American Community Survey - Journey to Work / Commuting Data

Every ten years the census counts heads. Between those years, the American Community Survey asks what those heads do - including how they get to work. It has run annually since 2005 as the Census Bureau's ongoing household survey, collecting social, economic, housing and demographic detail from a sample of households across the fifty states, the District of Columbia and Puerto Rico.

For passenger ground transportation the interesting slice is narrow and deep: table B08301, Means of Transportation to Work. Its universe is total workers 16 years and over, and its columns walk that universe through drove alone, carpooled, public transportation excluding taxicab, walked, bicycle, motorcycle, taxicab and worked from home. Under the transit subtotal sit bus, subway or elevated, long-distance train, light rail and ferryboat, so "transit" decomposes into modes an operator can actually act on.

Around that core sit companion tables that make it usable: travel time to work banded by minute, vehicles available crossed against mode, and worker flows pairing origin counties with destination counties. Get a sample of this dataset and Datadory hands you the resolved tables - fields populated, margins attached - instead of a documentation project.

Rows arrive as typed counts keyed to a geography you name, each carrying its margin of error. The verified frame looks like this:

• B08301_001E - total workers 16 years and over, the denominator everything else divides against.
• B08301_003E - drove alone; B08301_004E - carpooled.
• B08301_010E - public transportation excluding taxicab, with B08301_011E the bus tier beneath it.
• B08301_019E - walked.
• NAME - the area name, so a row reads as Los Angeles County, California rather than a FIPS string.

Two properties distinguish these numbers from modelled or secondhand alternatives. First, they are estimates with stated reliability: every E value has an M twin, so a thin block-group cell admits its uncertainty instead of posing as fact. Second, the grain scales without changing vocabulary - the same columns answer at national scope and at block-group scope, which is what makes a commute-mode panel built once usable everywhere.

Eight verified fields anchor the frame, listed in full in the field dictionary above. Three jobs organise them: measurement (the B08301 mode columns), identification (NAME), and selection (the geography predicate that scopes every row).

Beyond the dictionary sits the wider journey-to-work family - travel-time bands, vehicles available by mode, the remaining mode cells including bicycle and worked from home, the flow series, and person-level PUMS microdata for questions the pre-tabulated counts cannot answer. Those extend the frame rather than replace it, and they arrive populated with your sample, documented against delivered extracts. Name them when you request it.

Geographically, the survey covers every U.S. state, county, place, tract and block group. The one constraint is the trade between recency and resolution: one-year estimates publish only for geographies of 65,000+ population, while five-year estimates reach every geography including block groups. Analysts needing both currency and fine grain typically run the two releases side by side.

Temporally the series runs annually since 2005, which makes two decades of commuting behaviour available as a genuine time series rather than a snapshot. Current releases are the 2024 1-year ACS and the 2020-2024 5-year ACS, each accompanied by new PUMS files.

Granularity comes in two forms: pre-tabulated counts per geography, ready to join and rank, and person-level PUMS microdata, ready to re-tabulate. County-to-county worker-flow tables add the directional layer, pairing where workers live with where they go.

API, files, or your warehouse. Daily, weekly, or hourly. Datadory resolves the survey's tables into typed records with definitions attached, so the work arrives as columns you can join rather than a table-numbering scheme to decode first.

Samples come shaped to your question: name the geographies, the modes or the companion tables you care about and the sample reflects that scope, with the field dictionary attached and any additional fields populated. The schema you test is the schema you ship against.

Six jobs dominate. Transit planners read mode share at tract level to find where dependence already concentrates. Network developers pair origin-destination flows with those shares to argue corridors. Site-selection teams use travel-time bands and vehicle availability to describe labour catchments around candidate locations. Mobility product teams size the drove-alone pool before committing capital. Employers compare commute burdens across office options. Policy researchers cross vehicles available against mode to separate transit choice from transit necessity.

What unites them is the denominator problem solved at source: every share in this dataset divides against a counted universe of workers 16 and over, not a modelled population. That is why ACS-derived figures survive scrutiny in board decks, filings and peer review when softer numbers do not.

Market researchers and consultants get citable mode-share evidence. Data scientists get typed panels whose margins of error arrive attached, suitable as ML features without reconstruction. Urban planners get the same block-group numbers their agencies must cite. Investors and quants read commuting structure as a slow indicator behind real estate and logistics exposure. Journalists, academics and students cite the federal institution itself.

The common thread is durability. Commute behaviour moves slowly, so a panel built on this survey stays defensible for years - useful precisely because it does not need constant rebuilding.

Decide your grain first. If the question lives below county level, the 5-year release is the only option that reaches block groups; if the question needs last year, the 1-year release stops at populations of 65,000+. Most teams take both and let each answer what it is built for.

Second, decide whether pre-tabulated counts serve you or whether the question needs PUMS microdata. Counts answer ranked comparisons instantly; microdata answers cross-tabulations nobody has published yet. Third, name the companion tables you want - travel time, vehicles available, flows - so they land in the sample populated rather than in a later round. Samples precede any commitment.

This survey is the behavioural backdrop; the rest of the industry catalog supplies the operations. The National Household Travel Survey captures personal travel behaviour nationwide, the National Transit Database carries monthly and annual agency ridership, NYC TLC trip records and Chicago Taxi Trips capture individual journeys at trip level, and the MTA daily ridership file tracks New York boardings day by day. Pair this survey's mode share with any operational feed and you have both why people travel and what happened when they did.