2. 2
Motivation
• Transit ridership is decreasing in most
American cities. In New York, bus ridership
went down 5.5% from 2016 to 2017
697
671
668
678
667
651
638
603
2010 2011 2012 2013 2014 2015 2016 2017
NYCT Annual Bus Ridership (millions)
3. • Studies in NYC have explained higher bus
ridership with improvements in bus service:
◦ Real-time information
◦ Reliability and on-time performance
◦ Select Bus Service
• But service is not improving: average bus
speeds went down 4.5% in 2017-2018
• And these studies don’t consider attitudes or
perceptions of service.
• Where are those trips going and why are
they opting out of the bus?
3
Motivation
5. Introduction
• MTA Average weekday bus route ridership
decreased in 2016-2017
◦ 87% of bus routes have with less demand
• Only 25% of bus routes in NYC have an
acceptable service level. (Transit Center / Bus
Turnaround Coalition)
• The project:
◦ Conduct a survey of known bus riders asking
about satisfaction with different attributes of
bus service
◦ Include a stated choice experiment with
changed attributes of bus service
◦ Identify factors
◦ Estimate a DCM considering perception
attributes
5
6. Proussaloglou, Koppelman
(1989)
Attitudes of commuter rail riders were used as explanatory variables in a
mode choice model to determine contribution of perception and service
improvements
Kuppam, Pendyala,
Rahman (1999)
Perception of service as attributes has a higher contribution for explaining
mode choice than only sociodemographic variables
Popuri, Proussaloglou,
Ayvalik, Koppelman, Lee
(2011)
Attitudinal factors related to daily travel improved goodness-of-fit of mode
choice model in the Chicago Area. The study also highlights the importance
of publicity for communicating transit service.
Transportation Research
Board (2013)
Guidebook to measure user satisfaction and service quality of transit
systems
6
Literature review
7. Wan et. al (2015) Survey of SBS route riders reflected an improvement in the
perception of service quality and identified actions to improve
traveler satisfaction
Brakewood et. al (2015) Conducted a survey after SBS implementation and concluded that
an increase of up to 100 trips on average weekday could be
attributed to availability of real-time information.
Tyndall et al (2018) Review of real-time bus location data concluded that ridership could
increase when bus frequency, reliability and speeds were increased
Hu et. Al (2015) Estimated a perception-based mode choice model in Nanjing, China
evaluating traveler attributes and bus service perception and
determined which factors can contribute in a higher ridership
7
Literature review
8. 8
Bus routes
• At least two bus routes with varying
attributes that have experienced a significant
decrease in demand
• Not substitutable with subway
Route Type Δ 2016-2017
M86 SBS
-4499
(-18%)
Cross-town
B44
Local /
SBS
-2541
(-6.8%)
20% shared with
subway
B46
Local /
SBS
-1677
(-3.9%)
25% shared with
subway
9. • Respondents need to be recent bus riders
• Short length (2-3 min) to ensure high
response rate
• Paper survey on field (in vehicles and in bus
stops)
• Online survey to minimize response time and
incentivize reponses
9
Data collection
10. General attributes
Characteristics of the bus trip
Satisfaction of bus service
Mode shift intention
• Age
• Gender
• Bus route
• Location (At bus / At stop)
10
Data collection
11. • Trip purpose
• Frequency of taking the bus
• Availability of other modes
• Distance to bus stop
• Waiting time at bus stop
• Trip length
• Transfer at end of trip
11
Data collection
General attributes
Characteristics of the bus trip
Satisfaction of bus service
Mode shift intention
12. • Overall satisfaction
• Availability
◦ Distance to bus stop
◦ Convenience of payment
◦ Number of bus stops in bus route
• Reliability
◦ Frequency of service
◦ On-time performance
◦ Boarding time
◦ Travel speed
• Comfort
◦ Cleanliness of vehicles
◦ Comfort inside vehicles
12
Data collection
General attributes
Characteristics of the bus trip
Satisfaction of bus service
Mode shift intention
13. • Stated preference choice (taking the bus /
switching to another mode) under three
scenarios:
◦ “Your bus stop gets removed: you must walk (5-
12) more minutes to a stop”
◦ “Bus frequency decreases: you must wait (5-12)
more minutes at the stop”
◦ “Travel speed decreases: you travel (5-12) more
minutes in the bus”
13
Data collection
General attributes
Characteristics of the bus trip
Satisfaction of bus service
Mode shift intention
14. 14
Sample
• Sample (Brakewood, MacFarlane and
Watkins, 2015):
𝑠𝑠 = 𝑝𝑜𝑝 1 +
𝑝𝑜𝑝 − 1 𝑑2
𝑝 1 − 𝑝 𝑧 𝛼/2
2
Route Demand
Representative
Sample
M86 20247
382B44 34877
B46 41876
Personal attributes
Observations
(Total sample = 57)
Percentage
Gender
Male 30 53%
Female 27 47%
Age
16 – 25 13 23%
26 – 35 34 60%
36 – 55 10 17%
Trip purpose
Work 29 51%
School 21 37%
Entertainment 0 0
Errands 4 7%
Visiting 3 5%
Frequency
Less than 1 day 5 9%
1-2 days per week 6 11%
3-4 days per week 11 19%
5+ days per week 35 61%
15. 15
Sample
Personal attributes
Observations
(Total sample = 57)
Percentage
Gender
Male 30 53%
Female 27 47%
Age
16 – 25 34 60%
26 – 35 13 23%
36 – 55 10 17%
Trip purpose
Work 29 51%
School 21 37%
Entertainment 0 0
Errands 4 7%
Visiting 3 5%
Frequency
Less than 1 day 5 9%
1-2 days per week 6 11%
3-4 days per week 11 19%
5+ days per week 35 61%
0% 20% 40% 60% 80% 100%
B46
B44
M86
Dissatisfied Neutral Satisfied Very Satisfied
16. • Exploratory factor analysis is used to estimate
a model of proper latent (unobserved
variables) from a series of service attributes.
𝐹𝑖 =
𝑛=1
𝑁
𝑥 𝑛 × 𝑐 𝑛𝑖
16
Factor analysis
Reliability
Convenience
Comfort
Frequency
On-time
Boarding
Distance
Payment
Speed
Bus stops
Cleanliness
Comfort
Attribute F1 F2 F3
Distance
Frequency 0.995
On-time 0.503
Payment
Boarding 0.529
Cleanliness 0.709
Comfort 0.937
Speed 0.453
Stops 0.987
RMSR obtained: 0.04
17. • Mixed logit model from panel dataset
◦ 3 observations for each individual
◦ Compute trip length and evaluate significance in
model
• Binary choice
◦ Travel by bus
◦ Switch mode
• Latent factors as attributes for mode shift
intention
17
Mixed logit model
Travel by bus
Travel by other
mode
Your bus stop gets removed: you must walk 5
more minutes to a stop
( )
( ) Walk
( ) Subway
( ) Private Car
( ) Taxi/Uber/Lyft
( ) Bicycle/Citibike
( ) Other: _______
Bus frequency decreases: you must wait 5 more
minutes at the stop
( )
( ) Walk
( ) Subway
( ) Private Car
( ) Taxi/Uber/Lyft
( ) Bicycle/Citibike
( ) Other: _______
Travel speed decreases: you travel 5 more
minutes in the bus
( )
( ) Walk
( ) Subway
( ) Private Car
( ) Taxi/Uber/Lyft
( ) Bicycle/Citibike
( ) Other: _______
18. • Bus Utility:
𝑉𝑏𝑢𝑠 = 𝛽 𝑛 𝑥 𝑛
• High significance of reliability factor and
positive coefficient in the utility function
◦ Higher reliability decreases chance of switching
from the bus
• Travel time highly significant
◦ Walking time has the most impact when
choosing to travel by bus
18
Mixed logit model
β S Pr (>|z|)
Intercept -6.512** 2.294 0.0045
Travel time -0.153*** 0.0453 0.0007
Reliability 0.590* 0.258 0.0223
Convenience -
Obligatory
1.229
0.893 0.1689
SD (Travel
time)
0.141* 0.057 0.0146
19. • More significant discrete choice model when including perception variables (latent factors of
bus service)
• Reliability factor is the most significant when choosing to travel by bus
◦ Frequency of service
◦ On-time performance
◦ Boarding time
• Positive impact in utility of bus alternative: higher perception of bus service reliability implies
less likelihood of switching from the bus
• Efforts should be prioritized in improving reliability
◦ Traffic Signal Prioritizing for bus
◦ Minimize bunching of buses to minimize boarding time
• Higher walking times have most negative impact in mode shift intention
◦ Optimize bus stops to minimize walking time without compromising bus speeds
19
Conclusions
Editor's Notes
P is a decimal value expressing probability
D confidence level
Z-value for confidence level
P is a decimal value expressing probability
D confidence level
Z-value for confidence level