Brian Levine

Measurement of Subway Service Performance at New York City Transit: Case Study with Automated Train Supervision Track-Occupancy Data

A recurring challenge that faces transit managers today is the persistent question of how to do more with less: to maintain and to improve service despite deficits of historic proportions. New York City Transit (NYCT) responded by retooling performance measurement frameworks and procedures to capture the customers perspective better, to respond to management initiatives, and to incentivize proper operating decisions. NYCTs primary performance measure, Wait Assessment (WA), measured customer maximum wait times to board at stations. A reach and match algorithm was developed; this algorithm was defined as the percentage of headways between trains that did not exceed 125% of scheduled headways. The purpose of the algorithm was to account for NYCTs irregularly scheduled service and to ensure that the way customers experienced headways matched the specific, published scheduled headway in effect at that moment, regardless of which scheduled trip was supposed to arrive. Sample-based methods that gathered limited data manually were upgraded, and track-occupancy data were downloaded from the automated train supervision system for the Number 1 through Number 6 routes. This action provided 100% coverage, which resulted in much lower public reporting of time lags and in the ability to take near-term corrective action. The increase in data availability also allowed NYCT to provide better service through improved consideration of corridor-level and track-level WA standards for internal diagnostic purposes and analysis of train performance in shared-track territory, regardless of route designation.

Real-Time Estimation of Platform Crowding for New York City Subway

Amid significant increases in ridership (9.8% over the past 5 years) on the more than 100 year-old New York City Transit (NYCT) subway system, NYCT has become aware of increased crowding on station platforms. Because of limited platform capacity, platforms become crowded even during minor service disruptions. A real-time model was developed to estimate crowding conditions and to predict crowding for 15 min into the future. The algorithm combined historical automated fare collection data on passenger entry used to forecast station entrance, automated fare collection origin–destination inference information used to assign incoming passengers to a particular direction and line by time of day, and general transit feed specification–real time data to determine predicted train arrival times used to assign passengers on the platform to an incoming train. This model was piloted at the Wall Street Station on the No. 2 and No. 3 Lines in New York City’s Financial District, which serves an average 28,000 weekday riders, and validated with extensive field checks. A dashboard was developed to display this information graphically and visually in real time. On the basis of predictions of gaps in service and, consequently, high levels of crowding, dispatchers at NYCT’s Rail Control Center can alter service by holding a train or skipping several stops to alleviate any crowding conditions and provide safe and reliable service in these situations.

Time-Expanded Network Model of Train-Level Subway Ridership Flows Using Actual Train Movement Data

Subway ridership estimates are important to transit operators for both internal applications (e.g., setting service frequencies, prioritizing station upgrades) and external reporting (e.g., to the National Transit Database). New York City Transit (NYCT) is developing a new model that will accomplish three primary objectives: (a) estimating subway ridership at a train level for the first time, (b) basing path choice on actual train movements rather than on schedules so that uneven loadings can be accurately captured, and (c) running fast enough to be used daily and being sufficiently automated to run with minimal human intervention. The model integrates entry data from fare cards with actual train movement data from a wide range of electronic systems and schedules. The model assigns riders to trains by using a Frank–Wolfe approach, including Dijkstra’s algorithm for shortest paths, with customizations designed for transit. These customizations improve speed, enable the algorithm to model delays better, and allow for multiple types of riders with different preferences for transfers and crowding. The size and the complexity of the NYCT system make for a challenging test case computationally. Approximately 6 million trips are made on a busy weekday, and these are assigned to a time-expanded network containing more than 3 million nodes and 7 million arcs. The model is automated and runs fast enough that it can be used daily. Validation against manual counts indicates strong results, with the R2 for max load point volumes for the morning peak hour equal to .91.

Observed Customer Seating and Standing Behavior and Seat Preferences on Board Subway Cars in New York City

An observational sampling methodology was used to explore seat occupancy patterns in New York City subway cars. The study was performed under uncrowded conditions on the basis of special attributes of what otherwise were highly homogeneous plastic bench seats. Onboard seating patterns, measured as relative seat occupancy probabilities, were explained in terms of interactions between railcar design, layout, customer preferences, and resulting behavior. Earlier research focused in general on passenger distribution between cars within long trains or on the desirability of attributes common to all seats, rather than on passenger seating patterns within a single car. Results of the study reported here had their basis in seating- and standing-room occupancy statistics and showed that customers had a clear preference for seats adjacent to doors, no real preference for seats adjacent to support stanchions, and disdain for bench spots between two other seats. On cars that featured transverse seating, customers preferred window seats, but their preference was almost equal for backward- or forward-facing seats. No gender bias in all seated passengers was detected, but as load factor increased, the chance of standing was higher for men than for women. Use of 90% of the seats was achieved only at a 120% load factor. Customers who stood strongly preferred to crowd vestibule areas between doors (particularly in cars with symmetric door arrangements) and to hold on to vertical poles. These findings were consistent with published anecdotes. In future, cars should be designed with asymmetric doors, 2 1 2 1 2 partitioned, longitudinal seats, and no stanchions or partitions near doorways. To understand customer seating preferences further, research should be conducted in commuter rail vehicles with suburban layouts and booth seating and in the subways of other cities.

Transforming bus service planning using integrated electronic data sources at NYC transit

The installation of an Automatic Vehicle Location (AVL) system alongside existing Automated Fare Collection (AFC) data spurred development of an inferred bus boarding and alighting ridership model at New York City Transit (NYCT), allowing for 100% passenger origin-destination (O-D) data citywide. Analysis techniques that relied primarily on professional judgment due to lack of data were replaced by more sophisticated statistical techniques. This paper describes two case studies and the resulting service planning potential from having access to fully-integrated big data sources: a neighborhood-wide analysis of performance and ridership, where 100% data allowed planners to pinpoint specific, low-cost reroutes and stop changes to better serve riders, and identification of an optimal route split location for a long route with poor performance by minimizing passenger impact using modeled O-D data. In both examples, new data sources allowed for novel analysis throughout problem investigation as well as forecasting ridership and cost impacts of proposed service adjustments. As the agency’s ability to leverage these data improves, it will support Title VI obligations as well as performance monitoring.

Development of Application for Estimating Daily Boarding and Alighting Counts on New York City Buses: Implementation of Daily Production System

To support bus service scheduling and planning, New York City Transit (NYCT) put into production a ridership application to determine surface transit boarding and alighting locations for each of approximately 2.8 million daily passenger trips on 218 bus routes. The application combined data from an automated vehicle location (AVL) system, a multimodal entry-only nongeographic automated fare collection (AFC) system, and general transit feed specification schedule file streams. To accomplish this objective, NYCT developed a highly optimized network-generation tool to estimate bus link loads and boarding and alighting locations by creating a scaled-down custom network based on first-bus trajectory (from AVL boarding location data) and a few possible AFC- or AVL-inferred second-leg pickup stops. Solving for the shortest walking path on this subnetwork yielded connection and alighting points more efficiently than did solving for all 128 million potential origin–destination (O-D) pairs systemwide. The program executes in less than 3 h in an automated environment that can support the operations management need for next-day reporting and the monitoring of patterns and trends in ridership behavior. The program also allows the aggregation of multiple days of route-level program output for schedule-making purposes and thus provides a significantly more representative understanding of passenger loads than were obtained historically from a few labor-intensive onboard observations (ridechecks) collected over a multiple-year period. Results were validated and found to be consistent with manual ridechecks, limited O-D surveys, and other sources. The accuracy obtained was sufficient to achieve acceptance of AVL-AFC data in lieu of traditional onboard observations for NYCT schedule making.

Designing New York City Subways’ Key Performance Indicators to Improve Service Delivery and Operations

A balanced scorecard (BSC) is widely used in private industry and the public sector to monitor key performance indicators (KPIs) and to help achieve strategic outcomes. This concept is widely used in the transit industry for carrier–regulator contractual relationships and performance monitoring. After a fact-finding mission to Southeast Asia, New York City Transit (NYCT) adopted KPIs for continuous improvement of service delivery performance. Subway line-level KPIs based on BSC concepts were introduced in conjunction with a line general manager program and numerous initiatives for incremental performance management. After a reorganization that re-created functional departments—car equipment, stations, and rapid transit operations—BSC was applied at departmental levels and resulted in maintenance-oriented passenger environment survey (PES)-KPIs and operations-oriented service (S)-KPIs. Weightings of indicator sub components were assigned as a result of surveys of customer priorities. KPIs provided one number that represented overall performance, and they also made it possible to identify each subcomponents contribution. The KPI design processes generated public feedback; this feedback prompted NYCT to tighten underlying performance standards. Today, PES-KPIs and S-KPIs are reported monthly to the Metropolitan Transportation Authority Boards NYCT Committee. Advantages of these indicators include high-level visibility and ease of communication, timely report availability, and detailed diagnostics. These factors, together with a reinvigorated competitive spirit between divisions triggered by reorganizations, resulted in a much more proactive organization focused on using performance scores to take corrective action. Wait assessment, the principal component of the S-KPI, improved 2.5% on the heavily crowded 1 through 6 lines in 2012 compared with 2011, even as ridership increased steadily systemwide.