Skip to content
American Factbook
Menu

How the demographic figures are made

How the pages of the Demographics hub read what their publishers put out, and what that record can and cannot say. The method every topic shares — where a figure comes from, snapshots, freshness and corrections — is on the site’s methodology page.

Two Census products, kept apart

Populations come from the Census Bureau’s population estimates: the number of people on July 1 of each year, built between censuses from births, deaths and migration records, with no sampling error. Everything else (age, income, poverty, housing, origin) comes from the American Community Survey, a sample the Bureau averages over five years. A share on this site is always an ACS count over the ACS’s own total from the same table. It is never divided by an estimated population, because the two count different people at different times.

Every survey figure carries its margin

The ACS publishes each estimate with a margin of error at 90% confidence, and this site prints it beside the figure. A share’s margin is computed with the Bureau’s own formula for a proportion. A five-year estimate is named by both its years, 2020–2024, and never by the last alone.

Where an estimate is too uncertain for one number

The coefficient of variation is the margin’s standard error (the margin divided by 1.645, the Bureau’s factor for 90% confidence) as a share of the estimate. Where it is over 30%, this site prints the range, the estimate less and plus its margin, in place of the estimate, with a note that the Bureau’s estimate is too uncertain to state as one number. A range stops at zero, and a share’s at 100%. Such a figure is shown but not ranked: in any ordered list it sits beneath the ranked rows, and no finding names it as highest or lowest. No figure is left out for being uncertain. The 30% line is a convention among users of the survey, not a rule the Bureau sets.

Codes that are not numbers

Where the Bureau could not compute an estimate or a margin, its files carry a code such as −666666666 in place of the number. These codes become blanks with the Bureau’s reason attached, never figures. Where a median falls in the top or bottom interval, the file carries the interval’s edge (250,001 means “$250,000 or more”), and a page prints the interval, not the edge.

The first year is three months

The estimates start from the April 2020 census, so the first year’s births, deaths and migration run from April to July 2020. Figures on this site sum the components over the full years from July 2020, and the change they explain is the same five years.

Which places have a page

Every county has a page. A place or a New England town has one when the survey counts 5,000 people or more in it: 7,192 of them. Below about 5,000 the survey’s margins grow until most of its figures for a place are a range too wide to print as a number, and a page of ranges is a thin page. A town that is also a place has one page, the place’s.

A rank can be a band

The state rankings of survey figures print each state’s margin, and where two states’ ranges overlap the survey cannot say which is higher. Such a state’s rank is written as the range it could hold, 3–5, worked out from the margins: its best rank is one more than the states wholly above it, its worst the number not wholly below it. The population rankings are of estimates, not a sample, and have single ranks.

What this data cannot show

  • The American Community Survey is a sample. Every figure from it is an estimate with a 90% margin of error, printed beside it; for small places the margin can be as large as the estimate.
  • An ACS figure is a five-year average. The 2020–2024 estimates describe those five years together, not 2024, and every page names both years.
  • The population estimates are revised every year, back to 2020. A figure names its vintage, and a later vintage can move an earlier year.
  • The estimates' first year runs from the April 2020 census to July 2020, three months. Births, deaths and migration on this hub are summed over the full years from July 2020.
  • A share is an ACS count over the ACS total of the same table, never over an estimates population: the two count different people at different times.
  • Hispanic or Latino origin is not a race. The race and origin shares come from one table whose categories do not overlap.

Sources

Data retrieved