County-to-county commuting flows: every home-work pair, delivered as typed rows

A commuting flow is a count of workers aged 16 and over who live in one county and work in another: an origin-and-destination matrix, not a ridership total, a fare ledger or a timetable. It answers the question every ground-transportation model eventually asks - how many people does this corridor have to move, and in which direction - before a single boarding is counted.

The grain runs deeper than the name implies. Five-year estimates cover every geography down to the block group - roughly 85,000 tract geographies and about 840 county equivalents in the current 2020-2024 release - while one-year estimates stop at populations of 65,000 and over. Corridor work touching rural or micropolitan counties lives in the five-year file; metro trend work takes the 2024 one-year release alongside it.

Rows arrive typed and keyed to named geographies, every estimate travelling with its margin-of-error twin. The verified frame:

# Table B08301 - Means of Transportation to Work (verified variable set)
B08301_001E  Estimate!!Total:                                            -> int
             Total workers 16 years and over
B08301_003E  Estimate!!Total:!!Car, truck, or van:!!Drove alone           -> int
B08301_004E  Estimate!!Total:!!Car, truck, or van:!!Carpooled             -> int
B08301_010E  Estimate!!Total:!!Public transportation (excluding taxicab)  -> int
B08301_011E  ...bus specifically, one tier beneath the transit subtotal
B08301_019E  Estimate!!Total:!!Walked                                     -> int

# Reliability ships beside every count
B08301_001E = n   <->   B08301_001M = +/-m

# The flow layer puts two geographies on one row
residence county  x  workplace county  ->  workers making that trip, margin attached

Three properties decide whether these rows survive contact with a real model. Direction is recorded: the A-to-B count and the B-to-A count are different rows, because morning and evening demand are different questions. Reliability is recorded: a thin county-pair cell declares its uncertainty instead of posing as fact. And the vocabulary never changes with scale: the same columns answer nationally, by metropolitan area, by county and by tract, so a panel built once re-levels without remapping anything.

Get a sample of commuting flow data cut to your corridors and counties - the field dictionary travels with it.

The load-bearing columns stack into three jobs: measurement, identification and scope.

Beyond the dictionary sit the remaining B08301 cells - bicycle, motorcycle, taxicab, worked from home, plus the subway, long-distance train, light rail and ferryboat tiers - and person-level microdata for cross-tabulations nobody has published yet. Name them when requesting a sample and they arrive populated, documented against the delivered extract.

Labour-shed and trade-area sizing. Travel-time bands describe who can reach a site and how long it takes them; vehicles available says whether they own the car that would. Site-selection teams weigh candidate locations on distributions instead of map guesses.

Corridor cases. Origin-destination flows paired with mode share turn "this train seems busy" into a defensible service case: how many workers make the trip, which modes absorb them today, and what substitution rate a new option implies.

Market sizing for mobility products. Drove-alone counts at fine geography size the pool a carpooling app, commuter-benefits platform or micromobility launch can realistically convert - computed before a dollar is spent, off the same rows the business case will later be audited against.

Yes, and the cleanest test case is commuter rail around New York, where MTA Daily Ridership Data: Beginning 2020 tracks nine counters day by day - Subway, Bus, LIRR, Metro-North, Staten Island Railway, Access-A-Ride, bridges and tunnels traffic, plus the congestion-zone entries added in January 2025 - in roughly 17,700 rows covering March 1, 2020 through August 19, 2026, one row per mode per date. Because LIRR and Metro-North serve exactly the long-distance county pairs that dominate New York's inflow matrix, a corridor-level flow estimate can be checked against actual boardings on the same geography.

Where the two disagree, the suspects are knowable in advance. Vintage mismatch first: five-year files blend five survey years, so treat them as rolling averages rather than point-in-time counts when quoting a single year. Stale assumptions second: a share built on an old mode split will miss a corridor whose behaviour moved.

National Household Travel Survey (NHTS) supplies the behavioural half of the conversion. Nine waves reach back to 1969, and the 2022 public-use core holds 7,893 households, 16,997 persons and 31,074 weighted travel-day trips, each carrying mode codes, home-based-work purpose flags, trip distances, durations and the full-file survey weights. It explains why people travel the way the flow matrix says they do.

Its boundary matters as much as its content: household geography stops at region, division, CBSA and state FIPS, so NHTS calibrates the commuting shares rather than replacing them as a second flow matrix. Run together - flows for volume and direction, survey microdata for behaviour and purpose - and a demand model inherits both the population and the motives. The fuller workflow sits on the data scientists use cases page.

Not at county grain, no. The nearest structural analogue is Eurostat Passenger Transport Statistics (tran_hv_psmod / rail_pa), whose flagship table reports the share of inland passenger-kilometres taken by trains, buses and coaches, and cars annually from 1990 through 2024 across 37 geographies including the EU27 aggregates and EFTA countries. Harmonized methodology, national level only - no European record in the slice pairs one home region with one workplace region.

That difference belongs in any deliverable. US county-to-county flows answer how many people travel from A to B for work; Eurostat answers what fraction of all travel happens by each mode. Pairing the two lets a study state both how large a corridor is and how its modal mix compares internationally - but neither substitutes for the other, and mixing their units is the most common presentation error in cross-market decks.

Transit planners and network developers read mode share at tract and block-group resolution to find where dependence already concentrates, then argue corridors from origin-destination flows instead of anecdote.

Site-selection and location-strategy teams describe labour catchments around candidate sites - who can reach them, how long it takes, whether they own a car - before committing capital; see the market researchers use cases page.

Investors and quant analysts read commuting structure as a slow indicator behind real estate, logistics and workforce exposure; the workflow lives at investors quants use cases.

Data scientists and ML engineers get typed panels whose margins arrive attached - defensible features for catchment, demand and siting models; see data scientists use cases.

Policy and equity researchers cross vehicles available against mode to separate households that ride transit by choice from those that ride it because nothing else exists.

Commute behaviour moves over years, not weeks, so a panel built on these rows stays defensible without constant rebuilding - a rarity in any data catalog.

Because the hard part was never the first number - it is the tenth. Choosing the wrong release for the question's geography. Margins stripped by a naive pipeline until thin cells pose as facts. Flow tables that refuse to line up with the marginal mode tables without a mapping nobody documents. Vintages stitched mid-series with silent breaks. Each is survivable once; none is fun to re-solve in every new project.

Datadory handles them upstream of delivery: releases matched to the question, margins travelling beside their estimates, join keys stable across vintages, flow pairs and mode splits landed in one consistent frame - as API responses, scheduled files, or warehouse-native tables.

Name your counties and corridors when you request a sample and it arrives already cut to that scope. The production feed follows the same shape, so whatever you prototype on the sample survives delivery intact.

Start with American Community Survey - Journey to Work / Commuting Data sampled to your corridors; product-level detail - sample rows, the full field dictionary, coverage chips - lives on its dataset page.

This page is one thread of a wider map. The passenger ground transportation data guide walks all fourteen cataloged records in the slice, the passenger-ground-transportation data hub browses them as products, and the best passenger ground transportation datasets ranking sorts the leaders side by side. For neighboring threads: the commuting mode share by census tract walkthrough covers the tract-level frame, and transit ridership by agency monthly explains the boardings record those flows feed. When descriptions stop and real rows need to start, get a sample scoped to your counties - the field dictionary travels with it.