Transit Ridership by Agency, Monthly: The National Panel, Delivered

The query sounds modest - boardings, by operator, by month - and exactly one record answers it at national scale: National Transit Database (NTD) Monthly & Annual Ridership Datasets, the measurement layer of American public transit. Every operator that reports to the Federal Transit Administration files into it, because Congress requires them to. That makes it a census, not a survey: no sampling frame, no opt-in panel, no recall bias - every reporting agency appears.

The grain is the query's grain, verbatim. One row per agency x mode x type-of-service x month, carrying four measures: unlinked passenger trips, vehicle revenue miles, vehicle revenue hours and vehicles operated in maximum service. The complete monthly module verifies at roughly 370,000 rows across 834 distinct agency identifiers, running January 2002 through the most recent reported month (June 2026 at research time).

Quality checks out too. It scores a perfect 10 on our field-documentation rubric - a ceiling shared by just 145 of the 1,744 datasets we catalog - and it anchors a passenger ground transportation slice of 14 primary records where five hit the same mark. Get a sample of transit ridership data cut to your agencies, modes and months before anything else; everything below shows what arrives.

Illustrative rows in the delivered shape - one California operator, two modes, one recent month:

# City of Fairfield, California - ntd_id 90092 - month 2026-06
mode=MB   tos=PT   _3_mode=Bus      # fixed-route bus, purchased transportation
upt=9,284     vrm=24,788    vrh=1,990    voms=12

# same agency, same month, demand response
mode=DR   tos=PT
upt=1,781     vrm=19,406    vrh=1,900    voms=10

# Redding Area Bus Authority - ntd_id 90093 - Small Systems Reporter
mode=DR   tos=PT   month 2026-06
upt=7,351     vrm=43,122    vrh=2,540    voms=22

Read the anatomy rather than the digits - each value above is the documented example for its column. Three habits separate careful users from casual ones. First, upt counts unlinked trips: a rider who transfers boards again, so summing across modes overcounts unique people. Any total worth quoting states whether it is boardings or passengers. Second, the efficiency ratios come free of the rows themselves - trips per revenue mile, trips per revenue hour - because vrm and vrh ride beside the count on every line. Third, voms is the fleet signal: vehicles operated in maximum service, the number that sizes an operator's network before anyone reads the ridership at all.

One operator contributing several rows per month is the design, not noise. Fixed-route bus and demand response behave like different businesses wearing the same logo, and the mode ladder keeps them separable down to nineteen codes.

All thirteen documented monthly fields verify against the official per-column dictionary - the reason this record sits at the top of our documentation rubric against a catalog average of 7.81 across 1,744 datasets.

• Geographic: the United States end to end - 834 distinct agencies in the monthly module, keyed to urbanized area and state, drawn from the roughly 900-plus operators that report annually overall.
• Temporal: January 2002 through the most recent reported month, June 2026 at research time - a quarter-century panel that walks through the pre-pandemic peak, the 2020 collapse and the uneven recovery month by month. Annual metric tables cover report years 2022-2024, with older report years held in the program archive.
• Granularity: one row per agency x mode x type-of-service x month. Long format by design, which rewards forecasting and punishes spreadsheet eyes - more on that below.

One boundary deserves stating up front rather than discovered mid-analysis: the monthly module carries Full Reporters. Agencies running fewer than 30 peak vehicles and rural reporters sit outside it, their estimated figures living in the companion annual release instead. Roughly 900-plus agencies file overall against 834 in the monthly table, and the gap is precisely the small systems that dominate agency counts in rural maps. Ask for total-industry figures and the sample merges both releases; the caveat travels documented either way.

The federal census answers the national question, but five sibling records in the same slice answer adjacent ones better, and they divide cleanly by grain and geography. Lined up as of August 2026:

Yes, from two city systems, and each changes what a trend means.

CTA Ridership - Daily Boarding Totals compresses Chicago into 9,312 rows - one per service date since January 2001, split bus versus rail, flagged weekday, Saturday or Sunday-holiday. Twenty-five years of seasonality open in a spreadsheet, observed through June 30, 2026, which makes it the fastest way to prototype a day-type model before scaling to the national panel.

MTA Daily Ridership Data: Beginning 2020 goes finer still: nine modes - subway, buses, LIRR, Metro-North, Staten Island Railway, Access-A-Ride, bridges and tunnels, plus the two congestion-pricing counters - about 17,700 long-format rows covering March 1, 2020 onward. It is the only cataloged series where post-2020 recovery and congestion-zone effects resolve mode by mode, day by day.

Both join to the federal census on agency identity plus calendar date, so a monthly national claim can be stress-tested against daily city evidence without reshaping anything.

Direct agency-to-agency comparison stops at the border, because no other country publishes the agency-x-month grain at national scale. Two cataloged records give the international frame:

• Eurostat Passenger Transport Statistics (tran_hv_psmod / rail_pa) - harmonized inland modal split across 37 EU27-plus-EFTA geographies annually from 1990 through 2024, with rail passenger volumes running further. It benchmarks countries, not operators.
• UK Bus Statistics & Bus Open Data Service (BODS) - Great Britain patronage, mileage, fares, fleet and staff in the accredited BUS01-BUS09 tables reaching back toward the 1950s, paired with 7,207 operator-filed timetable datasets and 949 live vehicle-location feeds.

A practical pattern: build the US panel from the NTD, normalize per urbanized-area population using American Community Survey Journey to Work tables to size each agency's catchment, then set the result against Eurostat modal shares so the American trend reads in a European context. Three records, one argument.

Data scientists and ML engineers inherit a quarter-century panel with built-in seasonality and a documented structural break - training substrate for demand models that need real shock absorption; see data scientists use cases.

Investors and quant researchers read monthly ridership curves as urban-activity factors - footfall proxies for mobility, real-estate and retail theses, keyed to metro rather than sentiment; the workflow lives at investors quants use cases.

Market researchers and consultants size US public-transit markets by mode and metro off a congressionally mandated census rather than a vendor's opt-in panel; see market researchers use cases.

Competitive intelligence teams rank operators by unlinked trips to prioritize fare-collection, fleet-tech and service-planning outreach, with reporter_type and voms sizing each account before the first call; see competitive intel product teams use cases.

Journalists and academics cite filed figures whose provenance is a federal reporting mandate - the strongest kind of attribution a mobility claim can carry.

Shape is part of delivery, not your problem afterward. The native form is long and loves time-series work; samples ship pivoted wide per agency on request - one row per agency and month, a column per mode, keys preserved so both shapes still join. Mode codes arrive decoded beside their spelled-out names. Annual metric families arrive joined onto the monthly spine where the question needs money attached to boardings.

Because deliveries repeat on a schedule, each month's report becomes another point on a panel rather than another file to reconcile - captures accumulate under stable keys, and restatements stay traceable to the extract they came from.

Because the hard part was never the first extract - it is the tenth. Long-format rows that punish anyone expecting one column per mode. Mode codes (MB, DR, VP) that need a decoder ring. Annual metrics living in separate tables that need joining onto the monthly spine. Reporter classifications that quietly decide whether your totals cover the whole industry or only its largest players. Small-system exclusions that surface three notebooks deep, right when a rural map stops adding up.

Each is survivable once; none is fun to re-solve in every new project. Datadory handles them upstream of delivery: pivots cut to the shape your model wants, dictionaries traveling with every extract, annual families joined where useful, coverage boundaries surfaced explicitly rather than discovered mid-analysis.

Name your agencies, modes and months when you request a sample and it arrives already cut to that scope - the production feed follows the same shape, so anything prototyped on the sample survives delivery intact.

Start with the passenger ground transportation data guide, which maps all fourteen cataloged records in the industry and shows where the ridership census sits beside trip-level ledgers and feed catalogs. The passenger-ground-transportation data hub browses the same catalog as product pages with field dictionaries and sample rows, and the best passenger ground transportation datasets ranking scores the top ten side by side.

For depth on the threads named here: Mobility Database vs National Transit Database compared weighs the world's GTFS feed discovery layer against the federal measurement layer, the Mobility Database - Global GTFS & GTFS-Realtime Feed Catalog covers the supply side of the same networks, and the National Transit Database (NTD) Monthly & Annual Ridership Datasets product page documents every field row by row.