Service Reliability Measurement using Oyster Data - A Framework for the London Underground David L. Uniman MIT TfL January 29 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 Probability of Late Arrival T departure T departure T arrival T desired arrival Time of Day 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 ontime at one s destination with 95% certainty RBT = (95 th percentile 5 th percentile) O-D, AM Peak, LUL Period sample size 2 No. of Observations Actual Distribution 5 th perc. 95 th 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 (27) 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) 95th percentile [min] 35 3 25 2 15 1 5 Waterloo to Picc. Circus (Bakerloo NB) - February, AM Peak 4-Feb 7-Feb 1-Feb 13-Feb 16-Feb 19-Feb 22-Feb 25-Feb 28-Feb 2-Mar Weekdays % of journeys Comparison of Travel Time Distributions (normalized).3.25 February 14th.2 February 5th.15.1.5 5 6 7 8 9 1 11 12 13 14 15 16 17 18 19 2 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 Relate to reliability factors and strategies to address them Recurrent Reliability Day-to-day (systematic) performance Includes the effects of service characteristics and other repeatable factors (e.g. demand) Can be considered as the Underground s potential reliability under typical conditions Incident-Related Reliability Unpredictable or Random (unsystematic) delays Unreliability caused by severe disruptions, additional to inherent levels of travel time variation 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. 27 35 95th percentile [min] 3 25 2 15 1 5 4-Feb 7-Feb 1-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.3.3.3.25.2.15.25.2.15 P[No Disruption] = 17/2 days = 85%.25.2.15 P[Disruption] = 3/2 days = 15%.1.1.1.5.5.5 5 1 15 2 25 3 5 1 15 2 25 3 5 1 15 2 25 3 RBT Actual = 4-min RBT Recurrent = 3-min RBT Incident-related = 1-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 6 6 Travel Time [min] 5 4 % Journeys.3.3.25.25.2 5.2.15.15.1.5 1 11 12 13 14 15 16 17 18 19 2 21 22 23 24 25 26 27 28 29 3 31 32 33 34 35 36 37 38 39 4 4 % Journeys.1.5 1 11 12 13 14 15 16 17 18 19 2 21 22 23 24 25 26 27 28 29 3 31 32 33 34 35 36 37 38 39 4 3 Travel Time [min] 3 Travel Time [min] 2 2 1 1 7: AM 7:3 AM 8: AM 8:3 AM 9: AM 9:3 AM 1: AM 7: AM 7:3 AM 8: AM 8:3 AM 9: AM 9:3 AM 1: AM Departure Time Departure Time Indicative Incident Date Cause Code Result NAX's Start Incident End February 7 Fleet Train Withdrawal 2.6124 7:2am 8:2am 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 9.883 8:3am 8:6am February 14 Signals Train Delay 47.843 9:5am 9:16am Partial Line February 14 Signals Suspension 45.473 9:57am 11:19am 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. ERBT = (RBT Actual RBT Recurrent ) O-D, AM Peak, LUL Period sample size 2, cumulative baseline Num. of Observations Recurrent Distribution (II) Actual Distribution (I) Excess Unreliable Journeys 5 th perc. (I) 95 th perc.(ii) 95 th perc.(i) Travel Time RBT (II) 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. 27 2. Travel Time [min] 18. E_RBT 16. 14. B_RBT 12. 1. (8.55) 8. (7.1) 6. 3.62 2.8 4. 2.. 4.93 4.93 16.71 February November Median Travel Time Period 1
Application # 1 JTM Reliability Addition Actual W eighted RBT value estimation Contribution to Service Quality (i.e. Perceived Performance) Travel Time [min] 2. 18. 16. 14. 12. 1. 8. 6. 4. 2.. Comparison of Actual Reliability Buffer Time and Median Journey Time: Victoria Line, AM Peak, Feb/Nov 27 8.55 7.1 16.71 February November Median Travel Time Compare contribution of RBT to other JTM components through VOT FEB. 27 Unweighted Unweighted, JTM Weighted, JTM (VOT RBT = 1.) Proportions* Proportions (VOT RBT = 1.) (VOT RBT =.6) 21% 16% 9% 34% 34% 1% AEI 35% 66% * total 11% due to rounding 5th Perc. RBT 1% 8% PWT OTT CLRS RBT 36% 12% 12% 37% 11 7% AEI PWT AEI OTT PWT OTT CLRS CLRS B_RBT RBT E_RBT
Application # 2 Journey Planner Reliability Addition Better information reduces uncertainty by closing the gap between expectations and reality improve reliability of service 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 =.2 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.9.8.7.6.5.4.3.2.1 Journey Planner Oyster Median 1 15 2 25 3 35 4 45 5 55 6 O yster Travel Time [min] Oyster 95th perc. 12 Courtesy of Transport for London. Used with permission.
Application # 2 Journey Planner Reliability Addition Assessment: Journey Planner information is INCOMPLETE in 2 ways: 1. Journey Planner consistently underpredicts Oyster journey times - possibly AET & PWT leaves around 3-5% of journey to chance 2. Expected journey times not helpful for passengers concerned with on-time arrival (e.g. commuters) Difference in Travel Times - yster Data and Journey Planner: Difference in Travel Times - Data and Planner: 5 Largest O-D's, O yster Journey AM Peak, Nov. 27 5 Largest O-D's, AM Peak, Nov. 27 's D - - O 5. of ota o (t N ) f l 2 3 5 45 18 16 25 4 14 2 35 12 3 1 15 25 28 156 1 14 5 52 % 1 1-2 1% 23 1-42% 3 5 2-3% 4 6 75 3-4%8 6 9 4-5%1 7 (OystePrrMobeadbian (95th ilit Perc. Ty roav f OelnT-iTmimee- Travel Time AJorruivrnael / Journey uy Psinlang njeorutrnrav Planner eyepl Time) lainmne)r T [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 8:27 8:57 9:11 :3 (:41) View Arrive by Æ 2 8:19 8:49 9: :3 (: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 21 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.