregional personal the distribution of air party characteristics in the survey
travel. results and measures of zonal socio-economic characteris-
tics. This would require the development of an air passenger
Home-based work trips are synonymous with journey-to- trip generation model that predicts the number of air passen-
work trips and cover travel in both directions to the extent ger trip ends in each zone based on the zonal socio-economic
that the modeling process considers travel outside the morn- characteristics (such as population, household income, com-
ing peak period. These trips would include the journey to mercial space, and hotel rooms). The Monte Carlo simulation
work by airport employees. Air passenger travel would gen- would then generate a synthetic sample of trips of any
erally be included in home-based shop/other for airport desired size, assigning each trip to an analysis zone based on
access trips by area residents from their home and in non- the distribution of trip ends from the trip generation model
home-based for airport access trips by area residents from and assigning it air party characteristics by drawing ran-
other locations (e.g., their workplace) or visitor trips. Airport domly from an appropriate subset of the sample of air parties
access trips by visitors to the area that start from the home of in the original survey.
a relative or friend in the area should strictly also be included
in home-based shop/other. However, the trip generation rela- The extent to which existing models of home-based work
tionships for home-based trips do not typically consider trips trip generation, distribution, and mode choice adequately
by visitors to the household. account for airport employee travel is unclear and a potential
subject for future research. However, it seems likely that any
Given the very diverse trip purposes that will be included inaccuracy in the application of these models to airport
in home-based shop/other or non-home-based trips, any employee travel will be much less significant than for air pas-
consideration of air passenger trips in the trip generation senger travel.
models is likely to be swamped by other types of trips. Per-
haps of even greater concern, the distribution of those trips
to destination zones in the trip distribution step of the mod- CASE STUDIES
eling process is likely to be driven by measures of zonal
attraction that take no account of the nature of airport trips. The following four case studies present a range of ap-
Thus, the number of airport trips generated by the normal proaches that has been followed by different MPOs that have
modeling process will probably bear little or no relationship explicitly modeled airport trips in their regional travel mod-
to the actual number of air party trips. Although the number eling process. They have been chosen to illustrate a number
of trips attracted to the airport zone can be adjusted to cor- of different ways in which airport trips can be integrated
respond to the actual level of airport traffic by using artifi- within the general regional travel modeling process. These
cial values of the trip attraction variables (e.g., adjusting the approaches vary from a special-purpose sub-model within
amount of retail space assigned to the zone), this is obvi- the travel modeling process of the ARC, through the external
ously not a very satisfactory solution because the trips generation of airport trip tables that are combined with the
attracted to the airport zone are drawn away from the other trip tables generated by the regular travel modeling process
zones and there is no guarantee that the resulting distribution of the MWCOG, to two examples of the use of external
of airport trip ends in the region will bear any relationship to airport access mode choice models. In the case of the
reality. Indeed, because shopping trips are generally much Boston CTPS, an air passenger mode choice model that has
shorter than airport access trips, this approach is likely to been described earlier in this report was developed in-house
bias the distribution of airport access trip ends in favor of in cooperation with the airport authority. In contrast, the
zones closer to the airport. Southern California Association of Governments (SCAG)

74
utilizes a proprietary air passenger mode choice model, the ger surveys at the three airports to obtain information for use
output of which is then used as input to the regular regional in its aviation system planning and regional travel demand
travel modeling process. modeling activities. The most recent survey was undertaken in
2005 (2005 Washington–Baltimore . . . 2006a).
Atlanta Regional Commission The MWCOG travel demand modeling process distin-
guishes between “modeled” trips and “non-modeled” trips
The ARC is the MPO for the Atlanta region. The ARC has de-
(FY-2003 Models . . . 2003). Modeled trips are those esti-
veloped an APM as a component of its regional travel demand
mated from the traditional four-step transportation planning
model (Model Documentation . . . 2005; Travel Demand . . .
process and are based on travel data obtained from house-
2006). The ARC/APM enables air passenger travel to and from
Atlanta Hartsfield–Jackson International Airport to be included hold travel surveys. Although these trips may include some
in the overall regional travel modeling process. The details of air passenger trips, to the extent that households may have
the model are summarized in chapter three and presented in recorded air travel trips in the household travel survey,
more detail in Appendix D (web version only) of this report. most air passenger trips will not be reflected in the modeled
trips. Non-modeled trips are those types of trips that will
The model is programmed in Fortran and can be called not be reflected in the household travel surveys, such as
by the software used by the ARC to implement its regional trips by visitors to the region or many air passenger trips.
travel demand model. This was originally programmed Therefore, MWCOG estimates air passenger trip tables
using Tranplan software, but has recently been converted to from the air passenger survey data outside the travel
TP+/Cube Voyager software. The program reads highway demand modeling process, converts air passenger trips to
and transit travel times and costs from standard TP+/Cube vehicle trips, divides these trip tables into the three time-of-
skim files and generates binary trip tables of transit passen- day periods used in the regional travel demand model
ger and vehicle trips in TP+/Cube Voyager format that can be (a.m. peak, p.m. peak, and off-peak), and then combines the
combined with trip tables generated by the regional travel de- resulting vehicle trip tables with those for modeled trips and
mand model for other types of trips before assigning passen- other types of non-modeled trips (such as taxi trips or trips
ger trips to the transit network and vehicle trips to the high- passing through the region) before the traffic assignment
way network in the final step of the regional travel demand step of the modeling process.
modeling process.
Because the MWCOG regional travel modeling process is
based on an average weekday, the results of the air passenger
Boston Central Transportation Planning Staff survey are first converted into average weekday enplane-
ments. The resulting pattern of access trips is then transposed
CTPS is the MPO for the Boston metropolitan area. With
to give the pattern of egress trips and the two trip tables are
support from Massport, CTPS developed an air passenger
added together to give the flow in both directions. These are
airport access mode choice model for use in a range of airport
then divided into the time-of-day trip tables by assuming that
ground transportation planning studies at Boston Logan Air-
10% of the air passenger trips occur in the a.m. peak period
port. The details of the model are summarized in chapter
and another 10% in the p.m. peak period. The projection of
three and presented in more detail in Appendix D (web ver-
these trip tables to future years uses a Fratar-based approach
sion only) of this report.
that initially applies growth factors at the airport end of the
trip based on forecasts of airport enplanements and growth
The model is run separately from the regional travel
factors at the other ends of the trip based on regional fore-
demand model to generate estimates of air passenger and
casts of households and employment and then iteratively bal-
vehicle trips by regional TAZ. These trip tables are then used
ances the resulting flows (FY-2006 Development . . . 2006).
to include the air passenger trips in the regional transit as-
signment model and the vehicle trips in the regional highway
MWCOG has established the goal of improving the repre-
assignment model as the last pre-assignment step in the
sentation of special traffic generators, including airports, and
CTPS Regional Travel Forecasting Model Set for Eastern
has begun the process of developing a more formal model of
Massachusetts (Harrington 2003).
airport access demand, including potentially an airport choice
component (FY-2003 Models . . . 2003; FY-2005 Develop-
Metropolitan Washington Council of Governments ment . . . 2005; FY-2006 Development . . . 2006). As part of
this effort, MWCOG staff undertook a review of airport ac-
MWCOG is the MPO for the Baltimore–Washington metro- cess and airport choice modeling in eight large metropolitan
politan area, which includes three large commercial service air- regions (FY-2003 Models . . . 2003). The findings of this re-
ports: Ronald Reagan Washington National Airport, Washing- view largely parallel the information presented in this report,
ton Dulles International Airport, and Baltimore–Washington although some additional detail is provided on the modeling
International Airport. MWCOG conducts periodic air passen- approach in some of the metropolitan areas. In particular, the

75
review addresses airport choice modeling, which is outside version has the capability of modeling the distribution of air
the scope of this report. cargo demand among airports in the region and can consider
the effect of airport capacity constraints on the allocation of
demand among airports. It can also estimate vehicle-miles of
Southern California Association of Governments highway travel resulting from airport access and egress trips
and the associated vehicle emissions. However, the pub-
SCAG is the MPO for the six-county Southern California lished documentation on RADAM is fairly limited and does
region that includes the Los Angeles basin and contains six not include many technical details on the model form or
primary commercial service airports with significant levels structure. In particular, there is very little information on the
of airline service and several secondary commercial service ground access mode choice component of the model.
airports that have some regional airline service or support air
cargo activity. Since the early 1990s, SCAG has made use of The analysis zones used in the model (called RADAM
a proprietary air travel demand forecasting model called the zones) consist of aggregations of SCAG TAZs, with approx-
Regional Airport Demand Allocation Model (RADAM) to imately 3,000 SCAG TAZs grouped into approximately 100
support its aviation planning activities (Southern California RADAM zones. This would make it difficult to directly inte-
Association of Governments 2002). Although the model is grate the results of the RADAM analysis into the regional
primarily designed to allocate regional air travel demand to travel demand model trip assignment process and that does
airports, it contains an airport access mode choice component not appear to be how the model is used. Rather, the model is
and can predict vehicle trips by different modes from each run as an entirely separate activity and the aggregate results
analysis zone to each airport in the region. The model has (e.g., vehicle-miles of travel and vehicle emissions) are
undergone a number of extensions and enhancements over added to the regional travel demand model results for other
the years and, in addition to air passenger travel, the current types of trips.

76
CHAPTER NINE
CONCLUSIONS AND FURTHER RESEARCH
Airport ground access and egress mode choice models play a segments. However, no clear consensus has yet emerged as
critical role in airport landside planning studies and model- to what explanatory variables should be included or how the
ing traffic on the regional transportation system in the vicin- various modes and sub-modes should be nested. In addition,
ity of airports. The ability to predict how air passenger and even the most recent models still have not addressed a num-
airport employee access and egress mode use will change in ber of problematic issues in a meaningful way. These include
response to changes in the airport landside access system or how to treat rental car use by non-residents of a region who
other anticipated changes in the regional transportation sys- often rent a car to meet local travel needs other than the air-
tem is essential to the proper evaluation of proposed mea- port egress and later return access trip, and how best to
sures and projects. However, these decisions are influenced account for the role of household income in the mode choice
by very different factors from those affecting general re- process.
gional travel patterns and the range of transportation options
available to airport travelers are often quite different from In addition to these technical considerations, there has
those for other types of regional trips. Therefore, there is a been almost no attention given to how reliably existing mod-
need for specialized models that can represent these mode els predict air passenger access mode use when circumstance
choice decisions as well as the means to integrate these mod- change from those prevailing when the model development
els or their output into the regional traveling modeling data were collected. However, because the models are typi-
process. cally used to predict mode use under very different condi-
tions, including changes in the physical infrastructure or
Aside from the need to predict ground access mode use at ground transportation services, there is a pressing need for
a given airport, air passenger ground access mode choice also more research into this aspect, as well as continuing research
affects airport choice in regions served by multiple commer- directed at improving the current state of practice.
cial airports. The relative accessibility of each airport is one
of the key determinants in air passenger choice of airport (to- In contrast to air passenger mode choice models, there has
gether with the air service offered at each airport) and for any been very little effort directed at developing airport employee
given traveler the accessibility of an airport is influenced not mode choice models. The majority of metropolitan planning
only by the driving time needed to reach it, but also by the organizations model airport employee trips the same way as
alternative ground transportation options that are available. they model any other journey-to-work trips, if indeed they
In those cases where regional planners or local airport model them at all. The development of better airport em-
authorities wish to encourage the use of secondary airports ployee access models is a promising research opportunity.
serving the region, or to explore the feasibility of construct-
ing a new airport, improvements to the ground access system Finally, many existing regional travel models do not
may be one way to influence air passenger choice of airports explicitly model airport trips, but treat them as general re-
and in turn the willingness of airlines to provide or expand air gional travel. Because of the unique characteristics of airport
service at those airports. travel and the range of transportation options typically avail-
able at airports, this is likely to give fairly poor predictions of
As a result, the development of air passenger ground airport mode use and the resulting vehicle trips. However, the
access mode choice models has been the subject of ongoing likely magnitude of the error is not clear. Further research is
research for more than 30 years. Over this period, the state of needed to explore how well existing regional travel models
practice has slowly evolved from relatively simple multino- account for airport trips and to provide guidance on how best
mial logit models to more complex nested logit models to implement explicit modeling of airport access mode
involving several levels of nesting and four or more market choice in the regional travel modeling process.

77
REFERENCES
2005 Washington–Baltimore Regional Air Passenger Sur- Cambridge Systematics, Inc., Portland International Airport
vey, Metropolitan Washington Council of Governments, Alternative Mode Study, Prepared for the Port of Portland,
National Capital Region Transportation Planning Board, Portland, Ore., Oct. 1998.
Prepared in cooperation with Metropolitan Washington Cambridge Systematics, Inc., NCTCOG Mode Choice Model
Airports Authority and Maryland Aviation Administra- Estimation, Prepared for the North Central Texas Council of
tion, Washington, D.C., Jan. 26, 2006. Governments, Arlington, Tex., undated (file last modified
Atherton, T.J. and M.E. Ben-Akiva, “Transferability and Oct. 4, 2005) [Online]. Available: http://www.nctcog.org/
Updating of Disaggregate Travel Demand Models,” trans/modeling/documentation/index.asp [June 12, 2007].
Transportation Research Record 610, Transportation Cambridge Systematics, Inc., with Urban Analytics, Inc.,
Research Board, National Research Council, Washing- PSRC Travel Model Improvements: Final Report, Prepared
ton, D.C., 1976, pp. 12–18. for Puget Sound Regional Council, Seattle, Wash., Mar.
Badoe, D.A. and E.J. Miller, “Analysis of Temporal Transfer- 2003.
ability of Disaggregate Work Trip Mode Choice Models,” Chang, S.K.J. and C.-L. Hsu, “Intermodal Facilities—Mod-
Transportation Research Record 1493, Transportation eling Passenger Waiting Time for Intermodal Transit Sta-
Research Board, National Research Council, Washington, tions, Transportation Research Record 1753, Transporta-
D.C., 1995, pp. 1–11. tion Research Board, National Research Council,
BART–Oakland International Airport Connector, Final Washington, D.C., 2001, pp. 69–79.
Environmental Impact Report/Environmental Impact COG/TPB Travel Forecasting Models, Version 2.2: Specifi-
Statement, State Clearinghouse Number 99112009, U.S. cation, Validation, and User’s Guide, Metropolitan
Federal Transit Administration and San Francisco Bay Washington Council of Governments and National Capi-
Area Rapid Transit District, Oakland, Calif., Mar. 2002. tal Region Transportation Planning Board, Draft, Wash-
Baycast-90 Users Guide: San Francisco Bay Area Travel ington, D.C., Jan. 19, 2007.
Demand Model System (Cube/Voyager Version), Metro- Coogan, M.A., The Peer Airport Analysis Report, Prepared
politan Transportation Commission, Oakland, Calif., for the Port of Portland, Apr. 9, 1997. (Included as
Aug. 2004. Appendix D to Cambridge Systematics, Inc., Portland
Ben-Akiva, M. and S.R. Lerman, Discrete Choice Analysis: International Airport Alternative Mode Study, Prepared
Theory and Applications to Travel Demand, The MIT for the Port of Portland, Portland, Ore., Oct. 1998.)
Press, Cambridge, Mass., 1985. Dallas–Fort Worth Regional Travel Model (DFWRTM):
Bierlaire, M., “BIOGEME: A Free Package for the Estima- Model Description, Transportation Department, Model
tion of Discrete Choice Models,” Proceedings of the 3rd Development Group, North Central Texas Council of
Swiss Transportation Research Conference, Ascona, Governments, Arlington, Tex., Draft, Feb. 2007 [Online].
Switzerland, 2003. Available: http://www.nctcog.org/trans/modeling/docu-
Bondzio, L., “Study of Airport Choice and Airport Access mentation/index.asp [June 12, 2007].
Mode Choice in Southern Germany,” Airport Planning DKS Associates, Downtown/Natomas/Airport Corridor Al-
Issues, Proceedings of Seminar K, held at the Planning ternatives Analysis/Draft Environmental Impact State-
and Transport, Research and Computation European ment and Environmental Impact Report—Travel Demand
Transport Forum, Brunel University, England, Sep. 2–6, Forecasting, Service and Patronage Impact Assessment
1996, Vol. P 409, 1996. Methodology Report, Prepared for Sacramento Regional
Bowman, J.L., Portland PDX Airport Access Project Mode Transit District under subcontract to Parsons Brincker-
Choice Models, memorandum to Keith Lawton, Portland hoff Quade & Douglas, Inc., Sacramento, Calif., Apr. 15,
Metro, Cambridge Systematics, Inc., Cambridge, Mass., 2002.
July 28, 1997. DKS Associates, Downtown–Natomas–Airport: Travel
Boyle, D.K. and P.R. Gawkowski, “Public Transportation for Forecasting Documentation and Results, Draft, Prepared
Airport Employees: Q3 Extension into John F. Kennedy for Parsons Brinckerhoff Quade & Douglas, Inc., and
International Airport,” Transportation Research Record Sacramento Regional Transit District, Sacramento, Calif.,
1373, Transportation Research Board, National Research Nov. 2004.
Council, Washington, D.C., 1992, pp. 26–34. DMJM Aviation, Inc., Dallas/Fort Worth International Air-
Bradley, M., M.L. Outwater, N. Jonnalagadda, and E. Ruiter, port Rail Planning and Implementation Study—Major In-
“Estimation of an Activity-Based Microsimulation Model vestment Study, Prepared for North Central Texas Council
for San Francisco,” Presented at the 80th Annual Meeting of Governments, Dallas/Fort Worth International Airport,
of the Transportation Research Board, Washington, D.C., Dallas Area Rapid Transit, and Fort Worth Transportation
Jan. 7–11, 2001. Authority, Fort Worth, Tex., Dec. 18, 2002.

78
Domencich, T. and D.L. McFadden, Urban Travel Demand: and D. Reid, eds., Institute of Transportation Studies,
A Behavioral Analysis, North–Holland Publishing Co., Technology Transfer Program, University of California,
Amsterdam, The Netherlands, 1975, reprinted 1996. Berkeley, 2005, pp. 64–75.
Dowling Associates, Inc., San Jose International Airport Gosling, G.D., “Predictive Reliability of Airport Ground
Transit Connection Ridership, Final Report, Prepared for Access Mode Choice Models,” Transportation Research
San Jose International Airport, Lea+Elliott and Walker Record 1951, Transportation Research Board, National
Parking, Oakland, Calif., Jun. 2002. Research Council, Washington, D.C., 2006, pp. 69–75.
Dunlay, W.J., “Model of Employee Access Traffic,” Trans- Guilbault, et Associés, Navette du Centre-Ville à l’Aéroport
portation Engineering Journal, Vol. 104, No. TE3, May Montréal–Trudeau—Volet: Passagers (Shuttle from the
1978, pp. 349–361. City Center to Montréal–Trudeau Airport—Section: Pas-
Ellis, R.H., J.C. Bennett, and P.R. Rassam, “Approaches for sengers), Final Report, Prepared for Aéroports de Montréal,
Improving Airport Access,” Transportation Engineering Saint-Bruno-de-Montarville, Québec, Canada, July 29,
Journal, Vol. 100, No. TE3, Aug. 1974, pp. 661–673. 2005.
Flyvberg, B., “Measuring Inaccuracy in Travel Forecasting: Halcrow Group Limited with Cansult Ltd., Air Rail Link
Methodological Considerations Regarding Ramp Up and from Lester B. Pearson International Airport to Union
Sampling,” Transportation Research, Part A, Vol. 39, Station: Revenue and Ridership Study, Report T8080-01-
2005, pp. 522–530. 1213, Final Report, Prepared for Transport Canada, May
Flyvberg, B., “Policy and Planning for Large-Infrastructure 2002a.
Projects: Problems, Causes, Cures,” Environment and Halcrow Group Ltd., SERAS Surface Access Modelling,
Planning B: Planning and Design, advance online publi- Prepared for the Department of Transport, Local Govern-
cation, Mar. 9, 2007. ment and the Regions, South East and East of England
Flyvberg, B., M.K.S. Holm, and S.L. Buhl, “How (In)accu- Regional Air Services Study, London, England, July
rate Are Demand Forecasts in Public Works Projects?” 2002b.
Journal of the American Planning Association, Vol. 71, Harrington, I.E., The Logan Airport Passenger Ground
No. 2, Spring 2005, pp. 131–146. Access Mode Choice Model, Draft Memorandum, Central
Fratar, T.J., A.M. Voorhees, and M.S. Raff, “Forecasting the Transportation Planning Staff, Boston, Mass., Feb. 28,
Distribution of Interzonal Vehicular Trips by Successive 2003.
Approximations,” Proceedings of the Highway Research Harrington, I.E., J. McClennen, E. Pereira, and C.-Y. Wang,
Board, Highway Research Board, Washington, D.C., Summary of People Mover Study Passenger Mode Choice
Vol. 33, 1954, pp. 376–384. Models, Draft Memorandum, Central Transportation
FY-2003 Models Development Program for COG/TPB Planning Staff, Boston, Mass., May 17, 1996.
Travel Models, Metropolitan Washington Council of Harvey, G., “A Study of Airport Access Mode Choice,”
Governments and National Capital Region Transporta- Journal of Transportation Engineering, Vol. 112, No. 5,
tion Planning Board, Washington, D.C., Draft, June 30, Sep. 1986, pp. 525–545.
2003. Harvey, G., ACCESS: Models of Airport Access and Airport
FY-2005 Development Program for TPB Travel Forecasting Choice for the San Francisco Bay Region—Version 1.2,
Models, Metropolitan Washington Council of Govern- Prepared for the Metropolitan Transportation Commis-
ments, National Capital Region Transportation Planning sion, Berkeley, Calif., Dec. 1988.
Board, Washington, D.C., Draft, June 30, 2005. Hensher, D.A., “Measurement of the Value of Travel Time
FY-2006 Development Program for TPB Travel Forecasting Savings,” Journal of Transport Economics and Policy,
Models, Metropolitan Washington Council of Govern- Vol. 35, No. 1, 2001, pp. 71–98.
ments, National Capital Region Transportation Planning Hensher, D.A., J.M. Rose, and W.H. Greene, Applied Choice
Board, Washington, D.C., June 30, 2006. Analysis: A Primer, Cambridge University Press, Cam-
Galbraith, R.A. and D.A. Hensher, “Intra-Metropolitan bridge, England, 2005.
Transferability of Mode Choice Models,” Journal of Hess, S. and J.W. Polak, “Mixed Logit Modeling of Airport
Transport Economics and Policy, Vol. 16, No. 1, Jan. Choice in Multi-Airport Regions,” Journal of Air Trans-
1982, pp. 7–29. port Management, Vol. 11, No. 2, 2005, pp. 59–68.
Gosling, G.D., An Airport Ground Access Mode Choice Hess, S. and J.W. Polak, “Exploring the Potential for Cross-
Model, Technical Document UCB-ITS-TD-84-6, Insti- Nesting Structures in Airport-Choice Analysis: A Case
tute of Transportation Studies, University of California, Study of the Greater London Area,” Transportation
Berkeley, July 1984. Research, Part E, Vol. 42, 2006, pp. 63–81.
Gosling, G.D.; Cambridge Systematics, Inc.; and SH&E, Howard Humphreys and Partners, Heathrow Surface Access
Inc., SCAG Regional Airport Demand Model: Literature Study, Prepared for the U.K. Department of Transport,
Review, Prepared for the Southern California Association Leatherhead, Surrey, England, June 1987.
of Governments, Los Angeles, Calif., June 2003. ICF Kaiser Engineers, Inc., with Gannett Fleming, Inc., and
Gosling, G.D., “Surface Transportation,” In Airport Air KPMG Peat Marwick, Miami Intermodal Center: Travel
Quality: Approaches, Basics & Challenges, M. Kenny Demand Forecast Report, Prepared for Florida Department

79
of Transportation and Federal Highway Administration, Highway Administration and Florida Department of
Miami, Aug. 1995. Transportation, Tallahassee, Fla., Dec. 23, 1997.
Income, Poverty, and Health Insurance Coverage in the Model Documentation: Mobility 2030 Regional Transporta-
United States: 2005, Current Population Report P60-231, tion Plan, Atlanta Regional Commission, Atlanta, Ga.,
U.S. Census Bureau, Washington, D.C., Aug. 2006. Feb. 11, 2005.
Karasmaa, N., The Transferability of Travel Demand Mod- Mohring, H., J. Schroeter, and P. Wiboonchutikula, “The
els: An Analysis of Transfer Methods, Data Quality and Values of Waiting Time, Travel Time, and a Seat on the
Model Estimation, Transportation Engineering Publica- Bus,” Rand Journal of Economics, Vol. 18, No. 1, 1987,
tion 106, Dissertation for Degree of Doctor of Science in pp. 40–56.
Technology, Department of Civil and Environmental Monteiro, A.B. and M. Hansen, “Improvements to Airport
Engineering, Helsinki University of Technology, Espoo, Ground Access and the Behavior of a Multiple Airport
Finland, 2003. System: BART Extension to San Francisco International
Kumar, A., “Pivot Point Modeling Procedures in Demand Airport,” Transportation Research Record 1562, Trans-
Estimation,” Transportation Engineering Journal, Vol. portation Research Board, National Research Council,
106, No. TE6, Nov. 1980, pp. 647–660. Washington, D.C., 1996, pp. 38–47.
Lam, T.C. and K.A. Small, “The Value of Travel Time and Moreau, A., “Public Transport Waiting Times as Experi-
Reliability: Measurement for a Value Pricing Experi- enced by Customers: Marketing Research Involving the
ment,” Transportation Research, Part E, Vol. 37, 2001, Grenoble System,” Public Transport International, Vol.
pp. 231–251. 41, No. 3, 1992, pp. 52–68.
Leake, G.R. and J.R. Underwood, “An Inter-City Terminal Papacostas, C.S., Fundamentals of Transportation Engi-
Access Modal Choice Model,” Transportation Planning neering, Prentice–Hall, Englewood Cliffs, N.J., 1987, pp.
and Technology, Vol. 4, No. 1, Sep. 1977, pp. 11–21. 276–279.
Leigh Fisher Associates, with M.A. Coogan and Market- Parsons Brinckerhoff, with Lea+Elliott, Inc., JKH Mobility
Sense, TCRP Report 62: Improving Public Transporta- Services, and Infrastructure Services, Inc., Central Area
tion Access to Large Airports, Transportation Research Loop Study, Final Report, Prepared for the Ohio–
Board, National Research Council, Washington, D.C., Kentucky–Indiana Regional Council of Governments,
2000, 148 pp. Cincinnati, Ohio, Dec. 2001.
Lu, X.-Y., G.D. Gosling, S.E. Shladover, J. Xiong, and A. Parsons/SYSTRA Engineering, Inc., Lower Manhattan Air-
Ceder, Development of a Modeling Framework for Analyz- port and Commuter Access Alternatives Analysis, Final
ing Improvements in Intermodal Connectivity at California Report, Prepared for the Lower Manhattan Develop-
Airports, California PATH Research Report UCB-ITS- ment Corporation, New York, N.Y., n.d. [Online]. Avail-
PRR-2006-14, Institute of Transportation Studies, Univer- able: http://mta.info/mta/planning/lmlink/alternatives.htm
sity of California, Berkeley, July 2006. [Mar. 28, 2007].
Lunsford, M.E. and G.D. Gosling, Airport Choice and Pas, E.I. and F.S. Koppelman, “Comparative Analysis of the
Ground Access Mode Choice Models: A Review and Transferability of Disaggregate Automobile-Ownership
Analysis of Selected Literature, Working Paper UCB-ITS- and Mode-Choice Models,” Transportation Research
WP-94-5, Institute of Transportation Studies, University Record 987, Transportation Research Board, National
of California, Berkeley, Calif., June 1994. Research Council, Washington, D.C., 1984, pp. 40–48.
Mandel, B.N., “The Interdependency of Airport Choice and Pearson, D., Houston Intercontinental Airport Travel Survey,
Travel Demand,” In Taking Stock of Air Liberalization, Technical Memorandum on Houston–Galveston Travel
M. Gaudry and R. Mayes, Eds., Kluwer Academic, Surveys, Prepared for Houston–Galveston Area Council,
Boston, Mass., 1999, pp. 189–222. Texas Transportation Institute, College Station, Dec. 17,
McCoomb, L.A., “Analysis of the Transferability of Disag- 1996.
gregate Demand Models Among Ten Canadian Cities,” PDX Ground Access Study Model Summary, Prepared by the
Transportation Forum, Vol. 3, No. 1, pp. 19–32. Travel Forecasting Staff, Portland Metro, Portland, Ore.,
McFadden, D., A.P. Talvitie, and Associates, “Transferabil- May 1998.
ity of Mode-Choice Models of Urban Travel,” Part IV, Pels, E., P. Nijkamp, and P. Rietveld, “Access to and Com-
Chapter 5 of Demand Model Estimation and Forecasting, petition Between Airports: A Case Study for the San
Urban Travel Demand Forecasting Project, Phase 1 Final Francisco Bay Area,” Transportation Research, Vol.
Report Series, Vol. V, Special Report UCB-ITS-SR-77-9, 37A, No. 1, Jan. 2003, pp. 71–83.
Institute of Transportation Studies, University of Califor- Psaraki, V. and C. Abacoumkin, “Access Mode Choice for
nia, Berkeley and Irvine, 1977. Relocated Airports: The New Athens International Air-
Meyer, M.D. and E.J. Miller, Urban Transportation Plan- port,” Journal of Air Transport Management, Vol. 8,
ning: A Decision-Oriented Approach, McGraw–Hill, 2002, pp. 89–98.
New York, N.Y., 1984, pp. 237–238. Purvis, C.L., Travel Demand Models for the San Francisco Bay
Miami Intermodal Center Final Environmental Impact State- Area (BAYCAST-90): Technical Summary, Metropolitan
ment, Report FHWA-FLA-EIS-95-01-F, U.S. Federal Transportation Commission, Oakland, Calif., June 1997.

80
Request for Business Case: Air Rail Link from Toronto–Lester Spear, B.D., An Analysis of the Demand for Airport Bus
B. Pearson International Airport to Toronto Union Station, Services at Washington National and Dulles Airports,
Transport Canada, Ottawa, ON, Canada, May 2003. Report DOT-TSC-FAA-84-2, Transportation Systems
Resource Systems Group, Inc., O’Hare and Midway Airport Center, U.S. Department of Transportation, Cambridge,
Express Train Ridership Forecasting Study: Chicago Air Mass., May 1984.
Traveler Stated Preference Survey Report, Prepared for Tambi, J.E. and J. Falcocchio, “Implications of Parking Pol-
Chicago Department of Transportation, White River icy for Airport Access Mode Choice,” Presented at the
Junction, Vt., Jan. 2004. 1991 Annual Meeting of the Transportation Research
Ricard, D.M., “Challenges in Developing an Airport Board, National Research Council, Washington, D.C.,
Employee Commute Program: Case Study of Boston Jan. 13–17, 1991.
Logan International Airport,” Transportation Re- Train, K.E., Discrete Choice Methods with Simulation, Cam-
search Record 1506, Transportation Research Board, bridge University Press, Cambridge, England, 2003.
National Research Council, Washington, D.C., 1995, Travel Demand Model Documentation, Atlanta Regional
pp. 70–81. Commission, Atlanta, Ga., May 2006.
Rossi, T.F. and M.L. Outwater, “How Valid Is it to Transfer U.S. Federal Transit Administration, Travel Forecasting for
Mode Choice Model Parameters,” Proceedings of the New Starts, Presentation at a Workshop on Travel Fore-
Seventh TRB Conference on the Application of Trans- casting for New Starts Proposals, Minneapolis, Minn., June
portation Planning Methods, Boston, Mass., Mar. 7–11, 15–16, 2006 [Online]. Available: http://www.fta.dot.gov/
1999, Transportation Research Board, National Research printer_friendly/planning_environment_5402.html.
Council, Washington, D.C., Sep. 1999. Vanasse Hangen Brustlin, Determination of the State of the
Sacramento International Airport Transit Access Study, Practice in Metropolitan Area Travel Forecasting: Find-
Report No. SACOG-00-014, Sacramento Area Council of ings of the Surveys of Metropolitan Planning Organiza-
Governments, Sacramento, Calif., July 20, 2000. tions, Prepared for Committee B0090, Transportation
Small, K.A., Urban Transportation Economics, Harwood Research Board, National Research Council, Washing-
Academic Publishers, Reading, England, 1992. ton, D.C., Draft, revised Apr. 7, 2006.
Sobieniak, J., R. Westin, T. Rosapep, and T. Shin, “Choice Waddell, P., M. Outwater, C. Bhat, and L. Blain, “Design of
of Access Mode to Intercity Terminals,” Transportation an Integrated Land Use and Activity-Based Travel Model
Research Record 728, Transportation Research Board, System for the Puget Sound Region,” Presented at the
National Research Council, Washington, D.C., 1979, pp. 81st Annual Meeting of the Transportation Research
47–53. Board, National Research Council, Washington, D.C.,
Southern California Association of Governments (SCAG), Jan. 13–17, 2002.
RADAM Airport Demand Allocation Model—Basic De- Wilbur Smith Associates, in association with Resource Sys-
scription, A Discussion Paper for the SCAG Aviation tems Group, Inc., Airport Express Ridership and Revenue
Task Force RADAM Model Workshop, Los Angeles, Forecast, Prepared for Chicago Department of Trans-
Calif., Oct. 17, 2002. portation, Chicago, Ill., 2004.

81
BIBLIOGRAPHY
Borges, H., Air Rail Link: Pearson Airport–Union Station, portation Access to Large Airports, Transportation Re-
Presented to the Chartered Institute of Logistics and search Board, National Research Council, Washington,
Transport, Transport Canada, Ottawa, ON, Apr. 11, 2006. D.C., 2002, 90 pp.
Franz, J.D., 1995 Metropolitan Transportation Commission Metropolitan Transportation Commission, Report on Find-
Airline Passenger Survey, Final Report, Prepared for the ings of the Peer Review Panel for the Metropolitan Trans-
Metropolitan Transportation Commission, Oakland, portation Commission (MTC), December 2–3, 2004,
Calif., by J.D. Franz Research, Feb. 1996. Prepared for the U.S. Department of Transportation,
Gosling, G.D., “Analysis of Changes in Airport Ground Transportation Model Improvement Program, Oakland,
Access Mode Use,” Meeting the Challenge: Rebuilding Calif., Mar. 2005.
Inner City Airports, Proceedings of the 24th International PB Consult, Inc., in association with Mercer Management
Air Transportation Conference, Louisville, Ky., June 5–7, Consulting, Inc., Parsons Brinckerhoff Quade & Douglas,
1996, pp. 63–77. Inc., and Velma Butler & Company, Ltd., Express Airport
Leigh Fisher Associates, with M.A. Coogan and MarketSense, Train Service: Business Plan, Final Report, Prepared for
TCRP Report 83: Strategies for Improving Public Trans- Chicago Transit Authority, Chicago, Ill., Sep. 22, 2006.

82
GLOSSARY
Air party—group of air passengers traveling together. ratio of the probability of choosing one alternative to that of
Airport access trip—travel by air passengers, airport em- choosing a second alternative is equal to the difference in
ployees, or others to an airport from their home or other the value of the utility functions of the two alternatives.
local trip origin. Mode choice—process by which a traveler chooses the mode
Airport egress trip—travel by air passengers, airport em- of transportation to use for a trip.
ployees, or others from an airport to their home or other Mode of transportation—specific form of transportation
destination. (e.g., automobile, taxi, and public transit).
Alternative-specific constant (ASC)—constant term in the Model calibration—process of adjusting the formulation or
utility function for a specific mode in a mode choice model parameters of a model so that the predictions of the model
that reflects the attributes of the mode that are not accounted correspond to an observed outcome of the process being
for by the other variables in the utility function. modeled.
Calibration—see model calibration. Model estimation—process of determining the values of the
Causal variable—factor that is believed to influence the out- parameters of a model that result in the model giving the
come of some process, such as a characteristic of a deci- best fit to a given set of data describing the process being
sion maker or a measure of the situation faced by that de- modeled.
cision maker that influences the decision that is made. Monte Carlo simulation—an analysis technique in which a
Coefficient—terms in a model that assume fixed values and series of pseudo-random numbers are generated by a com-
interact with the independent variables to predict the value puter program (pseudo-random because numbers generated
of the dependent variable. by a computer program cannot be truly random) and then
Composite alternative—an alternative in a choice model used to select values from one or more defined statistical
(typically a nested choice model) that represents the com- distributions to generate a new dataset that conforms to the
bined effect of several discrete sub-choices. statistical distribution(s).
Continuous variable—variable in the utility function for a Multinomial logit (MNL) model—form of logit model with
specific mode in a mode choice model (or indeed a vari- more than two choice alternatives in which all the alterna-
able in any model) that can assume any appropriate value. tives enter into the choice process at the same level.
Dependent variable—term in a model the values for which Nested logit (NL) model—form of logit model in which
are predicted by the model based on the values of the in- some choice alternatives are considered to enter into the
dependent variables. choice process as sub-choices of a higher-level choice
Deplanements—number of passengers getting off aircraft at alternative.
an airport. Originations—number of passengers boarding aircraft at an
Disaggregate choice model—model explaining the choice airport who have traveled to the airport by ground and are
made by an individual (or group of individuals) in terms of starting their directional air trip.
the characteristics of that individual or group and the pre- Pivot-point analysis—a technique in which the coefficients of
vailing conditions under which the choice was made. a mode choice model are used to predict the proportional
Dummy variable—variable that can only take the value zero change in mode share from a change in some transportation
or one. service variable and then this proportional change is ap-
Enplanements—number of passengers boarding aircraft at an plied to the existing mode share to predict the new mode
airport. share.
Estimation—see model estimation. Revealed preference model—model of behavioral choice
Explanatory variable—independent variable describing process estimated from data on choices actually made by
some external factor or attribute of the process being mod- subjects together with data on factors believed to influence
eled that is believed to influence or has been shown to in- their choice process and descriptive data on alternatives
fluence the outcome of the process being modeled. available to subjects from which they made their choices.
Independent variable—term in a model that can take values Stated preference model—model of behavioral choice process
that can be varied independently of the values of other estimated from data in which subjects select options from
terms in the model. among hypothetical sets of alternatives for which they have
Logarithmic transformation—conversion of a variable to the been provided with comparative descriptive data.
logarithm (usually the natural logarithm) of its value. Utility function—mathematical expression combining the
Logit model—particular form of mathematical model describ- effect of several causal variables into a single measure of
ing the probability of choosing an alternative from among a the perceived attractiveness of a given alternative within a
set of possible choices, in which the natural logarithm of the set of possible choices.

83
APPENDIX A
Survey Questionnaires
Four versions of the survey questionnaire were prepared to tailor the questions to the four types of organizations that become involved in
airport ground access modeling:
• Airport Authorities
• Metropolitan Planning Organizations
• Airport Planning Consultants and Research Organizations
• State and Federal Government and Industry Organizations
Each of these questionnaires is shown in turn on the following pages.

84
ACRP SYNTHESIS SURVEY
Airport Authority Survey
Survey purpose: This survey is being performed as part of a study by the Airport Cooperative Research Program (ACRP) to docu-
ment the state-of-the-practice in airport ground access mode choice models. Formal models of airport ground access mode choice form
a key analytical component of airport landside planning, as well as airport system planning, and are used to predict how airport users
will change their access or egress mode in response to changes in the airport ground transportation system (such as changes in fares,
rates, or service levels) or the introduction of new modes (such as the extension of a light rail system to the airport). Airport access
trips are different from the typical trip accounted for in regional transportation planning models. Therefore, it is difficult to determine
the economic feasibility of proposed projects to improve airport ground transportation or effectively manage the existing airport
ground transportation system using traditional regional travel forecasting models. However, models designed to address airport
ground access travel decisions are highly specialized and their technical details are often not well understood by airport managers and
planners. The purpose of this survey is to gather information on the use of such models by airports and other organizations.
The final report, to be published by the Transportation Research Board, will document the technical details of existing airport ground
access mode choice models together with agency assessments of their effectiveness and reliability. This report will be extremely use-
ful to all airports and other agencies involved in airport ground transportation planning as they consider how best to analyze the effect
of changing conditions and circumstances on how air passengers, airport employees, and other airport users decide to travel to and
from the airport. All survey responses will be confidential.
Thank you for taking the time to participate. Instructions on returning the survey are included on the last page.
This form can be completed by typing information in the fill-in text fields and clicking the check
boxes to check them. Save the file when done and return by e-mail. Alternatively, the form
can be printed, filled in by hand, and returned by fax or mail.
RESPONDENT INFORMATION
Date: _______________________________________________________________________________________________________
Name and Title of Respondent: ___________________________________________________________________________________
Agency Name: ________________________________________________________________________________________________
Respondent Telephone Number: __________________________________________________________________________________
Respondent e-Mail Address: _____________________________________________________________________________________
AIRPORT GROUND ACCESS STUDIES
1. Has your airport undertaken or commissioned any studies in the past 10 years that have included some analysis of airport ground access
mode choice?
䡺 Yes. Please provide title of study or studies (add lines as necessary).
1. _____________________________________________________________________________________________________
2. _____________________________________________________________________________________________________
3. _____________________________________________________________________________________________________
䡺 No (Skip to Q.8)
2. Did any of these studies make use of formal analytical models of airport ground access mode choice?
䡺 Yes. Please indicate which (using numbers from Q.1): ____________________________________________________________
3. Are reports from any of these studies available on your organization’s website?
䡺 Yes. Please provide website address: _________________________________________________________________________
䡺 No

85
4. What was the source of air passenger survey data used in the development of these airport ground access mode choice models? (Please
check all that apply.)
䡺 Surveys performed specifically for the model development
䡺 Surveys performed for other purposes
䡺 Don’t know
5. Are reports on any of these surveys available on your organization’s website?

䡺 No
6. Based on your agency’s past experience with airport ground access mode choice models, how would you characterize the current state-
of-practice? (Please check all that apply.)
䡺 Adequate for our needs
䡺 Predictions from existing models are not reliable enough
䡺 Available models are too complex to use
䡺 Models are too costly to use except for very large studies
Please skip to Q.8
7. What was the reason that these studies did not include any analytical modeling of airport ground access mode choice? (Please check all
that apply.)
䡺 Scope of studies did not require it
䡺 Available models are not reliable enough
䡺 Available models are too difficult to use
䡺 There is a lack of information and guidance on the use of such models
䡺 Decision on analytical techniques to use was made by those performing the studies
䡺 Don’t know
8. Has your airport undertaken or commissioned any studies prior to the last 10 years that made use of formal analytical models of airport
ground access?
䡺 Yes
䡺 No
䡺 Not sure
AIRPORT CHOICE STUDIES
9. Has your airport undertaken or commissioned any studies in the past 10 years that have included some analysis of air passenger airport
choice?
1. _____________________________________________________________________________________________________
2. _____________________________________________________________________________________________________
3. _____________________________________________________________________________________________________
10. Did any of these studies make use of analytical techniques that took airport ground access mode choice into consideration as a factor
in airport choice?
䡺 No
STUDIES BY OTHER ORGANIZATIONS
11. Are you aware of any other airports or organizations that have made use of formal analytical models of airport ground access mode
choice in the course of studies or analysis addressing airport ground transportation or airport choice?
䡺 Yes. Please provide name of organization(s) (list up to three).
1. _____________________________________________________________________________________________________
2. _____________________________________________________________________________________________________
3. _____________________________________________________________________________________________________
䡺 No
86
GENERAL COMMENTS
12. Please feel free to add any comments about airport ground access mode choice models in general or the scope of this ACRP study in
particular:
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
Please return by February 23, 2007 to: Geoffrey Gosling

Principal
805 Colusa Avenue
Berkeley, CA 94707
phone: (510) 528-8741
fax: (510) 528-8745
e-mail [email protected]
We encourage you to return your completed survey via e-mail. If you have any questions on the survey or the project, feel free to contact
Geoff Gosling by e-mail or phone.
THANK YOU FOR YOUR HELP

87
Metropolitan Planning Organization Survey
Date: _______________________________________________________________________________________________________
Agency Name: ________________________________________________________________________________________________
1. Has your agency undertaken or commissioned any studies in the past 10 years that have included some analysis of airport ground
access mode choice, apart from your normal travel demand modeling activities?
4. _____________________________________________________________________________________________________
5. _____________________________________________________________________________________________________
6. _____________________________________________________________________________________________________
2. Did any of these studies make use of formal analytical mode choice models for airport ground access trips?

88
3. Were these mode choice models specifically developed for airport ground access travel or did you use or adapt mode choice models for
general urban travel behavior? (Please check all that apply.)
䡺 Studies used special-purpose airport ground access mode choice models
䡺 Studies adapted special-generator sub-models designed for other types of trips
䡺 Studies adapted mode choice models for general urban travel to airport trips
䡺 Studies used mode choice models for general urban travel for airport trips
䡺 No
5. What was the source of air passenger survey data (if any) used in the development of the mode choice models for airport ground access
travel? (Please check all that apply.)
䡺 Surveys performed for other purposes by this agency
䡺 Surveys performed for other purposes by another agency
䡺 No survey data used in the development of the models
䡺 Don’t know

䡺 No
7. Based on your agency’s past experience with airport ground access mode choice models, how would you characterize the current state-
of-practice? (Please check all that apply.)
Please skip to Q.9
that apply.)
䡺 Don’t know
9. Has your agency undertaken or commissioned any studies prior to the last 10 years that made use of formal analytical models of air-
port ground access?
䡺 Yes
䡺 No
䡺 Not sure
10. How do you model airport trips in your general regional travel modeling process? (Please check all that apply.)
䡺 Use a special-generator sub-model for air passenger trips
䡺 Use a special-generator sub-model for airport employee trips
䡺 Treat air passenger trips the same way as other home-based non-work trips
䡺 Treat air passenger trips the same way other non-home-based non-work trips
䡺 Treat airport employee trips the same way as other journey-to-work trips
䡺 Do not consider airport trips in regional travel modeling process
䡺 Other (Please describe): ___________________________________________________________________________________
_______________________________________________________________________________________________________

89
11. Has your agency undertaken or commissioned any studies in the past 10 years that have included some analysis of air passenger air-
port choice?
1. _____________________________________________________________________________________________________
2. _____________________________________________________________________________________________________
3. _____________________________________________________________________________________________________
in airport choice?
䡺 Yes. Please indicate which (using numbers from Q.11): ___________________________________________________________
䡺 No
13. Are you aware of any other planning agencies, airports, or organizations that have made use of formal analytical models of airport
ground access mode choice in the course of studies or analysis addressing airport ground transportation or airport choice?
䡺 Yes. Please provide name of organization (list up to three).
1. _____________________________________________________________________________________________________
2. _____________________________________________________________________________________________________
3. _____________________________________________________________________________________________________
䡺 No
GENERAL COMMENTS
14. Please feel free to add any comments about airport ground access mode choice modes in general or the scope of this ACRP study in
particular.
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________

Principal
805 Colusa Avenue
Berkeley, CA 94707
phone: (510) 528-8741
fax: (510) 528-8745

90
Consulting Firm and Research Organization Survey
Date: _______________________________________________________________________________________________________
Organization: ________________________________________________________________________________________________
1. Have you or your organization undertaken or participated in any studies in the past 10 years that have included some analysis of airport
ground access mode choice?
䡺 Yes. How many? _____
Please provide name of study sponsor and title of the most recent study or studies (list up to six).
1. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
2. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
3. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
4. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
5. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
6. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
91
3. Are reports from any of these studies available on your or another organization’s website?
䡺 No
䡺 Don’t know
5. Are reports on any of these surveys available on your or another organization’s website?
䡺 No
6. Based on your organization’s past experience with airport ground access mode choice models, how would you characterize the current
state-of-practice? (Please check all that apply.)
Please skip to Q.8
that apply.)
䡺 Don’t know
8. Have you or your organization undertaken or participated in any studies in the past 10 years that have included some analysis of air
passenger airport choice?
䡺 Yes. How many? _______________________________________________________________________________________
Please provide name of study sponsor and title of the most recent study or studies (list up to three).
1. Sponsor: _____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
2. Sponsor: _____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
3. Sponsor: _____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
in airport choice?
䡺 No

92
10. Are you aware of any other organizations that have made use of formal analytical models of airport ground access mode choice in the
course of studies or analysis addressing airport ground transportation or airport choice?
䡺 Yes. Please provide name of organization(s) (list up to three).
4. _____________________________________________________________________________________________________
5. _____________________________________________________________________________________________________
6. _____________________________________________________________________________________________________
䡺 No
GENERAL COMMENTS
particular:
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________

Principal
805 Colusa Avenue
Berkeley, CA 94707
phone: (510) 528-8741
fax: (510) 528-8745

93
Government and Industry Organization Survey
Date: _______________________________________________________________________________________________________
Agency Name: ________________________________________________________________________________________________
1. Has your organization undertaken or commissioned any studies in the past 10 years that have included some analysis of airport ground
access mode choice?
7. _____________________________________________________________________________________________________
8. _____________________________________________________________________________________________________
9. _____________________________________________________________________________________________________
䡺 No

94
䡺 Don’t know

䡺 No
6. Based on your organization’s past experience with airport ground access mode choice models, how would you characterize the current
state-of-practice? (Please check all that apply.)
Please skip to Q.8
that apply.)
䡺 Don’t know
8. Has your organization undertaken or commissioned any studies in the past 10 years that have included some analysis of air passenger
airport choice?
4. _____________________________________________________________________________________________________
5. _____________________________________________________________________________________________________
in airport choice?
䡺 No
10. Are you aware of any airports or other organizations that have made use of formal analytical models of airport ground access mode
choice in the course of studies or analysis addressing airport ground transportation or airport choice?
䡺 Yes. Please provide name of study sponsor and title of study (list up to five).
7. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
8. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
9. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________

95
10. Sponsor: ____________________________________________________________________________________________

Study: ______________________________________________________________________________________________
11. Sponsor: ____________________________________________________________________________________________
Study: ______________________________________________________________________________________________
䡺 No
GENERAL COMMENTS
particular:
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________
___________________________________________________________________________________________________________

Principal
805 Colusa Avenue
Berkeley, CA 94707
phone: (510) 528-8741
fax: (510) 528-8745

96
APPENDIX B
Summary of Survey Results
The following tables (B1–B6) summarize the responses to the survey questions by type of responding organization. Where more than one
individual responded from a given organization, their responses have generally been combined in the following statistics. For ease of refer-
ence, the question numbers in the tables generally follow the numbering in the Airport Authority survey questionnaire. The corresponding
questions may have different numbers in the questionnaires for other types of organizations owing to the addition of questions specific to
each type of organization. The two questions that were specific to metropolitan planning organizations are numbered with their original num-
ber in the relevant questionnaire.
TABLE B1
Airport State/Fed Consulting

Authorities MPOs Government Firms Research Total Percent
Survey Responses
32 24 19 23 7 105
(organizations)
Q.1: Number
reporting ground 21 8 2 20 6 57 54%
access studies
Number of studies
24 11 7 43 10 95
identified
Unduplicated 85
Q2: Number
reporting mode 8 5 2 13 6 34 60%a
choice model use
Number of studies
12 7 7 23 9 58
identified
Unduplicated 52
Q3: Study reports

1 2 0 0 2 5 15%b
available on web
Notes: a) Percent of organizations reporting ground access studies

b) Percent of organizations reporting use of mode choice models

TABLE B2
AIRPORT ACCESS MODE CHOICE MODEL DEVELOPMENT AND USE

Number reporting
mode choice model 8 5 2 13 6 34
use
Q.4: Source of air

Of those
passenger survey
using
data
models
Model
development 6 2 1 9 5 23 68%
Other purposes 4 3 2 10 4 23 68%
Don’t know 0 0 0 0 0 0
Q.5: Survey reports

1 0 0 2 2 5 15%
available on web
Of those
Q.7: Reasons
not
modeling not used
using
(where not)
models
Number reporting 12 3 7 22
Not required 8 1 6 15 68%
Not reliable - 2 - 2 9%
all all
enough reported reported
Too difficult to - 1 - 1 5%
model model
use use use
Lack of guidance - 2 2 4 18%
Decided by 2 - - 2 9%
others
Don’t know 2 - 1 3 14%
TABLE B3
CURRENT STATE OF PRACTICE WITH AIRPORT ACCESS MODE CHOICE MODELS

Number reporting
mode choice model 8 5 2 13 6 34
use
Q.6 Assessment of
current state of
practice
Adequate for 7 5 - 5 2 19 56%
needs
Not reliable 1 - 1 6 4 12 35%
enough
Too complex to - - 1 2 - 3 9%
use
Too costly to use 2 - 1 5 2 10 29%
TABLE B4
STUDIES INVOLVING AIRPORT ACCESS MODE CHOICE
MODELS PRIOR TO LAST TEN YEARS
Airport
Authorities MPOs Total Percent
Survey responses to
31 24 55
question
Q.8 Prior studies using

mode choice modeling
Yes 4 5 9 16%
No 15 14 29 53%
Not sure 12 5 17 31%

98
TABLE B5
STUDIES OF AIR PASSENGER AIRPORT CHOICE

Survey responses
32 24 19 23 7 105
(organizations)
Q.9: Number
reporting airport 13 2 6 14 2 37 35%
choice studies
Number of studies
11 4 3 13 3 34
identified
Unduplicated 32
Q10: Number
reporting mode 5 1 1 6 2 15 41%
choice model use
Number of studies
4 3 1 5 3 16
identified
Unduplicated 16
Note: Some respondents reported involvement in airport choice studies or use of mode choice models but did not
identify specific studies
TABLE B6
MODELING AIRPORT ACCESS TRIPS IN REGIONAL TRANSPORTATION
PLANNING
MPOs Percent
Number responding to Question 3 (MPO performed study

5
using airport access mode choice model)
Q.3 (MPO) Type of model used (a)

Special purpose airport access model 4 80%
Adapted special-generator model 1 20%
Adapted general urban travel model —
Used general urban travel model 1 20%
Number responding to Question 10 (all MPO respondents) 24
Q.10 (MPO) How airport trips are modeled in general

regional travel modeling process (b)
Special-generator sub-model for air passengers 15 65%
Model air passengers as home-based non-work trips 6 26%
Model air passengers as non-home-based non-work trips 5 22%
Special-generator sub-model for airport employees 5 22%
Model airport employees as regular journey-to-work trips 12 52%
Do not consider airport trips in modeling process —
Other 2 9%
Don’t know 1
Notes: a) Multiple responses allowed

b) Percent excludes “Don’t know” response

99
AIRPORT GROUND ACCESS STUDIES IDENTIFIED BY SURVEY RESPONDENTS
Studies Involving Airport Access Mode Choice Modeling
Sponsor Study
Atlanta Regional Commission Regional Travel Demand Model Update

Bay Area Rapid Transit District Oakland Airport Connector EIR/EIS
Chicago DOT/Chicago Transit Authority Chicago Airport Express Study
Dallas/Fort Worth International Airport Curbside Assignment Study
Dallas/Fort Worth International Airport DFW Zonal Activity Forecasts
Federal Aviation Administration New England Regional Airport System Plan
Federal Transit Administration Rail Access to Honolulu International Airport
Federal Transit Administration Rail Access to Minn.–St. Paul Int’l Airport
Federal Transit Administration Rail Access to New York Kennedy Int’l. Airport
Federal Transit Administration Washington–Dulles Access Study
Greater Orlando Aviation Authority Analysis of Terminal Curbside Roadways
Hartsfield–Jackson Atlanta Int’l. Airport Terminal Planning—Phases I–III
Lower Manhattan Development Corp. JFK to Lower Manhattan Transit Access Study
Massachusetts Port Authority Boston Logan AITC Environmental Assessment
Massachusetts Port Authority Logan Express Analysis
Massachusetts Port Authority Logan Airport Environmental Status and Planning Report
Massachusetts Port Authority Surface Access Study of Boston Area Airports
Metro–Portland, Oregon Airport Max (Red Line) Analysis
Miami–Dade County Transit Airport Access Study
New York MTA/NYC Transit LaGuardia Access Study
North Central Texas Council of Govts. DFW Rail Access Implementation Plan
Port Authority of NY & NJ Regional Air Service Demand Study
Port of Seattle SEA-TAC Comprehensive Development Plan
Joint Transportation Study
Sacramento Regional Transit District Sacramento Airport Light Rail Extension Study
San José International Airport Airport Transit Connector
Southern California Assoc. of Govts. 1998 Regional Transportation Plan
Southern California Assoc. of Govts. Regional Airport Demand Model
Unspecified O’Hare Commuter Rail Market Assessment
Privately Funded Studies
Denver East Corridor Rail Group, LLC Denver Airtrain Study

Resource Systems Group, Inc. Biannual Syndicated Air Traveler Surveys
Canada
Aéroports de Montréal Navette du Centre-ville à l’Aéroport Montréal–Trudeau/Volet: Passagers

Greater Toronto Airports Authority Toronto Pearson International Airport
Air Rail Link Revenue and Ridership Study
United Kingdom
BAA, Plc Heathrow Airport Surface Access Model (2003)

BAA, Plc Heathrow Airport Surface Access Model (2004–current)
Scottish Executive Rail Links to Glasgow and Edinburgh Airports

100
High Speed Rail and Maglev Studies Involving Mode Choice Modeling
Sponsor Study
California High-Speed Rail Authority Bay Area/California High-Speed Rail Ridership and Revenue Forecasting Study
Florida Department of Transportation Florida High Speed Rail Study
Maryland Department of Transportation Baltimore to Washington Maglev Study
Pennsylvania Dept. of Transportation Pittsburgh Maglev Study
University Research Involving Airport Access Mode Choice Modeling
University Study
University of California, Berkeley Planning for Improved Intermodal Connectivity at California Airports
University of Virginia Feasibility of Offsite Terminals in Landside Access
International
City University of Hong Kong Ground Access Mode Choices at Hong Kong International Airport
Imperial College, London Airport, Airline, and Access Choice in SF Bay Area
Imperial College, London Airport, Airline, and Access Choice in Greater London
National Technical University of Athens Access Mode Choice for Relocated Airports
National Technical University of Athens Demand and Mode Choice in Access to Airports
National Technical University of Athens Parking Capacity Requirements for Relocated Airports
University Research Involving Airport Ground Access Analysis
University Study
Virginia Transportation Research Council Evaluating Improvements in Landside Access for Airports
Other Airport Ground Access Studies
Sponsor Study
Anchorage International Airport Airport Roadway Traffic Analysis

E-470 Public Highway Authority Denver E-470 (Tollroad) Airport Forecasts
Dallas Area Rapid Transit Application of the North Central Texas Council of Governments Regional Travel
Demand Model to the planning for the Northwest Corridor light rail transit line
National Aeronautics and Space Administration Civil tilt rotor studies
Hartsfield–Jackson Atlanta Int’l. Airport CONRAC Study
Houston Intercontinental Airport Master Plan
I-95 Corridor Coalition Airport Ground Access Study
Indianapolis Metropolitan Planning Organization Intermodal Freight Study
Jacksonville Aviation Authority Jacksonville International Airport Master Plan 2002
Los Angeles World Airports LAX Advanced Planning
Los Angeles World Airports LAX Master Plan EIS/EIR
Los Angeles World Airports Ontario International Airport Master Plan—Phase 2
Metropolitan Washington Airports Authority North Area Road Landside VISSIM Simulation
Metropolitan Washington Council of Governments Washington/Baltimore Region Access Forecast Update
Puget Sound Regional Council Regional Airport Ground Access Plan (2004)
Raleigh–Durham Airport Authority
(with Triangle Transit Authority) Airport Rail Link Study
Reno–Tahoe Airport Authority Terminal Area Master Plan
Reno–Tahoe Airport Authority Terminal Programming Study (underway)
Sacramento County Airport System Sacramento International Airport Master Plan Update
San Diego Regional Airport Authority San Diego International Airport Master Plan
San Diego Regional Airport Authority San Diego Int’l. Airport Terminal 2 Expansion
San José International Airport Terminal Area Improvement Program

101
Sponsor Study
Southern California Assoc. of Govts. 2007 Regional Transportation Plan (underway)

Tampa International Airport Airport Master Plan Updates
West Virginia Public Port Authority Western West Virginia Regional Airport Feasibility Study
Canada
Greater Toronto Airports Authority Toronto Pearson International Airport
New Terminal Groundside Design
Ottawa Macdonald–Cartier International 2006 survey of key attributes of modal choice
Airport Authority
AIRPORT CHOICE STUDIES IDENTIFIED BY SURVEY RESPONDENTS
Studies Involving Airport Access Mode Choice Modeling
Sponsor Study
Federal Aviation Administration New England Regional Airport System Plan

Federal Transit Administration Washington–Dulles Access Study
New York State Dept. of Transportation Stewart Airport Access Analysis
Port Authority of NY & NJ Regional Air Service Demand Study
Resource Systems Group, Inc. Biannual Syndicated Air Traveler Surveys
Sacramento County Airport System Sacramento International Airport Master Plan Update
San Diego Association of Governments Maglev Study, Phase 1
Southern California Assoc. of Govts. Regional Airport Demand Model
Unspecified High Speed Rail Access to Third Chicago Airport
Canada
Greater Toronto Airports Authority Air Traveller Choice in Southern Ontario
University Research Involving Airport Choice and Access Mode Modeling
University Study
Imperial College, London Airport, airline, and access choice in SF Bay Area
Imperial College, London Airport, airline, and access choice in Greater London
Imperial College, London Stated preference studies of airport choice
Other Airport Choice Studies
Sponsor Study
Anchorage International Airport Airport Master Plan Update 2002

Hartsfield–Jackson Atlanta Int’l. Airport CONRAC Study
Jacksonville Aviation Authority Jacksonville International Airport Master Plan 2002
Los Angeles World Airports LAX Advanced Planning
Los Angeles World Airports LAX Master Plan
Los Angeles World Airports Remote Terminal Market Analysis for LAX
Massachusetts Port Authority Hanscom Airport Environmental Status and Planning Report
Minnesota DOT Office of Aeronautics State of Minnesota Leakage Study
North Central Texas Council of Govts. DFW Rail Access Implementation Plan
Reno–Tahoe Airport Authority 2005 Catchment Area Survey
Southern California Assoc. of Govts. 2007 Regional Transportation Plan (underway)
Tampa International Airport Air service development market studies
Tampa International Airport International Market Study, March 2006

102
APPENDIX C
List of Participating Agencies
AIRPORT AUTHORITIES
State Agency
AK Alaska Dept. of Transportation & Public Works (Anchorage International Airport)

AR Fort Smith Regional Airport
AZ Albuquerque International Sunport
CA City of San José Airport Dept. (Norman Y. Mineta San Jose International Airport)
CA County of Orange (John Wayne Airport)
CA Los Angeles World Airports
CA Reno–Tahoe Airport Authority
CA Sacramento County Airport System (Sacramento International Airport)
CA San Francisco International Airport
CO Denver International Airport
DC Metropolitan Washington Airports Authority
FL Greater Orlando Aviation Authority
FL Jacksonville Aviation Authority
FL Lee County Port Authority (Southwest Florida Regional Airport)
FL Tampa International Airport
GA Hartsfield–Jackson Atlanta International Airport
LA Louis Armstrong New Orleans International Airport
MA Massachusetts Port Authority (Boston Logan International Airport)
MN Metropolitan Airports Commission (Minneapolis–St. Paul International Airport)
MO Lambert–St. Louis International Airport
NC Raleigh–Durham Airport Authority
NY/NJ Port Authority of New York and New Jersey
PA Allegheny County Airport Authority (Pittsburgh International Airport)
PA Philadelphia International Airport
TN Memphis–Shelby County Airport Authority (Memphis International Airport)—State Agency
TX City of Dallas Department of Aviation (Dallas Love Field Airport)
TX Dallas/Fort Worth International Airport
WA Port of Seattle (Seattle–Tacoma International Airport)
WI General Mitchell International Airport (Milwaukee)
Canada
Quebec Aéroports de Montréal

Ontario Greater Toronto Airports Authority
Ontario Ottawa Macdonald–Cartier International Airport
METROPOLITAN PLANNING ORGANIZATIONS
State Agency
CA Sacramento Area Council of Governments

CA Southern California Association of Governments (Los Angeles region)
CT Capitol Region Council of Governments (Hartford)
DC Metropolitan Washington Council of Governments
FL First Coast Metropolitan Planning Organization (Jacksonville)

103
State Agency
FL Hillsborough County Metropolitan Planning Organization (Tampa)

FL Metroplan Orlando
GA Atlanta Regional Commission
HI Oahu Metropolitan Planning Organization (Honolulu)
IL Chicago Metropolitan Agency for Planning
IN Indianapolis Metropolitan Planning Organization
KY Kentuckiana Regional Planning and Development Agency (Louisville)
MA Central Transportation Planning Staff (Boston)
MD Baltimore Metropolitan Council
MN Metropolitan Council of the Twin Cities (Minneapolis–St. Paul)
NV Regional Transportation Commission of Southern Nevada (Las Vegas)
NY New York Metropolitan Transportation Council
OH Northeast Ohio Areawide Coordinating Agency (Cleveland)—State Agency
OH Ohio–Kentucky–Indiana Regional Council of Governments (Cincinnati)
OR Metro (Portland)
TX Houston–Galveston Area Council
TX North Central Texas Council of Governments (Dallas/Fort Worth)
VA Virginia Department of Transportation (travel modeling for MPOs in Virginia)
WA Puget Sound Regional Council (Seattle)
AIRPORT/TRANSPORTATION CONSULTING FIRMS
State Organization
CA Dowling Associates, Inc. (Oakland)

CA Mark Bradley Research and Consulting (Santa Barbara)
CA PB Consult (San Francisco)
CA Velocity Group (San Francisco)
CO PRT Consulting, Inc. (Franktown)
MA Cambridge Systematics, Inc. (Cambridge)
MA MarketSense Consulting, LLC (Charlestown)
MA SH&E, Inc. (Cambridge)
MD Vanesse Hangen Brustlin (Silver Spring)
NY Aviation Consulting (New York)
NY Earth Tech (New York)
PA GRA, Incorporated (Jenkintown)
SC Wilber Smith Associates (Columbia)
TX DMJM Aviation (Fort Worth)
TX TransSolutions, LLC (Fort Worth)
VA AECOM Consult, Inc. (Arlington)
VA HNTB Corporation (Alexandria)
VA Larry Kiernan, Consultant (Reston)
VA Ricondo & Associates (Alexandria)
VT Resource Systems Group, Inc. (White River Junction)—State Organization
WA HDR Engineering, Inc. (Bellevue)
WA URS Corporation (Seattle)
International
Australia Sinclair Knight Merz (Armadale, Victoria)

104
RESEARCH ORGANIZATIONS
State Organization
CA University of California at Berkeley

MI Michigan Tech University (Houghton)
VA University of Virginia (Charlottesville)
VA Virginia Transportation Research Council (Charlottesville)
International
Greece National Technical University of Athens

H.K. City University of Hong Kong
U.K. Imperial College (London)
STATE AND FEDERAL GOVERNMENT
State Agency
AZ Arizona Department of Transportation, Aeronautics Division (Phoenix)

CA California Department of Transportation, Division of Aeronautics (Sacramento)
FL Florida Department of Transportation, Aviation Office (Tallahassee)
IA Iowa Department of Transportation, Office of Aviation (Ames)
KY Kentucky Department of Aviation (Frankfort)
MN Minnesota Department of Transportation, Office of Aeronautics (St. Paul)
MS Mississippi Department of Transportation, Aeronautics Division (Jackson)
ND North Dakota Aeronautics Commission (Bismarck)
NE Nebraska Department of Aeronautics (Lincoln)
NJ New Jersey Department of Transportation, Division of Aeronautics (Trenton)—State Agency
NY New York State Department of Transportation, Aviation Services Bureau (Albany)
OH Ohio Department of Transportation, Office of Aviation (Columbus)
OK Oklahoma Aeronautics Commission
OR Oregon Department of Aviation (Salem)
TX Texas Department of Transportation, Aviation Division (Austin)
WA Washington State Department of Transportation (Arlington)
Federal
DC Federal Aviation Administration

DC Federal Transit Administration
MA U.S. Department of Transportation, Volpe National Transportation Systems Center

105
APPENDIX D
Mode Choice Model Technical Summaries
This appendix contains detailed technical summaries of selected Taxi 8%

airport ground access mode choice models discussed in the body of Limousine 2%
the report. The summaries document the technical details of the Public transit 10%
models, including the functional form adopted, market segmenta- Commercial shuttle/van 4%
tion, causal variables included in the model, estimated parameter Hotel/motel courtesy vehicle 4%
values, and goodness-of-fit measures where these have been re- Other <1%
ported in the literature or model documentation. The summaries Market Segmentation Residents—Business trips
also include statistical information on the airports for which the Residents—Non-business trips
models have been developed. Non-residents—Business trips
Non-residents—Non-business trips
The total annual passengers for each U.S. airport shown in the Explanatory Variables Off-peak highway travel time
summary table for each technical summary are obtained from FAA Private vehicle cost (operating, parking)
statistics for primary and nonprimary commercial service airports for Walk time (transit)
the calendar year 2005. These statistics typically differ slightly from Wait time (transit)
those published by each airport on their websites. The percentage of In-vehicle time (transit)
passengers at each airport for whom the airport is their origin or des- Transit fare
tination was estimated from data reported to the U.S. Department of Taxi fare
Transportation by U.S. airlines in the Airline Origin and Destination
(O&D) Survey and published by the Bureau of Transportation Statis-
Description
tics (BTS) on their TranStats website (http://www.transtats.bts.gov).
Because the origin-destination (O-D) data on the BTS website is
The Atlanta Regional Commission (ARC) Airport Passenger Model
restricted to U.S. domestic itineraries, the calculated O&D percent-
(APM) comprises a component of the overall regional travel demand
ages were adjusted to account for passengers connecting between
model that models air passenger trips to and from Hartsfield–Jackson
domestic and international flights. The sources for passenger statistics
Atlanta International Airport (H-JAIA) (Model Documentation . . .
for non-U.S. airports are noted on the summary tables. 2005). The model was originally developed in 2003 using data from
the H-JAIA Peak Week Air Passenger Survey performed in 2000 and
was updated in 2006 for new income groups for 2000 (Travel Demand
LIST OF TECHNICAL SUMMARIES . . . 2006). The reestimation of the mode choice model only changed
the alternative-specific constants (ASCs) for three of the modes and
D1 Atlanta Regional Commission Model did not affect the coefficients for the continuous variables, as ex-
D2 Boston Logan International Airport Model plained here.
D3 Chicago Airport Express Ridership Forecasting Study
D4 Miami Intermodal Center Travel Demand Forecast Study The ARC/APM consists of two components: a trip genera-
D5 Oakland International Airport BART Connector Study tion/distribution model that assigns the total originating air passen-
D6 Portland International Airport Alternative Mode Study ger traffic at H-JAIA to regional transportation analysis zones
D7 San José International Airport Model (TAZs) and a mode choice model that predicts the ground access
D8 Toronto Air Rail Link Revenue and Ridership Study mode use of those air passenger trips. A subsequent step converts
D9 United Kingdom SERAS Study Air Passenger Surface Access air passenger trips to vehicle trips for inclusion on the traffic as-
Model signment step of the regional transportation demand model. The trip
generation model (and subsequent prediction of air passenger trips
by mode and resulting vehicle trips) is based on the airport passen-
D1 ATLANTA REGIONAL COMMISSION MODEL ger traffic for an average day of the year, which is assumed to be the
same as the average weekday, and is discussed in more detail here.
Summary
The ARC/APM predicts air passenger and vehicle trips by four
Airport Hartsfield–Jackson Atlanta International market segments that distinguish between residents of the Atlanta
Airport (H-JAIA) region and visitors to the region (non-residents) and between busi-
Model Developer Atlanta Regional Commission ness and non-business (personal) trips, giving the following market
Date Developed 2003 (updated 2006) segments: resident business, resident non-business, non-resident
Market Addressed Air passengers business, and non-resident non-business.
Model Type Revealed preference data
Model Structure Nested logit The mode choice model considers five modes: air passengers
Survey Data Used 2000 H-JAIA Peak Week Survey dropped off by private vehicle, private vehicle parked at the airport
Airport Profile Total annual passengers (2005): 84.8 for the duration of the air trip (termed drive self), rental car, tran-
million sit, and taxi. Although not explicitly stated in the model documen-
Percentage O&D: 35% tation, it appears from the existing mode use data given in the
Ground access mode split (2005 H-JAIA documentation that the transit mode includes the Metropolitan
Peak Week Survey): Atlanta Rapid Transit Authority (MARTA) rail and bus services,
Private vehicle—drop off 25% commercial shared-ride shuttle van services, and other high-
Private vehicle—parked 31% occupancy shared-ride modes such as charter bus. The taxi mode
Rental car 16% appears to include exclusive ride limousine services and hotel and

106
motel courtesy vehicles, as well as conventional taxi use. This of the combined access, egress, and sidewalk times from the transit
grouping of modes is likely to have a significant effect on the esti- network, whereas the transit wait times consist of the initial wait
mated model coefficients, because it implies that the modes plus any transfer wait times from the network.
grouped together for the purpose of model estimation have similar
service characteristics. However, in reality this is clearly far from Private vehicle operating costs are assumed at 8.74 cents per
the case. Shared-ride shuttle van services provide a door-to-door mile, based on the off-peak highway distance from the regional net-
service for a significantly higher fare than regular transit, whereas work. Parking costs for vehicles parked for the duration of the air
hotel/motel courtesy vehicles provide a free service by definition, trip are based on half the daily long-term parking cost at H-JAIA
although their use is typically restricted to guests at the hotel or multiplied by the average trip duration in days, assumed as 4 days
motel providing the service. for business trips and 7 days for non-business trips. Thus, all air par-
ties with a given trip purpose are assumed to incur the same parking
The basic form of the mode choice model is a nested logit (NL) cost if they choose to park at the airport during their trip, irrespec-
model with a separate structure for trips by residents of the region tive of their actual air trip duration. Transit fares are obtained from
from that for non-residents, as shown in Figure D1. The resident the transit network (the fares are unaffected by the time of day).
model divides the modes into private vehicle trips and public modes Taxi fares were estimated from the off-peak highway distance as-
(termed non-auto modes). The private vehicle nest distinguishes suming a flag drop of $1.75 and a rate of $1.75 per mile.
between drop-off trips and those where the vehicle was parked for
the duration of the air trip. The public mode nest distinguishes There are no explanatory variables in the utility function for the
between use of transit and taxi. The non-resident model contains rental car mode, only an ASC. Thus, the model implicitly assumes
three modes at the top level: drop off by private vehicle, rental car, that the rental car alternative is perceived as providing the same
and public modes. As with the resident model, the public mode nest utility to all non-resident air parties, irrespective of their air party
distinguishes between use of transit and taxi. characteristics.
The model does not take the type of ground access trip origin Apart from estimating different model coefficients for the four
into account. Thus, all visitors to the region are considered to market segments and any differences in travel time and cost owing
include drop-off by private vehicle and rental car in their choice set, to the different trip origin locations, the model does not consider any
whether or not they are staying with residents of the region and thus air party characteristics, such as household income or air party size.
have someone who could drop them off at the airport or the need for Based on the model documentation, the treatment of air party size
a rental car during their visit. appears to be inconsistent, with costs for private vehicle and taxi
modes being calculated on an air party basis, but transit fares being
calculated on an air passenger basis.
Explanatory Variables
The explanatory variables consist of the travel times and costs for Model Coefficients
each mode. The travel times for the private vehicle and taxi modes
use the off-peak travel time from the highway network in the re- The adopted and estimated model coefficients for each of the mar-
gional travel demand model. Transit travel times are obtained from ket segments are shown in Table D1, and the corresponding implied
the a.m. peak transit network in the regional travel demand model. values are shown in Table D2.
Separate variables are defined for in-vehicle, walk, and wait times.
The transit in-vehicle times use the total in-vehicle time for the trip The coefficients for the continuous variables were not esti-
from the origin zone to the airport zone. Transit walk times consist mated from the data but rather adopted from other models. The
Residents
Private Non Auto
Drive Self Dropped Off Transit Taxi
Non-residents
Dropped Off Rental Car Non Auto
Transit Taxi
FIGURE D1 ARC/APM model choice model nesting structure.

[Source: Model Documentation: Mobility 2030 Regional Transportation Plan,
Atlanta Regional Commission (2005).]

107
TABLE D1
ATLANTA AIRPORT PASSENGER MODEL COEFFICIENTS
Resident Resident Non-Resident Non-Resident
Coefficient Business Non-Business Business Non-Business
Variables
Highway time (minutes) 0.071 –0.044 –0.068 –0.039
Transit in-vehicle time (min) –0.053 –0.031 –0.050 –0.029
Walk time (minutes) –0.093 –0.051 –0.089 –0.045
Wait time (minutes) –0.107 –0.077 –0.096 –0.071
Cost (cents) –0.00277 –0.002105 –0.00256 –0.001969
Constants
Private vehicle parked for trip 5.427 4.517 N/A N/A
Rental car N/A N/A –4.061 –3.153
Transita –0.738 0.159 0.720 –1.853
Taxia 8.366 2.876 7.753 3.408
Nest Coefficients
Private auto nest 0.3 0.3 N/A N/A
Public mode nest 0.3 0.3 0.3 0.3
Note: N/A = not applicable.

a
Includes nest constant term.
ASCs (termed model bias coefficients in the documentation) dents of the region on business trips is slightly lower than for non-
were then estimated to ensure that the model predicted the ob- residents on business trips, which is not unreasonable given the gen-
served mode shares. Therefore, the model bias coefficients not eral income levels in the Atlanta region compared with many other
only account for intrinsic attributes of the different modes not ex- parts of the United States. However, because these coefficients were
plained by the continuous variables but also any differences that apparently not estimated from the Atlanta data, this appears coinci-
would have existed between the continuous variable coefficients dental. The implied values of transit in-vehicle time are lower than
that were adopted for the model and the values that would have for highway travel time, as expected because those with a higher
been obtained had these coefficients been estimated from the value of time will tend to use other modes than transit. The implied
data. values of walk and wait time involved in transit trips are signifi-
cantly higher than for in-vehicle time, as is generally found in other
The adopted coefficients for the continuous variables all have mode choice models.
the expected signs. The implied value of highway travel time varies
between $12/h and $16/h, which is significantly less than values of The constant term for private vehicle parked for the duration
time typically found for air travelers in air travel demand models. of the air trip suggests that this alternative has a lower disutility
Although the implied values of time for non-business trips are lower than being dropped off by private vehicle after allowing for the
than those for business trips, as is normally expected, the differ- travel time and cost involved in the two alternatives, equivalent
ences are surprisingly small. The implied values of time for resi- to about $20 per air party. Because the utility functions for the
TABLE D2
IMPLIED VALUES OF ATLANTA AIRPORT PASSENGER MODEL COEFFICIENTS

Parameter Business Non-Business Business Non-Business
Travel Time ($/hour)
Highway time 15 13 16 12
Transit in-vehicle time 11 9 12 9
Walk time 20 15 21 14
Wait time 23 22 23 22
Constants ($)
Private vehicle parked for trip 20 21 N/A N/A

Rental car N/A N/A –16 –16
Transit –3 1 3 –9
Taxi 30 14 30 17

108
drop-off alternative do not consider the operating cost of the and those that originated from some other type of trip origin (pri-
vehicle for the return trip or assign any disutility to the time of marily hotels or motels), whereas relatively few non-resident busi-
the driver, this does not seem unreasonable, although in reality ness trips originated from a private residence. Therefore, the four
these effects are not likely to be constant across all air parties. market segments were further divided into six trip types based on the
Because the utility functions for the rental car alternative do not type of origin, as follows:
include any continuous variables (travel time or cost), the esti-
mated values of the constant terms ensure that the model predicts • Resident business trips from private residences
the observed use of rental car in the estimation dataset and have • Resident business trips from other types of trip origin
no intrinsic interpretation. • Resident non-business trips
• Non-resident business trips
• Non-resident non-business trips from private residences
Model Fit • Non-resident non-business trips from other types of trip origin.
The documentation on the model provided no information on the The division into these six categories was made on the basis of
overall fit of the model. the proportions in the 2000 H-JAIA air passenger survey. These
trips were then allocated to TAZs on the basis of either households
(for trips from a private residence) or total employment in a zone
Model Application (for trips from other origin types). The household allocation equa-
tions divided households into four income categories on the basis of
The ARC/APM has been designed to be used as an integral part of the 2000 census. The coefficients of the allocation equations were
the regional travel demand modeling process and generates trip estimated using linear regression from the results of the 2000
tables for air passenger trips and the associated vehicle trips that are H-JAIA air passenger survey. The form of each equation estimates
subsequently combined with other types of trips in the regional the number of trips in the survey data in each category from a given
travel demand model (Travel Demand . . . 2006). However, the re- zone in terms of the number of households in each income category
sults of the APM are saved as separate tables and can be presented in a zone or the total employment in the zone. These estimates must
or exported independently of the regional travel demand model re- then be converted to a percentage of all trips in that category from
sults for use in airport planning or related studies. the zone in question to allocate any other estimate of total trips by
category.
The ARC/APM is programmed in Fortran and can be called as
part of running the regional travel demand modeling software. This However, because the survey only gave the trip origin informa-
was originally programmed using Tranplan software but has re- tion by zip code, rather than TAZ, the allocation equations were
cently been converted to TP+/Cube Voyager software. The model estimated on a zip code basis. When these equations were applied to
can read regional travel demand model skim tree tables of highway TAZs, it was found that the trips from some parts of the region were
and transit travel times, highway distances, and transit costs directly over-estimated while those from other parts of the region were
for use in modeling air passenger trips. under-estimated. Therefore the region was divided into three differ-
ent groups of zones and different adjustment factors were applied to
each group of zones. Details of the allocation equations and adjust-
Trip Generation/Distribution Model ment factors are provided in the model documentation (Model
Documentation . . . 2005), but are not presented here because they
The first steps in the ARC/APM generate a sample of air passenger do directly affect the mode choice model.
trips with their associated air party characteristics, including the trip
origin TAZ, for use in modeling the air passenger ground access
mode choice. Although strictly not part of the access mode choice
model, these steps are sufficiently important to generating the over-
Integration with Regional Planning Process
all pattern of air passenger and associated vehicle trips that they
deserve a fairly detailed explanation. The mode choice model within the APM generates a table of air pas-
senger trips using each of the five defined modes from each TAZ.
The first step in the trip generation/distribution process is to obtain These tables of air passenger trips are then converted to associated
an estimate of the total air passenger enplanements at H-JAIA for the vehicle trips using assumed values of average vehicle occupancy
year in question. This is then adjusted to exclude connecting passen- for each mode. The number of vehicle trips for the private vehicle
gers and divided into the four market segments. In the case of histor- drop-off mode is doubled to allow for the return trip by the driver
ical data, the percentage of originating air passengers in each market dropping off the air party and the total number of vehicle trips dou-
segment can be obtained from air passenger survey information and bled again to allow for the traffic generated by egress trips, which
the proportion of connecting passengers can be obtained from airline are assumed to be symmetrical to the pattern of access trips.
data reported to the U.S.DOT. For the purposes of developing the
APM, the ARC used an estimate of connecting passengers from the The resulting tables of air passenger trips using transit and high-
H-JAIA Master Plan. When applying the APM to future years, it will way vehicle trips are then added to the trip tables for other types of
be necessary to adjust the proportion of connecting passengers and regional trips generated by the other components of the regional
the percentage of originating passengers in each market segment to travel demand model before the highway traffic assignment and
reflect any forecast changes in the traffic composition at H-JAIA. transit trip assignment steps of the overall modeling process.
The second step in the trip generation/distribution process allo- Although this approach provides a reasonable representation of
cates these trips to TAZs. Analysis of the 2000 H-JAIA air passen- vehicle trips by private vehicles and rental cars, the combination of
ger survey showed that there were significant differences between several distinct modes in the transit and taxi modes of the APM is
resident business trips that started from a residence and those that more problematical. Trips on public modes other than MARTA do
started from another type of origin, typically the place of work. not contribute to MARTA ridership, as implied by including these in
Almost all resident non-business trips originated from a private res- the transit assignment step, but rather generate additional highway
idence. Similarly, there was a significant difference between non- vehicle trips. Although the various modes included in the APM
resident non-business trips that originated from a private residence taxi mode each generate highway vehicle trips, the average vehicle

109
occupancy for hotel/motel courtesy vehicles is likely to be greater formed at Boston Logan International Airport. Separate sub-models
than that for conventional taxi and limousine, owing to the shared- were developed for resident business trips, resident non-business
ride nature of hotel/motel courtesy vehicles, resulting in an overesti- trips, non-resident business trips, and non-resident non-business
mate of vehicle trips for the taxi mode. However, the geographic trips. The two resident sub-models consist of a two-level NL model,
pattern of this error is likely to be very uneven, because most with separate second-level nests for door-to-door modes (taxi and
hotel/motel courtesy vehicle trips are from areas fairly close to the limousine) and automobile modes (drop-off, short-term parking,
airport, whereas taxi and limousine use is more widely distributed long-term parking, and off-airport parking). There are four shared-
throughout the region. ride public modes at the top level (regular transit, scheduled airport
bus, the Logan Express service to off-airport terminals in the region,
and the Water Shuttle between the airport and the downtown Boston
Documentation waterfront). The visitor sub-models are multinomial logit (MNL)
models and omit the long-term parking alternatives, but add a hotel
Model Documentation: Mobility 2030 Regional Transportation Plan, Atlanta shuttle mode.
Regional Commission, Atlanta, Ga., Feb. 11, 2005.
Travel Demand Model Documentation, Atlanta Regional Commission, This model includes a rail access mode, the Massachusetts Bay
Atlanta, Ga., May 2006. Transportation Authority (MBTA) regional rail transit system, and
off-airport terminals, the Logan Express service operated by the
Massachusetts Port Authority (Massport), the airport authority for
D2 BOSTON LOGAN INTERNATIONAL AIRPORT Logan Airport. The MBTA Airport Station is adjacent to the airport
MODEL and linked to the passenger terminals by a free shuttle bus service
operated by Massport. Unlike many other airport access mode
Summary choice models, the CTPS model treats rental car use as an indepen-
dent decision and excludes it from the mode choice decision
Airport Boston Logan International Airport process.
Model Developer Central Transportation Planning Staff,
Boston
Date Developed 1996
Market Addressed Air passengers Explanatory Variables
Model Structure Nested logit (resident), multinomial logit Independent variables include both in-vehicle and out-of-vehicle
(visitor) travel time, automobile access time to the public modes, the num-
Survey Data Used 1993 Boston Logan Air Passenger Survey ber of transfers, travel costs, and dummy variables for the type of
Airport Profile Total annual passengers (2005): 26.4 trip origin (residence or not), the amount of luggage, air party size,
million number of air trips in past year, and whether an employer was pay-
Percentage O&D: 90% ing travel expenses. Not all variables are included in all models, and
Ground access mode split (2003 Logan various combinations of the independent variables were estimated.
Air Passenger Survey): For some model variations, separate travel cost coefficients were
Private vehicle—drop-off 21% estimated for low-income and high-income travelers or for those for
Private vehicle—parked 11% whom their travel costs were paid by their employer. However, the
Rental car 17% definition of low-income and high-income travelers was not
Taxi 19% included in the model documentation. Travel times were measured
Limousine 7% in minutes and costs in dollars, based on 1993 rates.
Logan Express 5%
Scheduled bus/limo 4%
Public transit—MBTA subway 6% Model Coefficients
Water shuttle 1%
Hotel courtesy shuttle 6% Tables D3 to D6 show the estimated model coefficients for the
Charter bus 3% four market segment models. Values in parentheses are the t-
Other (including MBTA bus) <1% statistics of the estimates. With a few exceptions, most of the esti-
Market Segmentation Residents—Business trips mated coefficients are statistically significant at the 95% level or
Residents—Non-business trips better. The t-statistics for the ASCs for the non-resident non-business
Non-residents—Business trips model (Table D6) are as reported in the model documentation, but
Non-residents—Non-business trips appear to be incorrect. They are identical to those shown for the
Explanatory Variables In-vehicle time non-resident business model (Table D5), which would be surpris-
Out-of-vehicle time (walk, wait, transfer) ing, and three have incorrect signs (t-statistics are generally re-
Automobile access time ported with the same sign as the coefficient), suggesting that the
Travel cost (parking, tolls, automobile wrong values were reported in the model documentation.
operating cost, fares)
Dummy variables (employer pays cost, As can be seen from Tables D3 to D6, separate travel time and
luggage, air party size, non-residence trip cost coefficients were estimated from groups of modes. This has the
origin, household income, flights/year effect of giving different implied values of travel time for different
from Logan) modes, as shown in Table D7. Whereas it can be expected that trav-
elers choosing different modes will on average tend to have differ-
ent values of time (e.g., travelers choosing a taxi will tend to have a
Description higher value of time than those using the MBTA) that is an entirely
different issue from assuming that a given traveler will have a
This model was developed by the Central Transportation Planning different implied value of travel time when considering alternative
Staff (CTPS) in Boston using a 1993 air passenger survey per- modes (as implied by the models).

110
TABLE D3
BOSTON LOGAN RESIDENT BUSINESS MODEL COEFFICIENTS
Travel Time Coefficients Travel Cost Coefficients Dummy Variable Coefficients
Self- Self-
Pay Pay Non-
Tree Auto Low High Empl. Resident Empl. Luggage >6 Flights
Mode Const. Coeff. IVTT OVTT Access Income Income Pays Origin Pays >2 bags in Year
MBTA Rail 1.471 0.034 0.034 0.072 0.080 0.080 0.080 1.175
( 1.7) ( 4.9) ( 5.7) ( 0.8) ( 2.2)
Scheduled 0.437 0.034 0.034 0.072 0.080 0.080 0.080
Bus/Limo (0.8)
Logan 0.126 0.034 0.034 0.072 0.080 0.080 0.080
Express (0.4)
Water 2.851 0.034 0.034 0.072 0.080 0.080 0.080
Shuttle ( 2.6)
Door-to- 0.361 –0.503 1.337
Door Nest (2.9) ( 2.5) (4.3)
Taxi 1.279 0.173 0.173 0.295 0.101 0.101
( 3.4) ( 2.0) ( 2.2) ( 7.5)
Limousine 0.173 0.173 0.295 0.101 0.101
Automobile 0.290 0.72
Nest ( 0.9) (5.6)
Long-term 0.897 0.036 0.171 0.370 0.193 0.102 0.850
park on (2.4) ( 2.2) ( 2.9) ( 3.4) ( 6.1) ( 6.1) (3.7)
airport
Long-term 0.527 0.036 0.171 0.370 0.193 0.102 0.850
park off (0.8)
airport
Short- 1.491 0.070 0.171 0.370 0.193 0.102 0.794
term park ( 4.0) ( 3.8) ( 2.6)
at airport
Drop off 0.070 0.171 0.370 0.193 0.102 0.794
Note: t-statistics shown in parentheses (omitted for repeated values). IVTT = in-vehicle travel time; OVTT = out-of-vehicle travel time.
TABLE D4
BOSTON LOGAN RESIDENT NON-BUSINESS MODEL COEFFICIENTS
Travel Time Coefficients Travel Cost Coefficients Dummy Variable Coefficients
Self- Self-
Pay Pay Non- >2 Party
Tree Auto No. of Low High Empl. Resident Luggage Flights Size
Mode Const. Coeff. IVTT OVTT Access Transfers Income Income Pays Origin >2 Bags in Year >1
MBTA Rail 0.926 –0.027 –0.027 –0.092 –0.150 –0.232 –0.232 –0.232 –1.805
(2.9) (–4.7) (–8.1) (–0.9) (–2.9) (–5.2)
Scheduled Bus/Limo 3.799 –0.027 –0.027 –0.092 –0.150 –0.232 –0.232 –0.232
(4.4)
Logan Express 2.781 –0.027 –0.027 –0.092 –0.150 –0.232 –0.232 –0.232
(5.1)
Water Shuttle –0.213 –0.027 –0.027 –0.092 –0.150 –0.232 –0.232 –0.232
(–0.0)
Door-to-Door Nest –0.401 0.470
(–0.4) (3.2)
Taxi –0.957 –0.057 –0.057 –0.093 –0.073 –0.073 1.118
(0.3) (–1.7) (–4.6) (–4.1) (2.2)
Limousine –0.057 –0.057 –0.093 –0.073 –0.073
2.452
(4.3)
Automobile Nest 0.631
(4.7)
Long-term park 0.115 –0.036 –0.066 –0.259 –0.118 –0.118 1.139
on airport (1.4) (–1.8) (–1.0) (–6.5) (–5.4) (4.0)
Long-term park –0.075 –0.036 –0.066 –0.259 –0.118 –0.118 1.139
off airport (0.1)
Short-term park –0.074 –0.066 –0.259 –0.118 –0.118 –1.153
at airport (–3.6) (–3.5)
Drop off 0.604 –0.074 –0.066 –0.259 –0.118 –0.118 –1.153
1.109
(3.4)
(4.1)

111
TABLE D5
BOSTON LOGAN NON-RESIDENT BUSINESS MODEL COEFFICIENTS
Travel Time Coefficients Travel Cost Coefficients Dummy Coefficients
Self- Pay Self-Pay Non-
Auto No. of Low High Empl. Resident Luggage >2 Party
Mode Const. IVTT OVTT Access Transfers Income Income Pays Origin Bags Size >1
MBTA Rail 1.855 0.022 0.022 –0.039 –0.286 –0.091 –0.091 0.058 –0.508
(–3.7) (–4.2) (–4.3) (–1.8) (–7.9) (–6.9) (–1.9)
Scheduled 1.564 0.022 0.022 –0.039 –0.286 –0.091 –0.091 0.058
Bus/Limo (–3.8)
Logan 2.856 0.022 0.022 –0.039 –0.286 –0.091 –0.091 0.058
Express (–4.7)
Water 1.620 0.022 0.022 –0.039 –0.286 –0.091 –0.091 0.058
Shuttle (–4.8)
Taxi 0.039 0.039 0.091 –0.091 0.058
(–4.3)
Limousine 0.275 0.039 0.039 –0.091 –0.091 0.058
(–1.4)
Hotel 2.187 0.039 0.039
Shuttle (–11.4)
Short-Term 1.586 0.039 0.152 –0.058 –0.058 0.058 –2.105
Park at (–2.1) (–2.4) (–6.9) (–9.6)
Airport
Drop Off 0.376 0.039 0.152 –0.058 –0.058 0.058 –2.105 0.377
(–1.2) (1.6)
TABLE D6
BOSTON LOGAN NON-RESIDENT NON-BUSINESS MODEL COEFFICIENTS
Travel Time Coefficients Travel Cost Coefficients Dummy Coefficients

Self- Self-
Pay Pay Non-
Auto No. of Low- High- Empl. Resident Luggage Party
Mode Const. IVTT OVTT Access Transfers Income Income Pays Origin >2 Bags Size >1
MBTA Rail –1.066 –0.013 –0.013 –0.013 –0.213 –0.091 –0.091 –0.058 –0.508
(–3.7) (–2.5) (–2.5) (–2.2) (–1.2) (–7.9) (–6.9) (–1.9)
Scheduled Bus/Limo 0.155 –0.013 –0.013 –0.013 –0.213 –0.091 –0.091 –0.058
(–3.8)
Logan Express –2.020 –0.013 –0.013 –0.013 –0.213 –0.091 –0.091 –0.058
(–4.7)
Water Shuttle –2.352 –0.013 –0.013 –0.013 –0.213 –0.091 –0.091 –0.058
(–4.8)
Taxi –0.013 –0.013 –0.091 –0.091 –0.058
(–2.2) (–2.2)
Limousine 0.812 –0.013 –0.013 –0.091 –0.091 –0.058
(–1.4)
Hotel Shuttle –0.021 –0.013 –0.013
(–11.4)
Short-Term Park –0.229 –0.013 –0.152 –0.058 –0.058 –0.058 –2.105
at Airport (–2.1) (–2.4) (–6.9) (–9.6)
Drop Off 0.376 –0.013 –0.152 –0.058 –0.058 –0.058 –2.105 0.377
(–1.2) (1.6)

112
TABLE D7
IMPLIED VALUES OF BOSTON LOGAN MODEL COEFFICIENTS
Resident Resident Non- Non-Resident Non-Resident
Parameter Business Business Business Non-Business
In-vehicle
Shared-ride modesa
Self-pay/employer pays 26 7 15/23 9/13
Taxi/limousine
Low-income 35 37 26 9
High-income/employer pays 103 47 26/40 9/13
Auto park
Low-income 6 8 N/A N/A
High-income/employer pays 11/21 18 N/A N/A
Auto drop or park short-term
Low-income 11 17 40 13
High-income/employer pays 22/41 38 40 13
Auto access (shared-ride modes)
Self-pay/employer pays 54 24 26/40 9/13
Constants (minutes of IVT)b
MBTA 43 –34 84 82
Scheduled bus/limo –13 –141 71 –12
Logan Express 4 –103 130 155
Water Shuttle 84 8 74 181
Taxi 7 24 –– ––
Limousine –– 7 7 –62
Hotel shuttle N/A N/A 56 2
Automobile
Park long-term on airport –17 –3 N/A N/A
Park long-term off airport –7 2 N/A N/A
Park short-term at airport 25 –– 41 –29
Drop off 4 –8 –10 18
a
MBTA, scheduled bus/limo, Logan Express, Water Shuttle.
b
Equivalent minutes of in-vehicle time.
N/A = mode is not available for this market segment; –– = no alternative-specific constant estimated for this
mode; IVT = in-vehicle travel.
It makes no sense that given travelers will value their time at one no consistent relationship across the different market segments.
amount when considering a high-priced mode and a different amount For some market segments a given mode is significantly more at-
when considering a less expensive, but more time-consuming mode. tractive than another mode, whereas for other market segments the
Because the CTPS modelers were able to obtain a statistically sig- reverse is true. It is quite likely that these values are so distorted
nificant difference in the model coefficients for different modes sug- by the model specification problems that they have no intrinsic
gests that this is a result of specification problems with the models or interpretation.
problems with the model estimation data. In particular, the omission
of any air party size information in the utility functions for most
modes would ignore the distinction between costs that are incurred Model Fit
on a per person basis from those costs that are incurred once per air
party. Similarly, the use of the same travel cost coefficient for all The documentation on the model provided no information on the
air parties irrespective of income is likely to lead to differences in the overall fit of the model.
estimated coefficients for modes with widely different costs.
Given these problems with the data and the conceptual difficulty
with having different implied values of time for different modes, Model Application
there is no reason to expect any particular relationship between the
implied values of time for different market segments or different in- The model has been used as part of the regional transportation plan-
come levels. However, the implied values of time for higher income ning process for the Boston metropolitan area to forecast vehicle
travelers or those for whom their employer is paying their travel trips to and from Boston Logan International Airport, as well as for
costs are generally higher than those for lower income travelers, as specific planning studies by the Massachusetts Port Authority (the
could be expected. Similarly, for non-resident travelers the implied operator of Logan Airport), including an evaluation of a people-
values of time for business travelers are higher than the corre- mover link between the MBTA Airport Station and an assessment
sponding values of time for non-business travelers. Although this is of strategies to address an anticipated future parking shortfall in the
also true for some modes for resident travelers, business travelers on-airport parking facilities.
have a lower implied value of time than non-business travelers for
automobile users paying their own travel expenses.
Integration with Regional Planning Process
The implied value of the ASCs, expressed as equivalent min-
utes of in-vehicle time where a positive value indicates that the The direct output of the model consists of the number of air passen-
mode has a relative perceived disutility that would be offset by re- gers in 1993 air passenger survey using each mode by TAZ. These
ducing the travel time by that amount, show no obvious pattern and air passenger flows are then expanded to correspond to the desired

113
period of analysis (e.g., average weekday or annual demand) using a Limousine 14% 10%
survey expansion factor that relates the number of air passengers in Hotel/airport van 9% 4%
the survey sample to the O&D passenger traffic at the airport for the CTA train 4% 6%
period of analysis (accounting for forecast traffic growth for future Other 5% 4%
years where necessary). Air passenger flows are then converted into Market Segmentation Residents—Business trips
vehicle trips for exclusive ride modes using the number of people in Residents—Non-business trips
each travel party reported in each survey record. Non-residents—Business trips
Non-residents—Non-business trips
The resulting passenger and vehicle trip tables are incorporated Explanatory Variables Total travel time (weighted access, trans-
in the CTPS regional highway and transit assignment models. The fer, and waiting time)
following five trip tables are created for each of the four weekday Travel cost
time periods (a.m. peak, midday, p.m. peak, and night time): Availability of baggage check-in at down-
town terminal
• MBTA passenger trips Use of intermediate station on Airport
• Scheduled bus and limousine passenger trips Express rail link
• Logan Express passenger trips
• Water Shuttle passenger trips Description
• Highway vehicle trips.
The Chicago Transit Authority (CTA) in partnership with the city
The highway vehicle trips include automobile access and egress of Chicago has been pursuing the feasibility of establishing an
trips to and from the MBTA origin or destination stations, the express train service between downtown Chicago and O’Hare
scheduled bus and limousine stops, and the Logan Express termi- International and Midway Airports (PB Consult 2006). The pro-
nals for those air passengers using private vehicles to access those posed service would utilize the tracks of the existing Blue Line of
services, as well as taxi, exclusive-use limousine, and rental car the CTA rapid transit system between downtown Chicago and
trips. The calculation of vehicle trips includes two-way trips for pas- O’Hare Airport and the existing CTA Orange Line tracks between
sengers dropped off or picked up at the airport, MBTA stations, bus downtown Chicago and Midway Airport, although some additional
stops, or Logan Express terminals. passing sections could be provided to reduce travel times. In 2003,
the Chicago Department of Transportation (CDOT) retained
Resource Systems Group, Inc. (RSG) and Wilbur Smith Associates
Documentation (WSA) to undertake a ridership and revenue forecasting study of the
proposed Airport Express train service (WSA 2004). The study
Harrington, I.E., J. McClennen, E. Pereira and C.-Y. Wang, Summary of examined the potential use of the Airport Express by air travelers
People Mover Study Passenger Mode Choice Models, draft memoran- with ground origins in a study area that extended approximately
dum, Central Transportation Planning Staff, Boston, Mass., May 17, 9 miles north and south of the planned Airport Express terminal in
1996. downtown Chicago, as shown in Figure D2.
Harrington, I.E., The Logan Airport Passenger Ground Access Mode Choice
Model, draft memorandum, Central Transportation Planning Staff,
Boston, Mass., Feb. 28, 2003.
To identify the pattern of air passenger trip ends, travel party
characteristics, and existing access mode use, and to understand how
airport travelers might change their access mode choice if the Airport
Express train was available, two surveys of air travelers were under-
D3 CHICAGO AIRPORT EXPRESS RIDERSHIP taken by Resource Systems Group, Inc., at O’Hare and Midway
FORECASTING STUDY Airports in September 2003, an O-D survey and a stated preference
survey (RSG 2004). Both surveys were undertaken in the secure area
Summary of the airport terminals after passengers had cleared security screen-
ing. The O-D survey intercepted 6,789 air travelers, of whom 3,348
Airport Chicago O’Hare International Airport; were originating passengers and completed the survey. The O-D sur-
Chicago Midway Airport vey results were used to develop a profile of existing originating air
Model Developer Resource Systems Group, Inc. passenger characteristics, including trip purpose, trip origin, and
Date Developed 2004 ground access mode use, as well as current and future trip tables that
Market Addressed Air passengers predicted the number of trips by market segment originating in each
Model Type Combined revealed preference and stated of 145 TAZs within the study area.
preference data
Model Structure Nested logit The stated preference survey interviewed 1,110 air travelers in
Survey Data Used 2003 Chicago Air Traveler Stated Prefer- the two airports that a screening question had identified as having a
ence Survey ground access trip origin in the study area. Respondents were asked
Airport Profile O’Hare Midway about details of their trip to the airport, including their trip origin,
Total annual mode used, and travel times and costs, as well as their trip purpose
passengers (2005): 73.4 16.8 and whether they were residents of the Chicago region. They were
million million then asked to complete eight stated preference choice experiments
Percentage O&D: 45% 73% in which they were presented with a choice between three modes:
Ground access mode split the mode they had just used, the Airport Express train, and a third
(2003 O-D survey): mode. The characteristics of the Airport Express and third mode
Private were varied in the experiments and the respondents were asked
vehicle—drop off 22% 27% which option they would have chosen for their current trip had these
Private been available. The experiments varied the travel time on the main
vehicle—parked 15% 22% mode and service headway, the access mode used, the access and
Rental car 12% 13% egress time, travel cost, and availability of baggage check-in at the
Taxi 18% 15% Airport Express terminal.

114
nests, one for private modes and one for public modes, where taxi
and rental car were considered private modes. The nest structure
O'Hare differed between business and non-business trips in the treatment
of the Airport Express train. In the business trip model structure,
the Airport Express was treated as a separate alternative at the
O'
Ha
re
same level as the private and public mode nests, whereas in the non-
Bl business trip model it was considered one of the public modes in the
ue
Lin
e public mode nest.
In both cases the model defined a lower level nest of five access
sub-modes below the Airport Express mode. These included a free
van service in addition to walk, taxi, drive and park, and CTA tran-
sit. The free van alternative was assumed to only be available in
Block 37 certain downtown zones, principally those with a concentration of
hotels along Michigan Avenue, whereas the walk alternative was
restricted to zones within about a one-half mile of the downtown
terminal. The drive and park access alternative was excluded from
the available access modes at the downtown terminal, but was
ine
Airport Express g eL assumed to be available at intermediate stations where these were
an
Study Area Or included in the Airport Express service scenarios.
ay
dw
Mi
Explanatory Variables
Midway
The utility functions for each airport access mode included two con-
tinuous variables in addition to ASCs: total travel time and travel
cost. The estimated coefficients for each of these two variables
Figure 1: Proposed within a given market segment were constrained to have the same
Airport Express 0 2 4 6 value for each mode. Traveler income was not explicitly included in
Miles
the model but travelers were divided into two income categories on
the basis of household income and separate travel cost coefficients
estimated for each category, where low-income travelers were de-
FIGURE D2 Proposed Airport Express routes and fined as those with a household income under $100,000 and high-
study area. (Source: WSA 2004.) income travelers were defined as those with a household income of
$100,000 or more.
The results of the stated preference survey were used to esti- Although a single travel time variable was used in each utility
mate a mode choice model that defined nine airport access modes, function, weights were applied to various components of the total
as follows: travel time that was used in the model estimation to account for dif-
ferent disutility of access, transfer, and waiting time (as applicable)
• Private vehicle parked at the airport for the duration of the air for each mode. Egress time at the airport was considered to have the
trip (drive and park) same disutility per minute as in-vehicle time on the primary mode.
• Dropped off at airport by private vehicle (dropped off) The other travel time components were assumed to have the same
• Rental car weight with different values of this weight for business and non-
• Taxi business trips. These values were determined by iteratively adjust-
• Other private mode ing the weights in steps of 0.25 to find the value that gave the
• Airport Express train best overall model estimation result. This turned out to be 1.25 for
• Airport Bus business trips and 1.50 for non-business trips. It is surprising that
• CTA train travelers on business trips appear to be slightly less sensitive to
• Other public mode. access and waiting time compared with in-vehicle time than travel-
ers on non-business trips.
The “Other private mode” included limousine, hotel/motel cour-
tesy shuttle, and shared-ride airport van service, whereas the “Other In addition to the two continuous variables, a dummy variable
public mode” included PACE bus, METRA train service, regional for the availability of baggage check-in at the downtown terminal
bus and charter bus. This grouping of modes has the effect of com- was included in the Airport Express and Airport Bus modes. A sec-
bining modes with very different service characteristics, particu- ond dummy variable was included in the Airport Express mode for
larly limousine, shared-ride van, and hotel/motel courtesy shuttle. those passengers boarding at an intermediate station.
Similarly, charter bus is not usually a feasible option for most air
travelers, and those using this mode (such as tour groups) typically Estimation of the mode choice model coefficients used the travel
have the mode choice decision made for them, in contrast to users times and costs reported by the survey respondents for the mode
of regional bus and rail services. The Airport Bus alternative is not they had actually used and the values for the alternative modes pro-
currently available, but was included in the stated preference exper- vided in the stated preference experiments.
iments as a proposed new service from downtown Chicago as an
alternative to the Airport Express train.
Model Coefficients
The mode choice model used a NL form with a somewhat dif-
ferent nest structure for travelers on business and non-business trips, Separate model coefficients were estimated for business and non-
as shown in Figure D3. Generally, modes are grouped into two business travelers. Separate model coefficients were not estimated

115
Business Mode Choice Model Structure

Mode Choice
Public Transport Public Transport

Mode Airport Express Train
Mode
Express Train
Access Mode
Drove & Dropped Rental Other Airport CTA Other
Taxi Private Public
Park Off Car Bus Train
Mode Mode
Free Drive & CTA

Walk Taxi
Van Park Service
Non-Business Mode Choice Model Structure

Mode Choice
Public Transport
Mode Public Transport Mode
Drove & Dropped Rental Other Other

Taxi Private Airport Express Train Airport CTA
Park Off Car Bus Train Public
Mode Mode
Express Train
Access Mode
Free Drive & CTA

Walk Taxi
Van Park Service
FIGURE D3 Chicago Airport Express mode choice model nesting structure.

(Source: WSA 2004.)
for residents of the Chicago region and non-residents, although the income and high-income respondents, two implied values are given
available modes for these two market segments were different, as for each coefficient, corresponding to the two income groups.
noted earlier. The estimated model coefficients for travel time and
the dummy variables were constrained to be the same for O’Hare The implied values of travel time appear reasonable and broadly
and Midway airports, whereas separate travel cost coefficients were consistent with values of time for air passenger travel in other mod-
estimated for each airport. The ASCs for taxi and Airport Bus, as els. The values for business trips are higher than for non-business
well as those for the access modes to the Airport Express train, were trips as expected, although the value for low-income respondents on
constrained to be the same for both airports, whereas those for the business trips through O’Hare Airport appears surprisingly low com-
other modes were allowed to vary by airport. pared with the value for high-income respondents at a little over one-
third the value. Given the income category split at a household
After the model coefficients were estimated using the stated income of $100,000 per year and the general distribution of house-
preference data, the model was calibrated to correspond to the mode hold incomes in the population at large, one might expect a ratio of
shares for the study area obtained from the O-D survey data by ad- about 2.0, as observed for non-business trips at both airports. Simi-
justing several of the ASCs until the model predicted the observed larly, the value for low-income respondents on business trips through
mode shares from the study area. Midway Airport appears surprisingly close to that for high-income
respondents. The latter may be partly explained by the location of
The initial estimated model coefficients for each trip purpose and Midway Airport to the south of the downtown and the dominance of
each airport are shown in Table D8 (RSG 2004). The final calibrated low-fare carriers at the airport. Higher income individuals are more
model coefficients for each of the market segments at both airports likely to live in the more affluent suburbs to the north and northwest
are shown in Tables D9 and D10 (WSA 2004). It appears that further of the downtown. These individuals would find O’Hare Airport more
changes were made in all coefficient values between the estimated convenient and may be less concerned about using a low-fare carrier,
values reported in the results of the model estimation (RSG 2004) resulting in a different income distribution for business travelers
and the results of the model calibration (WSA 2004). The implied using O’Hare and Midway Airports. However, if this is the explana-
values corresponding to the calibrated model coefficients are shown tion, it does not appear to apply to non-business trips, where the
in Tables D11 and D12. Because the implied values depend on the implied value of travel time for high-income travelers using O’Hare
value of the travel cost coefficient, which varies between low- is slightly lower than for those using Midway.

116
TABLE D8
CHICAGO AIRPORT EXPRESS MODEL ESTIMATED COEFFICIENTS
OíHare O’Hare Midway Midway
Total travel time (minutes) –0.098 –0.080 –0.098 –0.080
(–4.5) (–5.0) (–4.5) (–5.0)
Travel cost—low income ($) –0.183 –0.212 –0.116 –0.239
(–4.4) (–5.0) (–4.1) (–4.9)
Travel cost—high income ($) –0.069 –0.096 –0.056 –0.117
(–3.8) (–4.0) (–3.5) (–3.8)
Dummy Variables
Downtown baggage check 0.566 1.788 0.566 1.788
(2.1) (3.9) (2.1) (3.9)
Use of intermediate station –1.640 –0.723 –1.640 –0.723
(–3.3) (–2.1) (–3.3) (–2.1)
Constants
Private vehicle parked for trip –0.789 –1.270 –0.209 1.453
(–1.3) (–1.7) (–0.4) (2.0)
Private vehicle dropped off –3.269 –3.236 –0.351 –1.482
(–3.9) (–3.7) (–0.6) (–2.2)
Rental car –2.911 0.825 2.001 0
(–2.7) (0.9) (2.6)
Taxi –2.730 –3.623 –2.730 –3.623
(–4.3) (–4.1) (–4.3) (–4.1)
Other private mode –0.588 –0.605 –1.488 –1.964
(–1.3) (–0.8) (–2.8) (–1.5)
Airport Bus –7.354 –4.669 –7.354 –7.087
(–4.7) (–4.8) (–4.7) (–4.9)
CTA train –0.545 –0.943 0.002 0.267
(–1.5) (–2.5) (0.0) (0.5)
Other public mode –4.426 0.028 –1.279 –0.524
(–2.5) (0.0) (–0.7) (–0.7)
Airport Express Access Modes
Free van 0.683 –0.025 0.683 –0.025
(2.1) (–0.1) (2.1) (–0.1)
Walk –0.573 –1.100 –0.573 –1.100
(–1.7) (–2.3) (–1.7) (–2.3)
Drive & park 2.261 2.431 2.261 2.431
(2.3) (3.1) (2.3) (3.1)
CTA bus or train –1.019 –0.943 –1.019 –0.943
(–2.9) (–2.5) (–2.9) (–2.5)
Nest Coefficients
Private mode nest 0.630 0.374 0.630 0.374
(8.0) (5.5) (8.0) (5.5)
Public mode nest 0.485 0.747 0.485 0.747
(5.5) (10.0) (5.5) (10.0)
These differences in the implied value of travel time across the proposed fares on the Airport Express train of only $10 per passenger,
market segments are troubling, because they directly affect the this suggests that the provision of downtown check-in would have a
trade-off between travel times and cost, which is central to the eval- similar effect in ridership to making the service free, at least for non-
uation of a new service that offers shorter travel times for a pre- business travelers. Those travelers using an intermediate station
mium fare. Given the obvious importance of household income in appear to find the Airport Express service significantly less attractive
explaining traveler behavior, both from the perspective of common than those using the downtown terminal, particularly travelers on busi-
sense and as demonstrated by the model estimation results, the ness trips, who would require a fare difference of somewhere between
classification of travelers into only two income categories must $9 and $26 to make the attractiveness of the service similar to that of
necessarily only provide a very approximate representation of the the downtown terminal, other things being equal. The upper end of this
role of income in access mode choice. Although better than ignor- range is of course significantly higher than the planned fare and it is
ing traveler income completely, it is likely to significantly reduce unclear why this should be so if the stated preference experiments had
the ability of the model to explain observed mode choice behavior correctly reflected the access times and costs to both the downtown
and could adversely affect predictions of the likely future use of terminal and intermediate stations.
new modes.
One possible explanation is that the access options were different
The implied values of the two dummy variables present some in- between the downtown terminal and the intermediate station, with
teresting implications. As might be expected, business travelers (who the drive and park access option only available at the intermediate
typically have less baggage than non-business travelers and may only station. It is noteworthy that the ASCs for drive and park access are
have carry-on bags) appear to value the availability of downtown of the opposite sign to the intermediate station coefficients and
baggage check-in at about half the value of non-business travelers. somewhat larger (slightly larger in the case of business trips and
However, the implied value for non-business travelers of between about twice as large in the case of non-business trips). If these ASC
about $7 and $18, depending on income, is surprisingly high. With values were overestimated, as discussed further later, the estimated

TABLE D9
CHICAGO AIRPORT EXPRESS MODEL CALIBRATED COEFFICIENTS: O’HARE AIRPORT
Dummy Variables
Constants
Private vehicle dropped off –1.433 –2.311 –1.022 0.925
Rental car N/A N/A 4.081 2.159
Taxi 2.750 3.658 6.620 7.017
Other private mode 3.147 3.767 6.461 6.515
Airport Bus –6.981 –5.556 –6.981 –5.556
CTA train –0.639 –2.320 –0.639 –2.320
Other public mode 3.989 1.702 5.582 4.788
Free van 0.467 –0.092 0.467 –0.092
Walk –0.620 –1.728 –0.620 –1.728
Drive and park 1.795 2.380 1.795 2.380
CTA bus or train –0.963 –1.133 –0.963 –1.133
Nest Coefficients
Public mode nest 0.523 0.775 0.523 0.775
N/A = not applicable.
TABLE D10
CHICAGO AIRPORT EXPRESS MODEL CALIBRATED COEFFICIENTS: MIDWAY AIRPORT


Dummy Variables

Constants

Private vehicle dropped off –2.767 0.189 0.157 0.335
Rental car N/A N/A 3.349 1.899
Taxi 0.625 3.532 5.588 5.849
Other private mode 0.492 8.708 5.003 5.548
Airport Bus –6.981 –8.768 –6.981 –8.768
CTA train –0.150 –0.182 –0.150 –0.182
Other public mode –2.242 –5.778 –2.506 –7.400
Free van 0.467 –0.092 0.467 –0.092
Walk –0.620 –1.728 –0.620 –1.728
Drive and park 1.795 2.380 1.795 2.380
CTA bus or train –0.963 –1.133 –0.963 –1.133
Nest Coefficients

Public mode nest 0.523 0.775 0.523 0.775

118
TABLE D11
IMPLIED VALUES OF CHICAGO AIRPORT EXPRESS MODEL COEFFICIENTS
O’Hare O’Hare Midway Midway
Total travel time 33/92 25/55 63/82 22/57
Dummy Variables ($)
Downtown baggage check 3/8 8/17 6/8 7/18
Use of intermediate station –9/–26 –5/–11 –18/–23 –4/–11
Constants ($)
Airport Express access modes
Free van 3/8 0/–1 5/7 –0.5/–1
Walk –4/–10 –10/–17 –7/–9 –7/–18
Drive and park 11/30 11/24 20/27 9/25
CTA bus or train –6/–16 –5/–11 –11/–14 –4/–12
Notes: The two values shown for each coefficient are for low-income and high-income respondents,
respectively.
The implied values of the dummy variables show the contribution of the attribute to the
perceived utility of the Airport Express train alternative.
The table shows the implied values of the alternative-specific constants for the Airport
Express train access modes relative to taxi (i.e., the perceived utility of the mode relative to
taxi if travel times and costs are equal).
TABLE D12
IMPLIED MODAL CONSTANT VALUES FOR THE CHICAGO AIRPORT EXPRESS MODEL
Modal Constants ($)
O’Hare Airport
Private vehicle parked for trip 34/ 95 36/81 N/A N/A
Private vehicle dropped off –9/–24 –11/–23 –6/–17 4/9
Rental car N/A N/A 25/68 10/22
Taxi 17/46 17/37 40/110 32/71
Other private mode 19/52 17/38 39/108 30/66
Airport Bus –42/–116 –25/–66 –42/–116 –25/–66
CTA train –4/–11 –11/–23 –4/–11 –11/–23
Other public mode 24/66 8/17 34/93 22/48
Midway Airport
Private vehicle parked for trip 25/33 22/59 N/A N/A
Private vehicle dropped off –31/–41 1/2 2/2 1/3
Rental car N/A N/A 38/50 7/20
Taxi 7/9 14/37 64/83 23/61
Other private mode 6/7 34/91 57/75 22/58
Airport Bus –79/–104 –35/–91 –79/–104 –35/–91
CTA train –2/–2 –1/–2 –2/–2 –1/–2
Other public mode –25/–33 –23/–60 –28/–37 –29/–77
Notes: The two values shown for each coefficient are for low-income and high-income respondents, respectively.
The table shows the implied values of the alternative-specific constants relative to the Airport Express train
(i.e., the perceived utility of the mode relative to the train if travel times and costs are equal).

119
value of the coefficient for the intermediate station dummy variable in the estimation and calibration of the ASC values can easily swamp
would attempt to correct for this. the effect of differences in travel times and costs between modes.
Interpreting the implied values of the ASCs for the access

modes to the Airport Express train requires some caution owing to Model Fit
the way that the associated costs and travel times were included in
the stated preference experiments. Each experiment presented an The model documentation presents t-statistics for each of the model
access time for the Airport Express option to the respondent and in- coefficients, although it does not provide an overall measure of the
dicated the associated access mode. The access times were derived fit of the model to the estimation data. Although the model calibra-
from the trip origin that the respondent had indicated; however, it tion process ensured that the predicted mode shares from the study
is unclear from the model documentation whether these access area corresponded to the observed mode shares, the model docu-
times varied by the access mode indicated. It is also unclear from mentation does not discuss how different the predicted mode share
the model documentation how the associated access costs were using the estimated coefficients was from that given by the O-D sur-
presented to the respondent, if indeed they were. The example sce- vey data, nor how much the mode share using the calibrated model
nario shown in the documentation for the stated preference survey coefficients differed from the observed mode share at the level of
(RSA 2004) gives the access time to the downtown terminal by each TAZ. Although differences are inevitable at the TAZ level, be-
CTA bus or train but makes no mention of the fare involved. In any cause the model is not a perfect predictor of individual behavior and
event, the respondents were only presented with one access option the O-D data itself will contain some error at the zonal level owing
for each Airport Express scenario and could not choose between to sample size issues, the issue of concern is whether these differ-
other options to access the service. Given the need to vary several ences exhibit any geographical bias that could affect the predictions
different parameters over the course of only eight stated preference of the use of the Airport Express train.
experiments, it is doubtful that the experiments provided much op-
portunity to examine tradeoffs between time and cost of different The t-statistics for travel time and cost coefficients are all highly
access modes as distinct from those of the primary modes. Thus, it significant. The dummy variable coefficient for downtown baggage
is likely that most respondents focused on the access time shown check is highly significant for non-business travelers and just statis-
rather than the other attributes of the access modes. tically significant at the 95% confidence level for business travelers.
The dummy variable coefficient for boarding at an intermediate sta-
This could explain some of the rather surprising implied values. tion shows the reverse pattern, being highly significant for business
It is counterintuitive that a free van service would be viewed as travelers and just statistically significant at the 95% confidence level
more attractive than a taxi if the taxi fare was already accounted for, for non-business travelers. The statistical significance of the esti-
given that the taxi provides a direct service with no waiting and does mated ASCs varies, with several having t-statistics less than 1.0 and
not have to be shared with other riders. It is not unreasonable that a few having t-statistics close to zero. However, it could be quite ap-
walking would have a higher disutility than the same time spent rid- propriate for an ASC to have a value close to zero, giving a very low
ing a taxi, although the travel time calculations apparently already t-statistic. What is of more concern is the absence of any consistent
include a weighting for this. The relatively high implied value of the pattern in the relative values of the ASCs for each mode between the
ASCs for drive and park mentioned earlier, in the range of $9 to two airports and for each trip purpose.
$30, could reflect that this access mode would have a much shorter
travel time than access by CTA bus or train, whereas the higher cost
involved may not have been fully apparent to the respondents. The Model Application
model documentation does not provide any screen shots of stated
preference scenarios involving the intermediate station or access The calibrated mode choice model for each market segment was
options other than CTA bus or train; therefore, it is unclear how applied to forecast ridership and revenue for the Airport Express
much information respondents had about the cost of different access train for various service scenarios for two future years, 2009 and
alternatives. 2020. These service assumptions included two different fare levels,
several different travel time assumptions for the service to O’Hare
The implied values for many of the ASCs for the primary ac- Airport (the travel time to Midway Airport was not varied), two
cess modes shown in Table D11 appear generally to have the ex- different growth rates for off-peak highway travel time, and whether
pected sign, but many of the values are surprisingly high. It is un- or not a free downtown shuttle or baggage check at the downtown
clear why the option of “other public mode” is viewed as terminal would be provided (WSA 2004).
inherently more attractive than the Airport Express in the case of
O’Hare Airport, by an amount equivalent to a cost difference as To apply the mode choice models it was necessary to calculate
high as $93 (in the case of non-resident business trips by high- travel times and costs on each mode from each TAZ. This issue did
income respondents), but inherently less attractive than Airport not arise in the model estimation, because the values were either
Express in the case of Midway Airport by an amount equivalent to provided by the respondents for their existing choice or generated
a cost difference of $77 (in the case of non-resident non-business hypothetically for the other options in the stated preference experi-
trips by high-income respondents). Similarly, it is unclear why ments. Highway travel times were obtained from highway network
Airport Bus would be viewed as inherently less attractive than the skims of the Chicago Area Transportation Study, whereas transit
Airport Express by an amount equivalent to a cost difference that travel times were obtained from similar skims from a CDOT/WSA
varies between $35 and $104. If travelers really valued the attri- transit network model developed for the CDOT Mid-City Transit-
butes of rail travel over bus travel by amounts of this magnitude, way Study (WSA 2004). Other travel cost and time assumptions
the challenge of financing rail transit systems would have been were made for various components of the airport access time and
solved long ago. cost by each mode. Because highway travel times vary by time of
day, separate highway travel times were calculated for peak and off-
These issues are a significant cause for concern because of the peak conditions and the models applied separately to each time
relative magnitude of the ASC values compared with the travel times period, with the results weighted by the proportion of airport access
and costs. Not only does this mean that the model predictions are rel- and egress travel that occurs during each period based on the profile
atively insensitive to changes in travel time and cost, but that errors of flight arrivals and departures.

120
Because the air passenger trip tables that were developed for the Public transit <1%
study did not include detailed air party characteristics, but only Hotel van 3%
counts of passengers from each TAZ by market segment, the model Market Segmentation Residents—Business trips
application was based on average air party size and average air trip Residents—Non-business trips
duration for each market segment in calculating travel costs, as well Non-residents—Business trips
as the average percentage of rental cars that were not needed for Non-residents—Non-business trips
other purposes than travel to and from the airport. Because logit Explanatory Variables Travel cost
choice models tend to exhibit strong nonlinearity with respect to the In-vehicle travel time
values of the explanatory variables, it is unclear what effect this Waiting time plus terminal time (if any)
might have on the results. Modes such as the Airport Express train
become less attractive to passengers traveling in larger parties,
because the cost increases in proportion to the party size in contrast Description
to modes such as taxi, limousine, or the use of private vehicles. The
average air party size used in the analysis was based on the O-D sur- The Miami Intermodal Center (MIC) is being planned as major
vey results and varied by trip purpose between 1.57 and 1.96. Thus, transportation interchange facility located immediately to the east
the mode choice analysis included no single-person parties, even of the Miami International Airport (MIA), as shown in Figure D4.
though these are by far the largest percentage of air travel parties The project is intended to provide an integrated terminal for several
(because the average air party size reflects a small number of quite intercity transportation services, including Amtrak, Tri-Rail com-
large parties). Similarly, the average number of days away by muter rail, Greyhound buses, and future High Speed Rail and
Chicago area residents varied between 3.8 and 5.5, depending on East–West Corridor rail lines, as well as Metrobus and Metrorail
the airport and trip purpose. The mode choice behavior of someone services (FHWA and Florida DOT 1997). An automated people
who is away for a period from 4 to 6 days is likely to be very dif- mover (the MIC/MIA Connector) will link the MIC to the airport.
ferent from someone who is only away for one or two days on the Provision has also been made in the planning for a future Air-
one hand or someone who is away for several weeks on the other, port/Seaport Connector rail link. The MIC will also accommodate
owing to the increasing cost of airport parking as the trip duration growth of MIA by providing expanded airport landside facilities,
increases. Whether these effects cancel each other out, giving rea- including rental car facilities and long-term parking facilities.
sonable estimates of the likely use of the Airport Express train or
result in biased estimates of ridership is unclear. As part of the planning for the MIC, a travel demand forecast was
prepared in 1995 (ICF Kaiser Engineers 1995). This incorporated an
airport access mode choice model that was used to forecast ridership
Documentation on the MIC/MIA Connector. The mode choice model was based on a
model originally developed by KPMG Peat Marwick for a study for
PB Consult, Inc., in association with Mercer Management Consulting, Inc., Newark International Airport in New Jersey. The Newark model was
Parsons Brinckerhoff Quade & Douglas, Inc., and Velma Butler & Com- calibrated as a simple binomial logit choice model of the use of tran-
pany, Ltd., Express Airport Train Service: Business Plan, Prepared for sit versus non-transit modes and was subsequently converted into a
Chicago Transit Authority, Chicago, Ill., Final Report, Sep. 22, 2006.
nested model by adding the choice between rail and bus to the transit
Resource Systems Group, Inc., O’Hare and Midway Airport Express Train
Ridership Forecasting Study: Chicago Air Traveler Stated Preference nest. Because of the large number of modes available at MIA and the
Survey Report, Prepared for Chicago Department of Transportation, complexity of the choices available as a result of the MIC project, the
White River Junction, Vt., Jan. 2004. mode choice model was expanded to include the following modes:
Wilbur Smith Associates, in association with Resource Systems Group, Inc.,
Airport Express Ridership and Revenue Forecast, Prepared for Chicago • Drop off by private vehicle (termed auto–kiss and ride)
Department of Transportation, Chicago, Ill., 2004. • Private vehicle parked for the air trip duration (termed
auto–park and ride)
• Rental car
D4 MIAMI INTERMODAL CENTER TRAVEL • Taxi
DEMAND FORECAST STUDY • Limousine
• Premium transit
Summary • Local transit
• Shared-ride van (termed Super Shuttle)
Airport Miami International Airport • Hotel courtesy shuttle (termed hotel van).
Model Developer Gannett Fleming, Inc. with KPMG Peat
Marwick The first five modes were grouped into a nest called Non-Group
Date Developed 1995 modes and the other four modes were grouped into a nest called Group
Market Addressed Air passengers modes. The resulting structure of the model is shown in Figure D5.
Model Structure Nested logit The model adopted the four market segments used in the
Survey Data Used 1991 Miami International Airport Air Newark model:
Passenger Survey
Airport Profile Total annual passengers (2005): • Resident business trips
30.2 million • Resident non-business trips
Percentage O&D: 55% • Non-resident business trips
Ground access mode split (1991 survey): • Non-resident non-business trips.
Private vehicle—drop off 45%
Private vehicle—parked 13% Air passenger trip tables were developed for each market seg-
Rental car 28% ment based on an air passenger survey of 3,002 respondents con-
Taxi 6% ducted at MIA by Landrum & Brown in 1990. This survey had coded
Limousine 2% the trip origins of survey respondents using a system of 19 zones.
Shared-ride van 3% The number of air passenger trip origins in each zone was weighted

121
FIGURE D4 Miami intermodal center location.

(Source: ICF Kaiser 1995.)
to represent the number of originating average weekday passengers Highway and transit travel times and highway distances from
in 1991 and doubled to account for arriving passengers. These air each TAZ to the airport or MIC were obtained from the travel time
passenger trips were then assigned to the much larger number of re- skim tables in the regional travel model. Transit travel times were di-
gional (TAZs) in the regional travel model using a simple prorating vided into in-vehicle time and waiting time from the transit network
process based on the type of trip origin given by the air passenger skim tables. Constant values for waiting and other out-of-vehicle
survey. Trips with home origins were allocated on the basis of pop- times were assumed for other modes. Transit fares were assumed
ulation, trips with a workplace origin were allocated on the basis of constant at $1.25. Private vehicle operating costs were assumed to be
employment, trips from a hotel or motel were allocated on the basis 27.5 cents per mile and parking costs were assumed to be $3 for
of the number of hotels and motels in the zone, and trips with other drop-off trips and $18 for vehicles parked for the duration of the air
types of trip origin were allocated on the basis of population. Air pas- trip. Taxi costs were estimated from the highway distance and meter
senger trips from external zones in the regional model were allocated flag drop and rate per mile, with a flat fare for airport area hotels and
on the basis of the total internal–external trips for the external TAZ the seaport. Limousine and shared-ride van fares varied by trip ori-
in the regional model. gin zone. The model documentation does not discuss how the size of
Miami International Airport (MIA)
Person Trips
Non-Group Group
Auto Auto Rental Premium Local Super Hotel

Taxi Limo
K/R P/R Car Transit Transit Shuttle Van
FIGURE D5 Miami air passenger ground access mode choice model structure.
(Source: ICF Kaiser 1995.)

122
the air party was addressed in calculating travel costs, if at all. The culation of the variable values differs for each mode. There is no
cost assumptions appear to have been made on the basis of the travel consideration given to the effect of income differences in the model.
cost for each air party. If these costs were then applied in the model Therefore, all respondents are assumed (in effect) to have the same
on the basis of each air passenger, this would significantly overstate value of travel time.
the cost of some modes for travel parties with more than one air pas-
senger. Similarly, the same cost for parking at the airport for the
duration of the air trip appears to have been used for each passenger, Model Coefficients
irrespective of the actual trip duration. This would tend to overstate
the cost of this alternative for passengers on shorter trips and under- The values for the coefficients for the continuous variables were not
state the cost for those on longer trips. estimated from the air passenger survey data, but rather adopted (or
“borrowed” to use the expression in the model documentation) from
Peak highway travel times were assumed for all air passengers, the values estimated for Newark International Airport in the original
irrespective of their actual flight time. This would tend to overstate model. There is no discussion in the model documentation about
the travel times for highway-based modes during most of the day whether any adjustments were made to reflect changes in income
and to the extent that transit travel times are less subject to peak levels and price values during the period of time that had elapsed be-
period congestion would bias the model in favor of transit use. cause the Newark model was estimated or there were differences in
income levels between Miami and the New York/New Jersey region.
All modes were assumed to be available to each market segment, The same coefficient values were used for all four market segments,
including parking a private vehicle for the duration of the trip, rental implicitly implying the same value of time for business and non-
car, and hotel/motel courtesy shuttle. Because visitors do not have business travelers and for residents of the Miami region and visitors.
the option of parking a car at the airport during their trip (if they had
access to a car during their visit they would not leave it at the air- Values for the ASCs were estimated from the air passenger sur-
port), and residents typically do not rent cars to access the airport or vey data to fit the model predictions of mode use to the observed
start their trip to the airport from a hotel or motel that provides a data. Estimated values were obtained for each mode, although usual
courtesy shuttle, it is likely that the estimated ASCs for these modes practice is to set the ASC for one mode to zero and express the other
are biased by the need to explain the low mode share by those mar- ASCs relative to that. Because the premium transit service did not
ket segments for which certain modes are not a viable option. exist at the time of the air passenger survey, the values for that mode
were based on the values of the ASCs for other modes and the abil-
ity of the model to predict transit ridership levels typically observed
Explanatory Variables in other U.S. cities with premium transit services similar to that en-
visioned for Miami.
The model has only three explanatory variables, apart from the
ASCs: in-vehicle travel time, out-of-vehicle travel time (including The assumed and estimated values of the model coefficients are
waiting time and terminal time, such as walking from the parking shown in Table D13 and the implied values of the coefficients are
lot to the airport terminal or returning a rental car), and travel cost. shown in Table D14. The values of the ASCs have been adjusted so
The same three variables are used for each mode, although the cal- that the ASC for drop off by private vehicle is zero. These values
TABLE D13
MIAMI INTERMODAL CENTER TRAVEL DEMAND MODEL COEFFICIENTS

Variables
In-vehicle time (minutes) –0.06383 –0.06383 –0.06383 –0.06383

Waiting/terminal time (min) –0.16077 –0.16077 –0.16077 –0.16077
Cost (cents) –0.00049 –0.00049 –0.00049 –0.00049
Constants
Private vehicle—dropped off 1.67737 2.82737 2.22737 2.22737

Private vehicle parked for trip 2.22737 2.22737 0.89737 1.37737
Rental car 2.71737 4.41737 6.05737 6.07737
Taxi 0.47737 1.47737 2.76237 1.63737
Limousine –4.27263 2.69737 3.80737 2.28737
Premium transit –0.7 0.8 –0.2 2.8
Local transit –5.9 –3.7 –8.2 –4.2
Shared-ride van –6.2 –6.3 –7.66 –6.86
Hotel courtesy shuttle –8.0 –7.0 –5.5 4.0
Nest Coefficients
Non-group modes nest 0.3 0.3 0.3 0.3

Group mode nest 0.3 0.3 0.3 0.3

123
TABLE D14
IMPLIED VALUES OF MIAMI INTERMODAL CENTER TRAVEL MODEL COEFFICIENTS

In-vehicle time 78 78 78 78
Waiting/terminal time 197 197 197 197
Constants ($)
Private vehicle—dropped off — — — —

Private vehicle parked for trip 11 –12 –27 –17
Rental car 21 32 78 79
Taxi –25 –28 11 –12
Limousine –121 –3 32 1
Premium transit –49 –41 –50 12
Local transit –155 –133 –213 –131
Shared-ride van –161 –166 –202 –185
Hotel courtesy shuttle –198 –201 –158 36
vary across the four market segments, reflecting difference in mode improved modes, because changes in the values of the continuous
use across the segments. variables will have a relatively small effect on the model predictions.
The implied value of in-vehicle time of $78/h appears reasonable The values of the nest coefficients were presumably also taken
for air passenger travel, particularly because it applies to both busi- from the Newark model. It is unclear that these would still be valid
ness and non-business travel. The implied value of waiting and ter- if the number of alternatives in each nest is increased to the extent
minal time is about two and one-half times the value of in-vehicle that was done.
time, which is somewhat higher than is commonly found in urban
travel models but not unreasonable, particularly given the time-
sensitive nature of airport access travel, where an unexpected delay Model Fit
could result in a missed flight. The implied values of ASCs vary
widely, with several implausibly large. The large negative values for The model documentation provides no information on the statistical
the hotel courtesy shuttle are most likely an artifact of this mode being significance of the estimated values of the ASCs or the overall fit of
available to all air passengers, rather than those starting their access the model.
trip from hotels that provide a courtesy shuttle service to the airport.
However, the large negative implied values for the shared-ride van is
surprising, because the assumed fares were based on the prevailing Model Application
fare structure for SuperShuttle and travel times are similar to other
highway modes. The travel time assumptions did not consider the The mode choice model was used to predict ground access mode use
time involved in any circuitry to pickup or drop off other passengers to MIA in 2020 and resulting peak-hour passenger trips using the
or schedule delay resulting from the frequency with which pickups MIC/MIA Connector. The 2020 air passenger trip tables were ob-
can be scheduled. However, the implied ASC values are equivalent to tained by factoring up the 1991 trip tables by the forecast growth in
more than 2 h of travel time, which is far greater than the time in- total airport traffic and the assumed proportion of locally originating
volved in waiting for a pickup or picking up other passengers. air passengers in 2020 provided by MIA staff. The peak-hour demand
on the MIC/MIA Connector was assumed to occur on a Friday after-
Although the difference in ASC between premium transit and noon between 3:30 and 4:30 p.m. and the air passenger component
local transit is in the expected direction, the magnitude of the dif- was assumed to comprise 14% of the average weekday air passengers
ference, an implied value of between about $90 and $160, seems other than cruise ship passengers. The forecast average weekday air
unrealistically high. The model documentation does not clearly passenger traffic was reduced to allow for cruise ship passengers
define premium transit, but presumably this mode is used to distin- using the Airport/Seaport Connector who would not use the MIC,
guish between Metrorail and local bus service. Although air travel- with the balance of the cruise ship passengers using the MIC. It was
ers may well prefer rail transit to bus service in situations in which assumed that 13% of the cruise ship passengers on an average Friday
the travel times on either mode are the same, it seems unlikely that using the MIC would do so in the peak-hour, together with 20% of the
this preference would be strong enough that the mode shares of the average weekday employees at the airport estimated to use the MIC.
two services would be equal if travel times were the same, but the
Metrorail fare was $160 more than the bus fare.
Documentation
The implications of the large ASC values is that the continuous
variables are not explaining the observed mode shares very well at ICF Kaiser Engineers, Inc., with Gannett Fleming, Inc., and KPMG Peat
all and the ASCs are having to adjust the model to enable it to pre- Marwick, Miami Intermodal Center: Travel Demand Forecast Report,
dict the observed mode shares. This in turn suggests that the model Prepared for Florida Department of Transportation and Federal High-
is not going to do a very good job of predicting the use of new or way Administration, Miami, Fla., Aug. 1995.

124
U.S. Federal Highway Administration and Florida Department of Transporta- The planned Connector would include two stops between the
tion, Miami Intermodal Center Final Environmental Impact Statement, airport and the Coliseum BART station and would reduce the cur-
Report FHWA-FLA-EIS-95-01-F, Tallahassee, Fla., Dec. 23, 1997. rent AirBART travel time of about 13 min to about 6 min. It was
assumed that service frequency would also improve from the cur-
rent AirBART frequency of every 10 min to a 3.2-min headway.
D5 OAKLAND INTERNATIONAL AIRPORT BART The goal of the mode choice analysis was to evaluate the effect of
CONNECTOR STUDY this service improvement on ridership on the Connector compared
with the AirBART service, as well as to examine an alternative im-
Summary proved bus service.
Airport Oakland International Airport The mode choice model addressed both air passenger trips and
Model Developer CCS Planning and Engineering, Inc. airport employee trips. The employee trips were treated as a sepa-
Date Developed 2001 rate market segment and this aspect of the model is discussed fur-
Market Addressed Air passengers ther later. The general form of the model is a MNL model with air
Model Type Revealed preference data passenger trips divided into four market segments as follows:
Model Structure Multinomial logit
Survey Data Used 1995 Metropolitan Transportation • Resident business trips
Commission Air Passenger Survey; 1999 • Resident personal trips
Survey of AirBART Passengers • Visitor business trips
Airport Profile Total annual passengers (2005): 14.1 • Visitor personal trips.
million
Percentage O&D: 95% Data on the air party characteristics for each market segment
Ground access mode split (2002 MTC were obtained from the 1995 Air Passenger Survey performed for
Air Passenger Survey): the Metropolitan Transportation Commission (MTC) at the three
Private vehicle—drop off 42% Bay Area airports, including Oakland International Airport (Franz
Private vehicle—parked 21% 1996). This information was supplemented by surveys of AirBART
Rental car 15% passengers performed by CCS Planning and Engineering, Inc., in
Taxi 3% December 1999 and May 2000 as part of the study. The mode choice
Limousine 2% analysis assigns airport trips among the following eight modes:
Hotel courtesy shuttle 2%
Shared-ride van 3% • Private vehicle (termed private auto)
Scheduled airport bus 3% • Rental car
Shuttle bus from BART 8% • Scheduled airport bus (termed scheduled shuttle bus)
Public transit bus 1% • Public transit
Other <1% • Shared-ride van (termed door-to-door shuttle)
Market Segmentation Residents—Business trips • Hotel courtesy shuttle
Residents—Personal trips • Taxi and limousine
Visitors—Business trips • Other.
Visitors—Personal trips
Explanatory Variables Travel time (private vehicles) Public transit included the use of BART through the AirBART
Travel time (rail transit) shuttle (or the Connector in the future), as well as local transit bus
Travel time (bus transit) service directly to the airport. However, these two modes were not
Walk distance separately identified in the model but rather the transit alternative
Wait time for travelers with trip origins in zones near the airport was as-
Travel cost sumed to be local bus, whereas the transit alternative for those
Household income from more distant zones was assumed to be BART. The model did
not distinguish between resident air passengers who were dropped
off by private vehicle and those who parked at the airport for the
Description duration of their air trip. Rather, all resident air passengers using
private vehicles were assumed to park at the airport during their
Oakland International Airport currently operates a shuttle bus link trip, whereas all visitors using private vehicles were assumed to be
called AirBART between the airport and the Coliseum station of the dropped off. The use of “Other” modes was not explicitly included
Bay Area Rapid Transit (BART) system, located about 2.5 miles in the model, but rather the use of those modes was assumed to
from the airport. Since the 1970s, a series of studies have been remain constant from the mode share observed in the 1995 Air
undertaken by the Port of Oakland (the operator of Oakland Inter- Passenger Survey.
national Airport), BART, and other agencies to explore the feasi-
bility of an automated people-mover connection between the airport The perceived operating cost of private vehicles was assumed to
and the Coliseum BART station (BART–Oakland International Air- be 15 cents per mile in 1999 dollars, whereas airport parking costs
port Connector . . . 2002, Executive Summary). The most recent of were calculated assuming a parking rate of $8 per day. However, the
these efforts commenced in 1999 with a public scoping meeting for parking cost was not calculated separately for each air party, but
the preparation of an Environmental Impact Report/Environmental rather the average trip duration and average air party size for each
Impact Statement (EIR/EIS) for a planned Oakland Airport Con- market segment was used to calculate a fixed parking cost for each
nector. The Draft EIR/EIS was distributed in August 2001 and the segment. The highway travel time for visitors dropped off at the
Final EIR/EIS was approved in March 2002. As part of the analy- airport (or picked up on their arrival) was increased by 50% to rep-
sis for the EIR/EIS, an airport access mode choice model was de- resent the inconvenience for the drivers dropping off or picking up
veloped and applied to generate ridership projections for the Con- air passengers. Appropriate costs and times were assigned to the
nector (BART–Oakland International Airport Connector . . . 2002, other modes based on published fares and schedules and assumed
Appendix B: Transit Ridership Procedures and Inputs). walking distances and waiting times.

125
Rental car costs were calculated on the basis of $50 per day fac- reimbursed for travel expenses. However, rather than using the ac-
tored by the average trip duration and average air party size for each tual income level of each air party, the average income for all air
market segment. Apart from that this assigned the same cost to all air passengers at Oakland International Airport found in the 1995 Air
parties, irrespective of their actual trip duration, it also assumed that Passenger Survey appears to have been used. If this is the case, then
the full cost of renting the car was attributable to the airport access this of course is a constant that applies to all air passengers and does
and egress trip. It is unclear from the model documentation whether not vary behavior by income level as implied by including this in the
this cost was divided between the access and egress trip or whether model formulation.
the costs and travel times for all modes were calculated on a round-
trip basis (the same issue applies to parking costs at the airport).
Model Coefficients
Highway and transit travel times and highway distances were
obtained from the MTC regional travel demand network model. The model coefficients for the continuous variables were adopted
However, the model was not run using the regional travel demand directly from an earlier airport ground access mode choice model
model TAZ system, but rather a system of 25 larger zones. A repre- for Bay Area developed by Harvey (1988). The values of the ASCs
sentative TAZ was chosen for each of the analysis zones to obtain were then estimated to fit the model to the mode use data from the
travel times and distances. Access times to transit for each zone 1995 MTC Air Passenger Survey. The resulting values of the as-
were estimated based on the size of the zone and assuming walk sumed and estimated coefficients are shown in Table D15. The cor-
access for the smaller zones and driving for the larger zones. responding implied values of the coefficient values are shown in
Table D16.
Explanatory Variables The implied values of travel times are fairly low compared with
values typically found in air passenger travel models. The model
The model utility functions included the following six variables: documentation notes that the average household income of the
highway travel time, travel time by rail transit, travel time by bus respondents to the 1995 MTC Air Passenger Survey was $75,000.
transit, walking distances, waiting times, and travel costs. Obvi- Assuming 1.7 wage earners per household (about the current U.S.
ously, not all variables applied to each mode. The distinction average)—that translates to an average wage rate of about $22/h.
between rail transit travel time and bus transit travel time allowed However, the values of time implied by the model coefficients for
the analysis to consider the effect of replacing the AirBART shuttle the continuous variables derive directly from the coefficients esti-
bus with the planned automated people-mover as well as account- mated by Harvey (1988), with those for personal travel factored up
ing for the different level of service between BART and local bus. by the change in income levels between 1985 (the base year used by
Harvey to estimate his model) and 1995 (the year of the MTC air
Household income was included in the model by dividing the passenger survey data used to estimate the current model). This also
costs for personal trips by the household income in thousands of explains why the implied values of time for business trips are
dollars raised to the power 1.5. This adjustment was not applied to approximately the same as for personal trips (or even slightly
business trips as it was considered that business travel decisions are lower) rather than higher as would be expected. Because the coeffi-
unaffected by income level because business travelers are usually cients for travel times and costs reflect the perceived value of time,
TABLE D15
OAKLAND AIRPORT CONNECTOR MODEL COEFFICIENTS
Resident Resident Visitor Visitor

Coefficient Business Personal Business Personal
Variables
Highway time (minutes) –0.071 –0.044 –0.068 –0.039

Rail transit time (minutes) –0.053 –0.031 –0.050 –0.029
Bus transit time (minutes) –0.093 –0.051 –0.089 –0.045
Walk distance (miles) –5.17 –3.28 –4.69 –2.94
Cost (cents) –0.00277 –1.04/ –0.00256 –0.973/
(HHINC)1.5 (HHINC)1.5
Constants
Private vehicle — — — —
Rental car –0.8 –4.2 0.7 –1.2
Scheduled airport bus –0.5 –1.4 0.0 –1.2
Public transit –1.5 –1.2 –1.0 –1.8
Shared-ride van 0.0 –0.9 1.0 –0.9
Hotel courtesy shuttle N/A N/A –3.2 –4.2
Taxi/limousine –0.2 –1.6 0.8 –0.8
Notes: HHINC = annual household income in thousands of dollars; N/A = mode is not available for this
market segment.

126
TABLE D16
IMPLIED VALUES OF OAKLAND AIRPORT CONNECTOR MODEL COEFFICIENTS
Resident Resident Visitor Visitor

Parameter Business Personal Business Personal
Highway travel time 15 16 15 16

Rail transit time 11 12 11 12
Bus transit time 20 19 19 18
Walk time 56 61 55 59
Wait time 23 29 21 28
Constants (dollars)
Private vehicle 0.7 10.0 –3.1 5.3

Rental car –13.0 –16.2 –0.4 –2.7
Scheduled airport bus –1.1 1.2 –3.1 –2.7
Public transit –4.7 2.5 –7.0 –6.7
Shared-ride van 0.7 4.4 0.8 –0.7
Hotel courtesy shuttle N/A N/A –15.6 –22.7
Taxi/limousine — — — —
Notes: Implied values of personal trips calculated for an annual household income of $75,000 per year.
Implied value of walk time based on a walking speed of 3 mph.
N/A = mode is not available for this market segment.
coefficient values estimated on data for one year cannot simply be ously perhaps, the negative value for the constant for private vehicle
applied to data for another year without adjustment for changes in (i.e., drop off) for visitor business trips reflects that those visitors stay-
real income levels (and hence the perceived value of time) between ing in a hotel or starting their trip to the airport from a business or sim-
the two years. ilar location will generally not have access to someone who can take
them to the airport by private vehicle.
Given these problems with the values of the coefficients for the
continuous variables, it follows that the estimated values for the Residents on business trips and visitors appear to find scheduled
ASCs are accounting for more than the differences in the perceived airport bus and transit less attractive than taxi (after accounting for
inherent attractiveness of the different modes. In addition, they have differences in travel time and cost), as might be expected, with
to correct for biases in the predicted mode use that result from the transit significantly less attractive than scheduled airport bus.
use of invalid coefficients for the continuous variables. Although Somewhat surprisingly, residents on personal trips appear to find
the ASCs have been estimated relative to the private vehicle mode, transit and scheduled airport bus more inherently attractive than
expressing the implied values in this way would give misleading taxi after accounting for differences in travel time and cost, al-
relative values of the coefficients across the different market seg- though the effect is not very large. Shared-ride van appears to
ments because of the assumption that the private vehicle mode rep- be inherently more attractive than scheduled airport bus, as might
resents parking for the duration of the air trip in the case of resident be expected owing to the door-to-door service. However, the dif-
trips but drop off in the case of visitor trips. Therefore, it would pro- ference in the perceived value of the constant is not very large,
vide a better comparison to express the implied values of the con- equivalent to just two or three dollars. The relatively high implied
stants relative to taxi, which offers the same service characteristics value for the private vehicle constant for residents making personal
across each market segment, as shown in Table D16. The values trips could be a result of assuming that all residents using private
shown in Table D16 indicate the amount by which the cost of a vehicles park for the duration of their air trip. In practice many are
mode would have to be greater than that for taxi (or less than if neg- dropped off and do not incur the parking cost assumed for this
ative) for travelers to consider the mode and taxi to be equally at- mode. Therefore, the constant will need to be large enough to off-
tractive if travel times were identical. set the assumed parking cost.
In addition to reflecting the inherent relative attractiveness of the Overall, whereas the implied values of the ASCs appear to be
different modes and any biases resulting from invalid coefficients of readily explainable from the way that the availability of the differ-
the continuous variables, the ASCs also have to correct for incorrectly ent modes has been assumed and the values of the service charac-
specified mode availability. The large negative values for the rental teristics for each mode defined, it also appears likely that the result-
car constant for resident trips undoubtedly reflects that most residents ing model, although replicating the observed mode shares for the
already have a private vehicle available and therefore do not need to base condition, will not correctly reflect the effect on mode share of
rent one. Similarly, the large negative value for the hotel courtesy the service improvements provided by the Oakland Airport Con-
shuttle for visitor trips reflects that this alternative is only available for nector. In particular, the low implied values of travel time will tend
those beginning their trip from a hotel near the airport that offers a to underestimate the effect on the attractiveness of using BART of
courtesy shuttle. Visitors staying in hotels elsewhere in the region or the reduced waiting and travel times involved in traveling between
with family or friends do not have this option available. Less obvi- the Coliseum station and the airport.

127
Model Fit income and vehicle ownership. Excluding these variables from the
model would change values of the other coefficients, because the
The model documentation did not provide any information on the behavioral explanation that they provide would now have to be ac-
statistical significance of the estimated values of the model coeffi- counted for by the remaining variables. The magnitude of this effect
cients or the overall fit of the model to the estimation data. would depend on the distribution of the employee characteristics
and is unclear without more detailed analysis. Furthermore, the
MTC mode choice model was also estimated using constant 1990
Model Application dollars for travel costs. Applying these model coefficients to costs
in 1999 current dollars without making any adjustments for inflation
The mode choice model was used to estimate transit ridership for would overstate the effect of cost in mode choice decisions. The
two future years, 2005 and 2020, and three project alternatives: a ASC for transit use does not correspond to the MTC model coeffi-
continuation of the AirBART shuttle (no-project alternative), an cients and appears to have been estimated to make the transit mode
improved shuttle bus alternative (termed Quality Bus), and the pro- share match the observed data for airport employees. Although this
posed automated people-mover (termed Automated Guideway will offset the effect of any bias in the model coefficients for the
Transit). Forecasts of future levels of air passenger traffic were base year, the net effect on the model predictions for any future year
interpolated from forecasts prepared as part of the Regional Airport is unclear.
System Plan and estimates of future airport employment levels were
provided by the Port of Oakland. These forecasts were used to fac- The implied values of travel time components correspond to the
tor up the air passenger and airport employee trip tables before values in the MTC model (not surprisingly, because the coefficients
applying the mode choice model. are unchanged). The documentation of the MTC model (Purvis
1997) notes that the value of time for journey-to-work trips has typ-
ically been found to be in the range of 25% to 50% of the wage rate.
Unlike air passenger business travel costs, journey-to-work travel
Airport Employee Mode Use costs are not generally reimbursed by employers and so ignoring
income in the utility function will fail to reflect the effect of income
The mode choice model included a market segment for airport em- on mode choice. This becomes a particularly important issue if the
ployees that considered only two modes, private vehicle and transit. model is to be used to forecast travel behavior in future years. With-
The employee model used the same coefficient value for both high- out knowing what assumptions were made for employee parking
way and transit travel time, including rail and bus. In addition to costs it is difficult to assess the reasonableness of the value of the
travel time, the utility functions included walking distance, waiting ASC for transit.
time, and cost. The coefficient values were not estimated from em-
ployee survey data but were adapted from the MTC regional travel
demand model for home-based work trips. The model documenta-
tion does not state what assumptions were made about the cost for Documentation
employee parking.
Franz, J.D., 1995 Metropolitan Transportation Commission Airline Passen-
ger Survey, Final Report, Prepared by J.D. Franz Research for the Met-
The model coefficients and implied values are shown in Table ropolitan Transportation Commission, Oakland, Calif., Feb. 1996.
D17. In this case, the implied value of the transit ASC is relative to Harvey, G., ACCESS: Models of Airport Access and Airport Choice for the
private vehicle, because taxi was not considered in the employee San Francisco Bay Region—Version 1.2, Prepared for the Metropolitan
mode choice model. Transportation Commission, Berkeley, Calif., Dec. 1988.
Purvis, C.L., Travel Demand Models for the San Francisco Bay Area
The coefficient values for the continuous variables were taken (BAYCAST-90): Technical Summary, Metropolitan Transportation Com-
directly from the MTC mode choice model (Purvis 1997, Table mission, Oakland, Calif., June 1997.
5.1), with the walk time coefficient converted to distance assuming U.S. Federal Transit Administration and San Francisco Bay Area Rapid
a walking speed of 3 mph. However, the MTC model includes seven Transit District, BART—Oakland International Airport Connector,
Final Environmental Impact Report/Environmental Impact Statement,
modes and several other explanatory variables, including household State Clearinghouse No. 99112009, Oakland, Calif., Mar. 2002.
TABLE D17
OAKLAND AIRPORT CONNECTOR AIRPORT D6 PORTLAND INTERNATIONAL AIRPORT
EMPLOYEE MODEL ALTERNATIVE MODE STUDY
Model Summary
Parameter Coefficient Implied Value
Airport Portland International Airport
Variables $/hour Model Developer Cambridge Systematics, Inc. and Portland
Metro
Travel time (minutes) –0.02683 11
Date Developed 1997–1998
Walk distance (miles) –1.1552 24 Market Addressed Air passengers
Wait time (minutes) –0.0418 17 Model Type Combined revealed preference and stated
preference data
Cost (cents) –0.001468 Model Structure Multinomial logit
Survey Data Used Port of Portland Air Passenger Survey
Constants $
Airport Profile Total annual passengers (2005): 13.6
Private vehicle — — million
Percentage O&D: 85%
Public transit –2.0 –14
Ground access mode split (2006 Cus-
Notes: Coefficients derived from regional travel demand model. tomer Satisfaction and Terminal User
Implied value of walk time based on a walking speed of 3 mph. Survey):

128
Private vehicle—drop off 36% transportation services and the lessons that might be applicable to
Private vehicle—parked 24% the Portland situation. Although there was no explicit comparison
Rental car 19% of the results of the mode choice analysis with the experience at
Taxi 4% other airports, this study nonetheless helped put the results of the
Limousine 2% mode choice modeling into a larger context and served to provide
Shared-ride van 4% some assurance of the likely validity of the modeling results.
Light rail transit 6%
Hotel shuttle 4%
Charter/tour bus 1% Explanatory Variables
Market Segmentation Residents—Business trips
Residents—Non-business trips The model included four explanatory variables: travel time, travel
Non-residents—Business trips cost, household income, and the time and cost of the driver dropping
Non-residents—Non-business trips off air passengers by private vehicle. Travel time combined in-
Explanatory Variables Travel time (in-vehicle time, wait, on- vehicle time, waiting time, and any on-airport time (such as the time
airport time) required to travel from a parking lot to the terminal). Travel costs
Travel cost/ln(household income) were divided by the natural logarithm of the average household
Drop-off driver time (as cost at assumed income for each origin zone.
value of time)
Drop-off automobile operating cost The direct costs of each mode (but not the operating costs and
(assumed) value of driver time of automobiles dropping off air passengers)
were divided by the logarithm of the average household income
for the trip origin zone (in thousands of dollars per year) for each
Description market segment, determined from the air passenger survey. This
gives values of time that vary with household income, as is to be
Soon after the Boston Logan model discussed earlier was devel- expected, but that have a nonlinear relationship that increases at a
oped, a similar model was developed for Portland, Oregon, as part declining rate at higher income levels. The use of the average
of a ground access study for Portland International Airport (PDX) household income for the zone resulted from the way that the
that was jointly undertaken by the Port of Portland and Metro, the model was applied, although this obviously fails to account for the
regional MPO, with the assistance of Cambridge Systematics, Inc. effect of variation in household income across survey respondents
The primary purpose of the model was to forecast the potential rid- from a given zone.
ership on a planned extension of the Portland MAX light rail system
to the airport, as well as other ground access enhancements. An air For the drop off alternatives, including air passengers dropped
passenger survey was performed at the airport that consisted of a re- off at the airport by private automobile (termed auto drop off), the
vealed preference (RP) survey that examined air passengers’ actual time of the driver (termed the chauffeur in the model documenta-
mode use and a stated preference (SP) survey that was designed to tion) was assigned a value of $20/h for business travelers and $10/h
determine travelers’ preferences for modes that were not then avail- for non-business travelers according to the model documentation
able, namely light rail, express bus, and shared-ride door-to-door (tables giving the final model coefficients indicate that $20/h was
vans (termed shared-ride transit). used for all trip purposes; however, this is assumed to be a typo-
graphic error). Automobile operating costs were assumed to be 12
An initial model estimation by Cambridge Systematics jointly es- cents per mile.
timated four MNL models using both the RP and SP data, with dif-
ferent modal alternative choice sets for residents and non-residents
of the region and separate coefficients for business and non-business Model Coefficients
travelers (Bowman 1997; Cambridge Systematics 1998). These
models were subsequently revised by the Metro staff to combine The final model coefficients are given in Tables D18 to D21 (the vari-
some of the choice alternatives and adjust the ASCs to recalibrate the ation in the number of decimal places of the coefficient estimates
models (Portland Metro 2001). The revised model coefficients re- reflect the model documentation prepared by Portland Metro). Sepa-
ported by Metro staff also included one change to the cost coefficient rate coefficients were estimated for the same four market segments as
for non-resident non-business trips. the Boston Logan model. In addition, separate ASCs were estimated
for each mode for trips originating within the Portland metropolitan
The final model included eight modes: private vehicle parked at area (termed internal trips) and those originating outside the metro-
the airport for the trip duration, drop off at the airport by private ve- politan area (termed external trips). Two different sets of model coef-
hicle, rental car, taxi and limousine (combined), hotel shuttle, ficients were estimated for each market segment. The first set (termed
shared-ride van and scheduled bus, light rail, and express bus. The Model 1) assumed that the ASCs for the light rail and express bus
use of private automobile parked at the airport for the trip duration modes would be the same as those for shared-ride van and RAZ bus (a
(termed auto park) was restricted to residents of the region, whereas scheduled bus service between the airport and downtown Portland lo-
the use of a rental car was restricted to non-residents of the region. cations operated by RAZ Transportation, a Gray Line affiliate). The
In the case of the light rail and express bus alternatives it was second set (termed Model 2) used the SP data to estimated separate
assumed that travelers would be dropped off at the station or stop by ASCs for the light rail and express bus modes. The documentation
a private vehicle. on the initial model estimation by Cambridge Systematics provides
t-statistics for the coefficient estimates, but the documentation of the
As part of the overall study of alternative airport access modes, final model does not.
a review of the experience of other U.S. airports with a range of air-
port ground access strategies was undertaken (Coogan 1997). This The model documentation does not explain why ASCs were not
report included statistics on the ground access mode shares of vari- determined for taxi and limousine use for resident business trips
ous airports that had implemented ground transportation services from external origins or for shared-ride van and RAZ bus use for
similar to those being considered for Portland, as well as a discus- non-resident business trips from external zones, but were deter-
sion of the operational experience of those airports with the ground mined for the other three market segments in each case. Indeed, it is

129
TABLE D18 TABLE D20

PORTLAND GROUND ACCESS STUDY RESIDENT PORTLAND GROUND ACCESS STUDY NON-RESIDENT
BUSINESS MODEL COEFFICIENTS BUSINESS MODEL COEFFICIENTS
Coefficient Model 1 Model 2 Coefficient Model 1 Model 2

Variables Variables
Drop off cost ($) –0.0195 –0.0195 Drop off cost ($) –0.0082 –0.0082
Travel time (minutes) –0.0176 –0.0176 Travel time (minutes) –0.0073 –0.0073
Cost/ln(income) $/ln($K) –0.2185 –0.2185 Cost/ln(income) $/ln($K) –0.0913 –0.0913
Constants (auto park base) Constants (rental car base)
Internal trips Internal trips

Auto drop off 0.85 0.85 Auto drop off –0.50 –0.50
Taxi and limousine –1.162 –1.272 Taxi and limousine –0.9135 –1.2335
Van, RAZ bus, and hotel shuttle –0.988 –1.258 Hotel shuttle –0.8865 –0.9965
Light rail (auto drop off) –0.988 –1.258 Van and RAZ bus –0.9365 –1.3965
Express bus (auto drop off) –0.988 –1.258 Light rail (auto drop off) –0.9365 –0.8009
External trips Express bus (auto drop off) –0.9365 –0.9960
Auto drop off –0.85 –0.85 External trips

Taxi and limousine N/A N/A Auto drop off –0.30 –0.30
Van, RAZ bus, and hotel shuttle 2.312 0.742 Taxi and limousine –1.0635 –2.2135
Light rail (auto drop off) 2.312 0.742 Van and RAZ bus N/A N/A
Express bus (auto drop off) 2.312 0.742 Light rail (auto drop off) –1.287 –1.4665
N/A = not available. Express bus (auto drop off) –1.287 –2.4165
N/A = not available.
TABLE D21
TABLE D19 PORTLAND GROUND ACCESS STUDY NON-RESIDENT
PORTLAND GROUND ACCESS STUDY RESIDENT NON-BUSINESS MODEL COEFFICIENTS
NON-BUSINESS MODEL COEFFICIENTS
Coefficient Model 1 Model 2
Coefficient Model 1 Model 2
Variables
Variables
Drop off cost ($) –0.0082 –0.0082
Drop off cost ($) –0.0235 –0.0235 Travel time (minutes) –0.0092 –0.0092
Travel time (minutes) –0.0264 –0.0264 Cost/ln(income) $/ln($K) –0.0716 –0.0716
Cost/ln(income) $/ln($K) –0.2170 –0.2170
Constants (rental car base)
Constants (auto park base)
Internal trips
Internal trips Auto drop off 0.10 0.10
Auto drop off –0.30 –0.30 Taxi and limousine –1.754 –1.574
Taxi and limousine –2.068 –1.538 Hotel shuttle –0.246 –0.046
Van, RAZ bus, and hotel shuttle –1.632 –1.362 Van and RAZ bus –0.596 –0.956
Light rail (auto drop off) –1.632 –0.3654 Light rail (auto drop off) –0.596 –0.914
Express bus (auto drop off) –1.632 –1.5281 Express bus (auto drop off) –0.596 –0.935
External trips External trips
Auto drop off –0.80 –0.80 Auto drop off –0.50 –0.50
Taxi and limousine –2.188 –2.188 Taxi and limousine –1.304 –2.054
Van, RAZ bus, and hotel shuttle 2.368 –0.652 Van and RAZ bus –0.346 –1.206
Light rail (auto drop off) 2.368 –2.3447 Light rail (auto drop off) –0.346 –1.206
Express bus (auto drop off) 2.368 –3.8869 Express bus (auto drop off) –0.346 –0.6862

130
not clear why the RAZ bus was included as an option for external worker and an annual income of $50,000 would have a wage rate of
trips at all or why the hotel shuttle was considered as an option for $25/h, whereas a household with two workers and an annual income
resident trips. of $150,000 would have an average wage rate of $37.50/h. There-
fore, the implied values appear to be in the general range of the
There are a number of counterintuitive or surprising values for wage rate.
the ASCs. Because the ASCs for taxi and limousine have a gener-
ally higher disutility than auto drop off suggests that the perceived The implied values of the ASCs, expressed as equivalent min-
cost of taxi and limousine fares have been underestimated. Also, it utes of travel time, appear implausibly large for many modes. For
is not clear why the perceived relative disutility of existing modes example, the relative disutility of most public modes for non-resi-
should change between Model 1 and Model 2 when the values for dent trips compared with auto drop off, apart from any differences
the light rail and express bus were adjusted using the SP data. The in cost and travel time, is equivalent to well over an hour of travel
large positive value of the ASC for shared-ride van and RAZ bus for time and more than 3 h of travel time in the case of taxi or limousine
resident trips from external zones seems inconsistent with the val- use for non-business trips from internal zones, or taxi, limousine, or
ues for internal trips. express bus use for business trips from external zones. The large dif-
ferences in the auto drop off constant compared with auto park (for
The implied values of the model coefficients for Model 2 are resident trips) and rental car (for non-resident trips) between busi-
shown in Table D22. Because the inclusion of household income in ness and non-business trips suggests that these constants are ac-
the cost term results in implied values of time that vary with aver- counting for more than just the inherent differences in the comfort
age household income, these values have been calculated for aver- and convenience of the various modes.
age annual household incomes of $50,000 and $150,000. Although
the resulting values of time seem consistent for resident and non- The ratio of the auto drop off cost coefficient to the cost coeffi-
resident travelers for each trip purpose, this is a consequence of the cient for all other costs suggests that the auto drop off costs (pri-
way the model was estimated, and the lower value of time for busi- marily the time of the driver) are valued at between about one-third
ness trips compared with non-business trips is counterintuitive. and one-half of the other costs. This is not unreasonable, because
some air travelers may consider being taken to the airport by others
The relatively small change in the value of time between a zone as essentially without cost to them. However, it is worth noting that
with an average annual household income of $50,000 and one with the assumed values of time for the drivers (twice as high for busi-
an average annual household income of $150,000 per year is a con- ness trips as for non-business trips) are inconsistent with the esti-
sequence of the use of the logarithmic transform. For comparison mated values of time for the air passengers, which are about half
with the implied values shown in Table D22, a household with one again higher for non-business trips than business trips.
TABLE D22
IMPLIED VALUES OF PORTLAND GROUND ACCESS STUDY COEFFICIENTS
Resident Resident Non- Non-Resident Non-Resident

Parameter Business Business Business Non-Business
$50,000 avg. h/h income 19 29 19 30

$150,000 avg. h/h income 24 37 24 39
Auto Drop Off Cost Ratio
$50,000 avg. h/h income 0.35 0.42 0.35 0.45

$150,000 avg. h/h income 0.45 0.54 0.45 0.57
Constants (minutes)
Internal trips
Auto drop off –48 11 68 –11
Taxi and limousine 72 58 169 171
Hotel shuttle 71 52 137 5
Van and RAZ bus 71 52 191 104
Light rail (auto drop off) 71 14 110 99
Express bus (auto drop off) 71 58 136 102
External trips
Auto drop off 48 30 41 54
Taxi and limousine N/A 83 303 223
Van and RAZ bus –42 25 N/A 131
Light rail (auto drop off) –42 89 201 131
Express bus (auto drop off) –42 147 331 75
Notes: avg. h/h income = average household income. N/A = not available.

131
Model Fit Private vehicle—parked 17%

Rental car 19%
The initial Cambridge Systematics model estimation results include Taxi 7%
t-statistics for the coefficients for each variable and measures of Limousine 1%
overall goodness-of-fit of the model, including the final value of the Hotel courtesy shuttle 2%
log likelihood and the improvement over the log likelihood with Shared-ride van 2%
zero coefficients or constants only. The overall improvement in the Shuttle bus from train 1%
goodness-of-fit of the model from the inclusion of the continuous Public transit bus <1%
variables is not particularly large. Other <1%
Market Segmentation Residents—Business trips
Residents—Personal trips
Model Application Visitors—Business trips
Visitors—Personal trips
As noted, the model was primarily developed to predict ridership on Explanatory Variables Travel time (private vehicles)
the planned extension of the Portland MAX light rail system to Travel time (rail transit)
serve the airport, as well as to examine alternative ground access Travel time (bus transit)
measures, including development of an express bus service and Walk distance
charging private vehicles a fee to drop off or pick up air passengers Wait time
at the airport. Travel cost
Household income
The approach to applying the model follows the traditional four-
step urban transportation planning approach, with the number of
trips generated by the airport determined from the airport traffic Description
forecasts. A trip distribution model (termed an origin location
model in the documentation) calculates the number of air parties This model was developed by Dowling Associates to estimate the
beginning their access trip in each zone. The mode choice model is ridership on a planned automated people-mover (APM) to connect
then applied to these trips to calculate the number of vehicle and air the airport to a nearby Santa Clara Valley Transportation Authority
passenger trips from each zone. (VTA) light rail line (Dowling Associates 2002). The model was
estimated using data from an air passenger survey performed at the
airport for the Bay Area Metropolitan Transportation Commission
Documentation (MTC) in 1995 (Franz 1996) and supplemented with the results of
stated preference surveys that were conducted as part of the study to
Bowman, J.L., Portland PDX Airport Access Project Mode Choice Models, determine how air passenger mode choice might be influenced by the
memorandum to Keith Lawton, Metro, Cambridge Systematics, Inc., availability of the people-mover, as well as to overcome the problem
July 28, 1997. that there were very few users of the light rail line in the 1995 survey
Cambridge Systematics, Inc., Portland International Airport Alternative sample. MNL models were estimated for the same four market seg-
Mode Study, Prepared for the Port of Portland, Portland, Ore., Oct. 1998. ments used in the Oakland Airport BART Connector Study: resident
Coogan, M.A., The Peer Airport Analysis Report, prepared for the Port of business trips, resident personal trips, visitor business trips, and vis-
Portland, Apr. 9, 1997. Included as Appendix D to Cambridge System-
itor personal trips. Each market segment model included six modes:
atics, Inc., Portland International Airport Alternative Mode Study,
Prepared for the Port of Portland, Portland, Ore., Oct. 1998. private vehicle, rental car, scheduled airport bus, shared-ride van
Portland Metro, PDX Ground Access Study Model Summary, Prepared by the (termed door-to-door shuttle), taxi, and public transit. In addition, the
Travel Forecasting Staff, Portland, Ore., undated (May 1998), revised visitor segment models included hotel shuttle.
June 2001.
The model also included an airport employee segment as dis-
cussed further here. This only considered two modes: private vehi-
D7 SAN JOSÉ INTERNATIONAL AIRPORT cle and public transit. Because the primary purpose of the model was
MODEL to estimate ridership on the planned APM, the model allowed up to
four connecting transit routes and developed separate fare and travel
Summary times for each. Those routes involving the use of the APM to access
the VTA light rail line were given a separate ASC from other transit
Airport Norman Y. Mineta San José International routes to reflect the greater attractiveness of the APM based on the
Airport stated preference survey.
Model Developer Dowling Associates, Inc.
Date Developed 2002
Market Addressed Air passengers Explanatory Variables
Model Type Combined revealed preference and stated
preference data Independent variables consisted of the automobile travel time,
Model Structure Multinomial logit transit travel time by rail, transit travel time by bus, waiting time,
Survey Data Used 1995 Metropolitan Transportation Com- walking distance, and cost. The cost variable for personal trips
mission Air Passenger Survey; Supple- was divided by the annual household income raised to the power
mentary stated preference survey at San 1.5. Only one set of ASCs for private car was presented in the re-
José International Airport port, making no distinction between air parties being dropped off
Airport Profile Total annual passengers (2005): 10.6 mil- and those parking for the duration of the air trip. This resulted
lion from a limitation in the 1995 air passenger survey, which also did
Percentage O&D: 91% not make this distinction. It was assumed in the model estimation
Ground access mode split (2002 MTC that residents using private vehicles parked at the airport, whereas
Air Passenger Survey): visitors were dropped off. The parking cost was included in the
Private vehicle—drop off 49% parking utility function for resident trips, whereas a “drop-off”

132
factor was included in the private vehicle utility function for visi- income of $55,000, which is stated in the study report to be the av-
tor trips to account for the inconvenience for drivers dropping off erage annual household income for potential transit users at San
air passengers (the details of this factor are not given in the report). José International Airport based on data for Santa Clara County
It is possible to use the estimated model to predict the choice of from the Association of Bay Area Governments (it is unclear what
resident air passengers being dropped off by including both modes “potential transit users” means in this context or how the Associa-
in the model and assuming that the ASC is the same for both drop tion of Bay Area Governments could determine the household in-
off and park. come of such users, but the value provides a reasonable point of
comparison).
Model Coefficients The implied values of the various components of travel time are
quite low by comparison with the values typically found in air pas-
The approach taken in estimating the model followed that used in the senger ground access mode choice models (and air travel models
Oakland International Airport BART Connector Study with the generally). However, because these implied times came directly
model coefficients for the continuous variables adopted directly from from the coefficients estimated by Harvey (1988) using 1985 data,
an earlier airport ground access mode choice model for Bay Area this is hardly surprising. Because the implied value of rail transit
developed by Harvey (1988). The values of the ASCs were then esti- travel time is lower than travel time by private auto is counterintu-
mated to fit the model to the mode use data from the 1995 MTC Air itive. Although the higher implied value for bus transit travel time
Passenger Survey. The estimated model coefficients presented in the is consistent with typical experience in urban travel models, the
study report are shown in Table D23. difference from travel time by private auto is surprisingly small,
particularly for visitor personal trips. Similarly, the implied values
As discussed earlier in the description of the Oakland Airport of the ASCs are quite low compared with those typically found in
Connector study model, the use of coefficients from a model that air passenger ground access mode choice models and the differ-
was estimated on much earlier data, without any adjustments for ences between the values for different modes are surprisingly small
changes in the implied values of time, introduces significant distor- and in several cases intuitively unreasonable. For example, it
tions in the model that are compounded when the model is used to makes no sense that the implied value of the ASC for transit or
predict future mode use. shared-ride van for resident business trips would be greater than
that for taxi, which provides significantly greater comfort and con-
The implied values of the estimated coefficients are shown in venience. Similarly, it seems quite implausible that scheduled bus,
Table D24. As with the Oakland Airport model, the implied values transit, or shared-ride van would be viewed by visitors on business
of the ASCs are expressed relative to taxi, because the private vehi- trips as more attractive than being dropped off at the airport by pri-
cle mode is different for residents and visitors. These implied val- vate automobile.
ues are expressed in dollars and represent the difference in cost be-
tween the mode and a taxi that would be required for travelers to be What is most likely distorting the values of the estimated coef-
indifferent between use of the two modes if travel times were the ficients is a failure to control for the need for and availability of
same. Because the implied values for personal trips depend on the different modes for different air parties. Visitors who are not stay-
household income, the values have been calculated for a household ing with residents of the area may not have anyone who can take
TABLE D23
SAN JOSÉ INTERNATIONAL AIRPORT MODEL COEFFICIENTS
Coefficient Resident Resident Visitor Visitor Personal
Business Personal Business
Variables
Auto time (minutes) –0.071 –0.044 –0.068 –0.039
Rail transit time (minutes) –0.053 –0.031 –0.050 –0.029
Bus transit time (minutes) –0.093 –0.051 –0.089 –0.045
Walk distance (miles) –5.17 –3.28 –4.69 –2.94
Cost (cents) –0.00277 –1.04/ –0.00256 –0.973/
(HHINC)1.5 (HHINC)1.5
Constants
Private vehicle — — — —
Rental car –2.9 –4.1 3.9 1.0
Scheduled bus –2.3 –2.7 1.2 –0.8
Transit (does not use APM) –1.3 –2.0 0.9 –0.4
Transit (uses APM) –1.2 –1.8 0.8 –0.3
Shared-ride van –1.2 –1.4 0.6 –0.1
Hotel shuttle N/A N/A 0.0 –3.1
Taxi –1.4 –1.3 1.1 0.1
Notes: HHINC = annual household income in thousands of dollars; N/A = mode is not available for this market
segment; APM = automated people-mover.

133
TABLE D24
IMPLIED VALUES OF SAN JOSÉ INTERNATIONAL AIRPORT MODEL
COEFFICIENTS
Parameter Resident Resident Visitor Visitor

Business Personal Business Personal
Auto time 15 10 15 10
Rail transit time 11 7 11 7
Bus transit time 20 12 19 11
Walk time 56 39 55 37
Wait time 23 18 21 18
Constants (dollars)
Private vehicle 5.1 5.1 –4.3 –0.4

Rental car –5.4 –11.0 10.9 3.8
Scheduled bus –3.3 –5.5 0.4 –3.8
Transit (does not use APM) 0.4 –2.8 –0.8 –2.1
Transit (uses APM) 0.7 –2.0 –1.2 –1.7
Shared-ride van 0.7 –0.4 –2.0 –0.8
Hotel shuttle N/A N/A –4.3 –13.4
Taxi — — — —
Notes: Implied values of personal trips calculated for an annual household income of $55,000 per year.
Implied value of walk time based on a walking speed of 3 mph.
them to the airport and therefore either rent a car to meet their local utility functions included walking distance, waiting time, and cost.
transportation needs or use public modes. To explain these choices The coefficient values were not estimated from employee survey
in a situation when the model has assumed that being dropped off data, but were adapted from the Santa Clara County countywide
by private vehicle is an option that is available, the values of the travel model for home-based work trips. The employee model doc-
ASCs have to be increased. Similarly, residents of the region gen- umentation does not state what assumptions were made about the
erally have access to a private vehicle and therefore do not need to cost for employee parking.
rent a car.
The model coefficients and implied values are shown in Table
D25. In this case, the implied value of the transit ASC is relative to
Model Fit private vehicle, because a taxi was not considered in the employee
mode choice model.
No goodness-of-fit statistics for the estimated coefficients or the
overall model were provided in the report. The implied values of travel time are extremely low, which is
most likely the result of the adoption of the model coefficients from
Model Application TABLE D25

SAN JOSÉ INTERNATIONAL AIRPORT EMPLOYEE MODEL
The model was developed as part of planning studies for a proposed
APM that will connect the passenger terminal at San José Interna- Model
tional Airport to a nearby stop on the Santa Clara VTA light rail line Parameter Coefficient Implied Value
that runs about a block to the east of the airport and connects in
Variables $/hour
downtown San José with other lines on the light rail network serv-
ing the Santa Clara Valley. A possible future extension of the APM Travel time (minutes) –0.02545 4.2
to the west side of the airport would also serve the nearest station
on the Caltrain commuter rail line that serves the Peninsula corridor Walk distance (miles) –1.17 9.6
between San José and San Francisco. Wait time (minutes) –0.05854 9.6
Cost (cents) –0.00366
Airport Employee Mode Use Constants $

Private vehicle — —
The mode choice model included a market segment for airport em-
ployees that considered only two modes, private vehicle and transit, Transit (does not use APM) –1.4 –3.8
although separate ASCs were developed for transit trips using the
Transit (uses APM) –1.3 –3.6
APM and those without the APM being available. The employee
model used the same coefficient value for both highway and transit Notes: Coefficients derived from countywide travel demand model.
travel time, including rail and bus. In addition to travel time, the Implied value of walk time based on a walking speed of 3 mph.

134
the Santa Clara countywide travel demand model. Although these Description
have similar values for travel time components (and walking dis-
tance) to those for the MTC regional travel demand model dis- In May 2003, Transport Canada issued a Request for Business Case
cussed in the description of the Oakland Airport Connector model for a public–private partnership to develop an Air Rail Link between
earlier, the coefficient for cost is significantly greater, suggesting Toronto Lester B. Pearson International Airport (LBPIA) and
that the countywide model is based on travel costs in even earlier Toronto Union Station (Request for Business Case . . . 2003). Subse-
constant year dollars than the MTC model. Without adjusting for quently, a private sector consortium, Union Pearson AirLink Group,
changes in prices and income, the direct use of these coefficients was selected to finance, design, construct, and operate the rail link in
and their implied value of time will tend to overstate the attractive- association with eight public and three private stakeholders, includ-
ness of transit. ing the Greater Toronto Airports Authority and several regional and
national rail and transit agencies (Borges 2006). In preparation for the
Request for Business Case, a revenue and ridership forecasting study
Documentation was undertaken in 2002 (Halcrow Group 2002). This study included
a stated preference air passenger survey and the development of a
Dowling Associates, Inc., San Jose International Airport Transit Connection mode choice model to predict the diversion of airport access trips
Ridership, Final Report, Prepared for San Jose International Airport, from existing modes to the proposed new rail link.
Lea+Elliott and Walker Parking, Oakland, Calif., June 2002.
Franz, J.D., 1995 Metropolitan Transportation Commission Airline The stated preference survey was carried out in February 2002
Passenger Survey, Final Report, Prepared for the Metropolitan Trans-
portation Commission, Oakland, Calif., by J.D. Franz Research, Feb.
in the terminal departure lounges, and interviewed 2,566 passengers
1996. of whom 1,927 (75%) were not connecting between flights. The sur-
Harvey, G., ACCESS: Models of Airport Access and Airport Choice for the vey collected data on the air party characteristics, ground trip origin,
San Francisco Bay Region—Version 1.2, Prepared for the Metropolitan and the ground access mode for the current trip. Some 807 respon-
Transportation Commission, Berkeley, Calif., Dec. 1988. dents were identified as potential air rail link users, based on their
ground trip origins, and completed a stated preference question-
naire. The results of the survey were then used to estimate a set of
D8 TORONTO AIR RAIL LINK REVENUE mode choice models.
AND RIDERSHIP STUDY
The Greater Toronto region was divided into a system of 34
Summary zones, with the 2 downtown Toronto zones further subdivided into
10 zones. Peak period highway and transit travel times between
Airport Toronto Lester B. Pearson International each of these zones and the airport and potential Air Rail Link sta-
Airport, Canada tions were extracted from the city of Toronto Greater Toronto
Model Developer Halcrow Group Limited Area transportation network model. Off-peak highway travel
Date Developed 2002 times were assumed to be 6% less than peak period travel times
Market Addressed Air passengers based on an analysis of the travel times reported in the stated pref-
Model Type Combined revealed preference and stated erence survey. Travel costs were largely derived from actual costs
preference data reported in the stated preference survey. Taxi fares for zones with
Model Structure Binomial logit diversion no reported data were derived from a distance-based regression
Survey Data Used February 2002 Air Passenger Survey using the reported fares from zones for which there were data. Pri-
Airport Profile Total Annual Passengers (2005): 29.9 vate vehicle fuel costs were calculated on the basis of 4.6 cents per
milliona kilometer.
Percentage O&D: 75%b
Ground access mode split (2005 air pas- The mode choice model consists of a set of binomial logit mod-
senger survey):c els (described in the model documentation as logistic, probabilistic,
Private vehicle—drop off 45% diversion models), each of which models the choice between an ex-
Private vehicle—parked 13% isting mode and the planned rail link. If such models had been de-
Rental car 9% veloped for every existing mode, they could have been combined
Taxi and limousine 24% into a MNL model. However, air travelers using a hotel bus or rental
Courtesy vehicles 6% car to access the airport were excluded from the diversion analysis,
Scheduled airport bus 2% because users of the hotel bus were assumed to have a door-to-door
Public transit 1% service that was effectively free, whereas those using a rental car
Source: Greater Toronto Airports were assumed to require the car for other purposes during their visit
Authority and thus not considering the use of other modes. The use of the bi-
aPassenger Statistics as of June 30, 2007. nomial logit diversion models has another consequence, namely
bEstimate quoted in Halcrow Group that the models only predict the diversion from existing modes to
(2002). the rail link and do not allow for future shifts in use between exist-
cPersonal communication, Marc Turpin, ing modes owing to changes in future values of travel time and cost
6/29/07. for the different modes or changes in travel patterns in the Toronto
Market Segmentation Residents—Business trips market. Although the purpose of the analysis was to predict the rid-
Residents—Non-business trips ership on the rail link, because this is calculated on the basis of the
Non-residents—Business trips diversion from existing modes, any errors in the predictions of the
Non-residents—Non-business trips future use of existing modes without the rail link will in turn affect
Explanatory Variables In-vehicle travel time the ridership projections for the rail link.
Service headway
Travel cost The model documentation does not explicitly discuss the modes
Dummy variables: travel party size that were included in the analysis; however, based on the presenta-
(accompanied/travel alone); checked tion of the survey results it would appear that the analysis was based
baggage on four modes: private vehicle parked at the airport for the trip du-

135
ration (termed drive & park), drop off by private vehicle (termed models for each existing access mode indirectly addressed some of
driven in car), taxi/limousine, and transit/airport bus. these effects, because air passengers being dropped off by private
vehicle would largely have begun their trip from a private residence,
In addition to estimating a formal mode choice model, the study while those using taxi or airport bus would be more likely to have
included a benchmark comparison analysis that examined the mode begun their trip from a hotel or place of business. The primary rea-
share of existing airport rail links in 24 cities in the U.S., Europe, son for including the type of trip origin in an airport access mode
and Australia. This analysis developed cross-sectional regression choice model is to account for the effect of this on available access
relationships that expressed the rail mode share in terms of a series modes (principally the availability of someone who can take the air
of market and geographical characteristics, such as the percentage party to the airport). By segmenting the mode choice analysis by the
of air passengers with central city origins, the distance of the airport existing access mode, this effect is already considered in the data.
from the central city, and the ratio of the rail travel time to the taxi However, this approach does have one drawback. Because the share
travel time from the city center. These relationships were then used of existing access modes in the absence of the rail link is required to
to predict the corresponding rail mode share for Toronto using the perform the analysis, which is necessarily derived from existing
same regional characteristics. The resulting range of rail mode conditions, the approach assumes that the distribution of air passen-
shares (which varied with the characteristic chosen) was compared ger trips across the different types of trip origin will remain un-
with the results of the formal mode choice modeling process, to pro- changed in the future.
vide a reality check on the modeling analysis. The details of this
analysis are presented in the model documentation (Halcrow Group The model utility functions also do not consider the household
2002). They are not presented here, because they are not directly income of the air travelers. Because the rail link will offer a differ-
relevant to the details of the mode choice model, but the results of ent combination of cost and travel time from existing modes, the
the comparison are discussed below. attractiveness of this service to air travelers from a given geographic
zone is likely to vary depending on their income. This in turn is
likely to have a significant effect when considering the impact on
Explanatory Variables revenue of changing fare levels.
The mode choice model relationships used three continuous vari-
ables: the in-vehicle travel time on each mode, the service headway Model Coefficients
for the mode, and the travel cost involved in using the mode. In addi-
tion, the utility function for the rail link included two dummy vari- The estimated model coefficients are shown in Table D26, with the
ables: one that indicated whether the air traveler was accompanied associated implied values of the travel time components and
(whether by other members of the air travel party or by well-wishers) dummy variables shown in Table D27.
or traveling alone and one indicating whether the air traveler(s) in-
tended to check any bags. These dummy variables were chosen based ASCs were initially included in the model utility functions; how-
on survey results that indicated that air travelers who traveled to the ever, these were found to not be statistically significant and were
airport alone or did not intend to check any bags were more likely to dropped from the model. This is surprising given the relatively sim-
use the rail link. This seems reasonable, because larger travel parties ple form of the utility functions and the absence from the model of
or those with significant amounts of baggage are likely to find other such factors as household income and the access time involved in
modes more convenient or cost-effective. In particular, being accom- reaching the rail link station.
panied to the airport by well-wishers will often mean that someone is
available to take the air party to the airport by private vehicle. The implied values of in-vehicle time (strictly the implied val-
ues of reduced in-vehicle times) appear reasonable. The values for
The model structure did not directly consider the type of trip business trips are approximately twice those for non-business trips,
origin. However, the approach of developing separate diversion which is not unreasonable, particularly given that income is not
TABLE D26
TORONTO AIR RAIL LINK MODEL COEFFICIENTS
In-vehicle time (minutes) –0.0494 –0.0665 –0.0435 –0.0462
(–4.2) (–7.6) (–2.8) (–3.9)
Headway (minutes) –0.1180 –0.0665 –0.1678 –0.0692
(–4.5) (–3.7) (–7.0) (–3.1)
Cost ($) –0.0557 –0.1382 –0.0368 –0.0820
(–5.9) (–8.6) (–7.3) (–4.5)
Dummy Variables
Accompanied to airport N/S –0.2736 N/S –0.7951
(–2.3) (–4.8)
Checked baggage N/S –0.4284 N/S –0.5120
(–2.8) (–2.5)
Number of cases 173 348 167 162
Notes: N/S = not statistically significant; t-statistics shown in parentheses.

136
TABLE D27
IMPLIED VALUES OF TORONTO AIR RAIL LINK MODEL COEFFICIENTS
In-vehicle time 53 29 71 34
Waiting time 254 58 547 101
Dummy Variables ($)
Accompanied to airport N/A 2 N/A 10
Checked baggage N/A 3 N/A 6
explicitly included in the model. Business travelers will tend to have dence level; highly so in the case of the cost terms and most of the
higher incomes on average than non-business travelers. The implied travel time terms.
values of in-vehicle time for non-residents are somewhat higher
than for residents (33% higher for business trips and 17% higher for
non-business trips). Because it can be assumed that a fairly high Model Application
proportion of non-residents are from the United States, this is also
not unreasonable. The binomial logit diversion models were used to predict the rider-
ship on the Air Rail Link for future years following the introduction
However, the implied values of waiting time appear surprisingly of service under a range of different service scenarios. These in-
high except for resident non-business trips, for which waiting time cluded potential intermediate stations between Union Station and the
is perceived as twice the inconvenience of in-vehicle time, which airport, as well as different fare and frequency assumptions. The pre-
corresponds to typical experience in urban travel demand models. dicted ridership on the Air Rail Link during the first two years after
The implied value for non-resident non-business trips is three times the start of service was adjusted to allow for an expected ramp-up in
the implied value of in-vehicle time, which is higher than would the diversion from other modes based on the experience following
normally be expected. The implied values of waiting time for busi- the start of service on the Heathrow Express rail link to London
ness trips are implausibly high, at almost five times the value of Heathrow Airport.
in-vehicle time for resident trips and almost eight times the value of
in-vehicle time for non-resident trips. Because the non-resident The models were applied to a range of different service scenar-
business travelers already had a fairly high value of in-vehicle time, ios, including a base case with an intermediate stop at the Dundas
the resulting implied value of waiting time is more than $500/h. Station of the Toronto Transit Commission Bloor–Danforth subway
Although business travelers may be particularly averse to waiting, line, as well as nonstop service from Union Station to LBPIA and a
it would be surprising if they found waiting that onerous compared reduced headway for both the nonstop and one-stop service with
with travel time. Indeed, such a ratio implies that they would be trains departing every 20 min rather than every 15 min. Other sensi-
willing to incur an additional hour of travel time to avoid a 10 tivity tests were performed to explore the effect of changes in rail
minute wait, which makes no sense. service travel times, changes in access times for travelers to reach the
rail link stations or egress times from the airport station to the pas-
This is not a trivial issue, because overstating the disutility of senger terminal, changes in highway travel times, changes in the
waiting time in the model will result in underestimating the demand value of time for air passengers, and the introduction of another in-
for the rail link if headways are increased and conversely overesti- termediate station at a proposed entertainment complex located at
mating the demand if headways are reduced. This will overestimate the site of the Woodbine Racetrack adjacent to the planned rail route.
the willingness of travelers, particularly business travelers, to pay a
higher fare to have a shorter headway and could result in an operat- To understand the potential uncertainty in the prediction of rider-
ing plan that runs too many trains at too high a fare. ship and revenues over the first 30 years of the project, a risk analy-
sis was performed that took into account the confidence in the mode
In contrast to the implied values of waiting time, the implied choice model relationships as well as the predicted future values of
values of the disutility of the rail link for travelers who are accom- the airport traffic, changes in market segments, trip end distribution,
panied to the airport or have checked baggage are relatively low, and transportation system variables. This analysis generated distrib-
varying between $2 and $10, and being higher for non-residents utions of predicted future levels of ridership and revenue that were
than residents. These values can be interpreted as the reduction in used to present both 20% and 80% confidence values for both mea-
fare that would be needed to offset the disutility and result in the rail sures. This recognized that private investors in the planned Air Rail
link being perceived to be as attractive as it is for those traveling Link would want to adopt a conservative view of the likely return on
alone with only carry-on bags. Thus, the fare reduction needed to investment, whereas design of the necessary facilities would need to
offset the combined disutility of being accompanied to the airport consider a more aggressive forecast of possible ridership.
and having checked bags is approximately $16 for non-residents
and $5 for residents. This would represent a significant percentage
of the proposed fare of $20. Comparison with Benchmarking Analysis
A particularly interesting aspect of the Air Rail Link Revenue and

Model Fit Ridership study is the comparison of the predictions of the mode
choice model and the findings of the benchmarking analysis. In
The model documentation presents t-statistics for each of the spite of both the very broad scope of the benchmarking analysis and
explanatory variables, as shown in Table D26. However, there is no the technical concerns with the mode choice model discussed
discussion of the overall fit of the model relationships to the survey earlier, the results of both approaches gave surprisingly consistent
data. The t-statistics for each of the variables included in the model results. The benchmarking analysis suggested that the likely mode
utility functions are all statistically significant at the 95% confi- share of originating air passenger traffic at LBPIA using the rail link

137
might varying between about 7% and 14%, with a central estimate Travel cost (fares, parking, automobile
of 9.3%. The forecasts of ridership applying the mode choice model operating cost)
to the base case service scenario gave a mode share of 9.3% for Driving distance
2001 traffic levels, rising to 9.7% for 2021 traffic levels. Interchange penalties
Documentation Description
Borges, H., Air Rail Link: Pearson Airport–Union Station, presented to the As part of the SERAS study undertaken for the U.K. Department of
Chartered Institute of Logistics and Transport, Transport Canada, Ottawa, Transport, Local Government and the Regions, a set of surface
ON, Apr. 11, 2006 access models were developed that included an air passenger mode
Halcrow Group Limited with Cansult Ltd., Air Rail Link from Lester B.
Pearson International Airport to Union Station: Revenue & Ridership
choice model, as well as an airport employee trip distribution model
Study, Report T8080-01-1213, Final Report, Prepared for Transport and an airport employee mode choice model.
Canada, Ottawa, ON, May 2002.
Request for Business Case: Air Rail Link from Toronto–Lester B. Pearson The structure of the air passenger mode choice model is stated to
International Airport to Toronto Union Station, Transport Canada, Ottawa, be the same as the Heathrow Surface Access Model (HSAM)
ON, May 2003. developed by the MVA Consultancy for the British Airports
Authority. This is a NL model that covers 12 defined ground access
modes and has separate coefficients for six market segments:
D9 UNITED KINGDOM SERAS STUDY AIR
PASSENGER SURFACE ACCESS MODEL • U.K. business passengers on domestic trips
• U.K. business passengers on international trips
Summary • U.K. leisure passengers on domestic trips
• U.K. leisure passengers on international trips
Airports London [Heathrow (LHR), • Non-U.K. passengers on business trips
area airports Gatwick (GTW), Stansted • Non-U.K. passengers on leisure trips.
(STN), Luton (LTN)]
Model Developer Halcrow Group Ltd. The 12 ground access modes consist of:
Date Developed 2002
Market Addressed Air passengers • Drop off by private automobile (termed Kiss & Fly)
Model Type Revealed preference data • Private automobile parked at airport (termed Park & Fly)
Model Structure Nested logit • Rental car (termed Hire Car)
Survey Data Used U.K. Civil Aviation Authority air passen- • Taxi
ger surveys 1992–1997 • Local bus and intercity coach
Airport Profile LHR GTW STN LTN • London Underground
Source: Total 66.8 31.8 21.7 8.9 • Coach links to British Rail stations (BR Coach)
United Kingdom annual m m m m • Dedicated premium rail service (Heathrow Express)
Civil Aviation passengers • New standard British Rail services
Authority, CAA (2005): • Alternative premium rail service
Passenger Survey Percentage 66% 84% 89% 94% • Charter coach (including hotel bus)
Report 2005 O&D: • Inter-airport transfer coach.
m = million.
Ground access mode split (2005): Although the term British Rail is used in the model documenta-
Private 34% 51% 48% 56% tion, these services are now provided by private companies
vehicle (e.g., Great Western Trains) and British Rail as such no longer exists.
Rental 3% 2% 4% 3% The Park & Fly mode was assumed to only be available to U.K. pas-
car sengers and was substituted by the Hire Car mode for non-U.K.
Taxi/ 26% 14% 9% 13% passengers. The Heathrow Express is a dedicated non-stop service
minicab between London Paddington Station and Heathrow Airport. The al-
London 13% — — — ternative premium rail service was assumed to be a similar service
Under- from another London station, whereas the new standard British Rail
ground service would provide direct rail service to the airport using conven-
Rail 11% 26% 25% 18% tional rail equipment with intermediate stops. The hotel bus service
Bus/ 13% 7% 14% 10% refers to a system of shuttle buses that serve local hotels near
coach Heathrow Airport. However, the use of charter coach, hotel bus, and
Other <1% <1% 1% <1% interairport transfer coach was not explicitly represented in the
Market Segmentation U.K. business passengers on domestic model, but instead the use of these services was determined inde-
trips pendently and the resulting vehicle trips added to those determined
U.K. business passengers on interna- using the mode choice model. Thus, the mode choice model for each
tional trips market segment consisted of nine modes.
U.K. leisure passengers on domestic trips
U.K. leisure passengers on international The nesting structure of the model is shown in Figure D6. There
trips are several levels of nest, particularly for the different rail modes.
Non-U.K. passengers on business trips The utility functions for each mode use a generalized cost approach
Non-U.K. passengers on leisure trips that considers the travel time and out-of-pocket costs (fares, park-
Explanatory Variables Travel time (in-vehicle time, walk access ing, and private automobile operating costs), as well as time penal-
time) ties for interchanges on public modes, and converts all costs to
Waiting time equivalent minutes of travel time. The utility function divides the

Segment 1—U.K. Business Domestic Mode Choice Structure Segment 3—U.K. Leisure Domestic Mode Choice Structure
0.14119 0.07445
Park & Fly

Coach Park & Fly
0.53079 0.13157 0.23525
0.28186
Rail Routes Coach Taxi Kiss & Fly
0.20884
Kiss & Fly Taxi Rail Routes
Segment 2—U.K. Business International Mode Choice Structure Segment 4—U.K. Leisure International Mode Choice Structure
0.09404
0.13338
Taxi Kiss & Fly Park & Fly

Park & Fly
0.10991
0.18127 0.41689
0.23545 Coach
Coach Taxi Kiss & Fly
0.40598
Rail Routes
Rail Routes
FIGURE D6 SERAS mode choice model nesting structure.

Segment 5—Non-U.K. Business Mode Choice Structure
(0.001)
0.12421
Hire Car
Kiss & Fly

0.19839
All Rail Routes
Taxi Bus/Coach
0.68082 BR Coach Direct Service
Rail Routes
London Premium New Standard
Underground Non -stop BR
Segment 6—Non-U.K. Leisure Mode Choice Structure
(0.001)
Heathrow Alternative
Express Premium
Service
Hire Car
0.11883
Taxi Bus Kiss & Fly
0.29196
Rail Routes
FIGURE D6 (Continued ).
140
generalized cost for the mode by the square root of the direct dri- uncertainty in travel times is an S-shaped distance function that is
ving distance to the airport. There are no calibration parameters as asymptotic to one at short distances and would only be applied to
such, although different values of time are assumed for each mar- private car, rental car, and taxi modes.
ket segment and different weights are applied to waiting time for
some market segments. Different automobile operating costs (in The utility functions for each mode are as follows:
pence per kilometer) are assumed for U.K. business and U.K.
leisure passengers. Because the models are applied to estimates of
c pd
air passenger trips that originate in each analysis zone, an average Tcar + D+
air party size and average trip duration are assumed for each mar- gv gv
U P&F =
ket segment. D
1
One of the most questionable aspects of the SERAS model is the Tcar + Ftaxi
use of an average value of time for each market segment. Although gv
U taxi =
this is a consequence of the use of aggregate trip generation data D
rather than applying the model to disaggregate air passenger survey
data, it will tend to under-predict the use of public transport modes 1
Tbus + αWbus + Fbus
by lower-income travelers and over-predict their use by higher- v
U bus = + τ1 I 1
income travelers. To the extent that higher- and lower-income trav- D
elers are not uniformly distributed geographically, this will result in
biased estimates of public transport mode use from any given zone, 1
Trail + αWrail + Frail
and hence for any particular service. v
Urail = + τ x X1 + τ1 I 1 + τ 2 I 2 + θ
D
Another questionable feature of the SERAS model is the division
of the computed generalized cost by the square root of the distance
in computing the utilities. To the extent that the same distance is used where Tm = in-vehicle time plus access walk time for mode m
in computing the utilities for each air party from a given origin zone, (minutes),
this simply scales the utility values, which implicitly assumes that D = direct driving distance to airport (kilometers),
the variance of the error term in the utility functions increases with c = perceived private car fuel cost (pence/kilometer),
distance from the airport, albeit at a declining rate. Although it is v = value of travel time (pence per minute),
likely that the uncertainty in highway travel times increases with dis- g = air party size,
tance from the airport, this is not true for out-of-pocket costs (such p = parking rate (pence per day),
as public transport fares and parking costs) or for travel times on rail Fm = fare for mode m (pence),
or intercity bus modes, which operate to a published schedule Wm = wait time for mode m (minutes),
(whereas intercity buses may get delayed in traffic congestion, X1 = number of cross-platform interchanges,
passengers are likely to base their mode choice decisions on the pub- I1 = number of full intra-modal interchanges,
lished schedule). Therefore, the effect of travel time uncertainty I2 = number of intermodal interchanges,
should play a greater role for private car, rental car, and taxi modes α = weighting of wait time relative to in-vehicle time,
than for public transport modes. Another concern with this approach τx = cross platform transfer penalty (minutes),
is that the scaling effect changes most rapidly at short distances. τ1 = intra-modal interchange penalty (minutes),
However, it is precisely at these distances that travel times are most τ2 = intermodal interchange penalty (minutes),
predictable. What would therefore provide a better reflection of θ = direct rail constant (minutes).
TABLE D28
SERAS MODE CHOICE MODEL PARAMETERS
U.K.
Business U.K. Business U.K. Leisure U.K. Leisure Non-U.K. Non-U.K.
Parameters Domestic International Domestic International Business Leisure
Value of Time (£/hour) 28.5 46.3 4.7 6.6 47.8 5.6
Vehicle Operating Cost (p/km) 9.40 9.40 8.14 8.14 n/a n/a
Average Air Party Size 1.36 1.36 1.99 1.99 1.56 2.08
Average Trip Duration (days) 2.57 8.50 6.43 18.66 N/A N/A
Wait Time Weighting Factor 1.0 1.0 1.0 1.9 1.0 1.35
Parking Adjustments 2.0 2.5 2.5 4.0 N/A N/A
Interchange Penalty (min)
Cross-platform 0.43 0.50 0.77 0.90 0.30 0.69
Intra-modal 2.13 2.52 3.86 4.48 1.48 3.45
Intermodal 2.48 2.52 3.86 5.40 1.48 4.19
HEX Constant (min)
Central London 6.70 9.10 17.93 15.54 5.88 16.90
Outer London 3.20 4.09 5.10 7.84 2.99 9.41
Notes: N/A = not applicable; p/km = pence/kilometer; HEX = Heathrow Express.

141
Model Coefficients results from the need for a longer journey on the Underground to
reach the Heathrow Express terminal at Paddington Station. How-
The values for the various model parameters that were used in the ever, because the ride on the Heathrow Express is the same duration
SERAS study are shown in Table D28. Air passenger value of time for all travelers, any measure of the higher utility of the Heathrow
and vehicle operating costs are given in 1998 pence. Most of the pa- Express service should be a constant for all travelers. Because these
rameter values were adopted unchanged from the 1991 version of interchange penalties and direct rail constants have been estimated
the Heathrow Surface Access Model. from air passenger survey data, this suggests that the model estima-
tion has underestimated the perceived disutility of the access journey
The values of time for business travelers appear reasonable, to Paddington Station, possibly owing to underestimated interchange
although those for leisure travelers appear surprisingly low (in 1998 penalties (from most parts of London, reaching Paddington Station
the pound was worth approximately 1.66 dollars). The interchange by Underground involves several changes of line or even changes
penalties appear too low, particularly for cross-platform connections. of mode).
In general, travelers will experience a wait of about half the headway
of the outbound service at an interchange, in addition to any walking Documentation
time involved. However, it is not clear from the documentation
whether these penalties are in addition to any waiting time or are Halcrow Group Ltd., SERAS Surface Access Modelling, Prepared for the
intended to account for it. The Heathrow Express constant (θ) reflects Department of Transport, Local Government and the Regions, South
the higher quality of service relative to the London Underground. The East and East of England Regional Air Services Study, London, Eng-
difference in value between central and outer London presumably land, July 2002.

Abbreviations used without definitions in TRB publications:
AAAE American Association of Airport Executives

AASHO American Association of State Highway Officials
AASHTO American Association of State Highway and Transportation Officials
ACI–NA Airports Council International–North America
ACRP Airport Cooperative Research Program
ADA Americans with Disabilities Act
APTA American Public Transportation Association
ASCE American Society of Civil Engineers
ASME American Society of Mechanical Engineers
ASTM American Society for Testing and Materials
ATA Air Transport Association
ATA American Trucking Associations
CTAA Community Transportation Association of America
CTBSSP Commercial Truck and Bus Safety Synthesis Program
DHS Department of Homeland Security
DOE Department of Energy
EPA Environmental Protection Agency
FAA Federal Aviation Administration
FHWA Federal Highway Administration
FMCSA Federal Motor Carrier Safety Administration
FRA Federal Railroad Administration
FTA Federal Transit Administration
IEEE Institute of Electrical and Electronics Engineers
ISTEA Intermodal Surface Transportation Efficiency Act of 1991
ITE Institute of Transportation Engineers
NASA National Aeronautics and Space Administration
NASAO National Association of State Aviation Officials
NCFRP National Cooperative Freight Research Program
NCHRP National Cooperative Highway Research Program
NHTSA National Highway Traffic Safety Administration
NTSB National Transportation Safety Board
SAE Society of Automotive Engineers
SAFETEA-LU Safe, Accountable, Flexible, Efficient Transportation Equity Act:
A Legacy for Users (2005)
TCRP Transit Cooperative Research Program
TEA-21 Transportation Equity Act for the 21st Century (1998)
TRB Transportation Research Board
TSA Transportation Security Administration
U.S.DOT United States Department of Transportation

Acrp

Uploaded by

Copyright:

Available Formats

Acrp

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Acrp

Uploaded by

Copyright:

Available Formats

THE NATIONAL ACADEMIES PRESS

This PDF is available at http://nap.edu/23106 SHARE   

Airport Ground Access Mode Choice Models

104 pages | | PAPERBACK

FIND RELATED TITLES

– Access to free PDF downloads of thousands of scientiﬁc reports

Copyright © National Academy of Sciences. All rights reserved.

AIRPORT COOPERATIVE RESEARCH PROGRAM

A Synthesis of Airport Practice

Research Sponsored by the Federal Aviation Administration

TRANSPORTATION RESEARCH BOARD

Copyright National Academy of Sciences. All rights reserved.

AIRPORT COOPERATIVE RESEARCH PROGRAM ACRP SYNTHESIS 5

Copyright National Academy of Sciences. All rights reserved.

Copyright National Academy of Sciences. All rights reserved.

ACRP COMMITTEE FOR PROJECT 11-03 COOPERATIVE RESEARCH PROGRAMS STAFF

Copyright National Academy of Sciences. All rights reserved.

Copyright National Academy of Sciences. All rights reserved.

11 CHAPTER ONE INTRODUCTION

14 CHAPTER TWO AIRPORT GROUND ACCESS MODE CHOICE

31 CHAPTER THREE REVIEW OF THE LITERATURE

39 CHAPTER FOUR USE OF AIRPORT GROUND ACCESS MODELS

43 CHAPTER FIVE STATE OF PRACTICE OF AIR PASSENGER

55 CHAPTER SIX AIRPORT EMPLOYEE MODE CHOICE

Copyright National Academy of Sciences. All rights reserved.

67 CHAPTER SEVEN TRANSFERABILITY OF AIRPORT GROUND ACCESS

72 CHAPTER EIGHT INTEGRATION OF AIRPORT GROUND ACCESS MODELS

76 CHAPTER NINE CONCLUSIONS AND FURTHER RESEARCH

83 APPENDIX A SURVEY QUESTIONNAIRES

96 APPENDIX B SUMMARY OF SURVEY RESULTS

102 APPENDIX C LIST OF PARTICIPATING AGENCIES

Copyright National Academy of Sciences. All rights reserved.

AIRPORT GROUND ACCESS MODE CHOICE MODELS

Copyright National Academy of Sciences. All rights reserved.

• Airport Ground Access Mode Choice Modeling Process

Copyright National Academy of Sciences. All rights reserved.

Copyright National Academy of Sciences. All rights reserved.

Ui = ai + b1x1 + b2x2 + . . . + bnxn + ε

Vi = ai + b1(travel time) + b2(waiting time) + b3(walk distance) + b4(cost/income)

Copyright National Academy of Sciences. All rights reserved.

Copyright National Academy of Sciences. All rights reserved.

• Use of Airport Ground Access Models in Airport Planning

• Air Passenger Mode Choice Models

• Resident business trips,

Copyright National Academy of Sciences. All rights reserved.

• Airport Employee Mode Choice

Copyright National Academy of Sciences. All rights reserved.

• Transferability of Airport Ground Access Mode Choice Models

Copyright National Academy of Sciences. All rights reserved.

• Airport Ground Access Models in the Regional Planning Process

Copyright National Academy of Sciences. All rights reserved.

• Conclusions and Further Research

Copyright National Academy of Sciences. All rights reserved.

BACKGROUND funding. As a result, no clearly recognized body or practice

In contrast to the current state of practice of urban trans- METHODOLOGY

Copyright National Academy of Sciences. All rights reserved.

organizations (MPOs), consulting firms and research orga- REPORT AUDIENCE

Copyright National Academy of Sciences. All rights reserved.

Copyright National Academy of Sciences. All rights reserved.