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.

� Passengers are concerned with average travel times, but also with certainty of on-time arrival

% of Trips

T’arrival TdepartureT’departure

Time of DayTdesired arrival

Probability of Late Arrival

Avg. Travel Time “Buffer” Avg. Travel Time - Adapted from Abkowitz (1978) 3

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 on-time at one’s destination with 95% certainty”

RBT = (95th percentile – 50th percentile)O-D, AM Peak, LUL Period sample size ≥ 20

No. of • Actual Distribution Observations

50th perc. 95th perc. Travel Time

RBT 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)

35 0.3 30

0.25 25

95th

per

cent

ile [m

in]

% o

f jou

rney

sFebruary 14th

0.2 20

15

10

February 5th

0.15

5 0.05

0 0

4-Feb 7-Feb 10-Feb 13-Feb 16-Feb 19-Feb 22-Feb 25-Feb 28-Feb 2-Mar 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Weekdays Travel Time [min]

0.1

5

Framework – Classification of Performance

• Propose to classify performance into two categories along two dimensions – degree of recurrence and magnitude of delays � 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

− Bakerloo Line Example: Waterloo to Piccadilly Circus – AM Peak, Feb. 2007

35 30

25 20 15

95th

per

cent

ile

[min

] 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

0.25

0.2

0.15

0.1

0.05

0

0.3

0.25

0.2

0.15

0.1

0.05

0

P[No Disruption] = 17/20 days = 85%

0.3

0.25

0.2

0.15

0.1

0.05

0

P[Disruption] = 3/20 days = 15%

0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

RBTActual = 4-min RBTRecurrent = 3-min RBTIncident-related = 10-min 7

40

50

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 0.25

0.25

% J

ourn

eys 0.2 50

% J

ourn

eys 0.2

0.15

0.1

0.15

0.1

400.05 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 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 Travel Time [min] 30 Travel Time [min]

20 20

10 10

0 0

7:00 AM 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 Departure Time D epart ure T ime

Indicative Incident Date Cause Code Result NAX's Start Incident End

February 7 Fleet Train Withdrawal 2.6124 7:02am 8:20am

February 7 Customer - PEA Train Delay 3.5756 9:16am 9:19am

Indicative Incident Incident Date Cause Code Result NAX's Start End

Customer -February 14 PEA Train Delay 9.8803 8:03am 8:06am

February 14 Signals Train Delay 47.8043 9:05am 9:16am

Partial Line February 14 Signals Suspension 45.4703 9:57am 11:19am

8

Trav

el T

ime

[min

]

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.”

ERBT = (RBTActual – RBTRecurrent )O-D, AM Peak, LUL Period sample size ≥ 20, cumulative baseline

50th perc. (I) 95th perc.(II)

RBT (II)

Excess Unreliable Journeys

Num. of Observations

Travel Time

• Recurrent Distribution (II) • Actual Distribution (I)

95th perc.(I)

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 E_RBT 16.00

B_RBT 14.00

12.00

10.00 (7.01) 8.00

3.62 6.00 2.08 4.00

2.00

0.00

4.93 4.93

16.71

(8.55)

February November Median Travel Time

Period 10

Trav

el T

ime

[min

]

Application # 1 – JTM Reliability Addition

• Actual W 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

16.71

7.01 8.55

0.00

2.00 4.00

6.00 8.00

10.00

12.00 14.00

16.00 18.00

20.00

February November Median Travel Time

Trav

el T

ime

[min

]

• Compare contribution of RBT to other JTM components through VOT – FEB. 2007

Unweighted Unweighted, JTM Weighted, JTM (VOTRBT = 1.0) Proportions* Proportions

(VOTRBT = 1.0) (VOTRBT = 0.6)7%

AEI 35%

21% 16%9%

1%1%34% 34% 35% 50th Perc. PWT

OTT8% CLRS RBT

RBT

66% 36%36%1% 12%

AEI PWT OTT CLRS RBT

AEI

PWT

OTT

CLRS

B_RBT

E_RBT

12% 37% 11

* total 101% due to rounding

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

12

− 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

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

10 15 20 25 30 35 40 45 50 55 60

O yster Travel Time [min]

Cum

ulat

ive

Prob

abili

ty

Journey Planner

O yster Median

O yster 95th perc.

Courtesy of Transport for London. Used with permission.

• Assessment: Journey Planner information is INCOMPLETE in 2 ways:1. Journey Planner consistently underpredicts Oyster journey times - possibly

AET & PWT – leaves around 30-50% of journey to chance

2. Expected journey times not helpful for passengers concerned with on-time arrival (e.g. commuters)

Assessment: Journey Planner information is INCOMPLETE in 2 ways: 1. Journey Planner consistently underpredicts Oyster journey times - possibly

AET & PWT – leaves around 30-50% of journey to chance

2. Expected journey times not helpful for passengers concerned with on-time arrival (e.g. commuters)

Difference in Travel Times - O yster Data and Journey Planner:Difference in Travel Times - O yster Data and Journey Planner:50 Largest O-D's, AM Peak, Nov. 200750 Largest O-D's, AM Peak, Nov. 2007

205045181640

ss'' ) 1435

DD )-- 00 1230

OO 5 5llf a 1025

. o

. of attoo

oo (t(t 208

NN 1561045200

0 0% 1 0-2 10% 3 10-420% 5 20-30%6 7 30-40%8 9 40-50%10

(O ystePrrMobeadbianilit Ty roavf OelnT-iTmimee- AJorruivrnael uy Psinlang nJeoru TrnraveyePl Tlainmnee)r [min]

Difference in Travel Times - yster Data and Journey Planner: O 50 Largest O-D's, AM Peak, Nov. 2007

30

25

s' ) D 20

- 0 O

5

f l 15

. o ato

o (tN 10

5

0 0 1 2 3 4 5 6 7

(95th Perc. Travel Time / Journey Planner Time)

Application # 2 – Journey Planner Reliability Addition

•

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:

Depart at… Æ

Arrive by… Æ

Route Depart Expected Arrival Latest Arrival Duration (up to) Interchanges

1

2

08:27

08:19

08:57

08:49 09:00

09:11 00:30 (00:41)

00:30 (00:41)

View

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 OpenCourseWarehttp://ocw.mit.edu

1.258J / 11.541J / ESD.226J Public Transportation SystemsSpring 2010

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.