How the census profiles are made
Each profile is built from the U.S. Census Bureau’s American Community Survey, compared with its state and the nation, with every figure’s source and margin of error on the page.
Updated November 17, 2026
Where the numbers come from
The American Community Survey (ACS) asks a sample of households each year. Its 1-year estimates, the most recent, cover areas of 65,000 people or more; its 5-year estimates cover every area, pooling five years of answers. A profile uses the 2025 1-year estimates where they exist and the 2020–2024 5-year estimates otherwise, measure by measure, and says which beside each figure. A state and the nation are always compared from the same survey as the place.
The figures come through Social Explorer’s tables, and each one names its table, so it can be found again. Where people work comes from the Census Bureau’s LEHD Origin-Destination Employment Statistics (LODES). Boundaries come from the Census Bureau’s TIGERweb. Each page gives the date its data was retrieved.
Margins of error
Because the ACS is a sample, every estimate has a margin of error. The profiles show it beside each figure as ±, at 90% confidence, the level the Census Bureau publishes. For totals, medians and single counts it’s the Bureau’s own published margin. The profiles take their figures from the Bureau’s detailed tables, which give margins for counts, not for percentages. So a percentage’s margin, and that of a row adding several counts together, such as ages 18 to 24, is worked out from the published ones with the Bureau’s own formulas (ACS General Handbook, chapter 8). The Bureau’s Data Profile and Subject tables publish some of these percentages with margins of their own, which can differ slightly. The survey’s results are also adjusted so its total population matches the Bureau’s own population estimates, the 2020 Census count carried forward with births, deaths and moves: for every county or group of small counties and, since 2009, for many cities and towns. Such a total has a margin of zero, or one under 0.1% of it. That margin measures only what’s left after the adjustment, not how far the estimate may be from the real population, so the profiles show none and call the figure the Census Bureau’s estimate. The same goes for a count of homes adjusted the same way. Other totals, such as a census-designated place’s or a ZIP code’s, come from the sample itself and carry a real margin. A margin too small to show at one decimal reads “±<0.1%”.
Two figures are said to differ only when the gap between them is wider than their margins together, the test the Bureau uses at 90% confidence. Otherwise the page says they’re about the same, or, when they’re far apart but the margins are wider still, that they’re not statistically different. A topic’s headline says one figure is higher or lower than another only when the gap passes this test. A † marks a figure too uncertain to lean on: one whose margin is large next to it, a coefficient of variation over 40%, or a small place’s median whose margin is implausibly small (below). Such a figure is left out of the page’s sentences and its description, and an answer about it says it’s too uncertain.
Medians past the end of a table
The Bureau works out a median from ranges of values, such as household incomes from “under $10,000” to “$200,000 or more”. When a median falls in the top or bottom range, it can only say which, and publishes no margin. A profile shows such a median the same way, as “$250,000 or more” or “under $2,500”, and doesn’t rank it or count its change over time.
Small places
A sample reaches few households in a small place, so its figures can swing widely from year to year. Below 10,000 residents a profile says so.
Below 1,000 residents, the ACS interviews a few dozen households at most. Its figures for such places were often far from the 2020 Census, which counted everyone: their population by more than a quarter for 27% to 70% of places, depending on size, and their median age by more than 5 years for 44% to 73%. So a profile of a place this small gives each figure as the range it likely falls in, its estimate plus or minus its margin, and compares none of them with its state or the nation. It shows no highlights, no rankings except its population’s, and no change over time, and it isn’t counted in other places’ rankings. Each topic points to the county, or for a small county its state, whose figures rest on many more households.
Even a range can miss. For places of every size, the survey’s range held the 2020 Census median age about three times in four. A small place’s median with a margin under 5% of it held the census figure far less often, from 6% to 33% of the time, because a margin that narrow usually means most of the few households sampled fell in one of the table’s ranges. Such a median is shown with a † and no range.
Below 2,000 residents, a profile doesn’t sum up the survey’s ten-year change in words, since its population was still off from the 2020 Census by more than a quarter for 15% to 20% of places of 1,000 to 2,000 people. The table and charts still show it.
Rankings
A place is ranked among the places of its kind in its state, or among all states, regions or divisions, on the 5-year estimates, the only ones that cover every area. Population is ranked among them all; every other measure only among those of 1,000 people or more.
Areas that change
ZIP code pages use the Census Bureau’s ZIP Code Tabulation Areas, which follow the Postal Service’s ZIP codes as closely as census blocks allow. Congressional and legislative districts, PUMAs and urban areas are redrawn from time to time, so their pages show no ten-year change.
How the words are written
Every sentence is written by fixed rules from the numbers, not by AI, so a page can’t state anything its data doesn’t. A sentence whose number is missing is left out rather than filled in.
Updates, reuse and corrections
The Census Bureau releases the 1-year estimates each September and the 5-year estimates each December, and the profiles are rebuilt after each release. Everything is free to read, cite and reuse: each profile has a CSV of its figures and margins, a citation to copy, and charts to embed. To report a mistake, contact us.