For people who rely on transit, a scheduled train is not an abstraction. It keeps life in motion.
Train waiting at the platform.
But we're too quick to forget what happened along the way.
SOURCE
MBTA General Transit Feed Specification (GTFS) Realtime archive, Train 870, May 27, 2026.
THE RECORD REFLECTS
SOUTH STATION
A LATE DEPARTURE
PROVIDENCE
A RIDER’S STORY
ATTLEBORO
World Cup schedules removed 43% of trains on the Providence/Stoughton Line. A fragile system lost its backup plan; passengers paid the price.
One event opened onto a network of uneven changes. Route level disparities were extreme. But, another social determinant also had influence here.
Outside Franklin/Foxboro, station catchments with higher transit reliance experienced larger losses in usable journeys during World Cup operations.
The World Cup was a logistical success for the MBTA. The agency and Keolis changed schedules and operations to manage event-related interruptions, moving record crowds to and from Foxboro while continuing to serve the broader commuter network. That operational achievement should not be understated.
Despite those efforts, the MBTA and Keolis could not overcome longstanding limits in infrastructure and service. Event trains and ordinary weekday travel still had to share a network with limited room to absorb schedule changes, service interruptions, and competing demands across the system.
Residents both depended on the system for more trips and faced larger service cuts. Car ownership can be both a fallback and a marker of social status—an advantage conventional socioeconomic measures may not fully capture. Wealthy transit-dependent communities were equally affected.
The study joins published schedules, the MBTA’s archived real-time records, station geography, and 2024 Census estimates. Choose a ticket to inspect the assumptions, construction choices, and limits behind the findings.
Begin with every inbound, direct trip from a station to North or South Station scheduled to depart from 5:00 through 9:59 AM.
Search forward from that promised departure for the first same-day direct train serving the same origin and downtown terminal that a rider could actually board.
Compare that train’s observed downtown arrival with the original trip’s promised arrival. The journey is timely when the difference is under 15 minutes.
The replacement can be the original train or a later train, including one on another route, as long as it is direct to the same downtown terminal. This measures whether the journey a rider planned remained usable—not merely whether a train that happened to run was punctual.
The operational input is the MBTA GTFS-Realtime archive paired with published GTFS schedules. The processed archive contains 485,836 final updates for scheduled trip-stop pairs at 143 stations on 12 lines, covering April 28 through August 18, 2026. Only the last archived update for each trip-stop record is retained so repeated feed messages are not treated as independent observations.
The primary weekday journey panel covers June 8 through July 12: six World Cup operation weekdays and 19 ordinary comparison weekdays, producing 70,446 posted direct morning journeys. Station context comes from MBTA station geography and 2024 American Community Survey five-year tract estimates.
The primary panel uses the six weekday operation dates: June 16, 19, 23, 26, and 29, plus July 9, 2026. June 13 and 14 are identified as event or dedicated-schedule dates in the project, but they are weekends and therefore do not enter the weekday fixed-effects comparison.
The route-level chart compares each operation weekday with that station and weekday’s median non-event service. The model-based analysis uses the complete 25-weekday panel instead of a simple before-and-after average.
A train is not usable if the trip or relevant stop is marked cancelled or skipped, if an observed origin departure or downtown arrival is unavailable, or if it departed more than one minute before its posted time. The search then continues to the next same-day direct train from the same station to the same terminal.
If no usable recovery is found within three hours, the journey is treated as unserved and receives a 180-minute penalty. Missing observed times can reflect incomplete reporting as well as unavailable service, so this is a conservative operational measure; it cannot fully distinguish those mechanisms.
Fifteen minutes is the primary rule for whether the original promise was still met after any waiting and replacement travel. The comparison uses the original trip’s scheduled downtown arrival—not the replacement train’s own timetable—so a later train does not reset the clock.
The analysis also repeats the outcome with stricter five-minute and looser 30-minute thresholds. The transit-reliance gradient remains negative under both alternatives, so the result is not created by a single cutoff.
Regular station-route-terminal combinations are crossed with every analysis weekday to create a balanced panel. When a journey normally exists but disappears from an event-day schedule, its station-day count is recorded as zero rather than dropped.
This matters because analyzing only trains that ran would make a withdrawn schedule appear reliable: the missing trips would no longer be eligible to fail. Posted opportunities capture the timetable; timely realized opportunities add whether operations kept those promises.
Each Massachusetts station receives a Voronoi catchment: nearby territory is assigned to the closest station, and every cell is capped at an eight-kilometer radius so a station does not represent distant areas simply because no other stop is nearby. Census tract representative points are spatially assigned to those cells.
Tract characteristics are aggregated using household population—total population minus group-quarters population—as weights. The result covers 139 Massachusetts stations. Providence, Pawtucket/Central Falls, T.F. Green Airport, and Wickford Junction are excluded because the project’s tract extract contains Massachusetts, not Rhode Island.
The socioeconomic-status index equally combines four standardized tract measures: log median household income (positive), unemployment (negative), deep poverty below 50% of the poverty line (negative), and population below 200% of the poverty line (negative). The ACS tables are B19013, B23025, and C17002.
Transit reliance is measured separately as the share of workers commuting by public transit. The share of households with no vehicle is a secondary measure of substitute access. These are characteristics of residents near a station, not measurements of the passengers aboard any particular train.
Station-day outcomes are estimated with ordinary least squares models containing station fixed effects and date fixed effects. Station effects absorb stable differences between locations; date effects absorb network-wide conditions on a given day. Standard errors are clustered by station.
The Franklin/Foxboro model interacts corridor membership with a World Cup weekday indicator. Outside that corridor, the distributional model interacts the event-day indicator with standardized transit reliance. Its coefficient is the change in journey opportunities associated with a one-standard-deviation increase in a station catchment’s public-transit commute share.
The analysis assigns “event” status to every possible set of ordinary dates with the same weekday composition as the six actual World Cup weekdays. It re-estimates the outside-corridor transit-reliance gradient for each of the 144 matched calendars.
This asks whether a gradient as large as the observed one routinely appears under comparable weekday groupings. The exact two-sided test adds one to both the numerator and denominator. It is a calendar-based falsification check, not a cure for every event-period confounder.
The announced Franklin/Foxboro service substitution provides the clearest intervention: the corridor comparison estimates how its direct South Station access changed relative to other stations and dates. Even there, the design is an observational fixed-effects comparison rather than a randomized experiment.
The rest of the network is treated as a spillover audit. Its transit-reliance coefficient describes how event-period losses were distributed outside Franklin/Foxboro; it does not claim that the World Cup independently caused every route-level schedule or operational change.
The records do not identify individual riders, origins, trip purposes, vehicle ownership, capacity constraints, crowding, denied boardings, transfers, fares, or door-to-door travel. Station catchments approximate nearby residential exposure; they do not reveal who was actually on a train.
The analysis also cannot infer operator intent, rank every possible express-versus-local service tradeoff, or cleanly separate an unreported train movement from an unavailable one. Those limits are why the project makes a distributional access claim rather than a claim about individual harm or agency motive.
Operational records come from the MBTA’s published schedule and archived GTFS-Realtime feeds. Demographic measures come from the 2024 American Community Survey five-year estimates. The project’s numbered notebooks document the complete pipeline from archive processing through catchment construction, journey matching, models, sensitivity checks, and figures.