Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The demand for reliable travel: evidence from Los Angeles, and implications for public transit policy
(USC Thesis Other)
The demand for reliable travel: evidence from Los Angeles, and implications for public transit policy
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
The Demand for Reliable Travel:
Evidence from Los Angeles, and Implications for Public Transit Policy
by
Sandip Chakrabarti
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
URBAN PLANNING AND DEVELOPMENT
August 2015
Copyright 2015 Sandip Chakrabarti
The Demand for Reliable Travel Chakrabarti (2015)
ii
Copyright 2015
by
Sandip Chakrabarti
All rights reserved.
The Demand for Reliable Travel Chakrabarti (2015)
iii
I dedicate this dissertation to –
My grandfather
Prafulla Kumar Chakraborty
(1911-1998)
The Demand for Reliable Travel Chakrabarti (2015)
iv
Acknowledgement
I am most grateful to my advisor and dissertation committee chair, Professor Genevieve
Giuliano, for supporting me through the Ph.D. program. This dissertation would not have been
possible without her help. Her valuable suggestions, critical feedback, attention to detail, and
demand for excellence not only helped improve my work, but also fundamentally contributed to
my development as a social science researcher. Professor Giuliano has been my greatest source
of inspiration. Her encouragement kept me motivated and focused. I feel honored to have her as
my mentor.
I would like to express my sincere gratitude to my dissertation committee members –
Professors Marlon Boarnet, Lisa Schweitzer and Jim Moore. They carefully reviewed many
drafts of the chapters and provided helpful feedback. I am indebted to Professor Schweitzer, one
of the best teachers I have ever had, for helping me improve my writing and public speaking.
I used traffic and transit system performance data from the Archived Data Management
System (ADMS) for my research. Thanks to Professor Giuliano for giving me the opportunity to
work on the project, and for allowing me to access the datasets. Many regional agencies shared
additional data. Thanks to Jesse Simon, Susan Phifer and Kali Fogel from the Los Angeles
County Metropolitan Transportation Authority (Metro), Hsi-Hwa Hu and Mana Sangkapichai
from the Southern California Association of Governments (SCAG), and Mohammad Assadi
from Caltrans. Special thanks to Udit Agrawal for his help in processing and analyzing large
datasets.
My parents and my wife were the sources of my strength. My son set a hard deadline to
finish my dissertation. Thank you Baba, Maa, Chandrani, and Aakash.
The Demand for Reliable Travel Chakrabarti (2015)
v
Overview
This dissertation uses data from the greater Los Angeles metropolitan region to demonstrate how
our demand for reliable or predictable travel conditions determines, in part, our transportation
mode choice. The research extends empirical literature on travel time reliability that has so far
analyzed how the demand for reliability affects trip scheduling and automobile route choice.
Through a series of independent empirical studies that use different methods, this dissertation
investigates how the reliability of public transit, independently and relative to other mobility
choices, influences the individual decision to consume transit and in turn aggregate service
consumption. Findings suggest that there is significant demand for reliable transit service, and that
good-quality transit can attract latent demand in specific situations such as high-demand congested
corridors during peak periods. In addition to contributing to the travel reliability literature, the
dissertation identifies the conditions that potentially make transit both competitive with the auto
mode and more productive as an industry.
The Demand for Reliable Travel Chakrabarti (2015)
vi
Contents
Chapter 1. Introduction............................................................................................................................................ 1
1. The Policy Context .......................................................................................................................................... 2
2. Research Overview ......................................................................................................................................... 5
3. Dissertation Structure ...................................................................................................................................... 6
4. References ....................................................................................................................................................... 9
Chapter 2. Demand for Reliable Travel – Past Empirical Evidence...................................................................... 10
1. Introduction ................................................................................................................................................... 11
2. The Value of Travel Time Savings ............................................................................................................... 12
3. The Value of Travel Time Reliability ........................................................................................................... 18
4. Directions of Future Research ....................................................................................................................... 33
5. References ..................................................................................................................................................... 38
Chapter 3. Demand for Transit Service Reliability: Analysis of Variation in Patronage across Bus Lines .......... 43
1. Introduction ................................................................................................................................................... 44
2. Transit Service Reliability: Theory and Practice ........................................................................................... 46
3. Methodology ................................................................................................................................................. 49
4. Results: Model Summaries and Observations ............................................................................................... 58
5. What about Rail? ........................................................................................................................................... 64
6. Discussions .................................................................................................................................................... 65
7. Conclusion ..................................................................................................................................................... 66
8. References ..................................................................................................................................................... 68
Chapter 4. Demand for Transit Service Reliability – Analysis of Variation in Boardings across Bus Stops ........ 73
1. Introduction ................................................................................................................................................... 74
2. Research Context ........................................................................................................................................... 75
3. Study Design ................................................................................................................................................. 78
4. Variables and Descriptive Statistics .............................................................................................................. 83
5. Results and Discussions ................................................................................................................................ 90
The Demand for Reliable Travel Chakrabarti (2015)
vii
6. Takeaways and Limitations ......................................................................................................................... 107
7. Conclusion ................................................................................................................................................... 108
8. References ................................................................................................................................................... 109
Chapter 5. The Reliability – Mode Choice Relationship ..................................................................................... 115
1. Introduction ................................................................................................................................................. 116
2. Literature Review ........................................................................................................................................ 123
3. Study Area and Approach ........................................................................................................................... 126
4. Transit Use for Commute Trips in LA County: Descriptive Analysis ........................................................ 128
5. Determinants of Car-owners’ Transit Mode Choice for Commute Trips .................................................... 143
6. Conclusion ................................................................................................................................................... 157
7. References ................................................................................................................................................... 162
Chapter 6. Conclusion and Takeaways................................................................................................................ 168
The Demand for Reliable Travel Chakrabarti (2015)
viii
List of Figures
Figure 3-1. Los Angeles Metro bus system map ......................................................................................................... 51
Figure 4-1. Metro time point bus stops ........................................................................................................................ 80
Figure 4-2. The Avg. OTP measure ............................................................................................................................. 84
Figure 4-3. The SD of schedule deviation measure ..................................................................................................... 86
Figure 4-4. Effects of peak-period reliability improvement across headway groups ................................................ 100
Figure 5-1.Transit in the U.S.: Government funding and service supply trends (1991-2012) ................................... 116
Figure 5-2. Public transit in the U.S.: Service effectiveness and productivity trends (1991-2012) ........................... 117
Figure 5-3. Percent of U.S. workforce using public transit for commuting (1990-2012) .......................................... 118
Figure 5-4. Study area: Los Angeles County ............................................................................................................. 127
Figure 5-5. Normal commute mode of individuals in car-owning and carless households (LA County) .................. 130
Figure 5-6. LA County census tracts in which car-owning transit commuters live and work ................................... 137
The Demand for Reliable Travel Chakrabarti (2015)
ix
List of Tables
Table 3-1. Variables in the patronage model ............................................................................................................... 53
Table 3-2. Descriptive statistics ................................................................................................................................... 57
Table 3-3. 3SLS (simultaneous-equations) models of bus line patronage ................................................................... 59
Table 3-4. Alternative OLS models of bus line patronage .......................................................................................... 62
Table 3-5. Reduced-form OLS models of bus line patronage ..................................................................................... 63
Table 4-1. Control variables ........................................................................................................................................ 87
Table 4-2. Descriptive statistics of key variables ........................................................................................................ 89
Table 4-3. Variables capturing stop neighborhood (i.e. constituent census tract) density ........................................... 91
Table 4-4. OLS regression models of stop-level bus line boardings – Avg. OTP effect ............................................. 93
Table 4-5. OLS regression models of stop-level bus line boardings – CV of OTP effect ........................................... 95
Table 4-6. OLS regression models of stop-level bus line boardings – SD of schedule deviation effect ..................... 96
Table 4-7. Summary of OLS regression models of peak-period stop-level bus line boardings – moderating effect of
headway ....................................................................................................................................................................... 99
Table 4-8. 2SLS regression models of peak-period stop-level bus line boardings .................................................... 106
Table 5-1. Change in estimated public transit trip share in select U.S. metros (2001-2009) ..................................... 120
Table 5-2. Alternative travel (journey to work) environments of commuters in carless households ......................... 133
Table 5-3. Household and personal characteristics of commuters in car-owning households ................................... 135
Table 5-4. Home and workplace neighborhood characteristics of commuters in car-owning households ................ 136
Table 5-5. Alternative travel (journey to work) environments of commuters in car-owning households ................. 142
Table 5-6. Independent variables ............................................................................................................................... 153
Table 5-7. Descriptive statistics ................................................................................................................................. 153
Table 5-8. Summary of binomial logistic regression models of car-owners’ transit mode choice for commute trips
................................................................................................................................................................................... 157
The Demand for Reliable Travel Chakrabarti (2015)
1
Chapter 1. Introduction
This chapter places the dissertation in the context of contemporary U.S. urban transportation policy
research. It highlights the expected contributions of the present study for both scholarship and
practice. The structure of the manuscript that includes contents of various chapters is outlined.
The Demand for Reliable Travel Chakrabarti (2015)
2
1. The Policy Context
The automobile has more critics than supporters within the U.S. urban planning community.
While planners may acknowledge that the automobile provides flexibility and convenience, they
advocate for transit because of its potential to support sustainability and other environmental
goals. So far, there is no data to show that planners have succeeded in convincing the traveling
public. Most U.S. residents continue to purchase and use automobiles, perhaps because there are
no better competitors in the passenger transport marketplace.
1
Urban planners have a point. Extreme automobile dependence in the U.S. causes
gridlock, degrades the urban fabric, deteriorates public health, promotes sprawl, and reinforces
social and environmental injustice.
2
Planners believe that these trends can threaten the future of
cities if left unaddressed. They seek to reduce auto use by altering the land use, regulatory and
market conditions that have made auto travel unreasonably cheap relative to its competitors.
Over the past few decades, planners have proposed, governments have supported, and
voters have approved higher investment in alternative travel infrastructures to reduce auto-
dominance (Boarnet, 2010). Direct market-driven pricing policies aimed at reducing auto
subsidies are often politically unpopular.
3
Therefore, high-density zoning laws, transit-oriented
mixed-use development incentives, generous public funding for new transit (particularly rail)
projects, and bike- and pedestrian-friendly environmental design regulations have emerged as
second-best strategies. The idea is to make alternatives to the auto more competitive.
Without undermining the achievements of other U.S. cities, Los Angeles (LA) has been a
frontrunner in the implementation of many new strategies that aim at creating an environment
1
Supporting data have been presented in Chapters 3, 4, and 5.
2
See Ewing (1997), Gordon and Richardson (1997), Handy (2005), and Schweitzer and Valenzuela (2004).
3
See Giuliano (2011).
The Demand for Reliable Travel Chakrabarti (2015)
3
where people can make efficient location and travel choices that benefit them and others, and
also promote sustainability. SB-375
4
mandates have paved the way for unprecedented efforts in
coordinated land use-transportation planning including large-scale context-specific investment
plans in public transit, light and commuter rail, travel demand management strategies (e.g.
congestion fees, parking charges, link/cordon pricing, etc.), Intelligent Transportation Systems,
etc. across California.
5
There seems to be significant political commitment towards improving
mobility, reliability, and safety of non-auto (along with auto) travel in the “car capital of the
world.”
Since 1980, the public transit, carpool, and bicycle lane networks have expanded
significantly across the Los Angeles metropolitan region. This includes five new rail lines, more
than 27 new rapid bus corridors, and over 500 miles of carpool lanes.
6
The voter-approved
Measure-R, a half-cent sales tax in Los Angeles County that took effect in July 2009, has
provided funding for new capital projects such as synchronization of traffic signals, extension of
light rail with airport connections, provision of clean-fuel buses, expansion of
subway/Metrolink/bus service, etc.
7
Transit projects worth over $20 billion are planned at the
time of writing this chapter. Metro,
8
through its Transit Oriented Development (TOD) Planning
Grant program has offered over $20 million to help local governments within the county create
TOD-supportive development environments.
9
The City of Los Angeles has partnered with Metro
4
Alignment of transportation and land use planning for meeting transportation-related greenhouse gas emission
reduction targets (refer http://www.leginfo.ca.gov/pub/07-08/bill/sen/sb_0351-
0400/sb_375_bill_20080930_chaptered.pdf; last accessed on 12/26/14).
5
Refer SCAG RTP 2012-2035 (http://rtpscs.scag.ca.gov/Documents/2012/draft/2012dRTP_ExecSummary.pdf; last
accessed on 12/26/14).
6
http://media.metro.net/projects_studies/images/final-2009-LRTP.pdf (last accessed on 6/6/2015).
7
Refer http://www.metro.net/projects/measurer/ (last accessed on 12/26/2014) for more details.
8
The Los Angeles County Metropolitan Transportation Authority.
9
Refer http://www.metro.net/projects/tod/ (last accessed on 12/26/2014) for more details.
The Demand for Reliable Travel Chakrabarti (2015)
4
to create Transit Neighborhood Plans around light rail stations that will establish new land use
regulations focusing on denser, pedestrian/bicycle-friendly built environments.
10
The Metro
ExpressLanes demonstration program, supported by a $210 million congestion-reduction grant
from the U.S. Department of Transportation, has introduced congestion pricing on sections of the
I-110 (since November 2012) and I-10 (since February 2013) freeways by converting High
Occupancy Vehicle (HOV) lanes to High Occupancy Toll (HOT) lanes.
11
The program has also
improved transit service and associated facilities (e.g. station up-gradation, park-and-ride,
bicycle storage, transit rewards programs, etc.), and funded alternatives to driving (e.g. carpool
and vanpool programs). Los Angeles Department of Transportation’s LA Express Park project
has implemented demand-driven variable pricing (with web-based real-time information) for
nearly 6000 parking spots in Downtown LA.
12
Numerous bike plans have been proposed and
implemented in the city and county of Los Angeles that aim at developing interconnected bike
corridors and support facilities.
13
The region celebrates active transportation by regularly
organizing government-supported, car-free street events such as CicLAvia.
14
Similar experiments are being conducted across the U.S. With large federal funding
support for value pricing pilot programs,
15
new fixed-guideway transit systems,
16
TOD
10
Refer http://www.latnp.org/ (last accessed on 12/26/2014) for more details.
11
Refer https://www.metroexpresslanes.net/en/about/about.shtml (last accessed on 12/26/2014) for more details.
12
Refer http://www.laexpresspark.org/about-la-expresspark/benefits-of-la-express-park/ (last accessed on
12/26/2014) for more details.
13
The 2012 Bicycle Master Plan by the County of Los Angeles Board of Supervisors, adopted on March 13, 2012 is
available at http://dpw.lacounty.gov/pdd/bike/masterplan.cfm; Biking infrastructure within the City of LA is
available at http://www.bicyclela.org/index.php; Metro and SCAG co-sponsor a Bike Count Data Clearinghouse
project, available at http://www.bikecounts.luskin.ucla.edu/ (links last accessed on 12/26/2014).
14
For more on CicLAvia, refer http://beta.ciclavia.org/thefaq (last accessed on 12/26/2014).
15
See http://www.ops.fhwa.dot.gov/congestionpricing/value_pricing/index.htm (last accessed on 12/26/2014).
16
See http://www.fta.dot.gov/12304.html (last accessed on 12/26/2014).
The Demand for Reliable Travel Chakrabarti (2015)
5
planning,
17
integrated corridor management demonstrations,
18
we should expect many new
planning-policy innovations.
I use the Los Angeles region as a laboratory for understanding public transit’s ability to
meet travel demand. By examining key transit service quality components, I advance practical
strategies to increase transit use.
2. Research Overview
This dissertation informs public transportation planning and policy. Using new technologies,
data and methods, I explore transit mode choice, analyze new investments/policies, and develop
realistic strategies targeted at increasing patronage and transit system productivity. The
dissertation builds on theories of the demand for reliable travel, and provides first empirical
evidence regarding the demand for reliable public transit service using data from Los Angeles. It
aims at providing the U.S. public transport industry, which has struggled with declining
productivity for decades, with efficient investment strategies for increasing its market share in
major metropolitan transport markets. By analyzing the determinants of ridership variation
across the Los Angeles Metro bus and rail system, and by investigating the sociodemographic
and spatiotemporal contexts in which people leave their cars behind and take transit to work, I
illustrate how public transit can attract patrons and increase productivity. I show that, while
system expansions and planned quality improvements must continue, transit operators can attract
passengers by improving service reliability along with strategically increasing service frequency
in areas with high latent demand. I demonstrate that increases in traffic congestion can only
result in mode shifts away from the automobile if alternative transit service is easily accessible,
17
See http://www.planetizen.com/node/71188 (last accessed on 12/26/2014).
18
See http://www.its.dot.gov/icms/pioneer.htm (last accessed on 12/26/2014).
The Demand for Reliable Travel Chakrabarti (2015)
6
highly reliable, frequent, and relatively fast. Using the case study of a natural experiment, the
implementation of Phase 1 of Metro Exposition light rail line in the dense core of Los Angeles, I
show how creatively designed rail service can, indeed, attract people out of cars, promote
ridership, and also reduce congestion locally.
3. Dissertation Structure
The dissertation has six chapters. The contents of Chapters 2 through 6 are outlined below:
Chapter 2 reviews past literature that provides context to, and serves as an introduction
of, the core topic of the dissertation. In this chapter, I review the theoretical underpinnings for
the value of travel time reliability, and explore its significance in travel behavior. I then analyze
the methods and results of past research on the demand for reliable travel, specifically in
departure time and auto route-choice. I also summarize studies underscoring the value transit
passengers place on reliable service. Finally, I discuss the limitations of, and gaps in, past
research to highlight the major contributions of my dissertation research.
Chapter 3 uses data from the Los Angeles Metro bus transit system to analyze the
variation in patronage across about 300 bus routes to demonstrate that service reliability
significantly influences patronage. The reliability effect is stronger in the weekday peak relative
to the off-peak. Based on the reliability measure used in this study, I provide the first empirical
evidence that interventions leading to better schedule adherence at time points (specifically, the
increase in frequency of on-time departures from schedule time points within a strict tolerance
range) can potentially promote patronage of fixed-route fixed-schedule transit systems in the
presence of latent demand. Policies encouraging reliability improvements may lead to
productivity gains for transit agencies.
The Demand for Reliable Travel Chakrabarti (2015)
7
As an extension of the study, I include both Metro bus and rail systems in the analysis.
Using statistical controls to distinguish between the groups, I show that, on average and all else
equal, Metro rail attracts significantly greater patronage than Metro bus. The extent to which this
effect can be attributed to rail’s near-perfect on-time performance, however, is not exactly clear.
I also do not generalize that rail is more effective than bus in promoting patronage. I conclude
that the Los Angeles Metro rail system is appropriately planned along corridors most primed for
rail investments; consequently patronage is high. Therefore, it is reasonable to infer that rail lines
extended to areas with high latent demand that significantly increase regional accessibility will
help promote patronage.
Chapter 4 explores the role transit service reliability plays in determining bus stop
boardings. Using highly disaggregate service supply, demand, and performance data from the
Los Angeles Metro bus system, I investigate whether the reliability of a line serving a stop
influences the number of people who board the line at that stop, controlling for various other
factors that are known to affect stop level boardings. This cross-sectional analysis of the
variation in boardings across more than 1300 sample bus stops served by about 300 directional
bus lines over a strategically selected six-month study period uses a unique historical archive of
real-time, geo-referenced automatic vehicle location data, and focuses on five different time
periods, peaks and off-peaks, of a typical weekday. By employing three different measures that
capture multiple dimensions of bus service reliability, and by estimating a series of regression
models, I find systematic evidence that higher average service punctuality (schedule adherence),
consistency in punctuality over time, and lower day-to-day variation in schedule deviation are
associated with greater ridership, all else equal. The reliability effect is most prominent during
weekday AM and PM peaks, when more travelers are expected to be risk-averse. Interestingly,
The Demand for Reliable Travel Chakrabarti (2015)
8
the effect of reliability on peak-period ridership depends on service headway; demand for
reliability seems to be highest in the 20-30 minute headway range. The findings indicate that
service reliability influences transit mode choice and route selection and suggest that system-
wide ridership gains can be expected from reliability improvements in the presence of latent
demand for transit travel. From an urban planning perspective, this study provides more proof
that good service quality can effectively compliment transformations in the urban fabric brought
about by coordinated land use – transit plans to promote transit use.
Chapter 5 uses data from the new 2012-13 California Household Travel Survey to
analyze the conditions/contexts under which workers belonging to car-owning households in Los
Angeles County normally use public transit for commuting. I analyze the household and personal
characteristics, home and workplace built environment factors, and alternative (auto vs. transit)
travel environments that are associated with the decision to take transit among choice riders. I
find that frequent and reliable transit service, presence of light rail or subway in route, fewer
transfers, and greater transit service/network density at origin and destination positively
influence transit mode choice among those who have a choice. It seems that, all else equal, the
relative demand for transit as a commute mode increases when relative travel time saving by the
alternative auto mode decreases, and also in contexts where the auto route is highly congested
and unreliable. It is possible that car-owners with pro-transit attitudes self-select into
neighborhoods with good transit access; for them, reliability may be an important service quality
factor to consider in location choice. Therefore, it seems plausible to conclude that service
reliability is key to restricting attrition of the existing choice-riders, and attracting any latent
demand through a wide range of supportive planning-policy interventions.
The Demand for Reliable Travel Chakrabarti (2015)
9
Chapter 6 concludes the dissertation with a brief discussion on broad implications for
planning scholarship and practice.
4. References
1) Boarnet, M. G. (2010). Planning, climate change, and transportation: Thoughts on policy
analysis. Transportation Research Part A: Policy and Practice, 44, 587–595.
2) Ewing, R. (1997). Is Los Angeles-Style Sprawl Desirable? Journal of the American
Planning Association, 63:1, 107-126.
3) Giuliano, G. (2011) “Transportation policy: Public transit, settlement patterns and equity in
the US,” chapter 28 in N. Brooks, K. Donaghy and G. Knaap, eds., Oxford University Press
Handbook of Urban Economics and Planning. New York: Oxford University Press.
4) Gordon, P. & Richardson, H. (1997). Are Compact Cities a Desirable Planning Goal?
Journal of the American Planning Association, 63:1, 95-106.
5) Handy, S. L. (2005). Smart growth and the transportation–land use connection: What does
the research tell us? International Regional Science Review, 28(2), 146–167.
6) Schweitzer, L., & Valenzuela, A. (2004). Environmental injustice and transportation: the
claims and the evidence. Journal of Planning Literature, 18(4), 383-398.
The Demand for Reliable Travel Chakrabarti (2015)
10
Chapter 2.
Demand for Reliable Travel – Past Empirical Evidence
This chapter reviews the empirical literature investigating how our demand for reliable travel
influences travel behavior – particularly trip scheduling decisions and automobile route choice.
The review serves as the basis for expecting a reliability – mode choice relationship in subsequent
chapters. I summarize findings on passengers’ perceptions about transit service reliability, and
these findings suggest that transit service reliability affects transit mode choice and, hence,
aggregate demand.
The Demand for Reliable Travel Chakrabarti (2015)
11
1. Introduction
Time is a scarce resource. Since travel time is generally unproductive, significant opportunity
costs are associated with travel. The behavioral intent of minimizing (or saving) travel time is
long established.
Research also suggests that travelers do not like unpredictable conditions. Travel time
unreliability adds additional costs due to uncertainty, and travelers respond by adjusting travel
time, mode, route, destination, etc. based on their perceptions (generated through experience, or
access to information) about the probability and severity of travel time variations across the
network (Fosgerau and Engelson, 2011). Response is assumed to be governed by the degree of
risk-aversion.
Past empirical research directly analyzing the effect of reliability on travel decisions has
focused only on private vehicle travel. We now know that drivers respond to unreliability
through careful departure time choice that may involve arriving early at their destination and
incurring additional time cost to avoid risk of being late. Drivers may also select routes that
involve paying additional out-of-pocket cost in order to get reliable travel conditions so that they
arrive at the destination on time. But it is also possible that travelers would choose modes in part
on the basis of reliability. For example, in heavily congested corridors, rail transit may draw
more travelers because of its greater reliability relative to private vehicle travel. Conversely,
public transit may discourage patrons if service is unreliable.
This chapter presents past empirical evidence regarding the demand for reliable travel. I
explore theoretical underpinnings of the value of travel time savings – the foundation of research
in this field, and travel time reliability. I critically analyze research methods, survey techniques,
and results of past research involving travel time reliability in departure time choice, route-
The Demand for Reliable Travel Chakrabarti (2015)
12
choice, and mode choice contexts. I end the chapter with evidence regarding the value of transit
service reliability, primarily based on passenger surveys. I argue that there is need to analyze the
relationship between mode choice and reliability in greater detail, and the effect of transit service
reliability on transit mode choice in particular.
2. The Value of Travel Time Savings
2.1 Context
Value of travel time savings (VTTS) is an important concept in travel behavior research.
Travelers are hypothesized to be utility maximizers, and to fundamentally value the prospect of
travel time savings. Note that this discussion does not include traveling as an activity (or
purpose) in itself, e.g. cases when people drive or ride for pleasure.
Theory suggests that for a given O-D (origin-destination) pair, trip purpose, land use
environment, and available travel choices, a traveler makes travel decisions (mode choices, route
choices, departure time choices, etc.) under constraints of:
Means (e.g. monetary budgets)
Responsibilities (e.g. time budgets – a function of personal, household, and job-
related constraints), and
Personal attitudes-preferences-habits.
Given a set of constraints, a traveler is assumed to rationally make travel decision (from
among available travel choices) that minimizes the generalized cost of travel in terms of time,
money, and inconvenience. Inconvenience includes discomfort and unpredictability of travel.
2.2 Concept
The fundamental concept of the value of travel time savings is not new. Hensher (1997) notes
that development of theories on the behavioral and resource values of travel time savings date
The Demand for Reliable Travel Chakrabarti (2015)
13
back to 1965 and G.S. Becker. Since time is a scarce resource, it affects the demand for goods
and services. Thus, travelers consider travel time as a cost, and positively value travel time
savings. Jiang and Morikawa (2004) observe that since traveling creates disutility, reductions in
travel time are preferred. Reductions in travel time provides opportunities for engaging in other
productive purposes (such as work for earning income), and minimizes the discomfort of travel.
DeSerpa (1971), building upon Becker’s (1965) theory of time allocation involving “time
budgets,” developed a theory of the economics of time that acknowledges time consumption
constraints (i.e., that time is a scare resource and thus time restriction is imposed on all
activities). According to the author, the amount of time spent in a given activity is partly
governed by choice, and partly by necessity. Time spent travelling is essentially a matter of
necessity, and is therefore “costly.” Hence, travel time savings is valuable.
Jiang and Morikawa (2004), summarizing DeSerpa’s (1971) theory, break the value of
time (VTT) into: 1) Value of Time as a Resource (VTR); 2) Value of Time as a Commodity
(VTC); and 3) Value of Time Savings (VTS). VTS refers to the value of reducing time spent in
an activity. The authors state that in general, VTS = (VTR – VTC). However, they note that
since travelling is an intermediate activity and usually produces no utility, the value of travel
time savings (VTTS) should be equal to VTR; unless if travel generates a sense of satisfaction,
relatively speaking. The relative satisfaction concept is important in travel behavior analysis. All
else equal, a traveler’s VTTS by personal car may be lower than that for public transit due to the
greater satisfaction (comfort) of travelling by car. Value of travel time savings (VTTS) is
extremely important in transportation research, since VTTS is a critical factor in the generalized
cost evaluating time spent traveling along a particular route, or by a particular mode (Jiang and
Morikawa, 2004).
The Demand for Reliable Travel Chakrabarti (2015)
14
Jiang and Morikawa (2004) further observe that VTTS for a particular travel activity is a
function of three factors: 1) Possibility of re-assignment of saved time to other necessary
activities (Hensher, 1997 refers to this factor as the opportunity cost component, or the “shadow
price” of time); 2) Travel environment (Hensher, 1997 refers to this factor as the relative
disutility component that reflects the alternative circumstances under which time consumption
occurs); and 3) Individual socio-economic and demographic characteristics. Their explanation is
useful, since it explains why, controlling for everything else, in-vehicle VTTS for a business trip
may be greater than a leisure trip; VTTS for public transit may be more than personal car; and;
in-vehicle VTTS for a business trip by a mother of a small child may be more than the mother of
a grown-up individual.
19
Thus, for a particular traveler, the value of a given trip (or the value placed on the activity
at the destination), the opportunity cost of travel, associated constraints, and the characteristics of
different alternative travel options determine the value of travel time savings (VTTS) for each
travel option and part thereof. These values determine the disutility associated with travel time.
VTTS is the willingness-to-pay for unit travel time saving in each travel option and part
thereof. In simple terms, the value of travel time savings is the monetary value that travelers
place on reducing travel time (Carrion and Levinson, 2012). Note that VTTS is often simply
referred to as VTT (value of travel time). Their common unit is monetary amount per unit time.
Using the empirical model of consumer behavior, Hensher (1997) explains that if a
traveler’s indirect utility function (controlling for factors such as trip purpose, income, gender,
19
The mother of a small child may be willing to pay more to reduce unit travel time and reassign that time to child
care activities, all else equal.
The Demand for Reliable Travel Chakrabarti (2015)
15
physical conditions, household characteristics, attitudes and preferences, time constraints, other
mode-specific attributes, etc.) is of the form:
𝑽 𝒊 = 𝑨 𝒊 − 𝜶𝑪
𝒊 − 𝜷 𝒊 𝑻 𝒊 … (Eqn. 1)
𝑉 𝑖 = the indirect utility associated with alternative i
𝐴 𝑖 = the mean of unobserved influences on the choice of alternative i
𝐶 𝑖 = the monetary cost associated with alternative i
𝑇 𝑖 = the travel time by alternative i
Then, the value of travel time savings (or value of time) of alternative i is given by
𝜷 𝒊 𝜶 . Note that
the value can vary for different segments of a trip (e.g. congested vs. free flow conditions;
walking, waiting, searching, etc.).
2.3 Findings
A large body of empirical research analyzes: 1) travelers’ willingness-to-pay for saving travel
time across different parts of a trip (e.g. walk, wait, in-vehicle, etc.) via different modes, and 2)
their choice of optimal departure time for minimizing travel time. Researchers have analyzed
variations in the demand for travel time savings by mode, travel purpose, travel distance, and
travelers’ personal characteristics.
Wardman (2001) observes that empirical research at the regional, corridor, or local levels
use disaggregate choice modeling within experimental settings. Abrantes and Wardman (2011)
note that stated-preference (SP) studies (particularly computer-assisted interviewing, pen-and-
paper based, and card based studies where 8-9 alternate hypothetical but realistic scenarios are
presented) are most common. They observe that most studies have tried to determine how
service quality factors relate to variations in travel time valuations. Wardman (2001) identifies
that studies focused on how people value walking and access time, waiting time, departure time
adjustments, time spent under congestion, search time, late arrival time, service headway,
The Demand for Reliable Travel Chakrabarti (2015)
16
interchanges, etc. A major share of past research efforts examines the automobile users, since
policies targeted at reducing automobile use is a high priority in the field (Abrantes and
Wardman, 2011).
Wardman (2001) found that controlling for everything else, people value walk and wait
times more than in-vehicle time in general, with wait time valued higher than walk time.
Moreover, business and commuting travel times are highly valued. Value of time for each part of
a trip varies by travelers’ sociodemographic and trip characteristics. Wardman (2004) estimated
that public transit users respectively value walk time and wait time twice and 2.5 times as much
as in-vehicle time. The author argues that although walk and wait time values are expected to
vary with individual socio-economic and situational factors, the results indicate a need to provide
high-frequency public transit services on core routes. He further notes that the results provide
evidence that faster, frequent, and accessible public transit services can reduce automobile use.
Abrantes and Wardman (2011) recently performed a meta-analysis of British studies, and
estimated that motorists value congested travel time 34% more than free flow travel time,
primarily owing to added anxiety and difficul of driving conditions. Increasing walk time by 20
minutes potentially increases the unit value of walk time by 34% and unit value of wait time by
25%. Car users value wait time 17% more highly than others; car users value travel time by bus
18% higher than car time and there is also a premium attached to time spent searching for
parking space.
Although drawing conclusions from meta analyses derived from studies conducted in
varied geographical contexts using different research methods is not advisable, the results
generally indicate motorists’ willingness-to-pay for congestion-free travel along fast (and tolled)
lanes, good transit access, better transit service, and parking convenience.
The Demand for Reliable Travel Chakrabarti (2015)
17
In the U.S., studies have mostly focused on motorists’ valuation of travel time and their
willingness-to-pay for faster travel along specific routes. For example, Lam and Small (2001)
used revealed preference data of commuters on SR-91 in Orange County, California, and
estimated that motorists (who had the choice of driving along a free or a variably tolled route)
were willing to pay $22.87 to save an hour of travel time. Brownstone et al. (2003) used revealed
preference data from a congestion pricing demonstration project on the I-15 in San Diego and
estimated a median willingness-to-pay to reduce travel time (by getting off the congested route)
of $30 per hour. They found that commuters, people belonging to high-income households,
women, and middle aged (35 to 45 years) individuals were more likely to pay and travel on the
fast lane (i.e. they value travel time via congested route more than others, all else equal). Both
results indicate the demand for speedy travel by commuters.
In general, various studies in the U.S. and the U.K. have established that travelers
inherently value travel time savings, and that the magnitude (in terms of monetary value placed
on travel time savings) varies according to the trip characteristics, travel environment, and
personal socioeconomic and demographic characteristics of the traveler.
20
Valuations vary by
study type (RP vs. SP; various SP presentation techniques).
Based on the findings of previous studies, some general conclusions may be drawn in the
context of public transit service. Controlling for everything else, high-frequency, fast
(comparable to automobile), comfortable, well-accessible (short access/egress distances with
good pedestrian-friendly environments) public transit services in relatively shorter routes might
attract choice (high-income, car-owning) riders. Also, Naess (2011) observes that high-quality
public transit service can potentially attract choice riders in off-peak periods for recreational
20
Refer to Börjesson and Eliasson (2014) for similar findings from Swedish studies.
The Demand for Reliable Travel Chakrabarti (2015)
18
trips. Thus, it seems that a necessary, but perhaps not sufficient, condition for inducing mode-
shifts away from the automobile is fast, high-frequency, comfortable, and convenient alternative
transit service.
But in addition to travel time savings, theory also suggests that travelers place
considerable value on reliable travel. According to Carrion and Levinson (2012), the concept
involving the value of travel time reliability is a “newcomer” in the field of transportation
research. Thus, it is important to explore the value of travel time reliability (hereafter referred to
as “value of reliability” or VOR), and how departure time choice, route choice, and mode choice
decisions are made considering reliability or the costs associated with uncertain travel
conditions.
3. The Value of Travel Time Reliability
3.1 Introduction
Fosgerau and Engelson (2011) argue that the societal cost imposed by delays associated with
recurrent congestion are captured though travelers’ valuation of travel time. But non-recurrent
congestion caused by incidents and imprudent operations/management of the multi-modal
transportation network cause unpredictable variations in travel time. Travelers’ knowledge or
perception regarding the probability of these variations influence decisions about departure time,
destination choice, mode choice, and whether to travel at all (Fosgerau and Engelson, 2011).
Asensio and Matas (2008) observe that user benefits brought about by transportation
investments include not only travel time savings, but a reduction in travel time variability.
Indeed, as the authors argue, projects reducing the variability or unpredictability of travel
without any significant impact on average travel time may also be valuable in themselves.
The Demand for Reliable Travel Chakrabarti (2015)
19
According to Noland et al. (1998), one of the greatest discoveries of transportation
analysis is recognition of the fact that travelers place significant importance on reliable travel. A
large number of qualitative and attitudinal studies have consistently found that travel time
reliability is important to commuter travel decisions (Bhat and Sardesai, 2006). Therefore, it is
necessary to understand travel time reliability to better understand travel behavior and mode
choice. Interestingly the amount of research involving the value placed by travelers on reliability,
and its effects on travel behavior, is relatively small compared to the effort that has gone into the
study of mean travel time savings (Bates et al., 2001; Asensio and Matas, 2008).
The omission of travel time reliability in travel demand models can have negative
consequences. For example, it is not possible to evaluate the effect of policies directed at
improving network reliability (as part of service quality); and that could lead to unreliable
estimates of other level of service (LOS) parameters in travel demand models, resulting in
inefficient policy formulations (Bhat and Sardesai, 2006). As a note of caution, however, Sweet
and Chen (2011) note that attention to travel time (un)reliability is warranted no doubt, but the
strength of its influence should not be overstated until more research is carried out.
3.2 Importance
Reliability is associated with the statistical concept of variability (Bates et al., 2001). It is
a measure of the variability in travel times over repeated journeys. Reliability can be conceived
to be a level of service parameter of a given mode-route combination for a journey connecting a
given origin-destination pair at a particular time. While value of time is the marginal rate of
substitution of travel time for money in a travelers’ indirect utility function, value of reliability
measures the willingness to pay for reductions in the day-to-day variability of travel times for a
particular type of trip (Brownstone and Small, 2005).
The Demand for Reliable Travel Chakrabarti (2015)
20
Travel time unreliability adds additional cost and uncertainty to travelers. Travelers
(assumed to have some information on the unreliability, and assumed to take rational travel
decisions) respond to uncertainty by adjusting time, mode, or route. In the face of travel time
variability, travel decisions are made under uncertainty or risk, and hence a traveler chooses the
option that maximizes “expected” utility (Li et al., 2010).
Why are travelers averse to unreliability? Why do they value reliable travel? Noland et al.
(1998) highlight two costs associated with unreliability that rational travelers try to minimize: 1)
Expected scheduling cost, or the desire to lower the likelihood of arriving at the destination at an
inconvenient time; and 2) Planning cost, or the problems associated with not being able to plan
activities owing to uncertainty about when a trip will be completed. Bates et al. (2001) suggest
two other reasons why travelers highly value reliable travel. First, travelers are sensitive to the
consequences associated with travel time variability, such as prolonged waiting times, missed
connections, and arrival at the destination either before or after the desired or expected arrival
time. Thus, a traveler chooses between travel alternatives, each of which is characterized by a
distribution of consequences, defined in terms of conventional generalized cost components
(cost, travel time, etc.), together with the impact on timing constraints (schedule delay). The
decision rule is based on the “maximum expected utility” theory, in which travelers are assumed
to choose the travel alternative that maximizes the expected value of an appropriately defined
utility function. Second, travelers place value on variability (or the uncertainty induced by
variability), independent of its consequences at the origin or destination. This is due to the
anxiety or stress caused by uncertainty. This explanation leads to a modeling approach in which
the conventional components of generalized cost are augmented by an extra term related to the
The Demand for Reliable Travel Chakrabarti (2015)
21
variability of travel time, usually taken to be the standard deviation (or variance) of the travel
time distribution.
According to Fosgerau and Karlström (2010), travel time uncertainty accounts for about
15% of time costs on a typical urban road, and considering the large share of individuals’ time
budgets that is spent on transport, it is clear that uncertainty of trip durations represents a
significant cost to society. People are often more concerned with travel time reliability than with
travel time itself. Most empirical studies have valued reliability at around 40% higher than travel
time (Lam and Small, 2001). Many studies of travel behavior have found that punctuality,
reliability and dependability of transport systems are rated by users as an important feature,
affecting both their perceptions and levels of use of different modes.
Past behavioral studies have found that an adjustment in departure time is the easiest
response to expected travel time variability. For example, Small (1982) empirically found that
urban commuters will shift their schedules by 1 to 2 minutes toward the early side, or by 1/3
rd
to
1 minute toward the late side when faced with travel time uncertainty. Noland et al. (1998) found
that people shift their schedules earlier in response to increases in travel-time dispersion for
offsetting the increased probability of being late resulting from unpredictability of travel time.
However, recent studies have found that risk-averse travelers would choose a transport mode
(also route) with less travel time variability, and even pay extra price (e.g. tolls) for reliable
travel (Li et al., 2010).
3.3 Methods
The methodological foundation for the empirical research involving travel time reliability centers
on schedule delay as formulated by Small (1982). Small (1982) postulated that travelers receive
The Demand for Reliable Travel Chakrabarti (2015)
22
disutility not only from travel time but also from arriving at their destination either earlier or later
than desired.
According to Bates et al. (2001), it is crucial to understand the concept of decision-
making under uncertainly to address choice problems with a reliability component. The standard
theoretical approach to choice problems under uncertainty is that of “minimum expected utility.”
A traveler is assumed to undertake the course of action which, bearing in mind the probabilities
of different outcomes, has the highest value of expected utility. This approach implies that a
traveler assesses all eventualities resulting from different possible outcomes. The utility function
is linear and identical for all travelers. Finally, the model assumes that we know the distribution
of travel time. Bates et al. (2001) also note that a common criticism of the approach is that
individuals cannot calculate all the probabilities of the outcomes and process the required
information to obtain the expected utility.
According to Asensio and Matas (2008), travel time variability is usually modeled
following two approaches. If it is assumed that travelers dislike the possibility of being early or
late, then the indirect utility function should include a measure of travel time variability together
with the usual components of travel costs. If it is assumed that travel time variability is valued
according to the consequences of being early or late, the modeling approach should consider the
consequences on the time restrictions of the individuals, such as early or late arrivals with respect
to the desired arrival time.
There are (at least) three different theoretical frameworks for modeling travel time
variability (Li et al., 2010; Carrion and Levinson, 2012).
The Demand for Reliable Travel Chakrabarti (2015)
23
Centrality-Dispersion or Mean-Variance model
The mean-variance model is used in route choice studies, and it assumes that travel time
variability (or unreliability) is a source of disutility similar to mean travel time. Variability is
measured by the variance or standard deviation of travel time, and utility is defined as a function
of the expected mean travel time and the expected variance. The traveler’s objective is to choose
a travel alternative that minimizes the sum.
According to Carrion and Levinson (2012), this approach is common in the context of
risk-return models in finance, where a decision-maker simultaneously aims at maximizing the
option’s return and minimizing its associated risk. The option’s return is generally represented by
the expected value and the risk by the variance. The authors explain that in this model
framework, a traveler is assumed to have prior information about the mean and variance of the
travel time distribution for each mode-route combination in the choice set between an origin-
destination pair. The traveler, with a certain degree of risk aversion, chooses the route that
minimizes a given objective function.
According to Li et al. (2010), the expected utility (the probability weighted average of the
utility of outcomes) function is of the form:
𝑬 ( 𝑼 )= 𝜷 𝑻 𝑬 ( 𝑻 )+ 𝜷 𝑺𝑫
𝑺𝑫 ( 𝑻 )+ 𝜷 𝑪 𝑪 + 𝜷 𝑿 𝑿 … (Eqn. 2)
E(T) and SD(T) respectively represent the mean and standard deviation of the travel time
distribution associated with a given alternative
C is the travel cost associated with a given alternative
X is a vector of sociodemographic variables that capture the heterogeneity in the value of
reliability (VOR)
In general, 𝑉𝑂𝑇 =
𝜷 𝑻 𝜷 𝑪 , 𝑉𝑂𝑅 =
𝜷 𝑺𝑫
𝜷 𝑪 , and the reliability ratio (RR) =
𝑽𝑶𝑹 𝑽 𝑻 𝑻
The Demand for Reliable Travel Chakrabarti (2015)
24
Bates et al. (2001) found that the reliability ratio (marginal rate of substitution between
average travel time and travel time reliability) is around 1.3 for car, and 2.0 for public transit.
That means that auto travelers are willing to pay 30% more for a 1-minute reduction in the
standard deviation of travel time compared to a 1-minute reduction in mean travel time. Transit
travelers are willing to pay double the amount.
In addition to out-of-pocket travel costs and socio-demographic characteristics, trip
purposes also go into utility functions to capture heterogeneity in the reliability value. Small et
al. (2005) used this model form by including toll difference, travel time difference, and
(un)reliability difference between alternate routes in an analysis of a value-pricing experiment in
Los Angeles. Lam and Small (2001) also adopted this approach in a route choice experiment in
Orange County, California, on a major commuting highway (SR91). Moreover, Brownstone and
Small (2005) used this approach in their analysis involving SR91 and Interstate 15.
Scheduling model
The scheduling model considers that disutility is incurred when a traveler arrives either
early or late at the destination. Small (1982) defined the difference between the preferred arrival
time (PAT) and actual arrival time as the schedule delay. He proposed a scheduling model to
understand how travelers choose departure times for on-time arrival at the destination.
Research suggests that the scheduling model provides a viable alternative to modeling
travel time reliability. This method has been traditionally used for departure time choice (or trip
scheduling) studies.
Small (1982) used the following indirect linear scheduling utility function to analyze
departure time choice of travelers who attempt to ensure on-time arrival (at PAT) at the
destination:
The Demand for Reliable Travel Chakrabarti (2015)
25
𝑼 ( 𝒕 𝒅 ; 𝑷𝑨𝑻 )= 𝜶𝑻 + 𝜷 ( 𝑺𝑫𝑬 )+ 𝜸 ( 𝑺𝑫𝑳 )+ 𝜽𝑫𝑳 + 𝝁𝑪 … (Eqn. 3)
Asensio and Matas (2008) note that PAT is exogenously determined
21
, and a traveler is
assumed to choose a departure time (including a “headstart” time) that minimizes disutility.
𝑡 𝑑 represents departure time choice (the decision variable),
T is the total travel time,
SDE and SDL represent schedule delay early and late respectively, such that
𝑆𝐷𝐸 = 𝑀𝑎𝑥 [0, 𝑃𝐴𝑇 − ( 𝑡 𝑑 + 𝑇 ) ] and 𝑆𝐷𝐿 = 𝑀𝑎𝑥 [0, ( 𝑡 𝑑 + 𝑇 )− 𝑃𝐴𝑇 ]
DL is a dummy variable that accounts for the extra cost of arriving late (DL=1 if SDL>0, and 0
otherwise),
C is the out-of-pocket travel cost,
All estimated parameters are assumed to be negative.
In this model formulation,
(𝑽𝑻𝑻 =
𝝏𝑼
𝝏𝑻
𝝏𝑼
𝝏𝑪
⁄ ), (𝑽 ( 𝑺𝑫𝑬 )=
𝝏𝑼
𝝏 ( 𝑺𝑫𝑬 )
𝝏𝑼
𝝏𝑪
⁄ ), and (𝑽 ( 𝑺𝑫𝑳 )=
𝝏𝑼
𝝏 ( 𝑺𝑫𝑳 )
𝝏𝑼
𝝏𝑪
⁄ )
Small’s (1982) model, however, is not for choice under uncertainty. Noland and Small
(1995) considered travel time and travel time variability to be uncertain, and dependent on the
departure time( 𝑡 𝑑 ) . Under this hypothesis, travel occurs under unpredictable conditions, the
expected utility function has been expressed as:
𝑬 [𝑼 ( 𝒕 𝒅 ) ] = 𝜶𝑬 [𝑻 ( 𝒕 𝒅 ) ] + 𝜷𝑬 [𝑺𝑫𝑬 ( 𝒕 𝒅 ) ] + 𝜸𝑬 [𝑺𝑫𝑳 ( 𝒕 𝒅 ) ] + 𝜽𝑷𝑳 ( 𝒕 𝒅 ) … (Eqn. 4)
PL represents the probability of late arrival, and is conditional on 𝑡 𝑑 .
Bates et al. (2001) and Li et al. (2010) suggest that the scheduling model works well for
the valuation of travel time reliability in the passenger car context, since drivers can continuously
adjust departure times in response to unpredictable travel conditions.
21
Meaning that PAT is not a function of travel conditions or choices.
The Demand for Reliable Travel Chakrabarti (2015)
26
Both Li et al. (2010), and Carrion and Levinson (2012) observe that the mean-variance
model is an approximation of the scheduling model if: 1) the travel time distribution is time-
independent; 2) no extra lateness penalty exists; and 3) departure time is continuous (not easily
applicable for transit). Carrion and Levinson (2012) further note that the scheduling model can
be modified by including risk attitudes, i.e. by defining SDE and SDL differently, etc. Bates et al.
(2001) suggest a “band of indifference” around the preferred arrival time that incurs no disutility.
The scheduling model has been extended beyond just departure time choice studies.
Asensio and Matas (2008) adopted the scheduling approach (in addition to the mean-variance
approach) for analyzing commute route choices in Barcelona, Spain (each alternative route
characterized by different levels of travel time variability). The authors found that time
constraints (e.g. maximum delay allowed at the workplace) significantly influence valuation of
reliability, all else equal, and that heterogeneity in valuation based on individual and trip
characteristics exist. In general, the value of travel time variability was found to be more than the
value of travel time savings.
Noland et al. (1998) used this approach for presenting a simulation model designed to
determine the impact of policies for dealing with travel time uncertainty on congestion. Fosgerau
and Karlström (2010) used a modified version of this approach to derive the value of reliability
in the scheduling of an activity of random duration, such as travel under congested conditions.
Mean Lateness model
The mean lateness model is a third method for measuring the value of reliability. In this
approach, unreliability is measured in terms of the mean lateness at the origin and destination.
Mean earliness is generally not considered.
The Demand for Reliable Travel Chakrabarti (2015)
27
According to Carrion and Levinson (2012), this approach has been used in the passenger
rail context in the UK. The authors observe that the mean lateness model consists of two
elements under the expected utility paradigm – schedule journey time (T) and the mean lateness
at destination. The former refers to the travel time between the actual departure time and the
scheduled arrival time, and the latter refers to the mean of the lateness. Lateness is defined as the
time between scheduled departure and actual departure (lateness at boarding), and time between
scheduled arrival and actual arrival (lateness at destination). The mean lateness model is of the
form (Carrion and Levinson, 2012):
𝑬 ( 𝑼 )= 𝜶 ( 𝑺𝒄𝒉𝒆𝒅𝒖𝒍𝒆𝒅 𝑻 )+ 𝜷 ( 𝑳 +
)+ 𝜸 ( 𝑩 +
)+ 𝜽𝑪 … (Eqn. 5)
𝐿 +
= expected mean lateness at the destination
𝐵 +
= expected mean lateness at the boarding point
C = out-of-pocket travel cost
Issues
Literature suggests that he estimated value of reliability and the reliability ratio are highly
sensitive to model specification (Bhat and Sardesai, 2006). The reliability parameter should
therefore be measured appropriately. Using standard deviation to measure unreliability assumes
that negative and positive deviations around the expected arrival time have similar effects. This
may not be accurate if travelers consider lateness as more costly. Thus, as Brownstone and Small
(2005) observe, many studies have relied on measures of the upper tail of the distribution of
travel times, such as the difference between the 90
th
and 50
th
percentile travel times. However,
the authors note that the way travelers respond to such distributions, and how they obtain
information about the randomness they would experience, is complex. Asensio and Matas (2008)
argue that traveler’s sociodemographic (and other personal) characteristics should be included in
The Demand for Reliable Travel Chakrabarti (2015)
28
the utility functions to better analyze the variation in value of reliability, value of travel time, and
their tradeoffs.
3.4 Survey techniques and results of past empirical research
Survey Techniques
Carrion and Levinson (2012) observed that “the differences among (past) studies span almost
every aspect such as: experimental design (e.g. presentation of reliability to the public in stated
preference [SP] investigations); theoretical framework (e.g. scheduling vs. centrality-dispersion);
variability (unreliability) measures (e.g. interquartile range, standard deviation; a requirement in
the centrality-dispersion framework); setting (or estimating) the preferred arrival time (e.g.
assuming work start time as preferred arrival time in the scheduling approach); data source (e.g.
revealed preference [RP] vs. state preference [SP]); and others. As a consequence, value of
reliability estimates also exhibit a significant variation across studies.”
The difference between SP and RP studies is that choices are assumed to be made under
hypothetical and real scenarios respectively (Li et al., 2010).
Most empirical studies that analyze and estimate the value of reliability are based on
stated preference (SP) data. The studies rely on surveys that ask respondents to make choices
under (or rank) hypothetical scenarios (Li et al., 2010). Each scenario is associated with the
choice under consideration (e.g. route, departure time, mode, etc.), expected travel time,
associated delays, etc. The questionnaires also capture many factors that contribute to the
heterogeneity in travel time variability valuation such as travelers’ sociodemographic
characteristics, trip purpose, etc. Li et al. (2010) observe that travel time variability (or the
stochastic aspect of travel time) is presented in two ways: 1) as the extent and frequency of delay
with respect to average expected travel time; and 2) a series of different arrival times that are
The Demand for Reliable Travel Chakrabarti (2015)
29
equally probable (including early, late, and on-time arrivals). Survey design depends on the
model(s) to be used for analysis.
Lam and Small (2001) observe that there are two possible reasons why most researchers
use data related to hypothetical scenarios. First, measuring the variability of travel times facing
actual travelers is difficult. Traffic data for RP studies need to be collected from GPS devices,
loop detectors, etc. Second, travel-time variability is highly correlated with mean travel time.
Second, collecting loop detector data is costly. Moreover, SP data helps simulate high levels of
variation in data, resulting in robust parameter estimates (Li et al., 2010). But Small et al. (2005)
argue that SP studies that describe hypothetical responses cannot always be trusted, because
behavior exhibited in hypothetical situations may not represent actual choices (Li et al., 2010
refer to this as “hypothetical bias”).
Stated choice studies
According to Carrion and Levinson (2012), most SP studies are based on the mean-
variance or the scheduling approaches. Li et al. (2010) provide a summary of past SP
experiments for valuing travel time reliability. The model forms of all experiments are mean-
variance, scheduling, mean-lateness, or a combination of mean-variance and scheduling. In this
context, it is worth highlighting the research by Asensio and Matas (2008), who conducted a SP
experiment in the Maresme corridor, north of Barcelona city, in Spain, for analyzing route choice
for home-to-work commute trips. The choice set comprised of two routes (tolled motorway and
free route), each characterized by monetary cost (vehicle operation costs plus toll in case of
motorway), average travel time (free flow plus time due to recurrent congestion), and a measure
of travel time variability due to unexpected congestion. The SP survey generated 2331 available
observations to estimate the choice model (binomial logit model). The authors included
The Demand for Reliable Travel Chakrabarti (2015)
30
socioeconomic variables to capture individual heterogeneity. The model results indicate that
travel time variability is valued on average 2.4 times more than travel time savings. The authors
found that delay time is valued at more than twice the savings of average travel times, although
the exact value depends on work time flexibility. Restrictions related to work start times (entry
flexibility and maximum allowed delays) have significant impacts on such valuations. Noland et
al. (1998) administered a SP survey of more than 700 commuters in the Los Angeles region to
empirically estimate the trade-offs among reliability, mean travel time, and scheduling decisions.
By estimating a binary logit model of route choice, the authors found that as the probability of an
incident is increased, commuters’ total travel costs increase. The authors argue that a way to
reduce the costs associated with unreliability is to encourage more flexible work schedules, since
late arrival and adherence to strict schedules seem to be the greatest source of stress related to
travel time unreliability. A review of Li et al. (2010) suggests that the results of past SP
experiments are conceptually fairly similar.
Revealed preference studies
Carrion and Levinson (2012) observe that most past RP studies involving VOR in the
U.S. have used data from value pricing experiments on SR91 and I-15 in California, and I-394 in
Minnesota. Value pricing, or congestion pricing, refers to variable road tolls that intend to reduce
or manage peak-period congestion. The cost of using tolled lanes is adjusted, based on corridor
demand and supply characteristics, in a way such that the average and variance of speed (also
volume/flow) are maintained at certain desired levels. System performance of the tolled lanes
are, on average, better than the free lanes. In theory, travelers (generally car users) having
relatively greater VOT or VOR choose tolled lanes to travel relatively faster and under more
predictable conditions during peak hours. Those who cannot afford tolls and also do not prefer
The Demand for Reliable Travel Chakrabarti (2015)
31
higher levels of recurrent and non-recurrent congestion adjust their travel times, frequency of
trips, destination, or route. Others choose the free lanes. Studies show that value pricing can
increase the efficiency of traffic flows (Henderson, 1974) if toll charges are set at optimal levels
(Small and Yan, 2001). The value pricing experiments provide opportunities to analyze
motorists’ willingness to pay for saving average travel time and reducing the expected variability
of travel time.
The mean-variance approach is commonly used in RP research. RP studies by Lam and
Small (2001) and Small et al. (2005) are noteworthy. Both studies use RP data to analyze
drivers’ responses to value pricing on SR91 in the Los Angeles area. Lam and Small (2001)
conducted a survey that asked travelers about their most recent weekday work trip. The authors
used loop-detector data to estimate mean travel time, median travel time, standard deviation of
travel time, and the difference between the 90
th
percentile travel time and median travel time
along free and tolled lanes. They then estimated a binomial logit model of route choice (tolled
lane vs. free lane) and found the value of reliability to be 39-46% more than the value of time. In
general, women and higher income motorists used the tolled lane more than others, suggesting
greater aversion for unreliability in those groups. Small et al. (2005) used survey data (collected
between 1999 and 2000) of a sample of motorists who participated in a value-pricing experiment
on the SR91. The authors used three samples: RP telephone survey by Cal Poly San Luis Obispo
researchers; RP mail survey by the Brookings Institution, and; SP mail survey by the Brookings
Institution presenting 8 SP scenarios. Respondents chose between two otherwise identical routes
with specified hypothetical tolls, travel times, and probabilities of delay. The authors analyzed
522 RP observations, and 633 SP observations from 81 different individuals. Motorists were
assumed to consider the central tendency (median) and dispersion (difference between the 80
th
The Demand for Reliable Travel Chakrabarti (2015)
32
and 50
th
percentiles) of travel times to choose between two alternate routes (tolled and free
lanes). The authors found that motorists with higher incomes are generally less responsive to
tolls. Women, middle-aged motorists, and motorists in smaller households were found to be
more likely to choose tolled lanes. Reliability was estimated to account for roughly one-third of
the attraction of the express lanes – less during early and middle parts of the rush hour, and more
during the later part. The authors concluded that travel time and its predictability are highly
valued by motorists. It is clear that RP techniques have been used in studies that are conducted in
an environment where a limited number of choices exist, and where the networks are relatively
simple. In a similar context, Liu et al. (2004) performed a route choice analysis using data from
the SR91 value pricing experiment. The authors used traveler survey results to capture choice,
price paid, and individual characteristics. Real time loop detector data was used to model the
travel environment (travel times). They too found that travel time variability is valued more than
travel time. It must be mentioned in this context that in a recent study, Carrion and Levinson
have used Global Position System (GPS) devices and transponders, and proposed an
experimental design to estimate value of time and value of reliability.
Valuations
Past SP and RP studies show that travelers value both travel time savings and reliability.
Valuations vary by model specification, survey design, and geographic context.
Results from studies focusing on the SR91 and I-15 corridors revealed median values of
travel time of $20 to $40 per hour, or 50-90% of the average wage rate (Brownstone and Small,
2005). The same studies valued reliability at roughly 95-140% as highly as median travel time.
Most studies have found a much higher value of reliability for women than men, possibly
The Demand for Reliable Travel Chakrabarti (2015)
33
because women have more child-care and other household responsibilities, which reduce their
scheduling flexibility.
According to Li et al. (2010), estimated reliability ratios vary across studies, with some as
high as 2.1 and others as low as 0.1. Bates et al. (2001) suggest that the ratio should be around
1.3 for car travel, and no more than 2.0 for public transport. However, the range of the estimated
value of reliability between 2009 US$0.79 and 2009 US$56.4 in recent travel time reliability
studies (see Li et al., 2010) raises serious concerns about research design and the reliability of the
studies themselves. Results of past research indicate that the value of reliability, and the
reliability ratio can be highly sensitive to survey technique and model specification (Bhat and
Sardesai, 2006). In general, occupations, sociodemographics, and time constraints explain the
variation in the value of travel time reliability to a large extent.
The high value placed on travel time variability has implications for transport policy,
both in terms of new infrastructure investment decisions and determination of pricing strategies
(Asensio and Matas, 2008).
4. Directions of Future Research
A major concern in the context of travel time reliability studies is the relatively small volume of
revealed preference (RP) research. The limitations of SP research are well established, and there
is a strong need for RP studies at a regional scale, to better understand metropolitan travel
behavior. Research involving value of travel time and value of travel time reliability need to be
expanded beyond local corridor-level studies (such as the SR91 and I-15). Moreover, past
research has focused mainly on commute trips. The variation of the value of reliability for
various trip purposes, and for various socio-economic and demographic characteristics of
travelers needs to be explored.
The Demand for Reliable Travel Chakrabarti (2015)
34
Past studies have demonstrated the high value travelers place on reducing the variability
of travel conditions. Empirical evidence suggests that reliable travel often contributes to
travelers’ utility much more than travel time savings. Travelers often pay a premium (e.g. tolls)
to ensure reliable travel and make travel time adjustments. However, since most studies have
focused on specific experiments, and only on the additional cost (out-of-pocket and time)
travelers are willing to incur for reliable travel, study results are primarily useful for evaluating
impacts of road- (or congestion-) pricing schemes on travel patterns.
Few researchers have considered travel time reliability in a mode choice context. I am
only aware of three studies. Bhat and Sardesai (2006) performed a combined RP and SP study
using primary data from a web-based survey of Austin, TX area commuters, and estimated a
mixed multinomial logit model of commuter mode choice. The authors found that the value of
reliability for commuters with an inflexible work schedule is about twice that for commuters
with a flexible schedule. Nam et al. (2005) conducted a study in Bangkok and developed a
multinomial logit model of mode choice. They found that values of reliability were higher than
values of travel time. Sweet and Chen (2011) used GPS-based travel survey data from Chicago
to study mode choice of workers in auto-owning households for work trips. They estimated that a
one standard deviation change in auto travel time unreliability, on average, is associated with
approximately a 23% reduction in the odds of using car, all else equal. The current dissertation
(particularly Chapter 5) attempts at exploring the network reliability (in a relative multi-modal
context) – mode choice connection in a new, possibly more comprehensive way.
Future research directions include analysis of the influence of unpredictable parking
availability and search time at the destination on mode choice, influence of incidents on various
dimensions of travel choice, and effect of advanced traveler information systems (ATIS) on the
The Demand for Reliable Travel Chakrabarti (2015)
35
perception of travel time reliability. According to Noland et al. (1998), measures that change the
degree/nature of (un)reliability, such as quick-response teams to clear up accidents, might make
people happier by reducing planning costs, or induce complex changes in the timing of people's
trips by changing the scheduling calculus. The same is true of measures, such as ATIS, that
improve information about travel conditions.
Finally, research on the demand for public transit service reliability (independently, and
relative to alternatives), and its impact on mode choice decisions and hence aggregate transit
travel demand has not been performed in the past. Since the current dissertation (Chapters 3, 4,
and 5) attempts at addressing this issue directly, it is useful to explore why transit service
reliability is considered to be important, and what we know regarding its value.
Numerous value of time studies in the US and UK show that all else equal, travelers
value (as a cost) waiting time more than in-vehicle travel time or even walking time. Longer than
expected waiting times (with or without real-time information about delays) impose large costs
(trip scheduling and planning costs) on travelers, and reduce the desirability of transit. Higher
probability of schedule delay (experienced or observed by users over repeated journeys) has been
linked to loss of risk-averse transit patrons (travelers may shift to other modes, or to other lines),
and to being a barrier for choice-riders to enter the transit market.
Reliability is long regarded as a critical service quality parameter of fixed-schedule
public transit systems (Perk et al., 2008). U.S. public transit agencies internally monitor various
dimensions of service reliability, and attempt at improving reliability through periodic schedule
changes and real-time system operations/management in order to meet predetermined targets.
The reason behind the significance of service reliability is obvious. Many transit users
need to travel under strict time constraints (i.e. within allocated time budgets), and therefore plan
The Demand for Reliable Travel Chakrabarti (2015)
36
their trips well in advance following published timetables,
22
with the goal of arriving at their
destinations at certain preferred times. The planning process is complex, since transit travel
generally involves multiple transfers across many lines, each running on its own unique
schedule. Although this coordinated trip-planning task is now performed by web-based
applications of transit operators
23
and other online service providers,
24
the risk of not being able
to execute a trip as planned remains. This is because buses and trains run ahead or behind
schedule all the time due to unpredictable variations in en-route traffic conditions, and therefore
fail to meet timetables. Also, service disruptions occur when scheduled trips are cancelled due to
a variety of reasons such as driver or vehicle unavailability, mechanical or electrical faults,
accidents, construction activities, etc. Patrons, as a result, are compelled to travel under risk of
experiencing longer-than-expected wait times, missing one or more connections, and incurring
penalties for deviating from their preferred arrival times. There is no doubt that real-time system
information that is increasingly becoming available via mobile devices has helped reduce
anxiety, adjust travel plans, and make alternative travel arrangements. However, costs associated
with service unreliability cannot be eliminated altogether. It is therefore not unreasonable to
expect that service unreliability determines in part the generalized cost of transit travel, the
22
“Timetables” generally contain estimated arrival/departure times of directional transit lines at every stop/station
they serve. Note, however, that drivers are generally responsible for adhering to arrival/departure times (referred to
as “scheduled” times) at certain designated stops/stations (referred to as “time points”) along the route. Therefore,
for passengers at stops/stations that fall in between time points, timetables provide a reasonable estimate (best guess)
of vehicle arrivals/departures. Sometimes, headways are specified instead of estimated times at the intermediate
stops/stations, and passengers learn approximate arrival/departure times through experience or web-based trip
planning applications. In such a case, if a given trip deviates from time point schedule, passengers at the
intermediate stops/stations observe irregularity in headway.
23
For example, the mobile application of the Los Angeles Metro transit system is available at
http://www.metro.net/mobile/metro-mobile-app/ (last accessed on 01/02/15).
24
e.g. Google Maps, Bing Maps, Yahoo Maps, Next Bus, etc.; Note that the General Transit Feed Specification
(GTFS), a standard format in which public transportation agencies share and regularly update their schedules and
associated geographic information with the public, helps programmers develop their own web-based applications
(e.g. GTFS data for the Los Angeles Metro transit system is available at http://developer.metro.net/, last accessed on
01/02/15)
The Demand for Reliable Travel Chakrabarti (2015)
37
overall transit travel experience, and consequently its demand relative to alternative modes. The
perception of service unreliability can indeed lead to patron attrition, and prohibit choice-riders
from entering the public transport marketplace. One would expect transit dependents to simply
bear the inconvenience cost; they may, however, choose alternative transit lines, routes, and
travel times to minimize expected uncertainty.
Transit user surveys underscore the importance of reliability. Wachs (1976), by
reviewing a large volume of passenger attitude surveys, found consistent evidence regarding the
high value placed on service reliability. He suggested that investing in reliability may have large
payoffs. Glascock (1997) surveyed 500 existing and potential customers of King County Metro
Transit in Seattle, Washington, and found that factors such as “the bus never leaves the stop
early” and “arrival time to work is no more than 5 min late” are rated highly. Eboli and Mazzulla
(2007) interviewed 763 University of Calabria students living in Cosenza, Italy, asking them to
rate 16 service quality attributes of a single transit agency in the region on a 1-10 scale; the
service reliability (schedule adherence) parameter emerged as one of the most significant
parameters of customer satisfaction. Tyrinopoulos and Antoniou (2008) performed a passengers’
perception (regarding service performance) survey in Athens and Thessaloniki, Greece, and
found that punctuality or schedule adherence is strongly preferred. Cantwell et al. (2009), using
an online stated preference survey instrument, found evidence that transit service reliability
improvements can substantially benefit commuters in Dublin, Ireland. dell’Olio et al. (2010)
analyzed perceptions of bus users in the city of Santander, Spain, and reported “service
reliability” as an important variable across all user groups. Iseki and Taylor (2010) observed that
frequent and reliable transit service in a safe environment is important for transit users in Los
Angeles, California, and suggested that these attributes can help attract and retain patrons. Nurul
The Demand for Reliable Travel Chakrabarti (2015)
38
Habib et al. (2011) found that transit users in Calgary, Canada value “reliability and
convenience” more than “ride comfort.”
Findings from the above studies indicate that transit service reliability could be a
significant determinant of transit travel demand, and that increases in reliability may attract latent
demand by influencing mode choice and/or line selection. It is possible that reliability
investments will help increase patronage and hence productivity of public transit systems.
ADMS
25
, with its capability of measuring reliability across modes, space, and time, provides the
opportunity to investigate the relationship in new ways, and inform future metropolitan
transportation decision-making. The following chapters (3-5) summarize my effort at providing
first empirical evidence on the reliability-ridership relationship. The chapters are independent
papers that present the research contexts, hypotheses, methods, data, analyses, discussions on
findings, and takeaways for policy.
5. References
1) Abrantes, P. A. L., & Wardman, M. R. (2011). Meta-analysis of UK values of travel time:
An update. Transportation Research Part A: Policy and Practice, 45(1), 1–17.
2) Asensio, J., & Matas, A. (2008). Commuters’ valuation of travel time variability.
Transportation Research Part E , 44, pp. 1074–1085.
3) Bates, J., Polak, J., Jones, P., & Cook, A. (2001). The valuation of reliability for personal
travel. Transportation Research Part E, 37, pp. 191-229.
4) Becker, G. (1965). A theory of the allocation of time. The Economic Journal, 75, 493–517.
25
The Metro-funded Archived Data Management System (ADMS) project archives real-time multi-modal
transportation system performance data in the Los Angeles region.
The Demand for Reliable Travel Chakrabarti (2015)
39
5) Bhat, C., & Sardesai, R. (2006). The impact of stop-making and travel time reliability.
Transportation Research Part B , 40, pp. 709–730.
6) Börjesson, M., & Eliasson, J. (2014). Experiences from the Swedish Value of Time study.
Transportation Research Part A: Policy and Practice, 59, 144-158.
7) Brownstone, D. & Small, K.A. (2005). Valuing time and reliability: Assessing the evidence
from road pricing demonstrations. Transportation Research Part A, 39, (4), 279–293.
8) Brownstone, D., Ghosh, A., Golob, T.F., Kazimi, C., & Van Amelsfort, D. (2003). Drivers’
willingness-to-pay to reduce travel time: evidence from the San Diego I-15 Congestion
Pricing Project, Transportation Research A, 37, pp. 373–387.
9) Cantwell, M., Caulfield, B., & O’Mahony, M. (2009). Examining the Factors that Impact
Public Transport Commuting Satisfaction. Journal of Public Transportation, 12(2), 1-21.
10) Carrion, C. & Levinson, D. (2012). Value of travel time reliability: A review of current
evidence. Transportation Research Part A, 46, pp. 720-741.
11) Dell’Olio, L., Ibeas, A., & Cecín, P. (2010). Modelling user perception of bus transit quality.
Transport Policy, 17(6), 388-397.
12) DeSerpa, A. (1971). A theory of the economics of time. The Economic Journal, 81, 828–846.
13) DeSerpa, A. C. (1973). Microeconomic Theory and the Valuation of Travel Time: Some
Clarification. Regional and Urban Economics, Vol. 3, No. 4, 401-410.
14) Eboli, L., & Mazzulla, G. (2007). Service quality attributes affecting customer satisfaction
for bus transit. Journal of Public Transportation, 10(3), 21.
15) Fosgerau, M., & Engelson, L. (2011). The value of travel time variance. Transportation
Research Part B: Methodological, Volume 45, Issue 1, 1-8.
The Demand for Reliable Travel Chakrabarti (2015)
40
16) Fosgerau, M., & Karlström, A. (2010). The value of reliability. Transportation Research
Part B: Methodological, 44(1), 38–49.
17) Glascock, J. (1997). Research on customer requirements for transit service design and
delivery. Transportation Research Record: Journal of the Transportation Research Board,
1604(1), 121-127.
18) Henderson, J. V. (1974). Road congestion: a reconsideration of pricing theory. Journal of
Urban Economics, 1(3), 346-365.
19) Hensher, D. (1997). Behavioral Value of Travel Time Savings in Personal and Commercial
Automobile Travel. In D. Greene, D. Jones, & M. Delucchi, The Full Costs and Benefits of
Transportation: Contributions to Theory, Method and Measurement (pp. 245-279). Springer.
20) Iseki, H., & Taylor, B. D. (2010). Style versus Service? An Analysis of User Perceptions of
Transit Stops and Stations. Journal of Public Transportation, 13(3), 23-48.
21) Jiang, M. & Morikawa, T. (2004). Theoretical analysis on the variation of value of travel
time savings. Transportation Research Part A, 38, pp. 551-571.
22) Lam, T.C., & Small, K.A. (2001). The value of time and reliability: measurement from a
value pricing experiment. Transportation Research Part E, 37 (2001), pp. 231–251.
23) Li, Z., Hensher, D., & Rose, J. (2010). Willingness to pay for travel time reliability in
passenger transport: A review and some new empirical evidence. Transportation Research
Part E, 46, pp. 384–403.
24) Liu, H.X., Recker, W., & Chen, A. (2004). Uncovering the contribution of travel time
reliability to dynamic route choice using real-time loop data. Transportation Research Part
A: Policy and Practice, Volume 38, Issue 6, 435-453.
The Demand for Reliable Travel Chakrabarti (2015)
41
25) Naess, P. (2011). “New urbanism” or metropolitan-level concentration: A comparison of the
influences of metropolitan-level and neighborhood-level urban form characteristics on travel
behavior. Journal of Transport and Land Use, 4(1), 25–44.
26) Nam, D., Park, D., & Khamkongkhun, A. (2005). Estimation of Value of Travel Time
Reliability. Journal of Advanced Transportation, Vol. 39, No. 1, pp. 39-61.
27) Noland, R., & Small, K. (1995). Travel-time uncertainty, departure time choice, and the cost
of morning commutes. Transportation Research Record 1493, 150–158.
28) Noland, R., Small, K., Koskenoja, P., & Chu, X. (1998). Simulating Travel Reliability.
Regional Science and Urban Economics, 28 pp. 535–564.
29) Nurul Habib, K. M., Kattan, L., & Islam, M. T. (2011). Model of personal attitudes towards
transit service quality. Journal of Advanced Transportation, 45, 271–285.
30) Perk, V., Flynn, J., & Volinski, J. (2008). Transit Ridership, Reliability, and Retention.
Tallahassee, FL: Florida Department of Transportation. Retrieved September 17, 2013, from
http://www.nctr.usf.edu/pdf/77607.pdf
31) Small, K., Winston, C., & Yan, J. (2005). Uncovering the Distribution of Motorists'
Preferences for Travel Time and Reliability. Econometrica, Vol. 73, No. 4, pp. 1367-1382.
32) Small, K.A. (1982). The scheduling of consumer activities: work trips. American Economic
Review, 72, 467–479.
33) Small, K. A., & Yan, J. (2001). The value of “value pricing” of roads: Second-best pricing
and product differentiation. Journal of Urban Economics, 49(2), 310-336.
34) Sweet, M. N., & Chen, M. (2011). Does regional travel time unreliability influence mode
choice? Transportation, 38(4), 625-642.
The Demand for Reliable Travel Chakrabarti (2015)
42
35) Tyrinopoulos, Y., & Antoniou, C. (2008). Public transit user satisfaction: Variability and
policy implications. Transport Policy, 15(4), 260-272.
36) Wachs, M. (1976). Consumer attitudes toward transit service: an interpretive review. Journal
of the American Institute of Planners, 42(1), 96-104.
37) Wardman, M.R. (2001). A review of British evidence on time and service quality valuations.
Transportation Research Part E, 37 (2001), pp. 107–128.
38) Wardman, M.R. (2004). Public transport values of time. Transport Policy, 11, pp. 363-377.
The Demand for Reliable Travel Chakrabarti (2015)
43
Chapter 3.
Demand for Transit Service Reliability: Analysis of Variation in Patronage across Bus Lines
This chapter analyzes the determinants of the variation in patronage across bus lines in Los
Angeles. Among other factors, service reliability and frequency and shown to positively affect
patronage. This study provides first empirical evidence that interventions leading to better schedule
adherence at time points can promote patronage of fixed-route fixed-schedule transit systems in
the presence of latent demand.
The Demand for Reliable Travel Chakrabarti (2015)
44
1. Introduction
The US public transit industry has experienced substantial growth in funding support over the
past several decades. For example, between 1992 and 2012, total annual government spending on
transit increased from $22 billion to $58 billion at an annual average inflation-adjusted growth
rate of about 2.5%.
26
The trend continues. Recently, the new federal surface transportation law,
MAP-21, authorized more than $20 billion just for federal transit programs over two fiscal years
(2013 and 2014).
27
Continued funding from federal, state and local governments have helped
expand service areas, improve quality levels and upgrade fleets, in addition to maintaining core
infrastructures and sustaining operations.
Patronage growth, however, has not kept pace with capital investments and rising
operating costs. And transit’s share of the US travel market continues to be small. Between 2002
and 2011, while total revenue miles of service increased by 14.2%, total subsidy per trip also
increased by 17.4% (NTD 2012). This indicates declining productivity. Moreover, less than 2%
of all trips within the US are made via transit (Santos et al. 2011), and my analysis of 2001 and
2009 NHTS
28
data shows that transit has even been losing share in some of its largest urban
markets (e.g. New York, Washington D.C., San Francisco, and Pittsburg) over the past decade.
Given these trends, it is important to consider how US public transit systems might attract
more patrons and increase productivity. A recent analysis of US public transit policy suggests
that strategies for increasing transit’s market share must include investment in the dimensions of
service quality that travelers value most (Giuliano 2011). Using empirical evidence from Los
26
US National Transit Database figures, available at http://www.ntdprogram.gov/ntdprogram/data.htm (accessed on
8/11/2014).
27
Refer http://www.apta.com/gap/legissues/authorization/Documents/APTA%20MAP-21%20Guide.pdf (accessed
on 8/11/2014).
28
National Household Travel Surveys (refer http://nhts.ornl.gov/; accessed on 8/11/2014).
The Demand for Reliable Travel Chakrabarti (2015)
45
Angeles, I explore whether investing in reliability
29
can be effective. I investigate the reliability-
patronage relationship by analyzing the variation in patronage across bus lines. The objective of
this study is to inform transit managers about the efficacy of reliability investment as a patronage
promotion strategy.
The cross-sectional system-wide study of the Los Angeles Metro
30
(Metro) bus system
reveals a significant positive association between a line’s service reliability and its patronage.
The reliability effect appears to be stronger during weekday peak periods relative to off-peak
periods. This implies that, all else equal, more reliable lines attract comparatively greater number
of peak-period riders across their service corridors as they are more frequently chosen over
alternate lines and competing modes. Based on the reliability measure (“on-time performance” or
OTP)
31
used in this study, I conclude that better schedule adherence, or increase in the frequency
of on-time (i.e. between 1 minute early and 5 minutes late with respect to schedule per Metro’s
convention) departures from time points, can positively influence patronage of fixed-route fixed-
schedule bus transit service. Investing in reliability can therefore be productive from an agency
standpoint, particularly in cities like Los Angeles that have extensive transit networks and high
latent demand for transit use.
The remainder of this chapter is organized as follows: Section 2 presents fundamental
concepts relevant in the context of this study; Section 3 explains the research methodology;
Section 4 summarizes empirical models and presents observations; Section 5 includes a
discussion of findings; and finally Section 6 concludes the chapter with policy implications.
29
Reliable service is broadly defined as one that consistently operates according to its schedule or plan.
30
The Los Angeles County Metropolitan Transportation Authority.
31
For Metro, OTP relates to the proportion of time a transit vehicle has been observed to have adhered to schedule
times (at designated time points) within predetermined tolerances – no more than 1 minute earlier and no more than
5 minutes later than schedule time.
The Demand for Reliable Travel Chakrabarti (2015)
46
2. Transit Service Reliability: Theory and Practice
2.1 The demand for travel time reliability
The travel time reliability literature is discussed in detail in Chapter 2. Theory suggests that risk-
averse travelers tend to avoid unpredictability associated with travel. Many studies, largely
focusing on the automobile mode, have analyzed how the demand for travel time reliability
affects trip scheduling (e.g. Small 1982), route choice (e.g. Noland et al. 1998; Lam and Small
2001; Liu et al. 2004; Small et al. 2005; Asensio and Matas 2008; Tilahun and Levinson 2010),
and mode choice (e.g. Nam et al. 2005; Bhat and Sardesai 2006; Sweet and Chen 2011).
However, the nature of the relationship between transit service reliability (independently, or
relative to other modes) and transit mode choice (and hence transit travel demand) is still
unclear.
2.2 Importance of transit service reliability
I have also highlighted (see Chapter 2) findings of several studies analyzing transit passengers’
attitudes and preferences. The studies find that unreliability (generally defined as punctuality of
service, i.e. how well the service adheres to expected schedule or plan) ranks among the top
inconvenience costs associated with transit travel. Studies by Wachs (1976), Glascock (1997),
Hensher et al. (2003), Tyrinopoulos and Antoniou (2008), Cantwell et al. (2009), Eboli and
Mazzulla (2009), Iseki and Taylor (2010), Eboli and Mazzulla (2010), dell’Olio et al. (2010),
Nurul Habib et al. (2011), and de Oña et al. (2013) are illustrative.
Given the importance of service reliability, many researchers have proposed creative
methods of measuring service reliability (e.g. Polus 1978; Nakanishi 1997; Camus et al. 2005;
Lin et al. 2008; Chen et al. 2009), analyzed factors that cause unreliability (e.g. Sterman and
Schofer 1976; Abkowitz and Engelstein 1983; Strathman and Hopper 1993; Strathman et al.
The Demand for Reliable Travel Chakrabarti (2015)
47
1999; Yetiskul and Senbil 2012), and recommended methods to improve reliability (El-Geneidy
et al. 2006; El-Geneidy et al. 2009; El-Geneidy et al. 2011; Xuan et al. 2011).
2.3 Transit on-time performance (OTP)
Based on how passengers perceive fixed-route fixed-schedule transit service to be unreliable,
researchers have proposed several measures of (un)reliability. In general, measures seek to
capture the (in)ability of a transit system to successfully complete scheduled trips, adhere to
schedules (i.e. reach the route-end and/or serve en-route time points around scheduled times),
maintain regular headways (across different segments of a route), and perform steady runs (i.e.
consistently maintain expected travel times across different segments of a route).
32
In sum,
unreliability is measured in terms of the variability in various dimensions of system performance,
observed over time (El-Geneidy et al. 2011).
The US Federal Transit Administration (FTA) recognizes reliability as a key transit
service quality factor (see the “Transit Capacity and Quality of Service Manual”
33
; Kittelson &
Associates, Inc. et al. 2013). It proposes several measures based on the source of unreliability,
magnitude of impact, and purpose of measurement.
For example, reliability impacts of major system breakdowns that cause service
disruptions are captured through measures such as: a) percent of scheduled trips that were
cancelled, b) percent of scheduled time operations were down, or c) average distance traveled
between mechanical breakdowns. Impacts of common service variations (generally considering
successfully completed trips only) caused by a combination of unpredictable/unmanageable
external factors (i.e. those that could not be accounted for during scheduling/planning decisions,
32
Refer Chen et al. (2009) for an exhaustive review.
33
The third edition is available online at http://www.trb.org/main/blurbs/169437.aspx (accessed on 8/11/2014).
The Demand for Reliable Travel Chakrabarti (2015)
48
and that could not be managed in real-time with available resources) and internal system
operations/management faults are reflected in performance measures such as: a) on-time
performance or OTP (proportion of total trips that served time points within and outside – before
or after – an acceptable tolerance range around the schedule time), b) headway adherence (a
metric of evenness of intervals between vehicle arrivals at designated stops/stations), and c)
excess wait time (or average schedule delay in departure from designated stops/stations).
Measures can be derived for a given line, or for a given line at a given stop/station, or system-
wide, averaged over a given time period (e.g. month), and aggregated across different times of
the day (e.g. peak and of-peak) and days of the week (e.g. weekdays and weekends).
There are different opinions regarding the strengths and limitations of the measures, and
the conditions under which each measure is useful. There is no single measure that best reflects
service quality.
In this study, I test the service (un)reliability – demand/patronage relationship using OTP
data. The measure is derived from Metro’s Advanced Transportation Management System
(ATMS) that includes GPS-based AVL (Automatic Vehicle Location) and APC (Automatic
Passenger Counter) as sub-modules. Metro routinely monitors OTP, and has shared data for this
research.
OTP data reflects the frequency (translated into expectation/probability from the
passengers’ point of view) of “on-time” (i.e. within the 1 minute early and 5 minutes late
window), “[>1 min]-early” (i.e. more than 1 minute early), and “[>5 min]-late” (i.e. more than 5
minutes late) departures from schedule time points. Better on-time service is related to better
punctuality (hence lower probability of missing a bus or waiting longer than expected), even
vehicle spacing, and uniform vehicle loading. OTP data cannot help determine average schedule
The Demand for Reliable Travel Chakrabarti (2015)
49
deviation, average earliness or lateness, or their variances in minutes. However, Cevallos et al.
(2011) theoretically demonstrate that schedule deviation and OTP data are inextricably linked;
efforts at improving schedule adherence by reducing the average and dispersion of the schedule
delay distribution will be reflected in better OTP measures.
3. Methodology
3.1 Approach, hypothesis, and expectations
I perform a cross-sectional analysis to investigate whether service (un)reliability
34
determines in
part the variation in demand (or patronage) across Metro bus lines. My research closely follows
Taylor et al.’s (2009) study that explored the determinants of aggregate transit patronage across
US urbanized areas.
I hypothesize that if service reliability influences the desirability of transit travel, then all
else equal, a relatively more reliable line should attract (and retain) greater number of users from
within its service corridor. This means that more choice riders may be drawn to that line, and
patrons may choose the line among other potential alternatives (if alternatives exist), at any given
segment of their journey. Although passengers can (and do) react to unreliability (experienced
over time) by building in extra time (or time cushion or buffer time) into their travel time
budgets, the resultant inconvenience can induce mode, route/line, or time shifts for those with
choice, particularly for time-critical travel (e.g. commute or business-related trips) and when
uncoordinated transfers are involved.
I expect to observe relatively higher patronage for a relatively more reliable line, all else
equal. Also, the reliability effect should be more prominent at times when travelers are more
34
I use a generic term “(un)reliability,” since while reliability refers to on-time service, unreliability refers to service
outside (before/after) the tolerance window.
The Demand for Reliable Travel Chakrabarti (2015)
50
sensitive to travel time variability – e.g. weekday peak periods. I should be able to pick up the
reliability effect in places such as Los Angeles that have a dense transit network, presumably
high latent demand, and multiple alternative transit routes connecting most origins and
destinations.
3.2 Time frame, units of analysis, and data sources
I select a six-month time period – December 2011 to May 2012 – for analyzing long-term
average measures of patronage and (un)reliability across bus lines. I assume that long-term
experience or observation of performance (e.g. how frequently service is on-time, and how often
it is not) is translated into a perception of the probability of on-time service. The perception
influences the generalized cost of transit travel and hence affects mode choice or route/line
selection. The six-month average also helps smooth out idiosyncratic and seasonal fluctuations in
the data. Choice of the time period is strategic: I ensured that no intermediate service/schedule
changes occurred, and also left a six-month period for patronage reorganizations to stabilize
following a prior June 2011 shake-up. Metro’s service area and constituent bus lines are shown
in Figure 3-1 for reference.
The unit of observation is a directional bus line (hereafter referred to as “line”). For
example, Metro route 37 consists of two lines – 37 East and 37 West. Note that lines in opposite
directions serving the same route at a given time of the day generally have different service
supply and performance characteristics due to the directional variation in travel demand and
traffic patterns. Their patronage levels are also different. I focus on weekdays only, and on four
time periods defined by Metro: AM peak (6 to 9 AM), PM peak (3 to 7 PM), Mid-Day (9 AM to
3 PM), and Night (7 PM to 12 AM).
The Demand for Reliable Travel Chakrabarti (2015)
51
I use data from three principal sources: Metro transit supply, patronage, and performance
data; ACS (American Community Survey) 2007-2011 data, and; regional employment and
transit network data from SCAG (Southern California Association of Governments).
Figure 3-1. Los Angeles Metro bus system map
Note: Map corresponds to the June 2011 service change, and was valid between June 2011 and June 2012; Locations of landmarks are tentative,
and intended for visual reference only.
3.3 Models and variables
Literature suggests that transit patronage is a function of characteristics of travelers, metropolitan
regions, substitute modes, and transit supply, with a high level of simultaneous association
between supply and demand (Taylor et al. 2009).
In the current study, the unit of observation is a bus line, and I am analyzing variation in
patronage across lines within a network in equilibrium. Constituent lines are not randomly
The Demand for Reliable Travel Chakrabarti (2015)
52
distributed across the region, and their supply characteristics (i.e. quantity and quality of service)
are not random either. Lines are strategically configured so that people are connected with jobs
and activity locations seamlessly via a well-integrated network. And supply is adjusted based on
demand signals so that system productivity can be maximized under constraints of resources,
service quality thresholds (e.g. load factor policy), overall network performance, and broader
social goals of providing lifeline service across the region.
Note that transit patrons travel along complex paths from origin to destination, often
transferring across multiple lines within the network. One may board a line to travel from origin
to destination directly, or for connecting to another line, or to move from one transfer point to
another, or for completing the last leg of a journey. All such boardings contribute to the
patronage of that line.
Considering the above, I use a system of simultaneous equations to analyze the
reliability-ridership relationship.
I first propose the demand-side model – the model of bus line patronage. In addition to
service (un)reliability, the variation in patronage across Metro bus lines will be governed by the
level of accessibility/connectivity offered by a line, and its planned service quality. I also need to
control for unique line characteristics (e.g. length, no. of stops, competition with other
city/municipal transit operators, etc.) that may influence patronage. The conceptual framework of
the line-level patronage model (for time period of the day “t”) is:
𝑷 𝒍𝒕
= 𝜷 𝟎 + 𝜷 𝟏 𝑹 𝒍𝒕
+ 𝜷 𝟐 𝑨 𝒍𝒕
+ 𝜷 𝟑 𝑺 𝒍𝒕
+ 𝜷 𝟒 𝑪 𝒍𝒕
+ 𝜺 ….. (Eqn. 6)
Where: P is a measure of line-level patronage; R is a vector of line-level service (un)reliability measures;
A is a vector of line-level accessibility/connectivity measures; S is a vector of line-level planned service
quality measures; and, C is a vector of other line characteristics (or controls), for line l at time period t. 𝜀
is the error term. Note that the parameter group R includes my explanatory variable(s) or variables of
interest. Read “t” as AM peak, Mid-Day, PM peak, and Night.
The Demand for Reliable Travel Chakrabarti (2015)
53
Variables included in the patronage model, along with descriptions, are given in Table
3-1. Variable selection follows the Taylor et al. (2009) study, and is also based on the theory of
mode choice/use (e.g. Badoe and Miller 2000).
Table 3-1.
Variables in the patronage model
Variable Construction/Measurement (Data source)
Dependent (patronage) variable, denoted by “P” in Eqn. 6
Average per-hour line boardings
Average (over the six-month study period) per-hour total bus line boardings
(summed across all stops), for each line for each time period of weekday. The
“per-hour” measure helps in comparing the relative influences of the
independent variables between the time periods consisting of unequal
35
number of hours. The dependent variables, although primarily derived out of
raw boarding counts, are averaged at multiple levels, and therefore can be
considered to be continuous (Metro data)
Explanatory variables, denoted by “R” in Eqn. 6
(Unreliability variables are used in the regression analyses)
[>1 min]-early performance
The frequency (% expressed as decimal) with which a bus line departs from its
schedule time points earlier than the tolerance window (observed over the
study time frame, averaged across all time points of a line, and available for
each time period of weekday ) – (Metro data)
[>5 min]-late performance
The frequency (% expressed as decimal) with which a bus line departs from its
schedule time points later than the tolerance window (observed over the study
time frame, averaged across all time points of a line, and available for each
time period of weekday ) – (Metro data)
Line accessibility/connectivity measures, denoted by “A” in Eqn. 6
Mean population density
Mean of the population densities (persons per sq. mi.) of all census tracts
served by a line (ACS 2007-2011)
Employment accessibility
Total jobs within quarter-mile buffer (either side) of a line (2008 employment
data from SCAG)
Planned service quality of line – denoted by “S” in Eqn. 6
Stops per mile
Count of total stops per unit length of a line; proxies average operating speed;
also note that rapid/express lines have significantly fewer stops per mile (June
2011 service change data from Metro)
Scheduled headway
Scheduled headway (in minutes) of a line; a key level of service parameter that
determines service frequency (June 2011 service change data from Metro)
Other line controls – denoted by “C” in Eqn. 6
Total stops
Count of total stops for a line; note that lines with more stops, holding stops
per mile constant, serve longer distances and also provide more opportunities
to transfer to other lines (Metro or non-Metro) – (June 2011 service change
data from Metro)
Transit alternatives
Count of all stops served by other regional transit operators, which can
potentially serve as alternatives to Metro, within quarter-mile buffer (either
side) of a line (2008 regional transit service data from SCAG)
Note: Line refers to a directional line; Study period refers to the Dec 2011 to May 2012 period; Tolerance window refers to the 1-minute early to
5-minute late window with respect to schedule; Weekdays are considered only
35
Weekday AM peak = 3 hours; Mid-Day = 6 hours; PM peak = 4 hours; Night = 5 hours.
The Demand for Reliable Travel Chakrabarti (2015)
54
Following are some clarifications/discussions regarding the patronage model.
Parameter groups A and S are not correlated in the current dataset. This is because two lines
in opposite directions serving the same corridor may traverse through identical spatial
contexts (hence have identical line-level population density and employment accessibility
measures), but their supply parameters (e.g. scheduled headway) will differ based on the
predominant direction of passenger traffic flow.
Second, (un)reliability (R) is exogenous to the model. Although, all else equal, aggregate line
patronage (P) can negatively affect line service reliability, scheduling takes into account
possible delays due to boardings/alightings in various volumes at various stops along a line.
Schedule adjustments (also line re-configurations) are made based on empirical data, often
targeted towards improving on-time service (or for meeting a pre-determined on-time service
target). Therefore, P cannot be assumed to systematically influence R if scheduling/service-
planning is performed efficiently.
There is some endogeneity/simultaneity between planned service quality (S) of a line and its
patronage (P), all else equal. In the patronage model, we should expect a line’s scheduled
headway to be highly correlated with (and to a large extent determined by) average per-hour
line boardings during a given time period of day, once other variables are held constant. This
is because headways may be adjusted periodically based on patronage signals to
accommodate people in buses more efficiently (i.e. maximize passenger loading per vehicle)
and consequently increase productivity (i.e. maximize boardings relative to investment)
under constraints (e.g. resource, coverage, quality-of-service, etc.). Since ignoring this
supply-demand endogeneity may cause bias, I specify a system of simultaneous equations.
The Demand for Reliable Travel Chakrabarti (2015)
55
The supply-side equation for scheduled headway (for time period of the day “t”) is
specified as follows:
𝑯 𝒍𝒕
= 𝜶 𝟎 + 𝜶 𝟏 𝑷 𝒍𝒕
+ 𝜶 𝟐 𝒁 𝒍𝒕
+ 𝜸 …................. (Eqn. 7)
Where: H refers to scheduled headway; P is the same measure of patronage as in Eqn. 6; Z is a
vector of line characteristics such that P signals line productivity once Z is controlled, for line l at
time period t. 𝛾 is the error term. Read “t” as AM peak, Mid-Day, PM peak, and Night.
The variables included in Z are: total stops (see Table 3-1); line length (in miles), and; an
indicator variable (1/0) for rapid bus line. Rapid buses have higher per-vehicle capacity
operational cost. Holding Z variables constant, and at a given headway, P indicates both per-
vehicle occupancy (load factor) and productivity (utilization factor) of a line, and hence informs
decisions regarding headway adjustments. Headway adjustments are typically made by moving
around buses in the existing fleet across lines, or by introducing new buses, or retiring existing
buses.
Eqn. 7 has some limitations. For example, it cannot explain why unproductive bus lines
are operated at policy headways during low demand times and at low demand places. It also
cannot explain why service frequency of some overcrowded, relatively productive lines cannot
be increased due to limited resources. Nevertheless, I expect a close supply-demand match in the
cross-sectional dataset, particularly during the peak period.
I estimate Eqn. 6 and Eqn. 7 together, using the three-stage least-squares regression
(3SLS) method that estimates all coefficients of the entire system simultaneously. 3SLS accounts
for the correlations of the errors across different equations of the system, and is a relatively more
efficient method compared to a two-stage least-squares approach (Zellner and Theil 1962). 3SLS
The Demand for Reliable Travel Chakrabarti (2015)
56
helps analyze the supply-demand simultaneous relationship explicitly. I use Stata
36
to estimate
the models.
I have analyzed correlations among independent variables, and ensured that regression
models do not suffer from significant multicollinearity bias. No pairwise correlation exceeds +/-
0.5.
3.4 Descriptive statistics
Summary statistics of the variables used in the regression models are given in Table 3-2. There
are several observations to be drawn about the Metro bus system.
First, there is great variation in average hourly line boardings – both across time periods
and across lines within each time period. Patronage is higher in the peak than the off-peak;
boarding numbers drop sharply after 7 PM. Transit activity is highest in the PM peak and lowest
at night.
Second, buses run [>1 min]-early only about 3-4% of the time on average; earliness is
therefore a rare occurrence. This is expected, because early departures can be avoided relatively
easily by dwelling at or near time points. Interestingly, service reliability seems to be highest
during the AM peak. At other times of the day, buses run [>5 min]-late around 20-26% of the
time. High service unreliability during the highly congested PM peak is expected, but poor
system-wide performance at night is surprising. The Metro bus system seems to have a large
room for improving on-time service by minimizing the frequency of [>5 min]-late departures
from time points.
36
Stata’s technical note on the 3SLS regression method is available at http://www.stata.com/manuals13/rreg3.pdf
(accessed on 8/11/2014)
The Demand for Reliable Travel Chakrabarti (2015)
57
Finally, on average, Metro bus lines serve corridors with very high population density,
and they also provide very high employment accessibility. This suggests high potential transit
demand in the Metro service area. However, the average scheduled headway is quite long (over
20 minutes) even during the peak. Long headways are inconsistent with high potential demand.
Table 3-2.
Descriptive statistics
Unit of observation: Directional Metro bus line
Variable N Mean SD Min Max Median
Average per-hour boardings
Peak
AM Peak 276 252.16 223.20 0.00 1008.67 177.83
PM Peak 280 281.30 267.12 0.00 1225.75 199.38
Off-peak
Mid-Day 283 223.95 217.23 0.00 1267.67 158.00
Night 283 62.28 79.86 0.00 590.60 31.80
[>1 min]-early performance
(% expressed as decimal)
Peak
AM Peak 267 0.03 0.03 0.00 0.26 0.03
PM Peak 270 0.04 0.04 0.00 0.47 0.03
Off-peak
Mid-Day 270 0.03 0.02 0.00 0.17 0.03
Night 266 0.03 0.03 0.00 0.23 0.02
[>5 min]-late performance
(% expressed as decimal)
Peak
AM Peak 267 0.14 0.07 0.01 0.37 0.12
PM Peak 270 0.26 0.10 0.03 0.62 0.25
Off-peak
Mid-Day 270 0.20 0.10 0.00 1.00 0.19
Night 266 0.26 0.14 0.00 1.00 0.25
Mean population density (persons per sq. mi.) 283 12864 4242 2057 28630 12400
Employment accessibility (no.) 283 233263 214970 13333 1085772 157288
Stops (no.) per mile 283 4.22 1.77 0.16 11.03 4.30
Scheduled headway (min)
Peak
AM Peak 274 23.55 15.53 5 90 18
PM Peak 278 24.51 16.94 5 90 18
Off-peak
Mid-Day 277 40.81 58.85 6 360 28
Night 281 71.06 73.23 7 360 45
Total stops (no.) 283 69.09 34.16 4 164 67
Transit alternatives (no.) 283 551.78 520.85 14 2548 350
Line length (mi) 283 17.13 6.50 4.15 40.44 16.54
Note: Data for all variables are not available for all lines; All lines do not operate at all times of the day; Weekdays are considered only.
The Demand for Reliable Travel Chakrabarti (2015)
58
4. Results: Model Summaries and Observations
4.1 Simultaneous-equations models
I estimate 3SLS models (Models A and B in Table 3-3) of average per-hour bus line boardings
for the weekday peak (observations in the AM and PM peaks are pooled) and off-peak
(observations for the mid-day and night periods are pooled) periods. I include time-period
indicator variables to control for differences in demand.
The Demand for Reliable Travel Chakrabarti (2015)
59
Table 3-3.
3SLS (simultaneous-equations) models of bus line patronage
Variable
Weekday Peak (Model A) Weekday Off-Peak (Model B)
Parameter
Estimate
Std. Est.
Parameter
Estimate
Std. Est.
Eqn. 6: Average per-hour bus line boardings
Line unreliability
[>1 min]-early performance -44.0961 -0.0040 -15.6738 -0.0020
[>5 min]-late performance -158.4622 ** -0.0680 -83.0181 ** -0.0520
Line accessibility/connectivity
Mean population density 0.0224 *** 0.3780 0.0146 *** 0.3440
Employment accessibility 0.0003 *** 0.2850 0.0003 *** 0.3700
Planned line service quality
Stops per mile -21.7915 *** -0.1540 -9.4082 *** -0.0920
Scheduled headway (endogenous) -7.4937 *** -0.4810 -1.2863 *** -0.4290
Controls
Total stops 2.3818 *** 0.3200 1.1364 *** 0.2080
Transit alternatives -0.0591 ** -0.1230 -0.0657 *** -0.1870
Time-period indicator for Model A
(Ref=AM peak)
PM Peak 37.4852 *** 0.0750
Time-period indicator for Model B
(Ref=Mid-Day)
Night -126.0666 *** -0.3450
Intercept 62.1912 31.2261
N 537 536
R-square 0.67 0.51
Eqn. 7: Scheduled headway
Endogenous patronage variable
Average per-hour bus line boardings -0.0406 *** -0.6330 -0.1590 *** -0.4770
Controls (line characteristics)
Total stops -0.0717 *** -0.1500 -0.4094 *** -0.2250
Line length 0.1533 * 0.0620 -0.0276 -0.0030
Rapid line (Indicator) -12.2686 *** -0.2700 -37.8262 *** -0.2210
Intercept 39.2626 *** 110.5577 ***
N 537 536
R-square 0.53 0.22
*p<0.10; **p<0.05; ***p<0.01
Unit of observation is a directional Metro bus line
The Demand for Reliable Travel Chakrabarti (2015)
60
Results of Models A and B support my hypothesis, and are generally consistent with
expectation. I find evidence that line service unreliability (specifically [>5 min]-late
performance) is negatively associated with patronage, on average, during both peak and off-peak
periods; effects are statistically significant at the 95% confidence level. Comparison of
coefficients shows that the effect size is relatively greater during the peak compared to the off-
peak. Lines that run [>5 min]-late (or depart more than 5 minutes late from time point stops) less
frequently are therefore able to attract more patronage, all else equal.
Per model estimates, [>1 min]-early performance does not have any influence on line
patronage (coefficients have expected signs, but are not statistically different from zero). This is
not surprising, and it does not imply that earliness is not onerous from passengers’ perspective.
For Metro, earliness is a rare occurrence, since if a bus arrives at or near a time point more than 1
minute early, it often has the option to dwell (or lay over) and meet the on-time departure
window. The average magnitude and variation of the [>1 min]-early performance variable across
Metro bus lines is insufficient to provide conclusive evidence regarding its importance. It is
however true that Metro has very little scope of increasing system-wide patronage by minimizing
its current level of [>1 min]-early service.
Table 3-3 reveals other significant issues. Lines traversing corridors with higher
population density and providing greater employment accessibility have significantly greater
patronage, all else equal. Also, frequent stop-making (captured through more stops per unit line
mile) seems to negatively affect patronage; consequently, rapid and express buses seem to draw
greater patronage, all else equal. On average, there is greater per-hour system-wide patronage in
the PM peak compared to the AM peak, and over the mid-day compared to night. All else equal,
lines with more stops attract greater number of passengers by accommodating more trips and
The Demand for Reliable Travel Chakrabarti (2015)
61
facilitating more transfer opportunities; lines traversing sub-regions with alternative/competing
city/municipal transit services have comparatively lower patronage.
By addressing endogeneity using the 3SLS approach, I show that while patronage signals
inform headway planning, headway also determines patronage, all else equal, across both peak
and off-peak periods. It seems that independent headway reductions can potentially promote
patronage by accommodating latent demand. This finding is consistent with prior literature, and
provides additional evidence for policy makers.
Observe that the R-square value of the supply-side equation in the off-peak period (Eqn.
7 of Model B in Table 3-3) is relatively lower (0.22). This is expected, since many more bus
lines operate at policy headways regardless of their low patronage in the off-peak to ensure
uninterrupted service coverage.
4.2 Reduced forms and alternative tests
Note that I estimated the system of simultaneous equations to derive unbiased estimates
(particularly for the scheduled headway variable) and establish causal connections.
One might argue, however, that headway is only used as a control in Eqn. 6. Therefore,
the headway-patronage association, and hence the biased headway parameter estimate, can
simply be ignored. To test this, I estimate alternative single-stage ordinary least-squares (OLS)
models of line patronage following the Eqn. 6 framework (see Models C and D in Table 3-4).
Conversely, one might argue that the headway variable should be dropped if the
headway-patronage endogeneity is significant. I therefore test reduced-form single-stage
ordinary least-squares (OLS) models of line patronage (see Models E and F in Table 3-5).
Service unreliability ([>5 min]-late performance) remains a significant determinant of bus
line patronage during both peak and off-peak periods across both specifications.
The Demand for Reliable Travel Chakrabarti (2015)
62
Table 3-4.
Alternative OLS models of bus line patronage
Variable
Weekday Peak (Model C) Weekday Off-Peak (Model D)
Parameter
Estimate
Std. Est.
Parameter
Estimate
Std. Est.
Eqn. 6: Average per-hour bus line boardings
Line unreliability
[>1 min]-early performance -9.8812 -0.0009 74.9967 0.0098
[>5 min]-late performance -130.6984 ** -0.0561 -93.5316 ** -0.0585
Line accessibility/connectivity
Mean population density 0.0256 *** 0.4314 0.0181 *** 0.4283
Employment accessibility 0.0003 *** 0.2264 0.0004 *** 0.4468
Planned line service quality
Stops per mile -28.9430 *** -0.2049 -17.1164 *** -0.1665
Scheduled headway -6.9338 *** -0.4451 -0.3496 *** -0.1166
Controls
Total stops 2.8533 *** 0.3829 1.7327 *** 0.3174
Transit alternatives -0.0532 * -0.1104 -0.0740 *** -0.2107
Time-period indicator for Model C
(Ref=AM peak)
PM Peak 47.1050 *** 0.0942
Time-period indicator for Model D
(Ref=Mid-Day)
Night -149.7420 *** -0.4095
Intercept 6.9467 -72.1510 ***
N 537 536
Adj. R-square 0.54 0.58
*p<0.10; **p<0.05; ***p<0.01
Unit of observation is a directional Metro bus line
The Demand for Reliable Travel Chakrabarti (2015)
63
Table 3-5.
Reduced-form OLS models of bus line patronage
Variable
Weekday Peak (Model E) Weekday Off-Peak (Model F)
Parameter
Estimate
Std. Est.
Parameter
Estimate
Std. Est.
Eqn. 6 (minus headway variable): Average per-hour bus line boardings
Line unreliability
[>1 min]-early performance -30.6511 -0.0028 84.3454 0.0110
[>5 min]-late performance -252.7600 *** -0.1085 -98.7119 ** -0.0618
Line accessibility/connectivity
Mean population density 0.0345 *** 0.5824 0.0194 *** 0.4584
Employment accessibility 0.0005 *** 0.4505 0.0004 *** 0.4807
Planned line service quality
Stops per mile -39.5694 *** -0.2801 -18.5745 *** -0.1805
Controls
Total stops 3.4124 *** 0.4580 1.9054 *** 0.3488
Transit alternatives -0.0789 ** -0.1639 -0.0777 *** -0.2210
Time-period indicator for Model E
(Ref=AM peak)
PM Peak 53.9618 *** 0.1079
Time-period indicator for Model F
(Ref=Mid-Day)
Night -156.8204 *** -0.4291
Intercept -293.7423 *** -113.6563 ***
N 537 536
Adj. R-square 0.54 0.57
*p<0.10; **p<0.05; ***p<0.01
Unit of observation is a directional Metro bus line
It is often argued that for short-headway (say <10 minutes) lines, schedule adherence is
less important and headway adherence is critical. While uniformity in spacing is certainly
important for these lines, on-time service may be too, since some passengers may need to make
highly time-critical transfers to long-headway lines. Also, erratic headways are a consequence of
schedule deviations. Nevertheless, I tested 3SLS models by excluding lines with scheduled
headway under 10 minutes. The unreliability effect (that of the [>5 min]-late performance
variable) remains fairly consistent across peak (Std. Est. = -0.06 at the 95% level) and off-peak
The Demand for Reliable Travel Chakrabarti (2015)
64
(Std. Est. = -0.05 at the 90% level) periods. In general, the effect size of the significant
unreliability variable is stable.
5. What about Rail?
Bus and rail are different systems. Rail lines are outliers in a combined (bus+rail lines) dataset.
The 16 rail observations (4 lines; 2 directions; 2 time blocks) per period (weekday peak and off-
peak) have very high patronage (10 times more than bus observations on average) and on-time
performance (over 95%). Rail lines can therefore bias results in favor of the unreliability
parameters if the difference between bus and rail systems is not controlled for.
In order to avoid bias, I use additional statistical controls. I add a line capacity
37
variable
to the set of variables used in the bus-only analysis. I also include a rail line indicator variable
that takes the value “1” for a directional rail line and “0” otherwise. The indicator variable is
definitely too restrictive, but it allows me to capture the many unobserved factors that determine
rail travel demand (such as on-board and station-level comfort, safety, and amenities), and also
account for the difference between bus and rail systems.
I test a complete set of regression models similar to the bus-only analysis and find that
the magnitude and significance of effect of the influential service unreliability (late performance)
variable is reduced across weekday peak and off-peak periods. The loss of significance of the
unreliability parameter after including rail observations can be attributed to the rail line indicator
variable. Estimating the same models without the indicator variable inflates the magnitude and
significance of the unreliability parameter; however, this leads to bias due to the many
(unobserved) omitted variables that influence rail travel demand, and that distinguish between
37
Average number of “standard” cars per equipment serving a line.
The Demand for Reliable Travel Chakrabarti (2015)
65
bus and rail lines. I do not report the models because they cannot isolate the marginal influence
of service (un)reliability on Metro’s system-wide patronage.
The significant large positive effect of the rail line indicator variable is valuable in itself.
It shows that on average, and all else equal, Metro rail attracts significantly greater patronage
than Metro bus. But again, the extent to which this effect can be attributed to rail’s near-perfect
on-time performance is unclear. Also, one should not generalize that rail is more effective than
bus in promoting patronage. It is possible that the Los Angeles Metro rail system is appropriately
planned along corridors most primed for rail investments; consequently patronage is high.
Therefore, it is reasonable to infer that rail lines extended to areas with high latent demand that
significantly increase regional accessibility will help promote patronage.
6. Discussions
This study of the Los Angeles Metro bus system reveals a statistically significant positive
association between line service reliability and line patronage. The association appears to be
stronger in the weekday peak compared to the off-peak. Peak-period travelers (under greater time
and tour-scheduling constraints on average, relative to off-peak travelers) are expected to be
more sensitive to schedule deviations, and less willing to accommodate unreliability by adding
time cushions. Consequently, it is possible that service reliability will have a bigger overall
impact on patronage in the peak period that presumably has higher latent demand as well.
Based on model estimates, and all else equal, 10-percentage point lower system-wide
average [>5 min]-late performance is associated with higher patronage in the range of 5-6%
(compared to the baseline). Therefore, to the extent that results from this cross-sectional study
that analyzes associations can be used to derive causal connections, Metro should expect
significant patronage gains from reliability investments. The actual effect size, however, may be
The Demand for Reliable Travel Chakrabarti (2015)
66
smaller or larger. Disproportionate increase in reliability across lines may lead to some
redistribution of patronage as existing patrons may shift from one line or route to another for
ensuring relatively more reliable travel. But again, increase in reliability even for a single line
can make many associated transit routes (linked chains) more viable and attractive, and
consequently lead to greater patronage gains than anticipated.
Although I conclude that increasing the on-time service frequency through better
scheduling decisions and efficient real-time system management has the potential of promoting
patronage, we must recognize that minimizing service disruptions is also critical. Riders can be
sensitive to even one (or just a few) bad experience(s). It is also probably important to
simultaneously aim at minimizing the average and variance of schedule deviation (in minutes),
and monitor that uniform headways and running times are maintained. In sum, there is
tremendous potential for promoting patronage by improving many dimensions of service
reliability.
Results should be generalized with caution. We must recognize that Los Angeles is
uniquely primed for both transit service provision and consumption. The high regional
population and employment density, sociodemographic diversity, and traffic congestion
collectively contribute to high latent demand for good-quality transit service. Metro, along with
many other regional operators, already provides extensive transit service coverage of reasonably
good quality. Consequently, we might expect that strategic improvements in service reliability
will more effectively attract patronage in Los Angeles than many other urban areas.
7. Conclusion
This analysis of the Los Angeles Metro bus system provides the first empirical evidence that
transit service reliability determines patronage. Transit mode choice (in places where latent
The Demand for Reliable Travel Chakrabarti (2015)
67
demand exists) and route selection (in transit-rich urban areas) are based, in part, on reliability.
Improvements in service reliability can make transit more attractive, and prepare it for competing
with alternate modes in the presence of latent transit travel demand. Reliability investments can
therefore help increase service consumption relative to supply, and make transit lines and
systems more productive.
This study does not demonstrate that service coverage expansions and other quality
improvements are less effective means of promoting patronage. Reliability is just one critical
dimension of transit service that should not be undermined during operating existing systems and
planning new projects.
Increasing on-time performance is complex. It must include external strategies that are
not solely within the realm of public transit policy or under control of transit managers. For
instance, bus-only lanes and signal preemption systems are theoretically straightforward means
of improving on-time service. However, implementation requires coordination among many
agencies responsible for managing multi-modal urban transportation networks. But there are
other internal strategies too. For example, efficient scheduling by analyzing historic archived
AVL/APC data, real-time rerouting of vehicles around incidents, better holding strategies, stop-
consolidation, driver training and incentive programs, periodic system maintenance, and real-
time information sharing via mobile devices can improve both system performance and users’
perception of service quality. There is no doubt that advances in intelligent transportation
infrastructure and information technology will help manage and operate transit systems more
efficiently, and consequently help improve service reliability.
This study is expected to inform public transit agencies in Los Angeles and other US
metropolitan regions as they continue to invest in the future.
The Demand for Reliable Travel Chakrabarti (2015)
68
8. References
1) Abkowitz, M. D., & Engelstein, I. (1983). Factors affecting running time on transit routes.
Transportation Research Part A: General, 17(2), 107–113.
2) APTA (2012). MAP-21, A Guide to Transit-Related Provisions. Washington, D.C.: American
Public Transit Association. Retrieved August 11, 2014, from
http://www.apta.com/gap/legissues/authorization/Documents/APTA%20MAP-
21%20Guide.pdf.
3) Asensio, J., & Matas, A. (2008). Commuters’ valuation of travel time variability.
Transportation Research Part E , 44, pp. 1074–1085.
4) Badoe, D. A., & Miller, E. J. (2000). Transportation-land-use interaction: empirical findings
in North America, and their implications for modeling. Transportation Research Part D:
Transport and Environment, 5(4), 235–263.
5) Bhat, C., & Sardesai, R. (2006). The impact of stop-making and travel time reliability.
Transportation Research Part B , 40, pp. 709–730.
6) Camus, R., Longo, G., & Macorini, C. (2005). Estimation of transit reliability level-of-
service based on automatic vehicle location data. Transportation Research Record: Journal
of the Transportation Research Board, 1927(1), 277-286.
7) Cantwell, M., Caulfield, B., & O’Mahony, M. (2009). Examining the Factors that Impact
Public Transport Commuting Satisfaction. Journal of Public Transportation, 12(2), 1-21.
8) Cevallos, F., Wang, X., Chen, Z., & Gan, A. (2011). Using AVL data to improve transit on-
time performance. Journal of Public Transportation, 14(3), 21-40.
9) Chen, X., Yu, L., Zhang, Y., & Guo, J. (2009). Analyzing urban bus service reliability at the
stop, route, and network levels. Transportation Research Part A, 43, pp. 722–734.
The Demand for Reliable Travel Chakrabarti (2015)
69
10) de Oña, J., de Oña, R., Eboli, L., & Mazzulla, G. (2013). Perceived service quality in bus
transit service: a structural equation approach. Transport Policy, 29, 219-226.
11) dell’Olio, L., Ibeas, A., & Cecín, P. (2010). Modelling user perception of bus transit quality.
Transport Policy, 17(6), 388-397.
12) Eboli, L., & Mazzulla, G. (2009). A new customer satisfaction index for evaluating transit
service quality. Journal of Public Transportation, 12(3), 21-37.
13) Eboli, L., & Mazzulla, G. (2010). How to capture the passengers’ point of view on a transit
service through rating and choice options. Transport reviews, 30(4), 435-450.
14) El-Geneidy, A. M., Strathman, J. G., Kimpel, T. J., & Crout, D. T. (2006). Effects of bus stop
consolidation on passenger activity and transit operations. Transportation Research Record:
Journal of the Transportation Research Board, 1971(1), 32-41.
15) El-Geneidy, A. M., Hourdos, J., & Horning, J. (2009). Bus transit service planning and
operations in a competitive environment. Journal of Public Transportation, 12(3), 39-59.
16) El‐Geneidy, A. M., Horning, J., & Krizek, K. J. (2011). Analyzing transit service reliability
using detailed data from automatic vehicular locator systems. Journal of Advanced
Transportation, 45(1), 66-79.
17) Giuliano, G. (2011) “Transportation policy: Public transit, settlement patterns and equity in
the US,” chapter 28 in N. Brooks, K. Donaghy and G. Knaap, eds., Oxford University Press
Handbook of Urban Economics and Planning. New York: Oxford University Press.
18) Glascock, J. (1997). Research on customer requirements for transit service design and
delivery. Transportation Research Record: Journal of the Transportation Research Board,
1604(1), 121-127.
The Demand for Reliable Travel Chakrabarti (2015)
70
19) Hensher, D. A., Stopher, P., & Bullock, P. (2003). Service quality––developing a service
quality index in the provision of commercial bus contracts. Transportation Research Part A:
Policy and Practice, 37(6), 499-517.
20) Iseki, H., & Taylor, B. D. (2010). Style versus Service? An Analysis of User Perceptions of
Transit Stops and Stations. Journal of Public Transportation, 13(3), 23-48.
21) Kittelson & Associates. (2013). TCRP Report 165: Transit Capacity and Quality of Service
Manual (3
rd
ed.) Washington, D.C.: Transport Research Board.
22) Lam, T.C., & Small, K.A. (2001). The value of time and reliability: measurement from a
value pricing experiment. Transportation Research Part E, 37 (2001), pp. 231–251.
23) Lin, J., Wang, P., & Barnum, D. T. (2008). A quality control framework for bus schedule
reliability. Transportation Research Part E: Logistics and Transportation Review, 44(6),
1086-1098.
24) Liu, H.X., Recker, W., & Chen, A. (2004). Uncovering the contribution of travel time
reliability to dynamic route choice using real-time loop data. Transportation Research Part
A: Policy and Practice, Volume 38, Issue 6, 435-453.
25) Nakanishi, Y. J. (1997). PART 1: Bus: Bus Performance Indicators: On-Time Performance
and Service Regularity. Transportation Research Record: Journal of the Transportation
Research Board, 1571(1), 1-13.
26) Nam, D., Park, D., & Khamkongkhun, A. (2005). Estimation of Value of Travel Time
Reliability. Journal of Advanced Transportation, Vol. 39, No. 1, pp. 39-61.
27) Noland, R., Small, K., Koskenoja, P., & Chu, X. (1998). Simulating Travel Reliability.
Regional Science and Urban Economics, 28 pp. 535–564.
The Demand for Reliable Travel Chakrabarti (2015)
71
28) NHTS (2014). Retrieved August 11, 2014, from National Household Travel Survey:
http://nhts.ornl.gov/.
29) NTD (2014). Retrieved August 11, 2014, from National Transit Database:
http://www.ntdprogram.gov/ntdprogram/data.htm.
30) NTD. (2012). 2011 National Transit Summaries and Trends. Available at
http://www.ntdprogram.gov/ntdprogram/pubs/NTST/2011%20National%20Transit%20Sum
maries%20and%20Trends-Complete.pdf: National Transit Database, Federal Transit
Administration, US Department of Transportation (Retrieved August 11, 2014).
31) Nurul Habib, K. M., Kattan, L., & Islam, M. T. (2011). Model of personal attitudes towards
transit service quality. Journal of Advanced Transportation, 45, 271–285.
32) Polus, A. (1978). Modeling and Measurements of Bus Service Reliability. Transportation
Research, 12(6), 253-256.
33) Santos, A., McGuckin, N., Nakamoto, H. Y., Gray, D., & Liss, S. (2011). Summary of Travel
Trends: 2009 National Household Travel Survey. Available at http://nhts.ornl.gov/: Federal
Transit Administration, US Department of Transportation.
34) Small, K., Winston, C., & Yan, J. (2005). Uncovering the Distribution of Motorists'
Preferences for Travel Time and Reliability. Econometrica, Vol. 73, No. 4, pp. 1367-1382.
35) Small, K.A. (1982). The scheduling of consumer activities: work trips. American Economic
Review, 72, 467–479.
36) Sterman, B. P., & Schofer, J. L. (1976). Factors Affecting Reliability of Urban Bus Services.
Transportation Engineering Journal of ASCE, 102(TE1), 147-159.
37) Strathman, J. G., & Hopper, J. R. (1993). Empirical analysis of bus transit on-time
performance. Transportation Research Part A: Policy and Practice, 27(2), 93-100.
The Demand for Reliable Travel Chakrabarti (2015)
72
38) Strathman, J. G., Dueker, K. J., Kimpel, T., Gerhart, R., Turner, K., Taylor, P., & Hopper, J.
(1999). Automated bus dispatching, operations control, and service reliability: Baseline
analysis. Transportation Research Record: Journal of the Transportation Research Board,
1666(1), 28-36.
39) Sweet, M. N., & Chen, M. (2011). Does regional travel time unreliability influence mode
choice? Transportation, 38(4), 625-642.
40) Taylor, B. D., Miller, D., Iseki, H., & Fink, C. (2009). Nature and/or nurture? Analyzing the
determinants of transit ridership across US urbanized areas. Transportation Research Part A:
Policy and Practice, 43(1), 60-77.
41) Tilahun, N. Y., & Levinson, D. M. (2010). A Moment of Time: Reliability in Route Choice
Using Stated Preference. Journal of Intelligent Transportation Systems: Technology,
Planning, and Operations, 14(3), 179-187.
42) Tyrinopoulos, Y., & Antoniou, C. (2008). Public transit user satisfaction: Variability and
policy implications. Transport Policy, 15(4), 260-272.
43) Wachs, M. (1976). Consumer attitudes toward transit service: an interpretive review. Journal
of the American Institute of Planners, 42(1), 96-104.
44) Xuan, Y., Argote, J., & Daganzo, C. F. (2011). Dynamic bus holding strategies for schedule
reliability: Optimal linear control and performance analysis. Transportation Research Part
B: Methodological, 45(10), 1831-1845.
45) Yetiskul, E., & Senbil, M. (2012). Public bus transit travel-time variability in Ankara
(Turkey). Transport Policy, 23, 50-59.
46) Zellner, A. & Theil, H. (1962). Three-stage Least Squares: Simultaneous Estimation of
Simultaneous Equations. Econometrica, Vol. 30, No. 1, pp. 54-78.
The Demand for Reliable Travel Chakrabarti (2015)
73
Chapter 4.
Demand for Transit Service Reliability – Analysis of Variation in Boardings across Bus Stops
This chapter uses schedule time point stop level data from the Los Angeles Metro bus system to
investigate whether reliability influences stop-level line boardings. I present evidence that higher
average service punctuality, consistency in the level of punctuality over time, and lower variation
in schedule deviation over time are associated with greater ridership, all else equal. The effect of
reliability on ridership is statistically significant during weekday AM and PM peaks. The effect of
reliability on peak-period ridership is moderated by headway; demand for reliability seems to be
higher for lines with relatively longer headways (highest in the 20-30 minute headway range).
From an urban planning perspective, this chapter provides more evidence that good service quality
can effectively compliment transformations in the urban fabric brought about by coordinated land
use – transit plans to promote transit use.
The Demand for Reliable Travel Chakrabarti (2015)
74
1. Introduction
Over the past several decades, U.S. planning authorities, local governments, and public transit
agencies have collectively worked towards coordinating innovative land use policies with
unprecedented transit service improvements for promoting transit use, and consequently
reducing various negative externalities associated with extreme automobile dependence.
Governments
38
have steadily increased funding for transit, and communities
39
have embraced
increased tax burdens for giving transit the opportunity of achieving its goals. Unfortunately,
however, transit’s share in the U.S. travel market continues to be relatively small
40
and
productivity of public transit systems across the nation continues to decline.
41
Consequently, the
search for magic planning-policy bullets to generate greater enthusiasm around transit continues.
Experts argue that key to increasing transit’s market share is to invest in those dimensions
of service quality that travelers value most (Giuliano, 2011). In this study, I explore whether
service reliability, a service quality factor that has received relatively low empirical attention in
the past, can boost ridership. My research helps transit managers formulate policies to rejuvenate
the U.S. public transit industry.
This study uses Los Angeles Metro
42
bus system data, collected as part of the ADMS
research project. I perform a cross-sectional analysis of the variation in average line boardings
38
Data from the National Transit Database (http://www.ntdprogram.gov/ntdprogram/) shows that between 1991 and
2012, total annual government (all levels combined) funding support for transit has increased from $22 billion to
$58.5 billion.
39
e.g. Measure R, a voter-approved half-cent sales tax in Los Angeles County for financing new transportation
projects and accelerating planned ones took effect in July 2009. A major portion of the tax receipts are channelized
towards the regional rail and rapid bus transit program.
40
Between 1990 and 2009, transit has consistently maintained an estimated mode share of less than 2% of all trips
made in the US (Santos et al., 2011).
41
Data from the National Transit Database (http://www.ntdprogram.gov/ntdprogram/) shows that unlinked
passenger trips per revenue vehicle hour (measure of “service effectiveness”) has declined from 46.5 to 39.4, and
farebox recovery ratio has dropped from 36.4 to 33.1, over the 1991-2012 period.
42
Transit service operated by the Los Angeles County Metropolitan Transportation Authority (MTA).
The Demand for Reliable Travel Chakrabarti (2015)
75
across around
43
1300 time point stops of about 300 directional bus lines for five time periods
(peaks and off-peaks) of a typical weekday. The analysis helps determine the marginal impact of
line service reliability at a stop on the number of passengers who board the line at the stop. I test
three measures of service reliability (including the industry’s standard schedule adherence
measure - “on-time performance
”
[OTP], and a new schedule deviation measure derived using a
historical archive of Metro’s real-time GPS-based vehicle location data feeds) at the [time point]
stop level and two modeling approaches (e.g. ordinary least squares regression and two-stage
least squares regression) in this study. The results provide reliable information to both
practitioners and scholars. To foreshadow briefly, I find systematic evidence that reliability
drives in part the demand for transit.
The rest of the chapter is organized as follows: Section 2 outlines the research context
and reviews relevant literature; Section 3 explains the research design; Section 4 describes
variables and their summary statistics; Section 5 presents various analyses, results of regression
models, and discussions on findings; Section 6 summarizes the broad takeaways and highlights
limitations; and Section 7 concludes the chapter with policy implications.
2. Research Context
2.1 Significance of transit service reliability
I have summarized theory and evidence on the demand for travel time reliability in previous
chapters. I have also summarized findings from past studies that have specifically explored the
significance of transit service reliability. Recall that passengers are found to consistently rank
unreliability among the top inconvenience costs associated with transit travel (e.g. studies by
Wachs, 1976; Glascock, 1997; Eboli and Mazzulla, 2007; Tyrinopoulos and Antoniou, 2008;
43
The actual sample size varies by time of day and by reliability measure used (due to data [un]availability).
The Demand for Reliable Travel Chakrabarti (2015)
76
Cantwell et al., 2009; dell’Olio et al., 2010; Iseki and Taylor, 2010; and Nurul Habib et al., 2011
are illustrative). And outputs from theoretical/simulation models demonstrate how service
unreliability [negatively] affects passengers’ wait times (e.g. Turnquist, 1978; Bowman and
Turnquist, 1981; Chen and Chen, 2009), departure time choice and travel cost (Benezech and
Coulombel, 2013), and overall transit network performance (e.g. Turnquist and Bowman, 1980).
In sum, past studies indicate that service reliability must matter to travelers and hence operators.
Many researchers have proposed new (and incrementally advanced) methods of
measuring transit service reliability (e.g. Polus, 1978; Nakanishi, 1997; Camus et al., 2005; Lin
et al., 2008; Chen et al., 2009) using available or state-of-the-art technologies.
44
Others have
assessed factors that contribute to (or cause) unreliability (e.g. Sterman and Schofer, 1976;
Abkowitz and Engelstein, 1983; Strathman and Hopper, 1993; Strathman et al., 1999; Chen et
al., 2009; Yetiskul and Senbil, 2012). And Abkowitz and Engelstein (1984), El-Geneidy et al.
(2006), El-Geneidy et al. (2009), El-Geneidy et al. (2011), and Xuan et al. (2011) have explored
and recommended innovative methods to improve reliability. It is established that efficient
network design, better system maintenance and human resource management, improved system
resilience, advanced system operations and management, coordinated multi-modal traffic
management, and real-time decision making and information sharing can enhance reliability.
Empirical investigation of the influence of service reliability, using any type of system
performance data or employing any type of measure, on demand (or ridership) has not been
performed in the past. The current study attempts to address this gap.
44
For example, Chen et al. (2009) suggested “punctuality index” based on routes, “deviation index” based on stops,
and “evenness index” based on stops as new reliability measures; and El-Geneidy et al. (2011) tested a transit run
time variation measure.
The Demand for Reliable Travel Chakrabarti (2015)
77
2.2 Measuring transit service reliability in practice
Although many original approaches to measuring transit service reliability have been proposed,
none have been tested beyond small corridors (specific routes/lines or segments thereof). Also,
different measures capture different dimensions of service reliability; there is no consensus
regarding a best or most comprehensive measure.
The U.S. Federal Transit Administration (FTA) recognizes reliability (refer the “Transit
Capacity and Quality of Service Manual”
45
) as a key dimension of transit’s quality of service.
FTA proposes several measures based on the source of unreliability, magnitude of impact, and
purpose of measurement. For example, service disruptions are captured through measures such
as: a) percent of scheduled trips that were cancelled, b) percent of scheduled time operations
were down, or c) average distance traveled between mechanical breakdowns.
46
Variants of these
measures are periodically reported to government funding/regulatory agencies.
47
More direct
measures of system unreliability affecting regular user-experience are captured through: a) on-
time performance (or OTP; most commonly the fraction of total trips that serve intermediate
schedule time point stops/stations between 1 minute early and 5 minutes late with respect to
schedule), b) headway adherence (or some metric of evenness of intervals between vehicle
arrivals at designated stops/stations), and c) excess wait time (or average schedule delay in
departure from designated stops/stations).
48
These measures, however, are comparatively
45
The third edition is available online at http://www.trb.org/main/blurbs/169437.aspx (accessed on 10/11/2014);
also see Kittelson & Associates. (2013).
46
Note that a), b), and c) can be measured for a given line or system-wide, averaged over a given time period (e.g.
month), and aggregated across different times of the day (e.g. peak and of-peak) and days of the week (weekdays
and weekends).
47
Refer National Transit Database, available online at http://www.ntdprogram.gov/ntdprogram/, for historic data.
48
Note that a), b), and c) can be measured for a given line, or for a given line at a given stop/station, or system-wide,
averaged over a given time period (e.g. month), and aggregated across different times of the day (e.g. peak and of-
peak) and days of the week (weekdays and weekends).
The Demand for Reliable Travel Chakrabarti (2015)
78
difficult to measure. For accurate and exhaustive data, GPS-based automatic vehicle location
(AVL) systems need to be installed in transit vehicles.
Currently, OTP, with some variation in definition, is most widely used as the reliability
indicator within the U.S. public transit industry. Despite its limitations (e.g. inability to capture
the average or variance of schedule deviation in minutes), it is a conceptually simple and
practically useful measure for transit planners. A low average OTP implies that a transit line is
mostly unable to adhere to schedules, and hence indicates associated problems such as frequent
(that may or may not be systematic) early or late arrivals/departures from stops/stations and
uneven headways (and consequently bunched vehicles) that may cause increased unpredictability
(that negatively affects demand) within the transit network.
Although a growing number of U.S. public transit operators are currently in the process
of adopting AVL technologies and developing historical archives of real-time system
performance information, the data is seldom used effectively for planning and decision-making
(El-Geneidy et al., 2011). This type of data can be a valuable resource for researchers, as this
current study demonstrates.
3. Study Design
The current research empirically analyzes the variation in line boardings across various time
point stops of various directional Metro bus lines over five different time periods (e.g. peaks and
off-peaks) of weekdays to determine the extent to which the observed variation can be attributed
to service reliability.
3.1 Study area
The greater Los Angeles metropolitan region constitutes the study area (see Figure 3-1 in the
previous chapter). Metro is the largest public transit operator in the region with a service area of
The Demand for Reliable Travel Chakrabarti (2015)
79
around 1500 sq. mi. 150+ Metro bus routes (300+ directional lines) carried more than 30
million
49
passengers per month over the December 2011-May 2012 study period.
3.2 Spatial units
The spatial unit of observation is a time point stop of a directional bus line. Since time points are
used for scheduling, schedule adherence/deviation measures can be estimated at the time point
stop level only (not at all stops on a line’s route). I exclude terminal time points considering data
quality.
Recall that the ridership measure of interest (say at a given time period of weekday t) is
the average number of boardings (say denoted by B) to a given directional line (say Lm) at a
given time point stop (say the unique stop ID is Sx). The unit of observation is therefore given by
B[Lm,Sx]
t
. It is possible that other directional lines Ln and Lp also serve the same stop Sx, and that
Sx is a designated time point stop location for those lines as well. In that case, B[Ln,Sx]
t
and
B[Lp,Sx]
t
are additional units of observations at the same stop location, in this study. In essence,
the number of passengers who get on to and get off from different lines at a given stop location
are different, and are determined by, for example, connectivity and quality factors of the
respective lines.
Figure 4-1 shows the locations of time point stops across various directional Metro lines,
all or some of which are used in subsequent analyses. A total of 1335 time point stops are
mapped; this constitutes about 10% of total Metro bus stops.
Hereafter, for simplicity, clarity and consistency, time point stops of each directional line
are referred to as “stops,” and directional lines are referred to as “lines.”
49
Total calendar-month boardings on average (refer http://www.metro.net/news/facts-glance/; accessed on
4/28/2014).
The Demand for Reliable Travel Chakrabarti (2015)
80
Therefore, each unit of observation is boardings to a particular line at a particular stop,
referred to as “stop-level line boardings” wherever relevant throughout the chapter.
Figure 4-1. Metro time point bus stops
Note: There is considerable overlapping of time point stops (i.e. dots) at this scale.
3.3 Temporal units, and the study period
Analysis of the variation in stop-level line boardings is performed for five time periods (defined
by Metro) of a typical weekday: Early AM (midnight to 6 AM), AM peak (6 AM to 9 AM), Mid-
day (9 AM to 3 PM), PM peak (3 PM to 7 PM), and Night (7 PM to midnight). We should expect
the marginal contribution of reliability in determining the variation in boardings to vary across
the time periods.
At a given level of temporal aggregation (say AM peak), the cross-sectional analysis is
performed by averaging service reliability and boarding measures for each line at the stop level
The Demand for Reliable Travel Chakrabarti (2015)
81
over a six-month study period (December 2011 to May 2012). All service supply parameters
(e.g. network configuration, schedules, etc.) for the Metro bus network were constant over that
period. This approach helps smooth-out idiosyncratic/seasonal fluctuations in the data, and
estimate effects of long-term average performance on long-term average ridership.
3.4 Conceptual model
The hypothesis is that all else equal, and relatively speaking, there will be more stop-level bus
line boardings when line reliability at the stop level is higher. I recognize that line boardings at a
stop includes both passengers making their origin boardings and those making a transfer
connection. These two groups cannot be identified/differentiated using available APC data.
However, since line reliability at a stop is expected to influence both groups by influencing
transit mode choice as well as line/route (and boarding/transfer point) selections, and hence to
influence aggregate boardings to the line at the stop, this constraint will not significantly affect
the study. Also, I carefully select independent variables (determinants of stop-level line
boardings) that help identify and control for the locations of common transfer points at major
intersections. Nevertheless, bias (that potentially causes underestimation of the reliability-effect)
arising out of the limitation of standard APC data are discussed in detail in Section 6.
In order to test the hypothesis, I account (or control) for the many factors that are also
expected to determine boardings (i.e. in addition to reliability).
Chu (2004) proposed an average weekday stop-level boarding model for the Florida
Department of Transportation that consists of six factors – catchment area sociodemographics,
transit level of service, pedestrian environment, population and employment accessibility,
interaction with other modes, and completion with other stops. The model is good starting point.
The Demand for Reliable Travel Chakrabarti (2015)
82
But an exhaustive review of extant literature has helped identify/justify critical factors, build the
conceptual model, and select variables most relevant to the current research.
The broad parameter groups identified are: stop neighborhood built environment factors
(e.g. based on work by Ewing and Cervero, 2010; Chatman, 2013; Estupiñán and Rodriguez,
2008; Kuby et al., 2004; Ryan and Frank, 2009, etc.); stop neighborhood socioeconomic and
demographic factors (e.g. based on work by Pucher and Renne, 2003; Buehler and Pucher, 2012;
Pulugurtha and Agurla, 2012; Taylor et al., 2009, etc.); bus line service quality factors (e.g.
based on work Taylor et al., 2009; Iseki and Taylor, 2010; Lai and Chen, 2011; Cirillo et al.,
2011; dell’Olio et al. 2012, etc.); and other factors (e.g. potential alternate/competing transit
services around a stop; line information availability and fare
50
; and the position/sequence of a
stop with respect to the line that serves it – e.g. stops towards the end of a line may attract
disproportionately fewer boardings).
I propose a linear model of stop-level bus line boardings as the starting point. The model
is of the form:
𝑷 𝒍𝒔𝒕 = 𝜷 𝟎 + 𝜷 𝟏 𝑩𝑬
𝒔 + 𝜷 𝟐 𝑺𝑫
𝒔 + 𝜷 𝟑 𝑸𝑳
𝒍𝒔𝒕 + 𝜷 𝟒 𝑶𝑭
𝒍𝒔𝒕 + 𝜺 ….. (Eqn. 8)
Where: P is average boardings; BE is a vector of stop neighborhood built environment measures;
SD is a vector of stop neighborhood socioeconomic and demographic measures; QL is a vector
of average planned service quality measures and average service reliability measure for the line
serving the stop; and OF denotes other factors assumed to influence stop-level line boardings, for
line l at stop s at time period t. 𝜀 is the error term. Read t as weekday Early AM, AM peak, Mid-
Day, PM peak, and Night. Average refers to the six-month average over the December 2011-
May 2012 study period. Variable selection under each parameter group is discussed in Section 4.
50
Since this chapter looks at a single system (Los Angeles Metro bus), there is no variation in information or fare
across cases.
The Demand for Reliable Travel Chakrabarti (2015)
83
3.5 Data sources
Data from three principal sources has been used for this research: Metro bus transit supply,
demand, and performance data available through the ADMS research project; ACS (American
Community Survey) 2007-2011 data;
51
and SCAG (Southern California Association of
Governments) employment, land use and regional transit network data.
52
4. Variables and Descriptive Statistics
4.1 Dependent variable
The dependent variable is an average (over the December 2011-May 2012 study period) estimate
of the number of boardings to a bus line at a stop for each time period (i.e. Early AM, AM peak,
etc.) of a typical weekday. Data is derived from automatic passenger counters (APC) installed on
Metro buses. Since the time periods consist of unequal number of hours, “average per-hour
boardings” is used; this helps compare the relative influences of the independent variables across
the different time periods. The dependent variable can be considered to be continuous, since it is
an average estimate. Descriptive statistics are given in Table 4-2.
4.2 Explanatory variables
Three explanatory (service reliability) variables are used in this study. They are:
Average on-time performance (Avg. OTP)
For a given line at a given stop, this is the average (over the study period) proportion of
time (considering all scheduled trips) that the line has departed “on-time” (“early” and “late” are
also tested), i.e. between 1-minute early and 5-minutes late with respect to schedule, within each
51
Census tract level sociodemographic data is collected from the American Community Survey 2007-2011 via
Social Explorer.
52
GIS-based locations of firms (including 2012 employment of each firm), and transit routes and stops (all regional
transit agencies; 2012 data) within the SCAG jurisdiction are collected through a contract with SCAG.
The Demand for Reliable Travel Chakrabarti (2015)
84
weekday time period (i.e. Early AM, AM peak, etc.). See Figure 4-2. The theoretical range of
this variable is 0-1. It is expected that average OTP is positively associated, and average early
and late performances are negatively associated, with the dependent variable. Processed data
was available as monthly averages directly from Metro.
Figure 4-2. The Avg. OTP measure
Note: This is a hypothetical distribution of bus line l departures from stop s by schedule deviation, observed over the study period, within time
period of weekday t. Average OTP of the line at the stop is given by (B+C)x100% / (A+B+C+D), where A, B, C, D denote areas of segments
under the curve.
Coefficient of variation (CV) of on-time performance (CV of OTP)
At a given stop, this is the coefficient of variation (or CV; standard deviation divided by
mean) of monthly average line OTP over the six-month study period. This measures the extent of
variability with respect to the average level of OTP, and attempts at capturing risk associated
with unreliability that can get concealed in averages. CV of OTP is dimensionless, and can take
any positive value. It is expected to be negatively associated with boardings.
The Demand for Reliable Travel Chakrabarti (2015)
85
Standard deviation of schedule deviation (SD of sched. dev.)
Access to a historical archive of Metro’s real-time AVL data feed via the RIITS
53
system
as part of the ADMS research project provided an opportunity to test a measure of the standard
deviation of line schedule deviation at the stop level (capturing risk imposed by
unpredictability/volatility of waiting time that jeopardizes trip plans) across different time
periods of weekday over the six-month study period. Figure 4-3 illustrates the concept of this
measure.
Metro’s AVL feed consists of periodically updated real-time locations of Metro buses
and corresponding schedule deviation estimates at time points that is primarily used for
providing real-time vehicle arrival/departure predictions to passengers via web based
applications. The historical archive helped derive distributions of schedule deviation across all
available time point stops for all available lines (considering all available scheduled trips) over
the study period to estimate the SD of sched. dev. measure by time period of weekday. Data
cleaning was performed to eliminate outliers (e.g. cases that are more than 20 minutes early or
over one hour late were dropped; consistent with Metro’s data cleaning approach) in the raw
data. A total of about (depending on time period of weekday) 800 line-stop combinations from
the AVL database could be accurately related to the boardings database (note: unmatched cases
are due to lack of standardization across Metro’s sub-systems/sub-modules).
The unit is minutes, and the measure can theoretically assume any positive value. It is
expected to be negatively associated with boardings.
53
Regional Integration of Intelligent Transportation Systems (RIITS) – see http://www.riits.net/ (accessed on
10/11/2014)
The Demand for Reliable Travel Chakrabarti (2015)
86
Figure 4-3. The SD of schedule deviation measure
Note: This is a hypothetical distribution of bus line l departures from stops s1 and s2 by schedule deviation, observed over the study period,
within time period of weekday t. Observe that SD of schedule deviation for the line is smaller at s1 than at s2.
4.3 Other independent variables (or controls)
Table 4-1 lists the other independent variables (or controls) chosen for estimating regression
models. Note that many other potential variables were tested, but later discarded considering
issues of multicollinearity.
The Demand for Reliable Travel Chakrabarti (2015)
87
Table 4-1.
Control variables
No. Variable name Variable description
A. Stop neighborhood built environment, socioeconomic, and demographic factors
1 Population density Population density (persons per sq. mi.) of the census tract in which a stop is
located.
2 Employment density Number of employments per square mile of the census tract in which a stop
is located.
3 Income Median household income (in 2010 inflation adjusted dollars) of the census
tract in which a stop is located.
4 Transit availability Number of stops of all transit operators (all transit modes) within quarter-
mile radius of a stop. Stops located at major transfer points/intersections are
expected to have a higher value of this variable.
B. Planned service quality of a line serving a stop
1 Line population density Mean of the population densities (persons per sq. mi.) of all census tracts
traversed through (and served by) line.
2 Employment
accessibility
Total number of jobs within quarter-mile buffer (either side) of a line,
divided by the length of the line (captures of en-route employment density).
3 Stops per mile Count of total stops per unit length of line (captures service speed and
impedance).
4 Line type Four indicator variables capturing all other unique characteristics (service,
corridor, equipment, amenities, etc.) of a line – Rapid line, Local CBD
connector, Local Non CBD connector, and Limited stop/express service;
Shuttle/circulator line is used as reference.
5 Headway Scheduled headway of line, in minutes. This line-level average measure is
not expected to be endogenous with the dependent stop-level line boardings
variable in theory. However, endogeneity issues are discussed and dealt with
in detail later in Section 5.
C. Other factors
1 Transit alternatives Captured through number of stops of all transit operators (all transit modes)
within quarter-mile walking radius of a stop (already mentioned above). A4
and C1 will be referred to as “Neighborhood transit services” in general.
2 Position of stop with
respect to line
Two indicator variables that capture whether the stop is “near line origin”
(i.e. whether its ordered sequence number lies within the lower 25%
considering all stops served by the line) or “near line end” (i.e. whether its
ordered sequence number lies within the upper 25% considering all stops
served by the line). Intermediate stops serve as the reference group.
Note: Stop refers to a time point stop, and line refers to a directional Metro bus line.
4.4 Descriptive statistics
Descriptive statistics are summarized in Table 4-2 for reference. Several broad observations can
be made about the Los Angeles Metro bus network. First, on average, intensity of Metro bus
ridership is highest in the weekday PM peak (12 line boardings per stop per hour), followed by
AM peak and Mid-day. Ridership is low during the night, and insignificant during Early AM.
The Demand for Reliable Travel Chakrabarti (2015)
88
Second, average service reliability of the Metro bus network is relatively low in the weekday PM
peak (in terms of all three reliability measures). This is not surprising, since we know from a
separate study
54
that arterial traffic speed in Los Angeles is lowest and most variable during the
PM peak. Third, population and employment densities of neighborhoods in which Metro bus
stops are located are high on average. Finally, Metro bus lines serve corridors with high
population and employment density; and as expected, bus service frequency is highest on
average during the peaks.
54
A regional traffic monitoring application as part of the ADMS research project.
The Demand for Reliable Travel Chakrabarti (2015)
89
Table 4-2.
Descriptive statistics of key variables
Variables N Mean SD Min Max
Per-hour line boardings at stop (avg.
no.)
Early AM 1335 0.90 1.87 0.00 32.67
AM Peak 1335 9.81 18.25 0.00 418.33
Mid-day 1335 8.59 12.39 0.00 188.17
PM Peak 1335 11.72 17.24 0.00 232.00
Night 1335 2.86 5.30 0.00 69.40
Avg. OTP (% expressed as decimal)
Early AM 1103 0.89 0.15 0.00 1.00
AM Peak 1220 0.83 0.10 0.39 1.00
Mid-day 1253 0.75 0.14 0.00 1.00
PM Peak 1257 0.69 0.14 0.00 1.00
Night 1262 0.71 0.18 0.00 1.00
CV of OTP (unit-less)
Early AM 1094 0.09 0.17 0.00 2.00
AM Peak 1220 0.06 0.06 0.00 0.82
Mid-day 1247 0.07 0.08 0.00 1.45
PM Peak 1251 0.11 0.11 0.00 1.45
Night 1250 0.15 0.21 0.00 2.45
SD of sched. dev. (min)
Early AM 785 2.17 1.60 0.16 14.62
AM Peak 869 2.81 1.15 0.81 12.46
Mid-day 894 3.73 1.41 1.00 13.58
PM Peak 872 4.41 1.48 1.08 11.64
Night 878 4.67 2.07 0.26 14.41
NH pop. density (persons per sq. mi.) 1335 12937 9187 13 89648
NH emp. density (no. per sq. mi.) 1335 10523 20993 106 282984
NH median income (2010 dollars) 1335 52169 26984 10107 196250
NH transit services (avg. no.) 1335 28 37 1 270
Line pop. density (persons per sq. mi.) 1335 12578 4075 2057 28630
Line emp. density (avg. no.) 1335 14594 13096 2280 60283
Line stops per mile (avg. no.) 1335 4.53 1.50 0.16 11.03
Line headway (min)
Early AM 1073 145.60 121.14 20.00 360.00
AM Peak 1220 23.52 15.50 5.29 90.00
Mid-day 1249 37.02 47.07 6.00 360.00
PM Peak 1257 24.13 16.50 5.14 90.00
Night 1262 57.10 50.66 7.66 360.00
Note: Spatial unit of observation is a time point stop on a directional Metro bus line; Note that all lines do not operate at all times of the day, and
complete data corresponding to all cases are not available for all time periods (consequently, cases used in regression models vary); OTP=on-time
performance, CV=coefficient of variation, SD=standard deviation, and NH=neighborhood; Weekdays are considered only; indicator variables are
not included in the table; outlier analysis was performed based on more detailed analysis of descriptive statistics, and observations were further
dropped from regression models.
The Demand for Reliable Travel Chakrabarti (2015)
90
5. Results and Discussions
5.1 Analysis of service reliability impacts on stop-level bus line boardings
This section summarizes ordinary least squares (OLS) regression model estimates of average
per-hour boardings to a bus line at the stop level for each time period of a typical weekday,
successively using the three reliability measures. Note that outliers and suspect records are
eliminated to derive reliable estimates. For example, for each time period, cases with zero
average boardings are filtered out. When OTP measures are included, cases with zero percent
average OTP are further dropped.
Also note that the stop neighborhood population and employment density variables are not
included as continuous variables. Rather, four indicator variable groups (with a fifth group as
reference) that capture neighborhood characters by using a combination of population and
employment density values are used (Table 4-3). Although density simply serves as control, this
approach helps associate neighborhood density to transit ridership in a conceptually meaningful
way, and provides insights for planners.
The Demand for Reliable Travel Chakrabarti (2015)
91
Table 4-3.
Variables capturing stop neighborhood (i.e. constituent census tract) density
Indicator variables
Employment density
(No. per sq. mi.)
High
(>5K)
Low
(<1K)
Population
density
(Persons
per sq. mi.)
High
(>17K)
Indicator 1
PD(High) & ED(High)
Indicator 2
PD(High) & ED(Low)
Low
(<6K)
Indicator 3
PD(Low) & ED(High)
Indicator 4
PD(Low) & ED(Low)
Notes:
1.
The thresholds are based on analyses of population and employment density distributions across all census tracts within Los Angeles
County; High: Upper 25th percentile; Low: Bottom 25th percentile.
2.
Stops belonging to neighborhood density groups that fall in the intermediate ranges (corresponding to both population and employment
densities) are used as reference.
3.
K denotes “thousand.”
The role of average on-time performance (Avg. OTP)
OLS regression models using average OTP as the reliability measure are summarized in
Table 4-4. Observe that all else equal, boardings to a bus line at a stop is positively associated
with average OTP of the line serving the stop in the AM and PM peaks (statistically significant
at the 0.01 level in the AM peak, and at the 0.05 level in the PM peak). In the AM peak, all else
equal, a stop served by a line with 10 percentage-point higher average OTP is expected to be
associated with 1.002 more per-hour boardings to the line (which is about 10% of the mean for
that time period); in the PM peak, the effect is about 5% of the mean. There is no evidence that
OTP affects ridership during off-peak periods.
The models (Table 4-4) highlight other important factors associated with bus ridership in
Los Angeles. For example, boardings are significantly higher at stops located in census tracts
with high population and employment densities across all time periods. These stops are in areas
The Demand for Reliable Travel Chakrabarti (2015)
92
where demand (perhaps also latent demand) exists, and where service is targeted. These stops
also serve as key transfer points at major intersections. It is not surprising that boardings are
relatively high in [predominantly residential] areas with high population but low employment
density during the AM peak (also Mid-Day) that mostly includes trips to work from home. All
else equal, lines that traverse high-density (both in terms of population and employment)
corridors and make less frequent stops attract greater number of boardings across their stops. A
stop near the end of a line attracts significantly lower number of boardings to the line due to its
location/sequence. Consistent with literature, service frequency is a major determinant of
ridership; lower (i.e. shorter) the service headway of a line, greater the number of line boardings
at the stop level, all else equal. Finally, neighborhood income is negatively associated with bus
ridership (except in Early AM and Night).
The Demand for Reliable Travel Chakrabarti (2015)
93
Table 4-4.
OLS regression models of stop-level bus line boardings – Avg. OTP effect
Dependent variable: Average per-hour boardings to a line at a stop
+ PD=Population density, ED=Employment density; Refer Table 4-3 for reference group and indicator variable definitions
++ Shuttle/circulator line type is used as reference
Note: Spatial unit of observation is a time point stop on a directional Metro bus line; OTP (on-time performance) data is in percent expressed as decimal (range between 0 and 1); NH refers to neighborhood;
Weekdays are considered only; P refers to P > | t |
Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef.
NH density (Indicator)+
PD(High) & ED(High) 0.958 0.00 0.160 5.808 0.00 0.148 8.985 0.00 0.232 9.110 0.00 0.196 3.890 0.00 0.244
PD(High) & ED(Low) 1.636 0.10 0.051 21.073 0.00 0.110 10.227 0.03 0.052 4.733 0.38 0.020 1.037 0.63 0.012
PD(Low) & ED(High) -0.563 0.02 -0.078 -1.505 0.16 -0.036 -0.538 0.60 -0.013 1.037 0.38 0.021 -0.131 0.79 -0.007
PD(Low) & ED(Low) -0.087 0.92 -0.003 2.261 0.32 0.027 3.515 0.10 0.043 2.391 0.34 0.025 0.300 0.80 0.007
NH median income -4.480E-06 0.19 -0.046 -7.720E-05 0.00 -0.160 -4.130E-05 0.00 -0.087 -2.800E-05 0.07 -0.050 3.310E-06 0.63 0.015
NH transit services 0.004 0.11 0.063 -0.046 0.00 -0.086 -0.027 0.06 -0.050 -0.001 0.96 -0.001 0.010 0.03 0.067
Line pop. density 5.550E-05 0.01 0.103 3.581E-04 0.00 0.111 0.001 0.00 0.273 0.001 0.00 0.252 2.926E-04 0.00 0.216
Line emp. density 2.550E-05 0.01 0.162 1.482E-04 0.00 0.145 3.373E-04 0.00 0.336 4.890E-05 0.38 0.040 4.230E-05 0.05 0.101
Line stops per mile -0.202 0.02 -0.139 -1.438 0.00 -0.169 -1.096 0.00 -0.128 -1.067 0.01 -0.106 -0.338 0.05 -0.090
Stop near line origin 0.381 0.02 0.074 -0.041 0.96 -0.001 -0.749 0.35 -0.024 -0.662 0.47 -0.018 -0.412 0.28 -0.030
Stop near line end -0.775 0.00 -0.119 -2.788 0.00 -0.084 -4.713 0.00 -0.141 -5.992 0.00 -0.153 -2.128 0.00 -0.140
Line type++
Rapid 0.182 0.77 0.022 4.215 0.10 0.084 5.377 0.02 0.108 3.964 0.16 0.065 0.924 0.44 0.043
Local CBD connector -0.211 0.67 -0.046 -1.602 0.43 -0.053 -1.571 0.41 -0.053 -0.335 0.88 -0.010 0.548 0.56 0.044
Other local 0.253 0.57 0.057 1.375 0.40 0.054 5.788 0.00 0.225 4.607 0.01 0.152 2.640 0.00 0.234
Limited/express 0.181 0.76 0.017 -0.739 0.74 -0.015 1.868 0.37 0.036 1.709 0.47 0.028 1.179 0.28 0.048
Line headway -0.005 0.00 -0.237 -0.274 0.00 -0.333 -0.066 0.00 -0.172 -0.335 0.00 -0.371 -0.042 0.00 -0.252
Avg. OTP -0.936 0.14 -0.055 10.022 0.01 0.079 2.338 0.39 0.023 6.299 0.03 0.057 1.388 0.19 0.040
Constant 2.481 0.01 11.626 0.01 -1.714 0.61 5.865 0.13 -0.787 0.64
N
Adj. R-square
Variable
Early AM AM Peak Mid-Day PM Peak Night
835 1144 1190 1203 1095
0.19 0.31 0.35 0.37 0.26
The Demand for Reliable Travel Chakrabarti (2015)
94
The role of CV of on-time performance (CV of OTP)
CV of OTP (refer Section 4.2 for definition) of a line at a stop has a statistically
significant (at the 0.01 level) negative influence on boardings to the line at that stop across the
peaks and during Mid-day (Table 4-5). This implies that, all else equal, if a stop is served by two
lines with identical average OTP, average number of boardings is expected to be greater to the
line with lesser month-to-month variation in OTP, further suggesting demand for service
predictability. Also, the higher the average OTP, greater is the acceptable range of variation. If
average OTP is low, then it needs to be highly stable across time, so that passengers can, under
certain circumstances, develop a consistent expectation of performance and plan/budget
accordingly.
The role of standard deviation (SD) of schedule deviation
Again, as explained in Section 4.2, a lower SD of schedule deviation of a line serving a
stop implies greater stability in performance (good or bad), and should theoretically have a
positive influence on boardings, all else equal, because of lower unpredictability/risk. Observe
that SD of line schedule deviation does have a statistically significant (at the 0.01 level) negative
influence on stop-level line boardings, again in the AM and PM peaks (Table 4-6). The
parameter is also significant during Mid-day (at the 0.05 level). The opposite effect in the Early
AM may just be due to lack of options/alternatives during this significantly low-service low-
demand period.
The Demand for Reliable Travel Chakrabarti (2015)
95
Table 4-5.
OLS regression models of stop-level bus line boardings – CV of OTP effect
Dependent variable: Average per-hour boardings to a line at a stop
+ PD=Population density, ED=Employment density; Refer Table 4-3 for reference group and indicator variable definitions
++ Shuttle/circulator line type is used as reference
Note: Spatial unit of observation is a time point stop on a directional Metro bus line; CV refers to coefficient of variation, OTP refers to on-time performance in percent expressed as decimal (range between 0
and 1), and NH refers to neighborhood; Weekdays are considered only; P refers to P > | t |
Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef.
NH density (Indicator)+
PD(High) & ED(High) 0.974 0.00 0.163 5.756 0.00 0.146 8.925 0.00 0.231 9.001 0.00 0.194 3.880 0.00 0.243
PD(High) & ED(Low) 1.616 0.11 0.051 20.426 0.00 0.107 9.972 0.03 0.051 4.450 0.40 0.019 0.932 0.67 0.011
PD(Low) & ED(High) -0.584 0.01 -0.081 -1.660 0.12 -0.040 -0.545 0.59 -0.013 0.993 0.40 0.020 -0.103 0.84 -0.006
PD(Low) & ED(Low) -0.071 0.93 -0.003 2.483 0.27 0.030 3.640 0.09 0.044 2.223 0.37 0.023 0.280 0.81 0.007
NH median income -4.340E-06 0.21 -0.045 -7.520E-05 0.00 -0.156 -3.920E-05 0.00 -0.082 -2.610E-05 0.09 -0.047 3.350E-06 0.63 0.015
NH transit services 0.005 0.08 0.077 -0.047 0.00 -0.087 -0.026 0.07 -0.048 -0.001 0.98 -0.001 0.009 0.06 0.063
Line pop. density 5.720E-05 0.01 0.106 3.398E-04 0.00 0.105 0.001 0.00 0.264 0.001 0.00 0.248 2.905E-04 0.00 0.214
Line emp. density 2.700E-05 0.01 0.171 1.514E-04 0.00 0.148 3.369E-04 0.00 0.336 4.590E-05 0.41 0.037 4.120E-05 0.06 0.098
Line stops per mile -0.194 0.03 -0.134 -1.421 0.00 -0.167 -1.126 0.00 -0.132 -1.029 0.01 -0.102 -0.342 0.05 -0.091
Stop near line origin 0.370 0.03 0.072 0.029 0.97 0.001 -0.885 0.25 -0.028 -0.314 0.72 -0.009 -0.287 0.44 -0.021
Stop near line end -0.725 0.00 -0.112 -2.916 0.00 -0.087 -4.650 0.00 -0.139 -6.350 0.00 -0.162 -2.252 0.00 -0.148
Line type++
Rapid 0.211 0.73 0.025 4.125 0.11 0.082 4.831 0.04 0.097 3.788 0.18 0.062 0.805 0.50 0.037
Local CBD connector -0.218 0.66 -0.047 -1.849 0.36 -0.062 -2.173 0.25 -0.073 -0.766 0.72 -0.022 0.549 0.56 0.045
Other local 0.272 0.54 0.062 1.262 0.43 0.049 5.240 0.00 0.203 4.212 0.01 0.139 2.591 0.00 0.230
Limited/express 0.259 0.66 0.024 -1.028 0.64 -0.020 1.228 0.56 0.023 1.337 0.57 0.022 1.078 0.32 0.044
Line headway -0.005 0.00 -0.241 -0.247 0.00 -0.300 -0.059 0.00 -0.154 -0.327 0.00 -0.363 -0.040 0.00 -0.246
CV of OTP 0.227 0.70 0.015 -32.754 0.00 -0.121 -17.180 0.00 -0.080 -9.470 0.01 -0.061 -0.403 0.64 -0.014
Constant 1.505 0.05 21.546 0.00 1.944 0.48 11.377 0.00 0.311 0.83
N
Adj. R-square
Variable
Early AM AM Peak Mid-Day PM Peak Night
835 1144 1190 1203 1095
0.19 0.31 0.36 0.37 0.26
The Demand for Reliable Travel Chakrabarti (2015)
96
Table 4-6.
OLS regression models of stop-level bus line boardings – SD of schedule deviation effect
Dependent variable: Average per-hour boardings to a line at a stop
+ PD=Population density, ED=Employment density; Refer Table 4-3 for reference group and indicator variable definitions
+ Shuttle/circulator line type is used as reference
Note: Spatial unit of observation is a time point stop on a directional Metro bus line; SD refers to standard deviation, and NH refers to neighborhood; Weekdays are considered only; P refers to P > | t |; The SD
of schedule deviation effects in the AM Peak (Std. Coef.= -0.117 at P=0.00), Mid-day (Std. Coef.= -0.085 at P=0.02), and PM Peak (Std. Coef.= -0.128 at P=0.00) periods remain unchanged if the mean of
schedule deviation is further controlled for.
Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef. Coef. P Std. Coef.
NH density (Indicator)+
PD(High) & ED(High) 1.062 0.00 0.194 6.942 0.00 0.166 6.930 0.00 0.218 8.777 0.00 0.193 3.008 0.00 0.206
PD(High) & ED(Low) 3.369 0.00 0.114 39.882 0.00 0.155 19.777 0.00 0.099 14.147 0.08 0.050 3.501 0.25 0.039
PD(Low) & ED(High) -0.392 0.08 -0.070 -2.196 0.08 -0.055 -0.967 0.30 -0.032 0.098 0.94 0.002 -0.517 0.30 -0.036
PD(Low) & ED(Low) -0.265 0.74 -0.014 2.430 0.35 0.032 3.940 0.05 0.066 4.058 0.14 0.048 0.781 0.51 0.025
NH median income -2.930E-06 0.40 -0.038 -7.990E-05 0.00 -0.166 -6.330E-05 0.00 -0.172 -5.330E-05 0.00 -0.103 -1.000E-05 0.18 -0.054
NH transit services -0.002 0.43 -0.035 -0.064 0.00 -0.118 -0.033 0.01 -0.086 -0.029 0.13 -0.049 -0.005 0.48 -0.027
Line pop. density 1.600E-05 0.46 0.035 2.041E-04 0.09 0.062 0.001 0.00 0.233 0.001 0.00 0.212 1.946E-04 0.00 0.164
Line emp. density 4.090E-08 1.00 0.000 9.950E-05 0.11 0.090 1.927E-04 0.00 0.231 2.790E-05 0.67 0.023 1.580E-05 0.53 0.040
Line stops per mile -0.034 0.70 -0.030 -1.097 0.01 -0.131 -0.714 0.04 -0.112 -1.133 0.02 -0.123 -0.284 0.13 -0.094
Stop near line origin 0.170 0.33 0.041 1.271 0.19 0.041 0.433 0.56 0.018 0.006 1.00 0.000 0.122 0.76 0.011
Stop near line end -0.387 0.08 -0.072 -1.421 0.18 -0.042 -3.530 0.00 -0.139 -5.076 0.00 -0.141 -1.417 0.00 -0.114
Line type++
Rapid 0.075 0.89 0.012 4.018 0.17 0.093 3.697 0.08 0.111 2.576 0.38 0.054 0.299 0.80 0.020
Local CBD connector 0.408 0.37 0.102 -1.118 0.64 -0.035 0.735 0.68 0.030 -0.446 0.85 -0.013 1.179 0.21 0.106
Other local 0.063 0.87 0.018 0.584 0.75 0.022 5.313 0.00 0.267 4.027 0.03 0.143 2.204 0.00 0.234
Limited/express 0.617 0.28 0.069 -0.461 0.86 -0.009 2.582 0.19 0.063 1.070 0.69 0.018 1.008 0.37 0.049
Line headway -0.004 0.00 -0.267 -0.280 0.00 -0.334 -0.047 0.00 -0.188 -0.335 0.00 -0.391 -0.041 0.00 -0.286
SD of sched. dev. 0.105 0.05 0.086 -1.388 0.00 -0.125 -0.516 0.03 -0.070 -1.015 0.00 -0.109 -0.090 0.31 -0.039
Constant 1.456 0.05 25.261 0.00 . 5.097 0.05 19.305 0.00 . 2.889 0.07
N
Adj. R-square 0.16 0.28 0.28 0.33 0.19
Night
548 812 822 834 737
Variable
Early AM AM Peak Mid-Day PM Peak
The Demand for Reliable Travel Chakrabarti (2015)
97
It is often argued that the demand for reliability is low for high-frequency (or short-
headway) service. To test this, I excluded cases with line headway less than 10 minutes
(industry-standard definition of high-frequency transit service in the U.S.) and re-estimated the
complete set of OLS regressions (i.e. those presented in Table 4-4, Table 4-5, and Table 4-6).
Evidence of the demand for reliability, particularly during weekday peak periods, remains.
5.2 Reliability impact on boardings: Headway as moderator
Next, I investigate how the effect of a line’s service reliability on stop-level boardings to the line
is moderated (or conditioned) by its headway. To perform the analysis, I had to interact two
continuous variables (headway and reliability), and therefore I implement a change in origin, and
center the headway variable around its own mean for each time period to generate a new MCLH
(i.e. mean-centered line headway) variable. Table 4-7 summarizes regression results for weekday
AM and PM peaks only (because all three reliability measures are consistently statistically
significant in the peaks based on Table 4-4, Table 4-5, and Table 4-6 findings).
Observe that the magnitude of marginal positive effect of average OTP, and the
magnitudes of marginal negative effects of both CV of OTP, and SD of schedule deviation
increase or decrease as line headway falls below or rises above the mean headway respectively,
in both AM and PM peaks. This generally implies that the marginal reliability effect is stronger
for relatively shorter-headway lines than longer-headway lines; or that on average, the absolute
number of stop-level line boardings will increase by a greater magnitude for unit improvement in
reliability (across any of the three reliability dimensions) in a shorter-headway line than the
corresponding improvement in a longer-headway line, all else equal.
The Demand for Reliable Travel Chakrabarti (2015)
98
The above finding does not contradict with expectations. On average, longer-headway
lines have lower baseline stop-level boardings than shorter-headway lines in the current dataset
(because service frequency is itself a very strong determinant of demand). Hence, although the
absolute magnitude of gain is relatively smaller, the expected percentage gain over baseline stop-
level boardings is relatively larger for longer-headway lines.
The series of graphs in Figure 4-4 illustrate this phenomenon, using parameter estimates
from Table 4-7. The effects of three [practical] reliability improvement strategies – a 10
percentage-point (or 0.1-unit) increase in average OTP, a 0.01-unit reduction in CV of OTP, and
a 1-minute reduction in SD of schedule deviation – on average stop-level line boardings are
plotted for the peaks across different headway groups.
Observe that for the short-headway (say less than 10 minutes) line group, expected
percentage gain in average stop-level line boardings with respect to baseline is smallest,
indicating relatively low demand for reliability in this group. As line headway increases
(progressively to the 10-20 min and 20-30 min range groups), identical reliability improvement
is expected to result in smaller absolute gain in average stop-level line boardings; however, the
corresponding percentage gain over baseline progressively increases, indicating increasing
demand for reliability with increasing headway (highest for the 20-30 minute headway group).
Interestingly, however, the demand drops abruptly when headway increases beyond 30 minutes.
This is because lines that operate at these policy headways during weekday peak periods cater to
those without many alternatives (due to their social or locational characteristics). In general, the
demand for reliability appears to be greater during the AM peak.
The Demand for Reliable Travel Chakrabarti (2015)
99
Table 4-7.
Summary of OLS regression models of peak-period stop-level bus line boardings – moderating
effect of headway
Dependent variable: Average per-hour boardings to a line at a stop
Note: MCLH=Mean-centered line headway (explained in Section 5.2); Spatial unit of observation is a time point stop on a directional Metro bus
line; OTP refers to on-time performance in percent expressed as decimal (range between 0 and 1), CV refers to coefficient of variation, and SD
refers to standard deviation; Weekdays are considered only; P refers to P > | t |; Set of all standard/common independent variables (refer Table
4-4, Table 4-5, and Table 4-6) are included in each of the 6 models estimated.
Coef. P Std. Coef. Coef. P Std. Coef.
Avg. OTP
(main effect; for avg. headway)
12.003 0.00 0.094 7.552 0.01 0.068
MCLH x OTP
(interaction effect)
-0.632 0.00 -0.646 -0.435 0.01 -0.343
N
Adj. R-square
CV of OTP
(main effect; for avg. headway)
-47.498 0.00 -0.176 -10.669 0.00 -0.069
MCLH x CV of OTP
(interaction effect)
2.839 0.00 0.302 1.177 0.00 0.194
N
Adj. R-square
SD of sched. dev.
(main effect; for avg. headway)
-1.169 0.00 -0.105 -0.922 0.00 -0.099
MCLH x SD of sched. dev (interaction effect) 0.062 0.01 0.211 0.036 0.04 0.187
N
Adj. R-square
Case 1: Avg. OTP effect
Case 2: CV of OTP effect
Case 3: SD of sched. dev. effect
0.34 0.38
0.29 0.33
1144 1203
0.31 0.37
1144 1203
Common set of variables are included in all 6 models
812 834
AM Peak PM Peak
Variable
The Demand for Reliable Travel Chakrabarti (2015)
100
Figure 4-4. Effects of peak-period reliability improvement across headway groups
Effect of 10 percentage-point higher average OTP on avg. stop-level bus line boardings
AM Peak
(dotted line indicates % increase over baseline)
PM Peak
(dotted line indicates % increase over baseline)
The Demand for Reliable Travel Chakrabarti (2015)
101
Effect of 0.01-unit lower CV of OTP on avg. stop-level bus line boardings
AM Peak
(dotted line indicates % increase over baseline)
PM Peak
(dotted line indicates % increase over baseline)
The Demand for Reliable Travel Chakrabarti (2015)
102
Effect of 1-minute lower SD of schedule deviation on avg. stop-level bus line boardings
AM Peak
(dotted line indicates % increase over baseline)
PM Peak
(dotted line indicates % increase over baseline)
Note: CV refers to coefficient of variation, SD refers to standard deviation, and OTP refers to on-time performance in percent expressed as
decimal (range between 0 and 1); Representative median headway in each headway group in each peak is used for estimations.
The Demand for Reliable Travel Chakrabarti (2015)
103
5.3 Additional tests and robustness check
Early vs. late performance
It is possible that early and late performances (average percentage of time a line has
served a stop earlier and later with respect to the [1-minute early to 5 minutes late] “on-time”
window around the scheduled time) have different effects on ridership. Regression estimates
show that average number of boardings are higher at lower levels of average early and late
performance, all else equal, and as expected. Note that statistically significant (at the 0.05 level)
standardized coefficients indicate relatively greater negative impact of early performance (Std.
coef. = -0.105 in the AM peak and -0.113 in the PM peak) than late performance (Std. coef. = -
0.051 in the AM peak and -0.063 in the PM peak). While this may suggest that missing a bus is
more onerous (which is probably true), earliness is actually rare (because drivers are generally
penalized more heavily for driving ahead of schedule, and also because drivers often have the
option to dwell at or near time points) and there is little scope of promoting Metro’s system-wide
ridership even by eliminating earliness entirely. The average early performance of the Metro bus
system in the study period for the weekday Early AM, AM peak, Mid-Day, PM peak, and Night
was 4%, 3%, 3%, 3%, and 2% respectively. Improvement of on-time performance by minimizing
late performance is therefore key.
Discussion on supply-demand endogeneity
This research is carefully designed such that the possibility of reverse causality in
regression models, i.e. demand (i.e. consumption) influencing supply, can be eliminated.
Although line headway (a potential endogenous independent variable) is often
55
adjusted by
transit operators based on empirical data on service consumption levels (including load factors)
55
Metro generally makes these supply-side adjustments, called shakeups, twice every year in June and December.
The Demand for Reliable Travel Chakrabarti (2015)
104
under coverage obligations and resource constraints, productivity measures across the entire
length (i.e. across all stops) of a line is expected to influence headway (including other service
quality factors) determination; average boarding count at any one stop (current unit of
observation) will not theoretically influence the average headway of the line serving the stop.
Therefore, endogenous relationship between the dependent stop-level line boarding variable and
the independent line headway variable is unlikely.
However, since I use time point stops located at major intersections that may contribute
to a significant proportion of total line boardings, it is useful to perform a robustness check by
statistically accounting for endogeneity and testing whether the peak-period reliability effect is
stationary. I adopt a two-stage least squares (2SLS) regression approach using an instrument for
the line headway variable.
First, average headway of the line serving a stop is modeled as a function of the average
56
productivity (average total per-hour boardings per revenue mile of service) of that line across its
entire length (i.e. summed over all stops, not just time point stops, along the line) for each peak.
However, I only include lines (and hence the associated line-stop cases) with headway ≤ 20
minutes (this is the peak-period headway limit for rapid bus service per Metro’s policy; all bus
line types/categories are included in this range), to ensure that productivity is most closely
correlated with, and hence can be assumed to serve as a good determinant of, line headway. This
eliminates all lines that operate at policy headways regardless of service consumption signals. I
also pool AM and PM peak observations together, and include an indicator variable (=1 for an
AM peak observation) that controls for the structural difference in system-wide ridership
56
Average over the six-month study period.
The Demand for Reliable Travel Chakrabarti (2015)
105
level/pattern across the two peaks. Pooling ensures sufficient observations for estimating models
using each of the three reliability measures.
Theoretically, the instrument (line productivity) should not influence a random stop-level
boarding to the line. And it is well correlated (correlation coefficient of -0.72 in the AM peak
and -0.70 in the PM peak) with line headway in the reduced sample.
In the second stage of 2SLS, estimated values of the line headway variable are plugged
into the standard full models (e.g. Table 4-4, Table 4-5, and Table 4-6). A summary of the peak-
period 2SLS regression models using the three reliability measures is presented in Table 4-8.
Since this [homogenous] set of high-frequency (relatively short-headway) lines all
traverse through the dense core of Los Angeles connecting high-density population and
employment centers, line level population and employment density measures are dropped from
the models due to insufficient variation across observations. The neighborhood transit
availability (including competition) variable is also dropped, since it does not contribute
meaningfully to the models.
The 2SLS approach confirms that service reliability of a line serving a stop has a
significant positive influence on boardings to the line at the stop during both AM and PM peaks.
All three reliability measures have the expected directions of influence, and are statistically
significant at the 0.01 level. Observe that the line headway variable retains its significance,
suggesting that a strong positive causal influence of service frequency on ridership might exist.
The Demand for Reliable Travel Chakrabarti (2015)
106
Table 4-8.
2SLS regression models of peak-period stop-level bus line boardings
Dependent variable: Average per-hour boardings to a line at a stop
Instrument for line headway at stop: Line productivity (avg. total per-hour bus line
boardings/line length)
Note: Spatial unit of observation is a time point stop on a directional Metro bus line; SD refers to standard deviation, CV refers to coefficient of
variation, OTP refers to on-time performance in percent expressed as decimal (range between 0 and 1), and NH refers to neighborhood;
Weekdays are considered only; P refers to P > | t |; Cases with line headway>20 min are not included.
Coef. P Coef. P Coef. P
Headway (endogenous) -1.812 0.00 -1.773 0.00 -1.566 0.00
NH density (Indicator)+
PD(High) & ED(High) 10.670 0.00 10.533 0.00 10.216 0.00
PD(High) & ED(Low) 16.104 0.00 15.636 0.00 25.970 0.00
PD(Low) & ED(High) -1.303 0.28 -1.357 0.25 -1.694 0.21
PD(Low) & ED(Low) -0.206 0.96 -0.452 0.91 4.725 0.26
NH median income -7.850E-05 0.00 -7.820E-05 0.00 -1.055E-04 0.00
Line stops per mile -1.515 0.00 -1.433 0.00 -1.499 0.00
Stop near line origin -0.953 0.35 -0.551 0.57 0.078 0.94
Stop near line end -7.737 0.00 -8.314 0.00 -4.928 0.00
Line type++
Rapid 7.041 0.03 7.031 0.03 6.111 0.08
Local CBD connector 1.253 0.61 0.896 0.72 1.834 0.47
Other local 6.169 0.01 5.955 0.02 4.886 0.05
Limited/express 4.386 0.27 4.237 0.29 3.265 0.49
Avg. OTP 9.528 0.02
CV of OTP -17.519 0.01
SD of sched. dev. -2.810 0.00
AM Peak (Indicator) -3.822 0.00 -3.196 0.00 -6.884 0.00
Constant 38.799 0.00 46.427 0.00 55.434 0.00
N
R-squared 0.26 0.27 0.29
Variable
Case 1
Avg. OTP effect
Case 2
CV of OTP effect
Case 3
SD of sched. dev. effect
1349 1349 829
The Demand for Reliable Travel Chakrabarti (2015)
107
6. Takeaways and Limitations
Analysis of time point stop level data from the Los Angeles Metro bus system reveals that
service reliability is a significant determinant of peak-period ridership; on average, more reliable
bus lines are successful in attracting significantly greater number of peak-period boardings
across their stops, relatively speaking, and all else equal. In simple terms, the demand for reliable
public transit service is demonstrated.
Demand for reliability appears to vary by headway, and is relatively higher for longer-
headway service. The findings are expected to enhance our general understanding of the
determinants of [the observed spatiotemporal variations in] transit ridership/demand, and to
inform public transit policy by underscoring the benefits of reliability improvement investments.
This study is based on data from the Los Angeles Metro bus system, and therefore results
are expected to be directly useful for Metro’s current operations and future planning/policy
actions; other transit agencies (particularly in the U.S.) can derive broad lessons. Similar context-
specific studies are required to estimate the degree to which reliability improvements can boost
ridership in other cities.
The extent to which service reliability improvements would result in system-wide
ridership gains is unclear. Reliability improvement of a particular line at a particular stop can
draw/steal riders from its [relatively unreliable] competitor; therefore, reliability investments can
lead to some redistribution of ridership, thereby reducing net gain. This is certainly possible in
transit-rich regions such as Los Angeles with a lot of transit alternatives serving the same
corridor in close proximity, and multiple routes connecting a given trip origin and destination.
Recall that the argument throughout the chapter is that reliability of a line at a given stop
influences boardings to the line at that stop. However, transit travel between a given origin and
The Demand for Reliable Travel Chakrabarti (2015)
108
destination often involves hopping/transferring across more than one line, and therefore boarding
multiple lines at multiple stops. Consequently, reliability of any one line (i.e. any one link in the
transit travel chain) at any one stop may influence decision regarding the entire journey, and
hence affect volumes at all other boarding points along the chain. On aggregate, this may result
in lower-than-expected boardings to a relatively reliable line. Since this phenomenon is not
accounted for in the current study, the marginal effect of reliability may have been
underestimated. The limitation can be addressed in future, for example, with information on
linked trips as opposed to unlinked trips using smart fare card data instead of APC data.
7. Conclusion
This research provides first empirical evidence that service reliability affects transit travel
demand. Improvements in various dimensions of reliability can make transit more attractive, and
prepare it for competing with alternate modes (particularly auto) in the presence of latent
demand. Reliability investments can therefore help increase service consumption relative to
supply, and make transit lines and systems more productive. I expect the reliability parameter to
get policy priority as Los Angeles and other U.S. metropolitan regions continue to invest in the
future.
Having said that, we must understand that increasing service reliability is complex. It
must include external strategies that are not solely within the realm of public transit policy or
under control of transit managers. For instance, bus-only lanes and signal preemption systems
are theoretically [fairly] straightforward means of improving reliability. However,
implementation requires coordination among many agencies responsible for managing multi-
modal urban transportation networks. But there are other internal strategies too. For example,
efficient scheduling by analyzing historic archived AVL/APC/Smart Card data, real-time
The Demand for Reliable Travel Chakrabarti (2015)
109
rerouting of vehicles around incidents, better holding strategies, stop-consolidation, driver
training and incentive programs, regular system maintenance, and real-time information sharing
via mobile devices can improve both system performance and users’ perception of service
quality. There is no doubt that advances in intelligent transportation infrastructure and
information technology will help manage and operate transit systems more efficiently, and
consequently help improve transit service reliability.
8. References
1) Abkowitz, M. D., & Engelstein, I. (1983). Factors affecting running time on transit routes.
Transportation Research Part A: General, 17(2), 107–113.
2) Abkowitz, M., & Engelstein, I. (1984). Methods for maintaining transit service regularity.
Transportation Research Record, (961).
3) Benezech, V., & Coulombel, N. (2013). The value of service reliability. Transportation
Research Part B: Methodological, 58, 1-15.
4) Bowman, L. A., & Turnquist, M. A. (1981). Service frequency, schedule reliability and
passenger wait times at transit stops. Transportation Research Part A: General, 15(6), 465-
471.
5) Buehler, R., & Pucher, J. (2012). Demand for public transport in Germany and the USA: an
analysis of rider characteristics. Transport Reviews, 32(5), 541-567.
6) Camus, R., Longo, G., & Macorini, C. (2005). Estimation of transit reliability level-of-
service based on automatic vehicle location data. Transportation Research Record: Journal
of the Transportation Research Board, 1927(1), 277-286.
7) Cantwell, M., Caulfield, B., & O’Mahony, M. (2009). Examining the Factors that Impact
Public Transport Commuting Satisfaction. Journal of Public Transportation, 12(2), 1-21.
The Demand for Reliable Travel Chakrabarti (2015)
110
8) Chatman, D. G. (2013). Does TOD need the T? On the importance of factors other than rail
access. Journal of the American Planning Association, 79(1), 17-31.
9) Chen, W., & Chen, Z. (2009). Service reliability analysis of high frequency transit using
stochastic simulation. Journal of Transportation Systems Engineering and Information
Technology, 9(5), 130-134.
10) Chen, X., Yu, L., Zhang, Y., & Guo, J. (2009). Analyzing urban bus service reliability at the
stop, route, and network levels. Transportation Research Part A, 43, pp. 722–734.
11) Chu, X. (2004). Ridership models at the stop level (No. NCTR-473-04, BC137-31; Available
at http://trid.trb.org/view.aspx?id=747775).
12) Cirillo, C., Eboli, L., & Mazzulla, G. (2011). On the asymmetric user perception of transit
service quality. International Journal of Sustainable Transportation, 5(4), 216-232.
13) Dell’Olio, L., Ibeas, A., & Cecín, P. (2010). Modelling user perception of bus transit quality.
Transport Policy, 17(6), 388-397.
14) Dell'Olio, L., Ibeas, A., Dominguez, A., & Gonzalez, F. (2012). Passenger preference
analysis: light rail transit or bus versus car. Transport, 27(3), 276-285.
15) Dueker, K. J., & Bianco, M. J. (1999). Light-rail-transit impacts in Portland: The first ten
years. Transportation Research Record: Journal of the Transportation Research Board,
1685(1), 171-180.
16) Eboli, L., & Mazzulla, G. (2007). Service quality attributes affecting customer satisfaction
for bus transit. Journal of Public Transportation, 10(3), 21.
17) El-Geneidy, A. M., Strathman, J. G., Kimpel, T. J., & Crout, D. T. (2006). Effects of bus stop
consolidation on passenger activity and transit operations. Transportation Research Record:
Journal of the Transportation Research Board, 1971(1), 32-41.
The Demand for Reliable Travel Chakrabarti (2015)
111
18) El-Geneidy, A. M., Hourdos, J., & Horning, J. (2009). Bus transit service planning and
operations in a competitive environment. Journal of Public Transportation, 12(3), 39-59.
19) El‐Geneidy, A. M., Horning, J., & Krizek, K. J. (2011). Analyzing transit service reliability
using detailed data from automatic vehicular locator systems. Journal of Advanced
Transportation, 45(1), 66-79.
20) Estupiñán, N., & Rodriguez, D. A. (2008). The relationship between urban form and station
boardings for Bogota’s BRT. Transportation Research Part A: Policy and Practice, 42(2),
296-306.
21) Ewing, R., & Cervero, R. (2010). Travel and the built environment: a meta-analysis. Journal
of the American Planning Association, 76(3), 265-294.
22) Giuliano, G. (2011) “Transportation policy: Public transit, settlement patterns and equity in
the US,” chapter 28 in N. Brooks, K. Donaghy and G. Knaap, eds., Oxford University Press
Handbook of Urban Economics and Planning. New York: Oxford University Press.
23) Glascock, J. (1997). Research on customer requirements for transit service design and
delivery. Transportation Research Record: Journal of the Transportation Research Board,
1604(1), 121-127.
24) Iseki, H., & Taylor, B. D. (2010). Style versus Service? An Analysis of User Perceptions of
Transit Stops and Stations. Journal of Public Transportation, 13(3), 23-48.
25) Kittelson & Associates. (2013). TCRP Report 165: Transit Capacity and Quality of Service
Manual (3
rd
ed.) Washington, D.C.: Transport Research Board.
26) Kuby, M., Barranda, A., & Upchurch, C. (2004). Factors influencing light-rail station
boardings in the United States. Transportation Research Part A: Policy and Practice, 38(3),
223-247.
The Demand for Reliable Travel Chakrabarti (2015)
112
27) Lai, W. T., & Chen, C. F. (2011). Behavioral intentions of public transit passengers—The
roles of service quality, perceived value, satisfaction and involvement. Transport Policy,
18(2), 318-325.
28) Lin, J., Wang, P., & Barnum, D. T. (2008). A quality control framework for bus schedule
reliability. Transportation Research Part E: Logistics and Transportation Review, 44(6),
1086-1098.
29) Nakanishi, Y. J. (1997). PART 1: Bus: Bus Performance Indicators: On-Time Performance
and Service Regularity. Transportation Research Record: Journal of the Transportation
Research Board, 1571(1), 1-13.
30) Nurul Habib, K. M., Kattan, L., & Islam, M. T. (2011). Model of personal attitudes towards
transit service quality. Journal of Advanced Transportation, 45, 271–285.
31) Polus, A. (1978). Modeling and Measurements of Bus Service Reliability. Transportation
Research, 12(6), 253-256.
32) Pucher, J., & Renne, J. L. (2003). Socioeconomics of urban travel: evidence from the 2001
NHTS. Transportation Quarterly, 57(3), 49-77.
33) Pulugurtha, S. S., & Agurla, M. (2012). Bus-Stop Level Transit Ridership using Spatial
Modeling Methods. Journal of Public Transportation, 15(1).
34) Ryan, S., & Frank, L. F. (2009). Pedestrian environments and transit ridership. Journal of
Public Transportation, 12(1), 39-57.
35) Santos, A., McGuckin, N., Nakamoto, H. Y., Gray, D., & Liss, S. (2011). Summary of Travel
Trends: 2009 National Household Travel Survey. Available at http://nhts.ornl.gov/: Federal
Transit Administration, US Department of Transportation.
The Demand for Reliable Travel Chakrabarti (2015)
113
36) Sterman, B. P., & Schofer, J. L. (1976). Factors Affecting Reliability of Urban Bus Services.
Transportation Engineering Journal of ASCE, 102(TE1), 147-159.
37) Strathman, J. G., & Hopper, J. R. (1993). Empirical analysis of bus transit on-time
performance. Transportation Research Part A: Policy and Practice, 27(2), 93-100.
38) Strathman, J. G., Dueker, K. J., Kimpel, T., Gerhart, R., Turner, K., Taylor, P., & Hopper, J.
(1999). Automated bus dispatching, operations control, and service reliability: Baseline
analysis. Transportation Research Record: Journal of the Transportation Research Board,
1666(1), 28-36.
39) Taylor, B. D., Miller, D., Iseki, H., & Fink, C. (2009). Nature and/or nurture? Analyzing the
determinants of transit ridership across US urbanized areas. Transportation Research Part A:
Policy and Practice, 43(1), 60-77.
40) Turnquist, M. A. (1978). A model for investigating the effects of service frequency and
reliability on bus passenger waiting times. Transportation Research Record, (663).
41) Turnquist, M. A., & Bowman, L. A. (1980). The effects of network structure on reliability of
transit service. Transportation Research Part B: Methodological, 14(1), 79-86.
42) Tyrinopoulos, Y., & Antoniou, C. (2008). Public transit user satisfaction: Variability and
policy implications. Transport Policy, 15(4), 260-272.
43) Wachs, M. (1976). Consumer attitudes toward transit service: an interpretive review. Journal
of the American Institute of Planners, 42(1), 96-104.
44) Xuan, Y., Argote, J., & Daganzo, C. F. (2011). Dynamic bus holding strategies for schedule
reliability: Optimal linear control and performance analysis. Transportation Research Part
B: Methodological, 45(10), 1831-1845.
The Demand for Reliable Travel Chakrabarti (2015)
114
45) Yetiskul, E., & Senbil, M. (2012). Public bus transit travel-time variability in Ankara
(Turkey). Transport Policy, 23, 50-59.
The Demand for Reliable Travel Chakrabarti (2015)
115
Chapter 5.
The Reliability – Mode Choice Relationship
This chapter uses 2012-13 California Household Travel Survey (CHTS) data to fundamentally
analyze the determinants of transit use for commuting among car-owners in Los Angeles County.
I seek to identify spatial contexts along with characteristics of the multi-modal transportation
network in which transit is successful in attracting discretionary riders. The study provides the first
direct empirical evidence regarding the association between service reliability and transit mode
choice.
The Demand for Reliable Travel Chakrabarti (2015)
116
1. Introduction
U.S. federal, state, and local governments have been investing heavily to expand and improve
public transport infrastructures across metropolitan regions. Figures from the Federal Transit
Administration’s National Transit Database (NTD)
57
reveals that between 1991 and 2012, the
total annual government spending on transit steadily increased from $22 billion to $58.5 billion
at an inflation-adjusted cumulative annual average growth rate (CAGR) of 2.2%. Over the same
period, total vehicle revenue miles of service increased from 2.5 to 4.0 billion miles at a CAGR
of 2.3% (Figure 5-1). A substantial increase in service coverage and density may have helped
connect more people with places at relatively low fares, and other service quality improvements
may have collectively enhanced the competitiveness of transit in the urban passenger
transportation marketplace.
Figure 5-1.Transit in the U.S.: Government funding and service supply trends (1991-2012)
Data source: U.S. National Transit Database (http://www.ntdprogram.gov/ntdprogram/; last accessed on 12/28/14).
57
NTD can be accessed online at http://www.ntdprogram.gov/ntdprogram/ (last accessed on 12/28/14)
The Demand for Reliable Travel Chakrabarti (2015)
117
However, increasing investments have not been proportionately translated into increased
ridership or productivity. For example, number of unlinked passenger trips per revenue vehicle
hour, the industry-standard measure of service effectiveness, decreased from 46.5 to 39.4 over
the 1991-2012 period (NTD data; Figure 5-2). Although better network planning may have
reduced required transfers by providing more direct connections between various origins and
destinations, the 15% decline is still large, and may indicate lower-than-expected return on
investment. Observe also that fare-box recovery ratio (fare revenue as % of total operating
expense) decreased from 36.4 to 33.1 over the same period (NTD data; Figure 5-2), indicating
declining productivity and increasing subsidy requirement for operations.
Figure 5-2. Public transit in the U.S.: Service effectiveness and productivity trends (1991-2012)
Data source: National Transit Database (http://www.ntdprogram.gov/ntdprogram/; last accessed on 12/28/14).
The Demand for Reliable Travel Chakrabarti (2015)
118
Aggregate figures must be interpreted with caution. It is possible that investments have
increased mobility of the transportation-disadvantaged, provided lifeline service across low
demand places and times, and promoted transit use in particular contexts. Therefore, large public
expenditures may have generated significant social benefits.
But transit remains a relatively small player in the U.S. travel market. For example, as
Figure 5-3 illustrates, repeated nation-wide estimations of mode use for commuting using
various survey instruments (Decennial Census, American Community Survey or ACS, National
Personal Transportation Survey or NPTS, and National Household Travel Survey or NHTS) over
the past decades have revealed that transit’s share is stagnant at about 5%. If all trip purposes are
considered, the share drops even further (refer Santos et al., 2011).
Figure 5-3. Percent of U.S. workforce using public transit for commuting (1990-2012)
Source: Census/ACS data obtained via Social Explorer (http://www.socialexplorer.com/), and NPTS/NHTS public datasets with analyses are
available online (http://nhts.ornl.gov/; Santos et al., 2011). The links were last accessed on 12/28/14.
Note: Figures refer to workers (age 16+) who normally used transit to go to work during the week before they were interviewed; definition of
transit varies across surveys.
Interestingly, public transit has lost share in some of its largest metropolitan markets over
the past decade. Table 5-1 lists the top 25 U.S. metros in terms of their estimated [unweighted]
public transit trip share (% public transit trips relative to total [unlinked] trips made by all modes
The Demand for Reliable Travel Chakrabarti (2015)
119
of travel for all purposes, derived from the respective one-day travel diary of each respondent,
age 5+) according to the 2001 NHTS survey, and tracks change by comparing with 2009 NHTS
data. Transit trip share has decreased in 24 out of 25 metros, with the highest percentage point
decreases in cities such as New York, Washington D.C., and New Orleans. Only Portland shows
a marginal increase in this group.
The change cannot be attributed to difference in survey instruments between 2001 and
2009. But it is possible that reorganizations in land uses and transit networks have resulted in
fewer transfers overall, and consequently fewer unlinked transit trips in the database. One-day
travel diaries are also sensitive to idiosyncratic behaviors and information recording/recollection
errors.
Nonetheless, it is difficult to argue that transit use has increased significantly in any part
of urban U.S. More importantly, if outliers such as the New York area, Washington D.C. area,
and the San Francisco Bay Area are excluded, transit struggled to touch the 2% share mark in
2009.
The Demand for Reliable Travel Chakrabarti (2015)
120
Table 5-1.
Change in estimated public transit trip share in select U.S. metros (2001-2009)
2001
Rank
CMSA
Public Transit Trips
(%)
Percentage
Point Change
(2001-2009)
NHTS
2001
NHTS
2009
1 New York-Northern New Jersey-Long Island., NY-NJ-CT-PA 7.91 4.72 -3.19
2 Washington-Baltimore, DC-MD-VA-WV 4.39 2.16 -2.23
3 Philadelphia-Wilmington-Atlantic City, PA-NJ-DE-MD 2.90 1.30 -1.61
4 San Francisco-Oakland-San Jose, CA 2.85 2.00 -0.85
5 New Orleans, LA 2.51 0.43 -2.08
6 Chicago-Gary-Kenosha, IL-IN-WI 2.51 1.49 -1.02
7 Pittsburgh, PA 2.46 1.70 -0.75
8 Boston-Worcester-Lawrence, MA-NH-ME-CT 2.30 1.53 -0.77
9 Miami-Fort Lauderdale, FL 2.15 1.27 -0.88
10 San Diego, CA 2.11 1.26 -0.85
11 Seattle-Tacoma-Bremerton, WA 1.98 1.55 -0.43
12 Los Angeles-Riverside-Orange County, CA 1.82 1.59 -0.23
13 Cleveland-Akron, OH 1.73 0.36 -1.37
14 Portland-Salem, OR-WA 1.48 1.71 0.23
15 Salt Lake City-Ogden, UT 1.42 1.25 -0.17
16 Buffalo-Niagara Falls, NY 1.33 0.63 -0.70
17 Denver-Boulder-Greeley, CO 1.30 1.12 -0.19
18 Rochester, NY 1.30 0.28 -1.02
19 Sacramento-Yolo, CA 1.29 0.84 -0.45
20 San Antonio, TX 1.20 0.82 -0.39
21 Indianapolis, IN 1.18 0.47 -0.71
22 Orlando, FL 1.04 0.60 -0.45
23 Greensboro-Winston-Salem-High Point, NC 1.03 0.22 -0.81
24 Minneapolis-St. Paul, MN-WI 1.01 0.59 -0.42
25 Hartford, CT 1.01 0.11 -0.89
Note: CMSA refers to Consolidated Metropolitan Statistical Area; Honolulu, HI (entire Oahu Island) CMSA is excluded because of
unavailability of 2009 NHTS data; 2001 rank relates to NHTS 2001 % of public transit trips; NHTS is the U.S. National Household Travel
Survey.
Given these trends, it seems imperative to re-think U.S. public transit policy so that
transit can increase its market share (measured in terms of, e.g., fraction of total [linked]
commute trips that are made by transit on average, or % of individuals normally using a transit
mode for commuting) across geographies, and consequently get on track to achieving its broader
sustainability goals in addition to improving its own level of productivity and fiscal
sustainability.
The Demand for Reliable Travel Chakrabarti (2015)
121
Key to increasing transit’s market share, particularly in the U.S. context, is attracting
people out of cars (and simultaneously containing attrition of transit riders who can own and/or
use an automobile). Effective goal-oriented policy formulation requires better understanding of
who among “people with choice” use transit and under what conditions they find transit use
feasible. This might help identify where latent [transit travel] demand exists, and how that latent
demand might be attracted to transit. The current study analyzes discretionary transit use, and
identifies strategies/policies that transit managers can adopt to encourage more discretionary
transit travel.
I use a straightforward approach. I isolate people with choice (specifically those among
car-owners who are expected to have the auto mode at their disposal), and investigate those
relatively rare situations in which some of them choose to use transit over auto. By statistically
controlling for factors that might restrict access of a car-owning individual to the household
vehicle, this approach helps identify spatial contexts, along with characteristics of the multi-
modal transportation network in which transit succeeds in attracting discretionary riders, in part
by governing location choice and perhaps also by inducing mode shifts of those who are not
completely averse to transit travel.
58
I focus on the Los Angeles (LA) County area and on commute (or work-) trips only – i.e.
I analyze normal/usual mode choice to work for persons (age 16+) living and working in LA
County. My research aims at providing more practical insights to transit managers regarding how
(and where) service quality and performance may be improved to increase patronage and
58
I could not explore characteristics of those who have the means but do not own an automobile, and choose to use
transit instead. While such individuals constitute an exclusive category of discretionary riders, they could not be
isolated using available data. Although presumably rare (based on the dataset used in this study), such cases can
provide much valued insights on qualities of the multi-modal environment that can limit auto ownership.
The Demand for Reliable Travel Chakrabarti (2015)
122
productivity. Note that access to a comprehensive historical archive
59
of regional real-time multi-
modal transportation system data provides a unique opportunity to offer the first empirical
evidence on how service quality differences among alternate modes available for the journey to
work influence, in part, transit mode choice.
To foreshadow briefly, my findings suggest that not all car-owners who use transit for
commuting are discretionary riders – absence of valid driving license and presence of more
workers than cars in the household are influential factors that limit access to auto and induce
choice of transit. But there is some indication that the transit users (some of whom could be
discretionary riders) are younger on average, and that the level/quality of transit availability and
connectivity influences their location and travel choices. A regression analysis examines the
association of fast (relative to auto), frequent, and reliable transit service availability (for the
journey from home to work) with discretionary transit commuting. Interestingly, there is no
evidence that increasing congestion or unreliability of the auto mode (again, for the journey from
home to work) increases the demand for discretionary transit use.
The rest of the chapter is organized as follows: Section 2 summarizes past empirical work
on factors determining transit mode choice; Section 3 introduces the study area and outlines the
research approach; Sections 4 and 5 present analyses with discussions; finally, Section 6
concludes the chapter with a summary of findings, policy implications, and limitations of the
study.
59
Los Angeles County Metropolitan Transportation Authority-funded Archived Data Management System (ADMS)
developed at the University of Southern California’s METRANS Transportation Center.
The Demand for Reliable Travel Chakrabarti (2015)
123
2. Literature Review
Based on consumer choice theory, travelers are assumed to rationally choose a travel mode by
evaluating (through experience/perception or by analyzing data/information) the characteristics
of various available competing alternatives in an attempt to maximize personal utility (de
Donnea, 1972; Domencich and McFadden, 1975). Since utility cannot be observed directly,
utility-based models estimate the probability that a given alternative (with an expected utility
based on its own attributes, and the attributes of the individual making the choice) will be chosen
by observing how people actually behave, or by capturing their stated choices in hypothetical
scenarios.
Transport policy relies on the idea that travel choices can be influenced/altered, and
broader social goals such as congestion- and pollution-reduction may be achieved, by
implementing policies that change relative costs of different modes. But mode choice
(particularly for routine travel such as the commute) is often a relatively longer-term (hence
rather inelastic) choice, made through careful assessment of conditions that are both internal (e.g.
travel time and cost budgets determined by personal, household, and work-related constraints)
and external (e.g. multi-modal transportation network factors) to the commuter, and that is
governed by a complex set of personal attitudes, preferences, habits, culture, lifestyle, and
physical (dis)abilities. We know that choice is often governed by inflexible personal attitudes
that may not be as sensitive to costs as we expect. Schneider (2013) notes that mode choice for
routine travel may be driven by habit. While this potentially makes auto to transit mode shifts
challenging, a study by Vredin Johansson et al. (2006) suggests that an enduring shift to more
socially desirable modes of travel may be possible even without direct economic incentives.
The Demand for Reliable Travel Chakrabarti (2015)
124
Perhaps, change in transport-related dogmas along with improved access to public transport at a
reduced cost can promote transit use (Collins and Chambers, 2005).
As a result, it is difficult to determine and generalize how travelers perceive costs
associated with particular modes, and how they might react to changes. Consequently,
identifying an exhaustive list of [cost] parameters determining mode choice, and collecting
appropriate data to measure them, has always been challenging (Cervero, 2002). Past mode
choice models lack consistency in specification, and data availability and research agenda drive
parameter choices in modeling.
The desirability of public transit and, consequently, its choice over other modes depends
upon its competitiveness in the urban passenger travel marketplace. In countries such as the U.S.,
where automobile travel has historically been cheap, shared modes of transportation have
struggled to attract users. In the U.S., public transit largely caters to the transportation
disadvantaged, particularly those with low incomes who have limited means of owning or using
an automobile (Pucher and Renne 2003). In other countries such as Great Britain (Giuliano and
Dargay, 2006) and Germany (Buehler, 2011), where auto transport costs are higher due to the
characteristic land use and pricing environments, alternative transport modes such as transit are
relatively more popular.
Consistent with expectations, empirical studies generally indicate that conditions that are
directly associated with relatively increasing the cost of auto travel (with respect to transit) –
such as high auto parking price (e.g. Gillen, 1977), road use charges (e.g. Washbrook et al.,
2006), lower levels of transit-to-auto travel time and/or out-of-pocket cost (e.g. Asensio, 2002;
Cervero, 2002; Frank et al., 2008; Basso and Jara-Díaz, 2012) – are also associated with greater
The Demand for Reliable Travel Chakrabarti (2015)
125
demand for transit, or at least lower demand for solo-driving. Direct interventions to decrease
time costs and improve transit service may also boost transit choice (Forsey et al., 2013).
Decades of research has shown that people use transit more in denser, mixed-use,
pedestrian-friendly, transit-accessible, vibrant station areas (e.g. Cervero, 1994; Frank and Pivo,
1994; Cervero, 2002; Zhang, 2004; Chen et al., 2008; Ding et al., 2014). The observed
association arguably operates through higher relative [generalized] cost of auto travel (e.g. Crane
and Crepeau, 1998). The observed effects also could be due in part to self-sorting of pro-transit
individuals into areas that have good quality transit where generalized costs of transit travel
relative to its competitor(s) may be low (for relevant discussions on the self-selection issue refer,
e.g.: Schwanen and Mokhtarian, 2005; Pinjari et al., 2007; Pinjari et al., 2011). The self-sorting
issue means that new transit supply may not change travel behavior. Nevertheless, the idea that
more transit supply will change mode choice has had major influence on local land use policy
with coordinated land use – transit planning. Chatman (2013) found that strategic integration of
high-density mixed-use developments with appropriate parking regulations and good transit
access can indeed promise lower automobile dependence and boost transit use.
Review of past empirical work suggests that some significant gaps exist in our
knowledge. For example, it is still unclear how some critical dimensions of transit and auto (e.g.
recurrent and non-recurrent congestion) network quality/performance contribute to determining
relative generalized transit travel cost, and hence how these might influence transit mode choice.
Among relevant studies, those by Outwater et al. (2011), Asensio (2002), and Sweet and Chen
(2011) deserve mention. Outwater et al. (2011) attempted to measure “premium transit service”
effects on mode choice using a stated-preference survey in Salt Lake City. They found that
awareness about service quality affects transit mode choice. Earlier, Asensio (2002) had
The Demand for Reliable Travel Chakrabarti (2015)
126
evaluated the effect of waiting time, transfer time, access-egress distance and service frequency,
in addition to time and cost, on commuter mode choice in Barcelona. And Sweet and Chen
(2011) investigated the effect of auto unreliability on mode choice using data from Chicago.
They found that unreliable auto travel conditions may induce mode shift to transit. While some
studies have investigated the effect of real-time information availability on transit mode choice,
policy implications are unclear.
This study advances knowledge in the field by using information from a new travel
survey, and by employing a comprehensive historical archive of regional real-time multi-modal
transportation system data. Granular data and a simple analytical approach helps understand how
transit can attract latent demand and reduce auto-reliance in a highly congested city.
3. Study Area and Approach
Los Angeles County (Figure 5-4) is the study area. Delineation is based on fine-grained multi-
modal transportation system data availability.
60
The area contains over 7,500 CHTS (California
Household Travel Survey) respondents whose commute travel choices have been analyzed.
60
Transportation system performance data is available from the ADMS that is limited to Los Angeles County.
The Demand for Reliable Travel Chakrabarti (2015)
127
Figure 5-4. Study area: Los Angeles County
Base map: Google
The analysis is divided into two parts. Part 1 (Section 4) is a comparative analysis that
describes contexts in which commuters belonging to car-owning households use transit. This
analysis identifies the sociodemographic and environmental factors that distinguish transit users
from auto users. Part 2 (Section 5) estimates multinomial logistic regression models to measure
how relative service quality and system performance measures (for the journey from home to
work) influence discretionary transit mode choice among car-owning commuters. A summary of
The Demand for Reliable Travel Chakrabarti (2015)
128
findings is given in Section 6. Data and methods are introduced in the respective sections for
clarity.
4. Transit Use for Commute Trips in LA County: Descriptive Analysis
In this section, I perform a descriptive analysis to explore transit use among commuters within
LA County. I focus on commuters belonging to car-owning (one or more cars) households;
commute mode choice of persons in carless households are briefly discussed for reference. This
section provides broad insights on factors that may be associated with discretionary transit travel.
4.1 Travel survey data (CHTS)
The 2012-13 CHTS includes 7,756 employed (age 16+) individuals who have both their
residential and primary workplace locations within LA County. These persons constitute the base
sample in this section.
The CHTS is the most recent statewide travel survey, aimed at developing and updating
regional transportation models.
61
It includes travel information from over 42,000 households
across 58 counties in California and three neighboring counties in Nevada. CHTS covered a one-
year period between February 1, 2012 and January 31, 2013.The recruit response rate was 4.9%,
and the retrieval rate was 67.3%.
I received the CHTS dataset from Caltrans. I submitted my research proposal and signed
a confidentiality agreement. The agreement includes policies for data storage, sharing and use. I
use the home and primary workplace locations,
62
personal/household socioeconomic and
61
The survey was led by the California Department of Transportation (Caltrans) and funded by the California
Strategic Growth Council, the California Energy Commission, and eight transportation planning agencies across
California.
62
Only census tract identifiers were made available by Caltrans due to strict confidentiality agreements with survey
respondents. Therefore, x-y coordinates of centroids of census tracts are used when exact locations need to be
identified, e.g. when home-to-work routes need to be simulated.
The Demand for Reliable Travel Chakrabarti (2015)
129
demographic profiles, and normal/usual commute mode choice variables from the dataset for this
study. I do not use the one-day travel diary (except for validation purposes, explained later in the
chapter) since the objective is to understand long-term mode choice to work.
4.2 Aggregate figures
Overall in LA County, about 8% of employed individuals normally use
63
transit, 88% use auto,
and the rest (around 4%) report using other non-motorized modes for commuting to their primary
workplace. Individuals using special
64
(including non-traditional) modes, and those who did not
report their normal commute travel mode are excluded from the above analysis, and also from all
subsequent analyses.
“Transit” involves public transit only. In addition to travel as driver of the household
vehicle (alone, or with household members, or with others), “auto” use includes travel as
passenger (in household vehicle, or others’ private vehicle), and travel via taxi or rental vehicle.
“Non-motorized” mode refers to walk, bike, or other non-motorized (per CHTS definition).
Figure 5-5 gives the commute mode choice distributions of individuals in car-owning
(97% of sample) and carless (rest 3%) households for reference.
63
Use refers to normal commute mode choice, identified in response to the question “How [do you] [does this
person] normally get to this primary job? That is, what method of travel is used for the longest distance?” This is
regardless of whether the person took that mode to work on the designated travel day (if applicable). In the context
of CHTS data, this definition of [normal] commute mode is consistent throughout the chapter.
64
Special/non-traditional modes, excluded from the [CHTS data] analysis, in all sections of this chapter, include:
Wheelchair/Mobility Scooter, Motorcycle/Scooter/Moped, Private shuttle (e.g. SuperShuttle, employer, hotel, etc.),
Greyhound Bus, Plane, “Other Private Transit”, School Bus, Dial-a-Ride/Paratransit (Access Services, etc.).
The Demand for Reliable Travel Chakrabarti (2015)
130
Figure 5-5. Normal commute mode of individuals in car-owning and carless households (LA
County)
Note: Percentages may not exactly add up to 100 due to rounding to two decimal places; Commute refers to travel from home to primary
workplace; Individuals of age 16+ living and working in LA County are considered only; Sample size=7675 (individuals using excluded modes
and those not reporting their mode choice are omitted); See Section 4.2 for components within each mode category.
4.3 Commuters in carless households
Individuals belonging to carless households (about 3% of the sample, with over half of the group
earning under $24,999 annually) are most likely to use transit or a non-motorized mode for
commuting. As expected, 61% of persons in this group use transit, and 25% use a non-motorized
mode. The remaining 14% either have access to an automobile (rented or borrowed; drive alone
or with other household workers), or use a carpool.
The Demand for Reliable Travel Chakrabarti (2015)
131
Analysis of their simulated travel environments (via their chosen mode, and via
alternative modes) from home (i.e. centroid of home census tract) to work (i.e. centroid of work
census tract) in a typical weekday AM peak provides useful insights.
Ideally, for a given individual, characteristics of travel environments along the chosen
and alternative modes can accurately be measured by using GPS-based probes through the
network over repeated trips from home to work. But the CHTS does not provide exact origin-
destination coordinates or GPS traces. Therefore, home-to-work travel environments via
alternative modes/routes have been simulated. For each person in the sample, three “best” (or
least total travel time) routes (in a typical weekday AM peak period, during which home-to-work
commute is assumed to take place) that are (or expected to be; relevant during the time in history
when they were surveyed) available to them are simulated from the centroid of home census tract
to centroid of [primary] workplace census tract – via transit, auto, and non-motorized mode
(walking/biking) – and expected alternative travel environment characteristics are identified.
The best transit route in the weekday AM peak period, valid for the time in history when
an individual was surveyed (unique time of survey for each respondent is specified in the
dataset), is identified by querying Google Maps using Google Developers
(https://developers.google.com/) API (Application Programming Interface) that provides
information (including such details as transit line number) on every segment of the transit route
(that may consist of multiple transfers and additional access/egress trips), similar to the output of
the Google Maps transit trip planner web service (https://www.google.com/maps).
The best auto route in the weekday AM peak period is identified using the 2012 regional
transportation network data (that consists of estimated loaded/congested travel time for
passenger vehicles at each road link) obtained from the Modeling & Forecasting group at the
The Demand for Reliable Travel Chakrabarti (2015)
132
Southern California Association of Governments (SCAG) as base, and by employing Esri
ArcGIS Network Analyst tool that provides information on travel time and also per-mile travel
time that measures level of [recurrent]auto route congestion.
Travel time along the best non-motorized mode-route is estimated assuming walking
speed of 3.5 mph and biking speed of 10 mph along an uncongested shortest auto path, the
distance of which is determined using a TRAVELTIME3 program in Stata (developed by
Bernhard, 2013) that extracts information from Google Maps using the web-based Google
Distance Matrix API (https://developers.google.com/maps/documentation/distancematrix/).
Note that routes could not be generated when home and workplace are in the same census tract
(and hence have overlapping centroids). Also, in some cases, Google could not suggest any
viable transit route. AM peak is generally defined as the 6AM to 9AM period by regional
transportation agencies.
Table 5-2 shows that those who use non-motorized modes live close to their workplaces
(15% live and work in the same census tract; the rest travel about 3 miles to work on average),
and that they could hardly save commute time to work using the transit alternative. Interestingly,
although the transit users are expected to locate in places with good transit connectivity to work,
they could potentially save over 75% of their commute time to work on average if they had
access to auto. Similarly, alternative home-to-work transit commute time of the small number of
auto users is four times larger.
While the transit users in this group mostly comprise of the existing transit dependents
and perhaps people with pro-transit (and/or anti-auto) attitudes, the auto users (potential transit
riders) constitute a very small fraction (0.4%) of the base sample, and hence the population.
Therefore, excluding carless commuters from the study may be justified by the small amount of
The Demand for Reliable Travel Chakrabarti (2015)
133
discretionary transit use (and hence latent transit travel demand) within this group. It is also not
possible to analyze characteristics of any discretionary transit riders in this group who can but do
not own a car. The lack of information in CHTS to identify such individuals limits the current
study.
Table 5-2.
Alternative travel (journey to work) environments of commuters in carless households
Variable All
Transit
user group
Non-motor
user group
Auto
user group
Total observations (N)
a
244 149 (61%) 61 (25%) 34 (14%)
Distance
b
(in mi; SD in parentheses)
7 (7) 8 (8) 3 (3) 9 (7)
Travel time (in min; Std. Dev. in parentheses)
via best transit route
c
50 (29) 24 (19) 59 (39)
via non-motorized mode
d
139 (135) 26 (36) 151 (126)
via best auto route
e
12 (8) 6 (4) 14 (9)
Note: Figures in bold denote chosen mode for a group; figures in italics denote alternate modes for a group. Travel time refers to total door-to-
door home-to-work time under weekday AM peak congested conditions.
a.
All usable cases, for which valid (transit/non-motor/auto) commute mode choice is reported.
b.
Distance along best (least total travel time) auto route in the weekday AM peak is used for comparison purposes; Cases where estimated
distance is zero (i.e. origin and destination in same census tract) are excluded here.
c.
Routes are excluded for cases where origin and destination are in same census tract, and unavailable where a viable transit route could not
be suggested by Google Maps.
d.
When non-motorized mode type is unknown (i.e. for cases where a non-motorized mode is not chosen, or when “other non-motorized”
mode is chosen), the average waking speed is assumed.
e.
Routes are excluded for cases where origin and destination are in same census tract.
4.4 Commuters in car-owning households
Figure 5-5 shows that less than 6% of individuals with at least one vehicle in their household
reported using transit as their normal commute mode. 91% use the auto mode, leaving the share
of non-motorized travel at below 4%. Transit use among car-owners is rare.
Recall that this study analyzes factors associated with discretionary transit use for the
commute – i.e. choice of transit among individuals who have the option to use multiple travel
The Demand for Reliable Travel Chakrabarti (2015)
134
modes(s). A direct approach is to focus on people who belong to car-owning households (some
of whom could be discretionary transit riders) and to focus specifically on auto vs. transit users,
to learn if and how latent demand can be attracted or mode shifts induced or discretionary rider
attrition arrested. I exclude the non-motorized travelers in this group because it is difficult to
ascertain whether transit can (or even needs to) present itself as a potential alternative for those
who choose to walk or bike or use other active travel modes for commuting.
Following (Table 5-3, Table 5-4, and Table 5-5) are a series of descriptive comparisons
between the auto user group and the transit user group, with the non-motorized traveler group as
reference when relevant.
As expected, Table 5-3 shows that the transit users are relatively younger, have lower
household income, and have fewer cars in their households in spite of marginally higher
household size on average, when compared to the auto users. Since there is no reason to expect
(nor is there a way to measure) any systematic difference in non-household carpooling
opportunity between the auto-user and transit-user groups, the figures suggest that some car-
owners may be compelled to use transit because the household vehicle(s) is/are not available to
them because of personal
65
or household
66
constraints. Although it is possible that some pro-
transit individuals choose not to be licensed and/or to purchase fewer cars, all transit users in the
sample of commuters in car-owning households may not be transit riders entirely by choice.
65
e.g. no driving license, and ridesharing with others is unfeasible
66
e.g. more workers than cars, and ridesharing with others is unfeasible.
The Demand for Reliable Travel Chakrabarti (2015)
135
Table 5-3.
Household and personal characteristics of commuters in car-owning households
Variable
Auto user
group
(1)
Transit user
group
(2)
Difference
(1-2)
Pr(|T| > |t|)
(Difference)
Non-motor
user group
(Reference)
Total observations (N)
a
6747 419 - - 265
Household characteristics
Household income
b
(annual; % by category)
$0 to $49,999 21.92 38.42
- -
28.98
$50,000 to $99,999 31.36 26.73 33.47
$100,000 to $149,999 19.03 16.47 17.14
$150,000 to $199,999 9.78 7.16 10.20
$200,000 or more 8.64 1.91 10.20
Number of household vehicles 2.23 1.62 0.61 0.00 1.71
Household size 3.07 3.23 -0.16 0.03 3.16
Members per household vehicle 1.51 2.35 -0.84 0.00 2.12
Workers per household vehicle 0.92 1.41 -0.49 0.00 1.32
Personal characteristics
% Female 48.77 46.06 2.71 0.28 39.25
% in the 16 to 35 age category 22.60 31.89 -9.29 0.00 34.00
% with more than 1 job 6.91 6.70 0.21 0.87 7.55
% with flexible
c
work schedule 59.13 58.68 0.45 0.86 69.11
% without driving license 1.96 27.58 -25.62 0.00 15.97
Note: Not all respondents who report valid (transit/non-motor/auto) commute mode have chosen to answer all other questions. Therefore, only
valid responses corresponding to each variable have been considered for statistical analysis.
a.
All usable cases, for which valid (transit/non-motor/auto) commute mode choice is available.
b.
Percentages do not add up to 100 because not all respondents chose to report their household income.
c.
At least “some flexibility” (CHTS definition).
Table 5-4 shows that the transit users live and work in census tracts that are denser – in
terms of population, employment, and transit service density – and have lower income levels on
average. It is possible that density induces auto to transit mode shifts; but perhaps it is only
appropriate to conclude that the discretionary transit users and those who cannot avail of an auto
are drawn to neighborhoods with characteristics that support transit service provision and
The Demand for Reliable Travel Chakrabarti (2015)
136
consumption. For readers familiar with the Los Angeles region, Figure 5-6 identifies the census
tracts in which the car-owning transit users in the sample live and work.
Table 5-4.
Home and workplace neighborhood characteristics of commuters in car-owning households
Variable
Auto user
group
(1)
Transit user
group
(2)
Difference
(1-2)
Pr(|T| < |t|)
(Difference)
Non-motor
user group
(Reference)
Total observations (N)
a
6747 419 N.A. N.A. 265
Population density (per sq. mi.)
Home 10,028 14,747 -4,719 0.00 13,631
Work 8,484 11,199 -2,715 0.00 10,384
Employment density (per sq. mi.)
Home 3,436 4,851 -1,415 0.00 6,558
Work 14,117 33,422 -19,305 0.00 15,548
Transit stop density (per sq. mi.)
Home 25.84 39.53 -13.69 0.00 38.69
Work 44.73 104.46 -59.73 0.00 52.68
Median income (2010 dollars)
Home 72,689 57,358 15,331 0.00 66,210
Work 64,795 51,216 13,579 0.00 63,365
Note: Home and work refer to home and workplace census tracts; it is possible that home and work census tracts are the same for some
respondents in the sample.
a.
All usable cases, for which valid (transit/non-motor/auto) commute mode choice is available.
The Demand for Reliable Travel Chakrabarti (2015)
137
Figure 5-6. LA County census tracts in which car-owning transit commuters live and work
Panel A. Home locations
Panel B. Workplace locations
Note: The sample of commuters is not selected to represent LA County. Therefore, census tracts are not weighted by the number of cases they
contain; Names of adjoining counties are given for visual reference.
The Demand for Reliable Travel Chakrabarti (2015)
138
Table 5-5 compares [simulated] home-to-work travel environments in the weekday AM
peak via the auto mode and the transit mode for the auto-user group (who chose auto but had
transit as an alternative) and the transit-user group (who chose transit, and [possibly] had auto as
an alternative).
Observe that on average, transit users do not live significantly closer to their workplaces
compared to auto users. But they do seem to have a more favorable transit travel environment
available, than that of auto users, for the journey from home to work – lower total travel time via
transit, lower transit/auto travel time (indicating comparatively lower travel time savings via the
auto mode), and fewer transfer requirement. Also, about 32% (a significantly larger proportion
compared to auto users) of the transit users have rail (light rail/subway/commuter rail) in their
shortest transit route to work, and over half (about 60%)
67
of them report rail as their normal
commute mode. The data suggest that the quality attributes of rail service and/or urban
characteristics associated with neighborhoods connected by rail, and often by good quality bus
along with rail, attracts some discretionary riders.
The descriptive statistics may indicate that, along with many other qualitative factors,
high-speed transit service with good network design that minimizes transfers promotes
discretionary transit travel, and that rail is particularly attractive. The observed effect could be
because persons in car-owning households with pro-transit attitudes and/or who do not have
67
Some respondents report normally using bus instead of rail for their commute, even when total travel time via the
bus-only route is greater than the simulated rail route (for the journey from home to work). This is possible if, for
them, the generalized cost of bus travel (including, e.g., factors such as service frequency, safety, access/egress
quality etc.) is lower than rail. This could also be due to routine household responsibilities (e.g. taking detour for
dropping children to school on the way to work) and personal preferences – factors on which there is no
information. However, note that since a transit journey often consists of two or more segments via different transit
modes (express bus, commuter rail, trolley-bus etc.), and since a CHTS respondent could only report one (the
longest-distance) transit mode (transit category), it is possible that there are cases where bus is reported even when
rail is included in the route.
The Demand for Reliable Travel Chakrabarti (2015)
139
access to a car consider the abovementioned factors when choosing household and/or workplace
locations.
Whether high levels of congestion (indicated by estimated per-mile travel time) and
unreliability (indicated by estimated buffer index) of the auto mode (journey from home to work)
could potentially explain transit mode choice among car owners in the LA region is unclear.
Note that buffer index is defined as the estimated fraction (expressed as %) of mean
travel time that should be budgeted (or added) in order to ensure on-time arrival at the
destination 95% of the time. Buffer time is the absolute magnitude of estimated additional time.
A higher value of buffer index implies greater level of travel time unreliability. For details, refer
http://ops.fhwa.dot.gov/publications/tt_reliability/brochure/ (last accessed on 12/29/14).
Auto route buffer index (for each journey from home to work case) for the typical
weekday AM peak commute is estimated by re-simulating home-to-work trips (shortest time
paths) though the SCAG network where each road link is loaded with 95th percentile travel time
(i.e. baseline congested link travel time plus a link-level buffer time), and then comparing the
output (i.e. 95th percentile commute travel time) with the previously obtained mean travel time.
Link-level buffer time is estimated by analyzing the distributions of 15-minute
aggregated average speed readings of over 3000 highway and 4700 arterial traffic sensors (i.e.
loop detector stations) across the Los Angeles region over the weekday AM peak periods of a
test month (November 2012). Data is available through ADMS, a comprehensive historical
archive of real-time multi-modal transportation system data (project sponsored by Metro). Road
links that do not contain a sensor are assigned average buffer times of “similar” links for which
data is available. Similarity is determined through a cluster analysis approach using no. of lanes,
AADT, speed limit, no. of feeder intersections and presence of HOV/HOT lanes as attributes for
The Demand for Reliable Travel Chakrabarti (2015)
140
highway links (3 highway clusters generated), and no. of lanes, link length and signalized
intersection density as attributes for arterial links (6 arterial clusters generated). The clustering
method was originally developed as part of a separate regional transportation system monitoring
application using ADMS. In essence, and in the present commuting context, buffer index
conceptually captures the estimated frequency and severity of non-recurrent congestion.
Note that auto route buffer indices may be overestimated, since in reality, all road links in
the home-to-work paths may not be affected by severe non-recurrent congestion simultaneously.
This limitation cannot be effectively addressed using disaggregate roadway traffic sensor
information. Accurate estimation requires repeated home-to-work total travel time data, e.g.
using probes through the network, in a carefully designed experimental setup.
There is no statistically significant difference in the average level of auto route
unreliability between the auto and transit user groups. On average, the auto users are found to
travel under marginally (but statistically significant) more congested conditions than the
alternative auto environments available to the transit users. This result is counter-intuitive. There
are several possible explanations.
First, the disproportionately small number of transit user cases may fail to adequately
capture the full range of conditions, particularly those situations where good-quality transit
alternatives exist and are used among car-owners traveling along highly congested corridors.
Indeed, the sample of commuters is not chosen in a way that appropriately represents the
underlying population within the study area; it is possible, for example, that there is less-than-
proportional representation of discretionary rail users. In that case, a descriptive comparison of
aggregate statistics across groups may not work. Hence, the marginal difference in mean
congestion may not have practical or policy significance. Note, however, that the range of auto
The Demand for Reliable Travel Chakrabarti (2015)
141
route congestion in the sample of car-owning transit users is fairly large – between 0.86 and 5.03
min per mile (buffer index ranges from 15-170%).
Second, most bus lines in the LA region do not have exclusive rights of way or signal
pre-emption/priority, and therefore experience similar levels of roadway congestion as the auto
mode. It is unlikely that the discretionary transit users in the sample, most of whom need to use
bus in one or more segments of their routes, would choose transit options in highly congested
contexts that are associated with lower transit travel speed, higher service unreliability, and
overcrowding on average.
Finally, in LA County, there are many high-demand corridors (connecting, e.g., desirable
residential locations with major employment centers) with high levels of peak-period congestion
and inadequate transit supply (a cause and consequence of high auto travel demand). A large
number of commuters tolerate large peak-period delays since they cannot find better locations
and/or modal alternatives based on their idiosyncratic demands. It is possible that the current
dataset includes substantial signal from such cases.
The Demand for Reliable Travel Chakrabarti (2015)
142
Table 5-5.
Alternative travel (journey to work) environments of commuters in car-owning households
Variable
Auto user group
(1)
Transit user
group
(2)
Difference
(1-2)
Pr(|T| < |t|)
(Difference)
Total observations (N)
a
6747 419 N.A. N.A.
Distance (mi)
b
12.35 13.18 -0.84 0.13
Transit travel environment
c
Travel time (min) 73.90 68.94 4.96 0.03
Transit/auto travel time 5.20 4.99 0.21 0.05
No. of transfers 1.11 1.00 0.11 0.02
% with rail (light/subway/commuter) in route 19.78 31.50 -11.72 0.00
Auto travel environment
d
Congestion level (per mile TT) 1.83 1.73 0.10 0.01
(Un)reliability (buffer index in %)
e
32.75 31.94 0.81 0.10
Note: All travel times are home-to-work door-to-door via the best (least total travel time) route, in a typical weekday AM peak, relevant during
the time in history when each respondent was surveyed, considering all intermediate [multi-modal] segments of a linked trip; Parking is
considered to be on-site at destination for the auto mode, and at-station in case of park-and-ride; Not all transit and auto environments could be
simulated for all origin-destination pairs.
Figures in bold denote expected travel environment along chosen mode for a given group – auto/transit users.
a.
All usable cases, for which valid (transit/non-motor/auto) commute mode choice is available.
b.
Distance along best (least total travel time) auto route in the weekday AM peak is used for comparison; Cases where estimated distance is
zero (i.e. origin and destination in same census tract) are excluded here.
c.
Routes are unavailable for cases where origin and destination are in same census tract, and where a viable transit route could not be
suggested by Google Maps.
d.
Routes are unavailable for cases where origin and destination are in same census tract.
e.
e.g. buffer index of 40% means that for a trip that usually takes 20 minutes, a traveler should budget an additional 8 minutes to ensure on-
time arrival 95% of the time.
The Demand for Reliable Travel Chakrabarti (2015)
143
5. Determinants of Car-owners’ Transit Mode Choice for Commute Trips
This section analyzes effects of select determinants of car-owners’ transit mode choice for
commute trips. The conceptual framework is based on the premise that, given where an
individual lives and works, the [expected] characteristics (i.e. home-to-work travel environment)
of each viable mode-route determines its (dis)utility, and that the individual chooses the mode-
route (from the available set of alternatives) that has minimum relative disutility. Each person’s
perceived (dis)utility for a given mode is governed by personal/household constraints including
attitudes and habits, and it is possible that home and/or workplace location choice is governed in
part by prior inclination towards a particular mode.
A cross-sectional approach is adopted, which can only investigate sociodemographic and
[relative] multi-modal transportation environment characteristics that are associated with car-
owners’ [transit] mode choice. I model normal/usual transit vs. auto mode choice for the journey
to work. Using statistical controls that account for factors that may limit access to the household
auto, I attempt to isolate factors associated with discretionary transit commute among car-
owners. Findings are expected to provide insights on how objective changes in system
performance through planning-policy actions can promote discretionary transit use.
5.1 Approach
I reduce the base CHTS dataset (refer Section 4.1) for this analysis. The condensed sample
consists of those employed car-owners living and working in LA County who have a direct (no
transfers) Metro
68
bus connection as their best (i.e. least total travel time) home-to-work transit
option and also report choosing either auto (as driver or passenger in household or any other
68
Bus service operated by the Los Angeles County Metropolitan Transportation Authority (LAMTA), the largest
public transit operator in the greater Los Angeles metropolitan area.
The Demand for Reliable Travel Chakrabarti (2015)
144
private vehicle) or transit (bus) as their normal commute mode.
69
The following paragraphs
systematically explain the rationale behind the step-wise filtering process.
First, I identify cases that have the option of a direct transit connection from home to
work. These are cases where transit can potentially serve as a close competitor to auto (although
transit travel may still require additional access/egress trips in most cases). Simulated transit
routes in these cases are most likely to be the routes actually chosen by those who use transit;
this has been verified using the information from the one-day travel diary in CHTS (refer
footnote 70), and it increases accuracy of estimated models. Eliminating transfers helps
accurately measure transit service quality/performance parameters for the entire home-to-work
journey; while we can estimate the negative effect imposed by [one or more] transfers on transit
mode choice, explicitly identifying the types and natures of risk associated with uncoordinated
transfers and measuring them accurately is difficult. Determination of average service quality
and performance measures for a chained transit trip may, therefore, be subject to large errors due
to the many assumptions necessary with standard line- or stop/station- or time point-level transit
system data available for this research. Since past literature does not provide any guidance, this
filtering step is useful.
Second, I eliminate cases that have any rail mode in the best direct home-to-work transit
route. There are two reasons: 1) Service quality and performance data at the desired level of
spatiotemporal disaggregation was not available for the regional rail network; and 2) A
comprehensive set of variables that differentiate rail and bus systems is difficult to identify and
measure; consequently, including rail results in biased estimates [in favor] of factors such as
69
Note: Individuals who have direct Metro bus connection as their best (least total travel time) home-to-work transit
travel option in the typical weekday AM peak (based on the regional transit network that was valid during the time
in history when each person was surveyed) are identified by simulating transit paths.
The Demand for Reliable Travel Chakrabarti (2015)
145
service frequency and reliability. Note that the positive influence of rail on car-owners’ transit
mode choice for commute trips has already been suggested by the descriptive analysis presented
in the previous section.
Third, I only consider cases where the direct bus line that constitutes the best home-to-
work transit connection is operated (directly, or through contract) by Metro. This is because
service quality and performance data at the desired level of spatiotemporal disaggregation is not
available for other city/local/municipal bus operations. Metro bus transit data used for this
analysis is available through the ADMS research project.
Finally, since I focus on transit vs. auto mode choice, I drop the small number of cases
where a non-motorized mode is chosen. Transit presents itself as a potential alternative to both
the remaining groups – auto users and transit users.
Note that the filtering helps extract effects of select [multi-modal] system performance
factors on transit mode choice by holding several other determining factors (e.g. transit travel
out-of-pocket cost, amenities, information availability, etc.) constant across cases.
A total of 1,381 persons in the CHTS dataset with at least one car in their household live
and work in LA County and had the option of a direct Metro bus (not rail, and not a non-Metro
bus) as their shortest duration transit trip from home to work in the weekday AM peak when they
were surveyed. Among them, 1,144 (83%) reported normally using auto, and 118 (9%) reported
normally using bus transit.
70
Therefore, the usable sample size for this analysis is (1144 + 118) =
70
There is no way to directly validate that a person who reported using bus as the normal commute mode actually
takes the same directional Metro bus line to work that has been simulated. CHTS did not require a person to report
the transit agency name or line number for any segment of a transit trip in the person-level characteristics file that
contains information on normal commute mode, and that is the reason for simulation requirement in the first place.
However, although I used census tract centroids as origins and destinations, it is unlikely that there are multiple
direct Metro bus lines having comparable total travel times between any given census tract pair (note that having a
rapid line and a local line as alternatives is common; but they have vastly different travel times) – so the simulated
line having least total travel time (i.e. the best transit connection) should match with the actually used bus line. Also,
The Demand for Reliable Travel Chakrabarti (2015)
146
1262. The remaining individuals either use a non-motorized mode, or a special/non-traditional
mode (refer footnote 64 for definition), or reported using a transit mode different from [the
simulated] Metro bus; they were therefore excluded from the analysis.
Note that for the sample of 1,262 persons, transit alternative automatically refers to a
single Metro bus line, since that is their best home-to-work transit commute option.
5.2 Model, variables, and descriptive statistics
The standard functional form for analyzing the influence of potential explanatory variables on a
binary (e.g. success vs. failure) dependent/outcome variable is the binomial logistic regression
model. In the mode choice context, the model fundamentally assumes that travelers attach a
(dis)utility to each mode available to them, and they choose the mode that provides maximum
utility (or minimum disutility). To the analyst, (dis)utility is unobserved, but choice is observed;
the problem is identifying and estimating effects of factors that may determine choice.
In the present context (i.e. transit vs. auto choice), if individual i in the sample attaches
utility Ui,transit to using the transit mode and Ui,auto to using the auto (i.e. not using transit)
alternative (U is determined by a set of m observable [X1… Xm; or vector X] and n unobserved or
random [ε1… εn; or vector ε] factors, including comparative generalized modal cost parameters
and individual-specific parameters, which capture the relative attractiveness of the two modes for
an individual) for her normal home-to-work trip, then:
The binary logistic regression model form for choice of transit (transit=1) is given by:
𝒍𝒏 𝑷 ( 𝒕𝒓𝒂𝒏𝒔𝒊𝒕 =𝟏 )
𝟏 −𝑷 ( 𝒕𝒓𝒂𝒏𝒔𝒊𝒕 =𝟏 )
= 𝒂 + 𝒃𝑿 + 𝜺 ... (Equation 9)
a 100% match (for agency name and line number) is found in case of those persons (i.e. bus commuters) who
reported their trips (via the one-day travel diary that includes transit line details) on the designated travel day and
also made a commute trip using transit.
The Demand for Reliable Travel Chakrabarti (2015)
147
Where:
a = coefficient on the constant term
b = coefficients on the observed independent variables
ε = error term
And the estimated probability of choice of transit for a given case is:
𝑷 ( 𝒕𝒓𝒂𝒏𝒔𝒊𝒕 = 𝟏 |𝑿 = 𝒙 )=
𝟏 𝟏 +𝒆 −( 𝒂̂+𝒃 ̂
𝒙 )
... (Equation 10)
Where:
𝑎 ̂ = parameter estimate on the constant term
𝑏 ̂
= parameter estimates on the observed independent variables
The dependent variable (or choice dimension) is the normal commute (journey to work
from home) mode choice of each of the 1,262 persons included in this part of the analysis.
Transit choice is 1; auto (base) is coded as 0.
The explanatory variables (parameters determining mode choice – transit vs. auto)
capture planned service quality and performance of the two alternative available modes (that
contribute to generalized costs of travel via the two modes from home to work), and also control
for select personal/household factors that are expected to have large independent influence on
choice.
Note that the variables are selected based on broad findings of prior research, research
focus of the current study, and general findings from the descriptive analysis presented in
Section 4. The strategic filtering/shrinking of the dataset increases homogeneity within the used
sample, and provides the opportunity to select variables most central to the current research
question. Also, the small proportion (9.5%) and absolute number (118 cases) of events (i.e.
choice of transit) imposes restrictions on the number of right-hand side variables.
Selection of explanatory variables is described below.
The Demand for Reliable Travel Chakrabarti (2015)
148
A. [Relative] expected travel times
Travel times of alternate available modes have been used in several standard commute
mode choice models in the past. In the present study, we should expect that, holding expected
total travel time via the best auto route constant, the expected total travel time via the best transit
route to be negatively associated with the probability of choice of transit as the usual commute
mode. I therefore use a ratio – transit/auto travel time – as an explanatory variable. Both travel
times refer to total expected door-to-door time from home (census tract centroid) to work (census
tract centroid) in a typical weekday AM peak including all associated intermediate segments
(e.g. access/egress to/from bus is/are included) of the journey. The variable is dimensionless
(ratio) and expected to be negatively associated with transit (Metro bus in this case) mode
choice. Note that I also include a transit travel time (in min) variable, since duration of transit
travel may have a separate influence on the (un)desirability of transit.
B. [Relative] travel time (un)reliabilities
Travel time reliability is an established determinant of travel behavior (refer Fosgerau
and Engelson, 2011), and we know that the demand for reliable travel can, under certain
circumstances, be even greater than the demand for travel time savings. A large volume of
empirical literature suggests that the demand for reliable travel influences trip scheduling
decisions (e.g. Small, 1982; Noland and Small, 1995) and auto route choice (e.g. Noland et al.,
1998; Lam and Small, 2001; Liu et al., 2004; Small et al., 2005; Asensio and Matas, 2008;
Tilahun and Levinson, 2010; Hainen et al., 2011; Carrion and Levinson, 2013). The reliability-
mode choice connection is relatively less explored, and standard passenger transport mode
choice models do not include the reliability parameter as a component of the generalized cost of
modal alternatives. Till date, only Sweet and Chen (2011) have analyzed the relationship. They
The Demand for Reliable Travel Chakrabarti (2015)
149
investigated how unreliability of the auto mode affects auto mode choice for commuters in
Chicago using GPS survey data. Their study suggests that unreliability of the auto mode may
increase demand for transit; but the authors find the influence of unreliability to be limited.
The influence of any dimension of transit service reliability on transit mode choice has
never been directly estimated, although passengers are found to rate reliability as one of the top
service quality factors of public transit (e.g. studies by Wachs, 1976; Glascock, 1997; Eboli and
Mazzulla, 2007; Tyrinopoulos and Antoniou, 2008; Cantwell et al., 2009; dell’Olio et al., 2010;
Iseki and Taylor, 2010; and Nurul Habib et al., 2011 are illustrative).
This study attempts at addressing methodological limitations of past research and
providing new insights by including measures of travel reliability to characterize travel
environments (and hence generalized costs) across alternative modes.
My model conceptually allows the relative reliability of a mode to influence in part the
probability of choice of that mode. To that extent, the model builds on the “centrality-dispersion”
or the “mean-variance” approach (refer Carrion and Levinson, 2012 and Li et al., 2010) used in
many past travel time reliability studies, that hypothesize that a decision maker (with a given
degree of risk aversion) in the transportation context, under a given set of conditions (personal,
household, and neighborhood characteristics; trip purpose etc.), aims at maximizing returns (e.g.
expected travel time savings) while minimizing associated risk (e.g. expected travel time
variability).
For the auto mode, buffer index (in the typical weekday AM peak) of home-to-work
travel is chosen as the (un)reliability measure.
For transit, two different measures of (un)reliability are tested. The first is the U.S.
industry standard service reliability measure -- on-time performance or OTP (refer the “Transit
The Demand for Reliable Travel Chakrabarti (2015)
150
Capacity and Quality of Service Manual”
71
). It is an indicator of schedule adherence or
punctuality.
In this study, OTP of the chosen/alternative [directional] Metro bus line for a particular
commuter is measured as the average percentage of time (considering all scheduled trips) that the
line has departed “on-time,” i.e. between 1-minute early and 5-minutes late with respect to
schedule, considering all time points along the line, within the weekday AM peak period over
three months preceding
72
the month in which the respondent’s choice was recorded. The
theoretical range of this variable is 0-100%. It is expected that OTP is positively associated with
transit (Metro bus in this case) mode choice, all else equal. Since this measure is standard within
the industry, processed data was directly available as monthly averages by directional bus line
directly from Metro. Note that OTP data for all Metro bus lines simulated for the sample of
commuters is not available.
Access to a historical archive of Metro’s real-time bus AVL (automatic vehicle location)
data feed via the RIITS (Regional Integration of Intelligent Transportation Systems) system as
part of the ADMS research project provided an opportunity to test a second measure, a measure
of service unreliability – the standard deviation (SD) of line schedule deviation. Lower SD of
schedule deviation is associated with reduction in unpredictability of travel.
Metro’s AVL feed consists of periodically updated real-time locations of Metro buses
and corresponding schedule deviation estimates at time points that is primarily used for
providing real-time vehicle arrival/departure predictions to passengers via web based
applications. The historical archive helps derive distributions of schedule deviation across all
71
The third edition is available online at http://www.trb.org/main/blurbs/169437.aspx (accessed on 12/24/2014);
also see Kittelson & Associates. (2013).
72
e.g., for a respondent who was surveyed in June 2012, data is averaged over March-May 2012.
The Demand for Reliable Travel Chakrabarti (2015)
151
available time point stops for all available lines (considering all available scheduled trips) at any
time of day over any period of interest. For a given commuter, the SD of schedule deviation
measure for the chosen/alternative [directional] Metro bus line is estimated considering all
constituent time point stops of the line, within the weekday AM peak, observed over three
months preceding the month in which the respondent’s choice was recorded. Data cleaning is
performed to eliminate outliers (e.g. records in the AVL database that are more than 20 minutes
early or over one hour late are dropped; consistent with Metro’s data cleaning approach) in the
raw data. The unit of this measure is minutes, and the measure can theoretically assume any
positive value. It is expected to be negatively associated with transit (Metro bus) mode choice,
all else equal. The measure could not be derived for all Metro bus lines.
Note that a direct ratio of transit and auto (un)reliability measures is not meaningful as
the measures are in different units.
C. Other service quality factors
For the best transit route, the service headway (in minutes) of the [directional] Metro bus
line in a typical weekday AM peak is used. For the best auto route, a level-of-congestion
measure, estimated per-mile travel time (in minutes) under recurrent congestion in a typical
weekday AM peak is included. Both measures correspond to the time in history when each
individual was surveyed. Theoretically, service headway is expected to be negatively associated
with, and auto congestion is expected to be positively associated with transit (Metro bus) mode
choice in the present context.
D. Access to transit, and built environment quality, at home and workplace
Access to transit at home and workplace is measured by census tract level physical transit
stop density (per sq. mi.; considering all modes of public transit of all agencies) variables. Since
The Demand for Reliable Travel Chakrabarti (2015)
152
there is a constraint in the number of independent variables that can be included, the stop density
variables are particularly useful. Considering the 1262 cases, home stop density is found to be
well correlated with average home population density, and workplace stop density with average
workplace employment density, all variables measured at the census tract level. This is not
surprising, because transit supply is targeted towards potential demand. The stop density
variables are therefore expected to pick up built environment effects such as walkability, parking
cost, etc.
E. Personal and household factors
Based on the descriptive statistics presented in Table 5-3, I select two key control
variables in this category: license holder (indicator variable, =1 if respondent holds a valid
driving license and =0 otherwise) and workers per household vehicle. I do not include income
and age, because the factors are not absolutely central to the current research, and particularly
because invalid responses on those variables lead to significant dropping (103 total and 16
events) of [the already small number of] usable cases for estimating regression models. The
combined effect of these variables, along with many unobserved influential variables (e.g.
personal attitudes and household/workplace responsibilities etc.), will be captured by the error
term in the regression model. Note that workers per household vehicle may be considered to be a
reasonably good proxy for household income.
I do not directly include out-of-pocket costs. A single Metro bus boarding costs the same
across cases ($1.50 at the time of survey); auto travel cost is picked up by auto travel time and
congestion level; and destination transit stop density proxies generalized cost of parking.
The independent variables are summarized in Table 5-6. Descriptive statistics of the
continuous independent variables are given in Table 5-7 for reference.
The Demand for Reliable Travel Chakrabarti (2015)
153
Table 5-6.
Independent variables
Variable
Expected relationship to
transit choice probability
Primary data source(s)
Transit travel time - Google Developers
Transit /auto travel time -
Google Developers and
SCAG model
Transit line OTP + Metro/ADMS
Transit line SD of sched. dev. < 3 min
(Indicator)
+ Metro/ADMS
Buffer index (auto route) + ADMS and SCAG model
Transit line service headway - Metro/ADMS
Per-mile travel time (auto route) + SCAG model
Home CT transit stop density + SCAG
Workplace CT transit stop density + SCAG
Driving license holder
(Indicator)
- CHTS
Workers per household vehicle + CHTS
Table 5-7.
Descriptive statistics
Variable Unit N Mean SD Min Max
Transit travel time min 1262 39.11 18.06 7 111
Transit /auto travel time - 1262 4.40 1.80 0.95 11.93
Transit line on-time performance % 1208 75.03 6.14 48 92
Transit line SD of schedule deviation min 1261 3.74 0.82 1.40 7.34
Buffer index (auto route) % 1262 32.04 9.88 6.99 117.36
Transit line service headway min 1262 19.18 12.95 5 90
Per-mile travel time (auto route) min 1262 2.19 0.73 0.85 5.91
Home CT transit stop density no./sq. mi. 1262 34.03 32.63 0 361
Workplace CT transit stop density no./sq. mi. 1262 55.49 65.77 0 361
Workers per household vehicle no. 1209 1.00 0.46 0.25 4
Note: CT refers to census tract; Transit refers to Metro bus.
The Demand for Reliable Travel Chakrabarti (2015)
154
5.3 Analyses, results, and discussions
Table 5-8 summarizes binomial logistic regression model estimates for car-owners’ transit mode
choice for commute trips, under the set of conditions described in Section 5.1.
Models 1A and 2A are estimated by the unconditional logistic regression (maximum-
likelihood) approach, and use OTP and SD of schedule deviation as measures of transit line
(un)reliability respectively. Corresponding Models 1B and 2B are presented for robustness
check, and are estimated using Firth’s penalized-likelihood approach to address possible small-
sample bias, particularly because of the small number of events (i.e. transit mode choice) in the
sample (refer Firth, 1993; Coveney, 2008).
SD of schedule deviation is not included as a continuous variable; rather, an indicator
variable that separates highly reliable service (with SD of schedule deviation < 3 min; i.e. under
the 25
th
percentile considering 1261 observations on the variable) from relatively more unreliable
ones (with SD of schedule deviation ≥ 3 min). This helps derive conceptually clearer estimates of
transit reliability that have practical relevance. Note that all cases for which complete data
corresponding to all variables were available have been used for estimating regression models.
Parameter estimates suggest that the quality of home-to-work transit travel environment
has a statistically significant, positive association with the probability of choosing transit.
Controlling for factors that may increase the likelihood of transit-dependence (i.e. no driving
license, and more workers than vehicles in household), transit travel among car-owners is more
prevalent in contexts where, on average, travel time savings via the alternate auto mode is
relatively smaller and where transit service is frequent, reliable, and well-accessible (at both trip
ends). Observe that, all else equal, the total duration of transit journey does not have a
statistically significant independent effect on choice. In general, the results suggest that good
The Demand for Reliable Travel Chakrabarti (2015)
155
service quality facilitates discretionary transit use by lowering the generalized costs of transit
travel significantly.
The estimated odds-ratios indicate the following statistically significant (at p=0.05 or
better) magnitudes of increase in odds of transit mode choice (all else equal): 21-27% for unit
reduction in the transit-to-auto travel time ratio; 6-7% for 1 percentage point increase in average
transit line on-time performance; about 150% for bringing down the average SD of transit line
schedule deviation to under 3 minutes; about 3% for 1 minute reduction in transit service
headway. The effect sizes seem realistic. Note that a valid driving license holder has 90% lower
odds of using the transit mode compared to one without a license, all else equal; and an
individual in a dual-worker one-car household has over 300% higher odds of using transit
compared to an identical individual in a dual-worker two-car household, all else equal. The
findings suggest that discretionary transit use may be a rare phenomenon, but service quality and
performance of transit still plays a significant role in attracting discretionary riders for commute
trips.
Interestingly, congestion and unreliability of the auto mode does not have statistically
significant impacts on transit mode choice. This result is consistent with that of the descriptive
analysis in the previous section (see Table 5-5). Deterioration of the auto travel environment
(within the range of conditions captured in this study) alone, all else equal, cannot increase
generalized cost of auto travel relative to transit on average to a level that might prompt
substantial auto-to-transit mode shift across the region. But it is possible, and the current study
does suggest, that simultaneous improvement (over baseline conditions) in transit service can
attract latent transit travel demand.
The Demand for Reliable Travel Chakrabarti (2015)
156
Note that the regression analysis has been performed without considering cases that
require one or more transfers for the home-to-work transit journey. Although transit managers try
to design networks in a way that increases direct connections between major residential and
activity locations on average, transfers are often practically unavoidable. The dataset used in the
current study shows that most commuters need to make at least one transfer, on average, based
on where they live and work and given the existing regional transit network. Increase in average
speed, service frequency, and reliability for linked or chained transit trips (for the journey from
home to work) could be challenging and costly, but they may be worth doing. Additional
reliability improvements could include priority busways and timed and positive transfers.
The Demand for Reliable Travel Chakrabarti (2015)
157
Table 5-8.
Summary of binomial logistic regression models of car-owners’ transit mode choice for commute
trips
Variable
Odds Ratios
Model 1A Model 1B Model 2A Model 2B
Transit travel time 1.008 1.008 1.005 1.006
Transit /auto travel time 0.771 * 0.787 * 0.732 ** 0.749 *
Transit line OTP 1.066 * 1.063 *
Transit line SD of sched. dev. <
3 min (Indicator)
2.594 ** 2.554 **
Buffer index (auto route) 0.982 0.984 0.982 0.984
Transit line service headway 0.972 * 0.975 * 0.965 * 0.967 *
Per-mile travel time (auto route) 0.824 0.843 0.731 0.748
Home CT transit stop density 1.009 ** 1.009 ** 1.010 ** 1.010 **
Workplace CT transit stop
density
1.005 ** 1.005 ** 1.005 ** 1.005 **
Driving license holder
(Indicator)
0.101 ** 0.107 ** 0.122 ** 0.127 **
Workers per household vehicle 4.209 ** 4.038 ** 4.450 ** 4.267 **
Constant 0.002 * 0.003 * 0.439 0.370
N 1155 1155 1208 1208
Pseudo R-square 0.33 - 0.33 -
**p<0.01; *p<0.05
Model 1A: Unconditional logit with OTP effect; Model 1B: Penalized logit with OTP effect; Model 2A: Unconditional logit with SD of schedule
deviation effect; Model 2B: Penalized logit with SD of schedule deviation effect.
Note: CT refers to census tract; OTP refers to on-time performance; Transit refers to Metro bus; Parameter coefficients are natural logarithms of
the odds ratios.
6. Conclusion
This study of car owners in Los Angeles shows that discretionary transit use for commuting,
although rare, is associated with good quality transit service, and that targeted investments at
improving transit travel environments have the potential to attract and retain discretionary riders.
The Demand for Reliable Travel Chakrabarti (2015)
158
Results suggest that carefully-designed transit (particularly light rail and subway) systems
can compete with the auto mode and attract latent demand created by deteriorating roadway
traffic conditions; however, transit managers should be prepared for lower-than-expected growth
in generalized cost of auto travel on average, due to phenomenal integrated corridor management
efforts using intelligent transportation systems (ITS) infrastructures across U.S. metros. We can
expect increasing latent demand as demographic shifts occur (e.g. increase in percentage of
population in relatively young and low-income cohorts), and as auto travel gets costlier (e.g.
increase in out-of pocket costs due to market-priced parking, road use charges, etc.); but the rate
of increase may be moderated by shifts in economic or geopolitical conditions (e.g. increase in
auto-affordability on average, decrease in fuel prices, etc.). Therefore, if transit has to attract
discretionary riders, it must provide exceptional service. It is critical to invest in the right
dimensions of service at the right places, and perhaps expect modest outcomes.
Competing with the auto mode is challenging. Considering the level of freedom and
flexibility offered by a personal vehicle, taking into account the many technological innovations
that are increasing efficiencies and reducing inconveniences associated with the automobile,
recognizing the politics behind subsidized auto travel, and accepting the practical limits of transit
service provision even under a hypothetical environment of limitless resources, it is unrealistic to
expect that transit can win over auto in a large number of circumstances. On average, the cost of
upgrading transit service may not be justified by the volume of (and volatility in) latent demand
that can potentially be attracted, at least in the short term.
But there are contexts in which it may be feasible for transit agencies to provide
competitive transit service. In these situations, significant latent demand, including, large number
of discretionary riders may actually be attracted, and the costs of building and/or delivering high-
The Demand for Reliable Travel Chakrabarti (2015)
159
speed, high-frequency, and highly reliable transit service may be offset by increases in
patronage. For example, there is evidence the LA Metro Exposition light rail line (Phase 1) that
connects West Los Angeles (Culver City) with Downtown LA, operating every 10 minutes
during the weekday peak periods at an average speed of 20 mph and an on-time performance
level of over 95% along the highly congested I-10 (Santa Monica) freeway has boosted transit
ridership across its service corridor (See Chapter 6), reduced auto use among residents in close
proximity to stations (Boarnet et al., 2013), and probably attracted many discretionary riders (as
evidenced by high occupancy rates at park-and-ride facilities), since it resumed service in June
2012. But such examples are relatively rare, since it is often difficult to retrofit rail in high-
demand corridors, and since buses with shared rights of way are not independent of roadway
traffic congestion and hence cannot reach desired levels of speed and reliability.
Nonetheless, descriptive tables and estimated models provide new insights for public
transit planners and policy makers. First, I show that the independent role of transit service
reliability in determining transit mode choice of car-owners is significant; therefore, there is
value in delivering transit service on-time most of the time. Reliable service increases user
confidence by minimizing unpredictability and hence reducing the generalized cost of transit
travel. Therefore, transit managers could see a good bump in ridership by utilizing real-time
multi-modal system information that are increasingly becoming available due to institutional,
technical, and functional coordination across agencies and systems for service planning and real-
time management. From the policy perspective, incentives that promote reliability improvements
may help increase system-wide patronage and hence productivity. Second, planning actions that
increase average speed of transit travel may be meaningful. Examples include network
reconfigurations that minimize transfer requirement, installation of transit signal priority
The Demand for Reliable Travel Chakrabarti (2015)
160
systems, designation of bus-only lanes, consolidation of stops, and investments in light rail and
subway. And third, channelization of resources to places and corridors with high potential
demand for transit travel is critical (without, of course, compromising on the broader social goals
of public transit).
There are several limitations of this study. First, the results of this cross-sectional analysis
may not be directly used to predict/forecast magnitudes of change in transit travel demand for
commute or other trips in response to service changes. This is because we cannot be certain
regarding the extent to which observed effects are due to individuals who prefer transit (or at
least are not averse to transit) self-selecting into areas that have good quality transit access. There
is no doubt, especially in this information age, that people have the opportunity to explore
alternative travel environments and make informed decisions regarding locations. If there is
significant intentional sorting such that most of who prefer transit systematically have more
favorable transit environments to work than transit-indifferent individuals, then we cannot say
for certain from this study that improving transit service will automatically induce mode shifts
within the latter group.
Second, the study does not consider alternative travel environments for the work-to-home
journey. It is possible, for example, that transit service quality and performance in the typical
weekday PM peak (when homebound trips are most likely to occur) negatively affect
discretionary transit use for commuting, regardless of journey-to-work conditions. If that is
indeed true for many cases, and if the AM peak (onward) and PM peak (return) conditions are
not significantly correlated on average across cases, then the positive association of transit
service quality and performance with discretionary transit use may have been underestimated in
the current study.
The Demand for Reliable Travel Chakrabarti (2015)
161
Third, there are many missing variables that could potentially explain discretionary
transit use better and also influence transit policy. Employer-provided financial incentives
(subsidy) for transit use, parking availability and price at or around origin transit stop/station,
parking availability and price at or around workplace, and perceptions regarding key parameters
(e.g. safety) of transit are a few examples. It would be useful if future national/regional personal
travel surveys include questions to capture such data.
Finally, due to unavailability of x-y coordinates, actual trip paths and actual travel
conditions, I had to perform simulations. The study had to be simplified in order to make realistic
assumptions and approximations. ADMS data, although highly granular, has many limitations,
and is not a substitute for GPS trace data. Therefore, there could be large errors in this study. The
results should therefore be used only for understanding potential determinants of transit mode
choice and the value of investing in transit service quality. Parameter estimates may not be used
for forecasting or for accurately estimating effects of quality/performance improvements.
In sum, my research suggests that service quality improvements can help attract new
riders and arrest attrition of existing discretionary riders. Although this is an important policy
goal, the real benefit of service quality investments could be improved utility for transit
dependent populations. The net benefit of these investments on the travelling public may
therefore be larger than what this study estimates.
In addition to informing public transit policy, this study advances knowledge about the
demand for travel time savings and travel time reliability, and how that demand affects mode
choice. The research is expected to contribute to our understanding of travel behavior.
The Demand for Reliable Travel Chakrabarti (2015)
162
7. References
1) Asensio, J. (2002). Transport mode choice by commuters to Barcelona's CBD. Urban
Studies, 39(10), 1881-1895.
2) Asensio, J., & Matas, A. (2008). Commuters’ valuation of travel time variability.
Transportation Research Part E , 44, pp. 1074–1085.
3) Basso, L. J., & Jara-Díaz, S. R. (2012). Integrating congestion pricing, transit subsidies and
mode choice. Transportation Research Part A: Policy and Practice, 46(6), 890-900.
4) Buehler, R. (2011). Determinants of transport mode choice: a comparison of Germany and
the USA. Journal of Transport Geography, 19(4), 644-657.
5) Cantwell, M., Caulfield, B., & O’Mahony, M. (2009). Examining the Factors that Impact
Public Transport Commuting Satisfaction. Journal of Public Transportation, 12(2), 1-21.
6) Carrion, C. & Levinson, D. (2012). Value of travel time reliability: A review of current
evidence. Transportation Research Part A, 46, pp. 720-741.
7) Carrion, C., & Levinson, D. (2013). Valuation of travel time reliability from a GPS-based
experimental design. Transportation Research Part C: Emerging Technologies, 35, 305-323.
8) Cervero, R. (1994). Transit-based housing in California: evidence on ridership impacts.
Transport Policy, 1(3), 174-183.
9) Cervero, R. (2002). Built environments and mode choice: toward a normative framework.
Transportation Research Part D: Transport and Environment, 7(4), 265-284.
10) Chatman, D. G. (2013). Does TOD need the T? On the importance of factors other than rail
access. Journal of the American Planning Association, 79(1), 17-31.
11) Chen, C., Gong, H., & Paaswell, R. (2008). Role of the built environment on mode choice
decisions: additional evidence on the impact of density. Transportation, 35(3), 285-299.
The Demand for Reliable Travel Chakrabarti (2015)
163
12) Collins, C. M., & Chambers, S. M. (2005). Psychological and situational influences on
commuter-transport-mode choice. Environment and behavior, 37(5), 640-661.
13) Coveney, J. (2008). "FIRTHLOGIT: Stata module to calculate bias reduction in logistic
regression," Statistical Software Components. S456948, Boston College Department of
Economics.
14) Crane, R., & Crepeau, R. (1998). Does neighborhood design influence travel?: A behavioral
analysis of travel diary and GIS data. Transportation Research Part D: Transport and
Environment, 3(4), 225-238.
15) de Donnea, F. X. (1972). Consumer behaviour, transport mode choice and value of time:
some micro-economic models. Regional and Urban Economics, 1(4), 355-382.
16) dell’Olio, L., Ibeas, A., & Cecín, P. (2010). Modelling user perception of bus transit quality.
Transport Policy, 17(6), 388-397.
17) Ding, C., Lin, Y., & Liu, C. (2014). Exploring the influence of built environment on tour-
based commuter mode choice: A cross-classified multilevel modeling approach.
Transportation Research Part D: Transport and Environment, 32, 230-238.
18) Domencich, T. & McFadden, D. (1975). Urban Travel Demand: A Behavioral Analysis.
North-Holland, Amsterdam.
19) Eboli, L., & Mazzulla, G. (2007). Service quality attributes affecting customer satisfaction
for bus transit. Journal of Public Transportation, 10(3), 21.
20) Firth, D. (1993). Bias reduction of maximum likelihood estimates. Biometrika, 80(1), 27-38.
21) Forsey, D., Habib, K. N., Miller, E. J., & Shalaby, A. (2013). Evaluating the impacts of a
new transit system on commuting mode choice using a GEV model estimated to revealed
The Demand for Reliable Travel Chakrabarti (2015)
164
preference data: A case study of the VIVA system in York Region, Ontario. Transportation
Research Part A: Policy and Practice, 50, 1-14.
22) Fosgerau, M., & Engelson, L. (2011). The value of travel time variance. Transportation
Research Part B: Methodological, Volume 45, Issue 1, 1-8.
23) Frank, L. D., & Pivo, G. (1994). Impacts of mixed use and density on utilization of three
modes of travel: single-occupant vehicle, transit, and walking. Transportation research
record, 44-44.
24) Frank, L., Bradley, M., Kavage, S., Chapman, J., & Lawton, T. K. (2008). Urban form, travel
time, and cost relationships with tour complexity and mode choice. Transportation, 35(1),
37-54.
25) Gillen, D. W. (1977). Estimation and specification of the effects of parking costs on urban
transport mode choice. Journal of Urban Economics, 4(2), 186-199.
26) Giuliano, G., & Dargay, J. (2006). Car ownership, travel and land use: a comparison of the
US and Great Britain. Transportation Research Part A: Policy and Practice, 40(2), 106-124.
27) Glascock, J. (1997). Research on customer requirements for transit service design and
delivery. Transportation Research Record: Journal of the Transportation Research Board,
1604(1), 121-127.
28) Hainen, A. M., Wasson, J. S., Hubbard, S. M., Remias, S. M., Farnsworth, G. D., & Bullock,
D. M. (2011). Estimating route choice and travel time reliability with field observations of
Bluetooth probe vehicles. Transportation Research Record: Journal of the Transportation
Research Board, 2256(1), 43-50.
29) Iseki, H., & Taylor, B. D. (2010). Style versus Service? An Analysis of User Perceptions of
Transit Stops and Stations. Journal of Public Transportation, 13(3), 23-48.
The Demand for Reliable Travel Chakrabarti (2015)
165
30) Kittelson & Associates. (2013). TCRP Report 165: Transit Capacity and Quality of Service
Manual (3
rd
ed.) Washington, D.C.: Transport Research Board.
31) Lam, T.C., & Small, K.A. (2001). The value of time and reliability: measurement from a
value pricing experiment. Transportation Research Part E, 37 (2001), pp. 231–251.
32) Li, Z., Hensher, D., & Rose, J. (2010). Willingness to pay for travel time reliability in
passenger transport: A review and some new empirical evidence. Transportation Research
Part E, 46, pp. 384–403.
33) Liu, H.X., Recker, W., & Chen, A. (2004). Uncovering the contribution of travel time
reliability to dynamic route choice using real-time loop data. Transportation Research Part
A: Policy and Practice, Volume 38, Issue 6, 435-453.
34) Noland, R., & Small, K. (1995). Travel-time uncertainty, departure time choice, and the cost
of morning commutes. Transportation Research Record 1493, 150–158.
35) Noland, R., Small, K., Koskenoja, P., & Chu, X. (1998). Simulating Travel Reliability.
Regional Science and Urban Economics, 28 pp. 535–564.
36) Nurul Habib, K. M., Kattan, L., & Islam, M. T. (2011). Model of personal attitudes towards
transit service quality. Journal of Advanced Transportation, 45, 271–285.
37) Outwater, M. L., Spitz, G. M., Lobb, J., Campbell, M., Sana, B., Pendyala, R. M., &
Woodford, W. (2011). Characteristics of premium transit services that affect mode choice.
Transportation, 38, 605–623.
38) Pinjari, A. R., Pendyala, R. M., Bhat, C. R., & Waddell, P. A. (2007). Modeling residential
sorting effects to understand the impact of the built environment on commute mode choice.
Transportation, 34(5), 557-573.
The Demand for Reliable Travel Chakrabarti (2015)
166
39) Pinjari, A. R., Pendyala, R. M., Bhat, C. R., & Waddell, P. A. (2011). Modeling the choice
continuum: an integrated model of residential location, auto ownership, bicycle ownership,
and commute tour mode choice decisions. Transportation, 38(6), 933-958.
40) Pucher, J., & Renne, J. L. (2003). Socioeconomics of urban travel: evidence from the 2001
NHTS. Transportation Quarterly, 57(3), 49-77.
41) Santos, A., McGuckin, N., Nakamoto, H. Y., Gray, D., & Liss, S. (2011). Summary of Travel
Trends: 2009 National Household Travel Survey. Available at http://nhts.ornl.gov/: Federal
Transit Administration, US Department of Transportation.
42) Schneider, R. J. (2013). Theory of routine mode choice decisions: An operational framework
to increase sustainable transportation. Transport Policy, 25, 128-137.
43) Schwanen, T., & Mokhtarian, P. L. (2005). What affects commute mode choice:
neighborhood physical structure or preferences toward neighborhoods?. Journal of Transport
Geography, 13(1), 83-99.
44) Small, K., Winston, C., & Yan, J. (2005). Uncovering the Distribution of Motorists'
Preferences for Travel Time and Reliability. Econometrica, Vol. 73, No. 4, pp. 1367-1382.
45) Small, K.A. (1982). The scheduling of consumer activities: work trips. American Economic
Review, 72, 467–479.
46) Sweet, M. N., & Chen, M. (2011). Does regional travel time unreliability influence mode
choice? Transportation, 38(4), 625-642.
47) Tilahun, N. Y., & Levinson, D. M. (2010). A Moment of Time: Reliability in Route Choice
Using Stated Preference. Journal of Intelligent Transportation Systems: Technology,
Planning, and Operations, 14(3), 179-187.
The Demand for Reliable Travel Chakrabarti (2015)
167
48) Tyrinopoulos, Y., & Antoniou, C. (2008). Public transit user satisfaction: Variability and
policy implications. Transport Policy, 15(4), 260-272.
49) Vredin Johansson, M., Heldt, T., & Johansson, P. (2006). The effects of attitudes and
personality traits on mode choice. Transportation Research Part A: Policy and Practice,
40(6), 507-525.
50) Wachs, M. (1976). Consumer attitudes toward transit service: an interpretive review. Journal
of the American Institute of Planners, 42(1), 96-104.
51) Washbrook, K., Haider, W., & Jaccard, M. (2006). Estimating commuter mode choice: A
discrete choice analysis of the impact of road pricing and parking charges. Transportation,
33(6), 621-639.
52) Zhang, M. (2004). The role of land use in travel mode choice: evidence from Boston and
Hong Kong. Journal of the American Planning Association, 70(3), 344-360.
The Demand for Reliable Travel Chakrabarti (2015)
168
Chapter 6.
Conclusion and Takeaways
This chapter concludes the dissertation. Findings from the empirical studies are summarized.
Broad takeaways for planning scholarship and practice are discussed.
The Demand for Reliable Travel Chakrabarti (2015)
169
My research using data from the Los Angeles region helps advance knowledge in the field of
travel time reliability studies. There is a large body of literature showing that the demand for a
reliable travel environment influences automobile route choice and travel time scheduling.
Survey data shows that travelers are willing to pay significant amounts for improving
predictability of travel or reducing travel time variation across various trip segments and modes.
And transit passengers have been consistently rating unreliability as one of the top
inconveniences associated with transit travel. The value of, and demand for, reliability has been
studied quite extensively. Interestingly, however, the travel time reliability – mode choice
connection remains largely unexplored. Particularly, the effect of transit service (un)reliability on
transit mode choice and/or route selection has not been studied in the past. I address this gap in
literature and provide first empirical evidence on how transit service reliability influences transit
travel demand and consequently affects system consumption and productivity.
This dissertation provides new information to transit managers by demonstrating the
demand for reliable transit service. I show that there is value in delivering service that does not
frequently or randomly depart from plan or schedule, such that the unpredictability of transit
travel is minimized for its consumers. I hope transit managers will realize that unreliability is
costly, and make strategic reliability investments to improve productivity.
In this dissertation, I systematically present evidence that relatively more reliable transit
systems are successful in attracting significantly greater patronage as they are chosen over
competing lines and alternate modes. I also find that discretionary transit travel occurs in
contexts where service reliability is significantly high on average. Reliability investments seem
to be able to facilitate choice of transit (including auto-to-transit mode shifts) and consequently
enhance system productivity. We may expect the reliability effect to be particularly strong during
The Demand for Reliable Travel Chakrabarti (2015)
170
peak periods when travelers are sensitive to unreliability and when congestion and unreliability
of the alternate auto mode are high. Various measures of service reliability are tested, and effects
of reliability increases on transit mode choice and system consumption are estimated. Strategies
that transit managers can adopt to improve reliability are also discussed, along with the role of
transit policy in promoting reliability investments.
Reliability improvement may be a cost-effective means for increasing transit ridership in
the presence of latent demand. But we should not overstate the importance of reliability. There
are several other parameters that govern generalized transit travel cost, and hence determine
choice of transit. My research shows that factors such as speed, service frequency, access, and
number and nature of transfers are highly influential. Other factors such as safety, comfort,
information availability, etc. could be as important. Therefore, quality and performance
improvement along multiple dimensions of service are required to design competitive systems.
We conducted a case study of the new Metro Exposition light rail line (Phase 1) in Los Angeles
as part of the Metro-funded ADMS research project to show that good-quality well-managed rail
investments in dense cores of metro areas along highly congested corridors have the potential to
attract discretionary riders and promote aggregate transit use by providing a viable alternative to
the automobile.
Results presented in this manuscript show that delivering exceptionally high-quality
transit service including highly reliable service is critical for attracting people out of cars.
Travelers can be more tolerant to unreliability of the auto mode because of the many advantages
of private auto travel, such as comfort, convenience and flexibility. And unprecedented advances
in intelligent transportation systems and information technology are making auto travel
increasingly fast, safe, reliable and flexible. Competing with the auto mode is becoming ever
The Demand for Reliable Travel Chakrabarti (2015)
171
more challenging for the bus and the train. While reliable web-based real-time information can
reduce unpredictability associated with transit travel, risks associated with service unreliability
cannot be eliminated using such indirect means, especially in transfer-intensive networks, and
particularly for travelers with few alternatives.
I think it is important that public transit policy incentivizes reliability investments. It is
also critical to make service reliability measurement, monitoring, and reporting mandatory.
Standard measures that capture quality of service from the passengers’ perspective should be
used. The idea of linking government funding availability with reliability levels or targets may
be explored.
I expect that my dissertation research will be useful for transportation planners and
policy makers in congested, auto-reliant cities with well-developed but underutilized and
underproductive transit systems. Findings will also help design new projects and programs aimed
at attracting more discretionary transit ridership. For cities in developing contexts that are
currently experiencing rapid transition from public to private modes of transport, my research
provides ideas to make transit travel more attractive.
Abstract (if available)
Abstract
This dissertation uses data from the greater Los Angeles metropolitan region to demonstrate how our demand for reliable or predictable travel conditions determines, in part, our transportation mode choice. The research extends empirical literature on travel time reliability that has so far analyzed how the demand for reliability affects trip scheduling and automobile route choice. Through a series of independent empirical studies that use different methods, this dissertation investigates how the reliability of public transit, independently and relative to other mobility choices, influences the individual decision to consume transit and in turn aggregate service consumption. Findings suggest that there is significant demand for reliable transit service, and that good-quality transit can attract latent demand in specific situations such as high-demand congested corridors during peak periods. In addition to contributing to the travel reliability literature, the dissertation identifies the conditions that potentially make transit both competitive with the auto mode and more productive as an industry.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
The built environment, tour complexity, and active travel
PDF
The flexible workplace: regional tendencies and daily travel implications
PDF
Spatial and temporal expenditure-pricing equity of rail transit fare policies
PDF
The long-term impact of COVID-19 on commute, employment, housing, and environment in the post-pandemic era
PDF
Congestion pricing with an unpriced time period and with heterogeneous user groups
PDF
Household mobility and neighborhood impacts
PDF
Essays on congestion, agglomeration, and urban spatial structure
PDF
The impact of demographic shifts on automobile travel in the United States: three empirical essays
PDF
Lessons from TAP implementation: obstacles and solutions to improve the transit users experience
PDF
Unraveling decentralization of warehousing and distribution centers: three essays
PDF
Environmental justice in real estate, public services, and policy
PDF
Productive frictions and urbanism in transition: planning lessons from traffic flows and urban street life in Ho Chi Minh City, Vietnam
PDF
Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health
PDF
Active travel, outdoor leisure, and neighborhood environment: path analysis, Los Angeles County
PDF
Population and employment distribution and urban spatial structure: an empirical analysis of metropolitan Beijing, China in the post-reform era
PDF
Location of warehouses and environmental justice: Three essays
PDF
Urban air pollution and environmental justice: three essays
PDF
Children’s travel behavior in journeys to school
PDF
Who learns where: understanding the equity implications of charter school reform in the District of Columbia
PDF
Urban universities' campus expansion projects in the 21st century: a case study of the University of Southern Calfornia's "Village at USC" project and its potential economic and social impacts on...
Asset Metadata
Creator
Chakrabarti, Sandip
(author)
Core Title
The demand for reliable travel: evidence from Los Angeles, and implications for public transit policy
School
School of Policy, Planning and Development
Degree
Doctor of Philosophy
Degree Program
Urban Planning and Development
Publication Date
07/10/2015
Defense Date
05/04/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
OAI-PMH Harvest,public transit policy,ridership,Transportation,travel behavior,travel time reliability
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Giuliano, Genevieve (
committee chair
), Boarnet, Marlon (
committee member
), Moore, James Elliott, II (
committee member
), Schweitzer, Lisa (
committee member
)
Creator Email
chakrabarti.sandip@gmail.com,sandipch@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-591864
Unique identifier
UC11301560
Identifier
etd-Chakrabart-3586.pdf (filename),usctheses-c3-591864 (legacy record id)
Legacy Identifier
etd-Chakrabart-3586.pdf
Dmrecord
591864
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Chakrabarti, Sandip
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
public transit policy
ridership
travel behavior
travel time reliability