Service Reliability Measurement using Oyster Data
- A Framework for the London Underground
David L. Uniman
MIT – TfL
January 2009
1
Introduction
• Research Objective
To develop a framework for quantifying reliability from the perspective of
passengers using Oyster data that is useful for improving service quality on the
Underground.
• How reliable is the Underground?
- how do we think about reliability?
- how do we quantify it?
- how do we understand its causes?
- how do we improve it?
2
Background – What is Service Reliability?
• Reliability means the degree of predictability of the service attributes
including comfort, safety, and especially travel times.
ƒ Passengers are concerned with average travel times, but also with certainty of
on-time arrival
% of Trips
Probability
of Late
Arrival
T’departure Tdeparture
T’arrival
Tdesired arrival
Time
of
Day
Avg. Travel Time
“Buffer”
- Adapted from Abkowitz (1978) 3
Avg. Travel Time
Framework - Reliability Buffer Time Metric
• Criteria for Reliability Measure
• Representative of passenger experience
• Straightforward to estimate and interpret
• Usefulness and applicability – compatible with JTM
• Propose the following measure: Reliability Buffer Time (RBT) Metric
“The amount of time above the typical duration of a journey required to arrive ontime at one’s destination with 95% certainty”
sample size ≥ 20
RBT = (95th percentile – 50th percentile)O-D, AM Peak, LUL Period
No. of
Observations
• Actual Distribution
50th perc.
95th perc.
RBT
Travel Time
4
Framework – Separating Causes of Unreliability
•
Two types of factors that influence reliability and affect the applicability
& usefulness of the measure:
1. Chan (2007) found evidence for the effects of service characteristics on
travel time variability – impact on aggregation (e.g. Line Level measure)
2. In this study, observed that reliability was sensitive to the performance of a
few (3-4) days each Period, which showed large and non-recurring delays
(believe Incident-related)
Waterloo to Picc. Circus (Bakerloo NB) - February, AM Peak
Comparison of Travel Time Distributions (normalized)
0.3
30
0.25
25
% of journeys
95th percentile [min]
35
20
15
10
5
February 5th
0.15
0.1
0.05
0
4-Feb
February 14th
0.2
7-Feb
10-Feb
13-Feb
16-Feb
19-Feb
Weekdays
22-Feb
25-Feb
28-Feb
2-Mar
0
5
6
7
8
9
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Travel Time [min]
5
Framework – Classification of Performance
• Propose to classify performance into two categories along two
dimensions – degree of recurrence and magnitude of delays
ƒ
Relate to reliability factors and strategies to address them
Recurrent Reliability
Incident-Related Reliability
Day-to-day (systematic)
performance
Unpredictable or Random
(unsystematic) delays
Includes the effects of service
characteristics and other
repeatable factors (e.g. demand)
Unreliability caused by severe
disruptions, additional to inherent
levels of travel time variation
Can be considered as the
Underground’s potential reliability
under “typical” conditions
Together with performance under
“typical” conditions, makes up the
actual passenger experience
• Methodology – use a classification approach based on a Stepwise
Regression to automate process
6
Framework – Classification of Performance
9 5 th percenti l e [m i n]
− Bakerloo Line Example: Waterloo to Piccadilly Circus – AM Peak, Feb.
2007
35
30
25
20
15
10
5
0
4-Feb
7-Feb
10-Feb
13-Feb
16-Feb
19-Feb
22-Feb
25-Feb
28-Feb
2-Mar
Weekdays
T.T.Actual = P[No Disruption]*T.T.Recurrent + P[Disruption]*T.T.Incident-related
0.3
0.3
0.3
0.25
0.25
0.2
0.2
0.15
0.15
0.1
0.1
0.1
0.05
0.05
0.05
0
P[No Disruption]
= 17/20 days
= 85%
0
0
5
10
15
20
25
RBTActual = 4-min
30
0
5
10
15
20
25
RBTRecurrent = 3-min
30
P[Disruption]
= 3/20 days
= 15%
0.25
0.2
0.15
0
0
5
10
15
20
25
30
RBTIncident-related = 10-min
7
Framework – Validation with Incident Log data
• Validation of non-recurrent performance with Incident Log data (from
NACHs system) confirmed Incident-related disruptions as the primary cause
Brixton to Oxford Circus (Victoria NB) - February 7, AM Peak
Brixton to Oxford Circus (Victoria NB) - February 14 - AM Peak
60
60
0.3
0.3
50
0.2
0.15
0.1
40
0.05
0.25
% Journeys
40
% Journeys
T ravel Tim e [m in]
50
0.25
0.2
0.15
0.1
0.05
0
0
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
30
30
Travel Time [min]
20
20
10
10
Travel Tim e [m in]
0
0
7:00 AM
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
7:30 AM
8:00 AM
8:30 AM
9:00 AM
9:30 AM
10:00 AM
7:00 AM
7:30 AM
8:00 AM
8:30 AM
9:00 AM
9:30 AM
10:00 AM
D ep art ure T ime
Departure Time
Indicative
NAX's
Incident
Start
Incident End
Train Withdrawal
2.6124
7:02am
8:20am
Train Delay
3.5756
9:16am
9:19am
Indicative
NAX's
Incident
Start
Incident
End
Train Delay
9.8803
8:03am
8:06am
Signals
Train Delay
47.8043
9:05am
9:16am
Signals
Partial Line
Suspension
45.4703
9:57am
11:19am
Date
Cause Code
Result
February 7
Fleet
February 7
Customer - PEA
Date
Cause Code
Result
February 14
Customer
PEA
February 14
February 14
8
Excess Reliability Buffer Time Metric
• Using these 2 performance categories, we can extend our reliability
measure by comparing the actual performance with a baseline value
• Propose the following: Excess Reliability Buffer Time (ERBT) Metric
“The amount of buffer time that passengers need to allow for to arrive on-time
with 95% certainty, in excess of what it would have been under disruption-free
conditions.”
sample size ≥ 20,
ERBT = (RBTActual – RBTRecurrent )O-D, AM Peak, LUL Period
cumulative baseline
Num. of
Observations
• Recurrent Distribution
(II)
• Actual Distribution (I)
Excess
Unreliable
Journeys
50th perc. (I)
RBT (II)
95th perc.(II)
95th perc.(I)
Travel Time
ERBT
9
Application # 1 – JTM Reliability Addition
• Use measure for routine monitoring and evaluation of service quality –
propose to include it within JTM as an additional component.
• RBT form is compatible is JTM – units (min, pax-min), aggregation (Period,
AM Peak, O-D), estimation (Actual, Scheduled, Excess & Weighted)
TPT | AEI | PWT | IVTT | C&I | RBT
•Apply RBT measure to Victoria Line – AM Peak, Feb. & Nov. 2007
20.00
18.00
Travel Time [min]
16.00
14.00
E_RBT
B_RBT
12.00
10.00
8.00
6.00
(8.55)
(7.01)
3. 62
16. 71
2. 08
4.00
2.00
4. 93
4. 93
February
November
0.00
Period
Median Travel Time
10
Application # 1 – JTM Reliability Addition
• ActualW eighted RBT value estimation
• Contribution to Service Quality (i.e. Perceived Performance)
Comparison of Actual Reliability Buffer Time and Median Journey Time:
Victoria Line, AM Peak, Feb/Nov 2007
20.00
Travel Time [min]
18.00
16.00
16.71
14.00
12.00
10.00
8.00
6.00
8.55
7.01
4.00
2.00
0.00
February
November
Median Travel Time
• Compare contribution of RBT to other JTM components through VOT – FEB. 2007
Unweighted
(VOTRBT = 1.0)
Unweighted, JTM
Proportions*
Weighted, JTM
Proportions
(VOTRBT = 1.0)
(VOTRBT = 0.6)
16%
21%
50th Perc.
RBT
66%
AEI
9%
34%
34%
7%
8%
AEI
PWT
OTT
CLRS
RBT
1%
1%
35%
35%
36%
36%
1%
* total 101% due to rounding
AEI
OTT
PWT
CLRS
OTT
CLRS
B_RBT
RBT
E_RBT
12%
12%
37%
PWT
11
Application # 2 – Journey Planner Reliability Addition
• Better information reduces uncertainty by closing the gap between
expectations and reality – improve reliability of service
¾ Propose more COMPLETE information through Journey Planner
SIMPLE EXAMPLE: David’s morning commute – Bow Road to St. James’ Park
− Probability of arrival on-time using Journey Planner = 0.02
− Access/Egress and Wait Times (?) … must “guess” 17% of trip
− Journey Planner time must be increased by a FACTOR of 1.64 to be 95% reliable
1
Cumulative Probability
0.9
0.8
0.7
0.6
0.5
0.4
Journey
Planner
0.3
0.2
O yster
Median
O yster
95th perc.
0.1
0
10
15
20
25
30
35
40
45
50
55
60
O yster Travel Time [min]
Courtesy of Transport for London. Used with permission.
12
Application # 2 – Journey Planner Reliability Addition
• Assessment: Journey Planner information is INCOMPLETE in 2 ways:
1. Journey Planner consistently underpredicts Oyster journey times - possibly
1.
AET & PWT – leaves around 30-50% of journey to chance
2. Expected journey times not helpful for passengers concerned with on-time
2.
arrival (e.g. commuters)
N o. of
of O - D 's
(tota l 5 0 )
Difference in Travel Times - O yster Data and Journey Planner:
Difference in Travel Times - OO yster Data and Journey Planner:
50 Largest O -D's, AM Peak, Nov. 2007
50 Largest O-D's, AM Peak, Nov. 2007
20
50
30
45
18
25
16
40
14
35
20
12
30
10
25
15
20
8
10
15
6
10
4
5
2
5
0
0 00%
1
1
02 10%
23
3
10-420%
5
4 6
20-30%
75 30-40%
8 6
710
9 40-50%
Travel
Time
/ rn
Planner
(O ysteP
rrM
o(95th
beadbian
ilitPerc.
yTroav
fO
elnT
-iTm
im
ee
- AJo
rruiv
aJourney
l yusPin
e
l an
gn
Je
orurTnrav
eyeTime)
Pllai
Tnm
nee)r [min]
13
Application # 2 – Journey Planner Reliability Addition
• Possible representation of new journey information:
SIMPLE EXAMPLE: David’s morning commute – Bow Road to St. James’ Park
Chose to:
Route
Depart
Expected Arrival Latest Arrival Duration (up to)
Interchanges
Depart at… Æ
1
08:27
08:57
09:11
00:30 (00:41)
View
Arrive by… Æ
2
08:19
08:49
09:00
00:30 (00:41)
View
14
Courtesy of Transport for London. Used with permission.
Conclusions & Recommendations
1. Reliability is an important part of service quality, relative to average
performance, and should be accounted for explicitly.
9 Monitor and evaluate reliability through JTM Extension
2. Incidents have a large impact on service quality through unreliability,
which may be underestimated through focus on average performance.
9 Use Oyster and Reliability Framework to improve monitoring and
understanding through NACHs (measurement vs. estimate)
3. The impacts of unreliability on passengers can be mitigated through
better information.
9 Update travel information and include reliability alternative in Journey
Planner
15
Conclusions & Recommendations
4. In order to manage performance, we need to be able to measure it first.
9 Framework contributes to making this possible, and sheds light on some
of the broader questions…
• How reliable is the Underground?
- how do we think about reliability?
- how do we quantify it?
- how do we understand its causes?
- how do we improve it?
16
Thank You
• Special thanks to people at TfL and LUL that made this research possible
and a memorable experience
17
MIT OpenCourseWare
http://ocw.mit.edu
1.258J / 11.541J / ESD.226J Public Transportation Systems
Spring 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.