How we estimate any shape

Most public data is published for official areas, such as counties, census tracts and block groups. Your shape, such as a 10-minute drive around a store, rarely matches them. Here's how Kadrin gets the numbers for exactly your shape, and how we check them.

Where people actually live

When your shape covers only part of an official area, we don't assume people are spread evenly across it. We look at where people actually live inside it, using the 2020 Census's counts for census blocks: the smallest areas the Census counts, often a single city block.

Say your shape covers 60% of a block group's land, but that 60% is a park and a lake. Spreading people evenly would put 60% of the households inside your shape. Looking at the blocks puts almost none of them inside, which is right.

Each measure uses the right count to decide its share: people for measures about people, such as age; households for measures about households, such as household income; and homes for measures about homes, such as rent. That keeps dorms, prisons and nursing homes from counting as households. This way of estimating is called apportionment.

The right rule for each kind of number

Numbers can't all be combined the same way:

• Counts, such as population, are added up.
• Rates and averages, such as the poverty rate, are worked out again from their parts: we add up the people in poverty and the people counted, then divide.
• Medians, such as median household income, are worked out again from the full spread of incomes across your shape.
• Jobs are published for each census block, so they're added up directly.
• Businesses are counted inside your shape.
• New homes approved are published only for counties, so your shape gets each county's number shared out by where its homes are. When your shape covers only part of a county, the number carries the warning coarser_resolution.
• Jobs, wages, hiring, workplaces and business openings and closings are published only for counties or ZIP codes, so your shape gets each one's number shared out by where its jobs are, using jobs counted by census block. When your shape covers only part of a county or ZIP code, the number carries coarser_resolution.
• Building permits from cities are counted inside your shape over the last 12 full months, or their new homes and costs added up. Only the cities whose permits we have count. Each number's coverage says what share of your shape's homes are in those cities. A shape partly outside them carries the warning partial_coverage, and one outside them all gets no number rather than zero.
• Oil and gas wells are counted inside your shape by where each well stands at the surface, sorted by what the state's records say: active, idle, orphaned (unplugged, with no owner on record) or plugged. People and homes near wells are added up from census blocks: each block counts by the share of its area within 1 kilometer (about 0.6 miles) of an active or idle well. New drilling permits are counted where each well is to be drilled, over the state's last 12 full months. Only the states whose well records we have count, so these work like building permits: coverage says what share of your shape's homes are in those states, and a shape outside them all gets no number rather than zero.

We never average percentages or medians across areas. It's the most common mistake in other tools, and it can be far off.

Official areas and single points skip this. Ask for a county by its GEOID (the official ID number the government gives each area) and you get exactly the county's published numbers. A single point gets the numbers for the block group it falls in, and for numbers published only for counties, that block group's share of its county.

Margins of error

Many numbers come from the American Community Survey, which asks some households rather than everyone, so they're estimates. Each comes with a margin of error: how far the true number could reasonably be from our estimate. The true number is very likely within that distance of the estimate.

When we combine areas, we combine their margins of error with the Census Bureau's standard formulas. These are close but not exact, so such numbers carry the warning approximate_moe. A number with a large margin of error for its size carries high_cv: treat it as a rough guide.

The margin of error covers the survey, not our estimate of where people live inside the areas your shape cuts. When that matters a lot, because spreading people evenly would give a very different answer, the number carries the warning method_disagreement. See Errors and warnings for every warning.

How we check

Before each release, we check the estimates automatically:

• An official area's own outline gives back exactly its published numbers.
• Shapes that cover a county without overlapping add up to the county's total.
• The shares of each area add up to the whole area.

How accurate it is

The Census Bureau publishes its own numbers for ZIP code areas and towns. We estimate those same places from block groups, exactly as we estimate your shape, then compare. Here is how we did.

We last checked this on October 10, 2026, over 2,016 towns and cities and 2,035 ZIP code areas across the whole country.

Counts show most clearly why we estimate the way we do. The same shapes, three ways:

Reading the table:

• Typical error is the middle one: half our estimates were closer than this, half further.
• Within the Census Bureau's own margin is the share of our estimates that landed inside the margin of error the Bureau publishes for that place. Their own numbers are estimates too, so this is the fairest test.
• The largest errors are on the smallest places. A town of 300 people can be off by a third while a town of 30,000 is off by a few percent, because one apartment block either way changes the answer. Rates and medians hold up better than counts for the same reason: they don't grow with the size of the area.

Most shapes people actually send, such as a few miles around a store or a 10-minute drive, cover many block groups, so they sit at the better end of these numbers.

Where building permits come from

Building permits come from the cities that publish them as open data, and only cities that pass our checks are used. Everywhere else has no building permit numbers yet.

No city's building permits are in the current release yet.

Where oil and gas well numbers come from

Oil and gas wells come from each state's oil and gas regulator (and, in federal waters off the Gulf and Pacific coasts, from the Bureau of Ocean Energy Management), and only states whose records pass our checks are used: every well counted in the state's own totals, nearly all of them placed on the map, and, where a state's coordinates need converting to today's map datum, a sample checked against the state's own map.

No state's oil and gas wells are in the current release yet.

What to keep in mind

• Where people live comes from the 2020 Census, so it misses what has been built since. Where an area had nobody in 2020 but has people now, we fall back to the share of its land inside your shape.
• The 2020 Census adds a little deliberate noise to small blocks to protect people's privacy. It mostly cancels out across many blocks, so we use blocks only to decide shares, never as numbers on their own.
• A very small shape inside one block group gets that block group's numbers scaled to its share, which can be less reliable: look for high_cv.
• Building permits from cities count work a city approved, not work finished. Cities don't all record the same things: where too many permits leave out what the work cost, renovation spending has no number, and where a city's permits don't say which work is an EV charger or a heat pump, those have no number there either.
• A city's permits are used only once they pass our checks: most permits placed on the map and inside the city, the kinds of work matching a hand-checked sample, new homes close to the Census Bureau's own count for the city, and the city's data kept up to date.
• Oil and gas wells are counts and distances, not a measure of health risk: wells differ widely in what they release. States don't all mean the same thing by active and idle: Texas, for one, shows an unplugged oil or gas well as active until it's marked shut-in or plugged. People near wells assumes people are spread evenly across their census block. A drilling permit approves drilling; not every permitted well gets drilled. Each state's wells are as of the date in the table above: where the only public copy of a state's records is older than our download, a value's vintage gives that older date.
• New homes approved come from the Census Bureau's survey of the offices that issue building permits. A permit counts work a local government approved, not homes finished, and the survey fills in estimates for offices that didn't report. A few rural counties have no permit office in the survey; a shape that reaches one gets no number rather than zero.