Who Actually Lives Here: Four Census Tables to Pull

Type a candidate town's name into the search bar at data.census.gov and you get back more places than exist. An incorporated place with that name. A county subdivision with the same name. Sometimes a county with the name again. Two ZIP Code Tabulation Areas that straddle the edges. A dozen census tracts, none of them labeled with anything a human recognizes.

All of them are real. All of them have different populations. The one you click decides every number that follows.

Nobody warns you about that part. The tables are not hard. The geography is.

The town you typed is probably three different towns

The cheapest way I know to settle it takes about fifteen seconds. Feed the exact street address — the one on the listing you are staring at — to the Census Bureau's geocoder:

https://geocoding.geo.census.gov/geocoder/geographies/onelineaddress
  ?address=STREET,+CITY,+ST+ZIP&benchmark=Public_AR_Current
  &vintage=Current_Current&format=json

I ran it against 1600 Pennsylvania Ave NW on 18 August 2026 to confirm the field names are still what I remembered. Among the geography buckets that came back:

• Incorporated Places — "Washington city", GEOID 1150000
• County Subdivisions — "Washington city", GEOID 1100150000
• Counties — "District of Columbia", GEOID 11001
• Census Tracts — "Census Tract 9800", GEOID 11001980000
• 2020 Census Blocks — "Block 1034", GEOID 110019800001034

Look at the first two. The name is byte-for-byte identical and the GEOIDs are not, and they are separate rows in every table you will ever pull. The name is not the identifier. The GEOID is. So copy the place GEOID and the tract GEOID into your notes. Those two strings are what you paste into the search box later, and they end all argument about whether you clicked the right Springfield.

One trap deserves its own sentence. An address outside city limits belongs to no incorporated place at all — it sits in a county subdivision or unincorporated county land. Compare it against city-limits figures and you are reading a different town. Same failure that wrecks crime rates, which I worked through in reading the FBI's Crime Data Explorer: wrong denominator, confident number, nothing on screen to tell you.

That case is easy to produce on demand. Send 3600 S Las Vegas Blvd, Las Vegas, NV 89109 through the same request and the Incorporated Places bucket comes back empty — the address sits in unincorporated Clark County, county subdivision "Las Vegas CCD" (GEOID 3200394504), Census Tract 67 (32003006700). Four miles north, in downtown Las Vegas, 495 S Main St, Las Vegas, NV 89101 returns "Las Vegas city", GEOID 3240000, tract 32003000700. Both say Las Vegas on the envelope; only one of them is inside a city whose published figures describe it.

Four table numbers instead of forty

The American Community Survey covers 40-plus topics across thousands of tables. You do not need thousands. For a first pass, four Data Profiles carry almost all of it:

Search the profile ID together with the geography — DP04 Census Tract 9800, District of Columbia — instead of browsing menus. Then do the step most people skip: select "Dataset" in the toolbar above the table and choose "2024: ACS 5-year Estimates." That instruction is the Census Bureau's own wording, from its January 2026 release announcement. A table quietly showing an older vintage looks identical to one showing the newest.

Two of those rows do double duty. DP04's heating fuel line tells you what you will burn to keep the house warm, which feeds straight into a cost sheet — the whole Real Cost of Living side of this site runs on lines like that. And vehicles-per-household is the earliest cheap signal that a town quietly assumes two cars.

If you would rather read prose than a grid, Narrative Profiles exist for the nation, states, counties, county subdivisions, places, census tracts, ZCTAs and metro areas. They are genuinely readable. They are also a summary, so I use them to orient myself and then go back to the numbered table.

For a sixty-second look, QuickFacts gives a one-screen profile. Its limit is coverage. States and counties are all in it; below them it stops at the population floor set in its own coverage note, 5,000 residents. Smaller towns are simply not there — and that is exactly the size of town this kind of search keeps landing on.

The number after the plus-or-minus is the number

Every ACS figure is a survey estimate carrying a margin of error, at a 90% confidence level by default. At state level the MOEs are small enough to ignore. At tract level they routinely erase whatever difference you thought you had found.

Invented but entirely typical arithmetic. Tract A shows median household income of $71,400 ±$9,800. Tract B shows $64,200 ±$8,100. Read the point estimates and A looks 11% richer. Read the intervals — $61,600 to $81,200 against $56,100 to $72,300 — and they overlap across most of their range. You have not measured a difference. You have measured noise. And you were about to cut a town for it.

My rule is deliberately crude: if the intervals overlap, write "no difference found" and move on. When a comparison genuinely matters, the Census Bureau publishes a statistical testing tool that does the test properly at 90%, 95% or 99%.

Three habits follow from that.

Prefer counts and shares over medians at tract level; "62% owner-occupied" survives a wide MOE far better than a median value does. Never set a tract estimate against a state or national figure without checking both margins. And write the MOE into your sheet, not just the estimate — copy the bare number and in three weeks you will have no idea which of your own comparisons were real.

The invented arithmetic above is not an exaggeration. In the 2020-2024 5-year file, Census Tract 67 in Clark County, Nevada — the tract under that Las Vegas Boulevard address — reports a median household income of $96,250 with a margin of error of $25,947 (table B19013), while Clark County as a whole reports $76,472 ±$643. The point estimates say the tract is 26% richer. The intervals — $70,303 to $122,197 against $75,829 to $77,115 — put the county figure inside the tract's range, which means you have found nothing.

What is running late right now, and why it changes your reading

This section will date faster than the rest, so: read 18 August 2026.

The newest local data covering the whole country is the 2020-2024 ACS 5-year, released 29 January 2026, available down to block group. The matching PUMS microdata followed on 5 March 2026. data.census.gov also got an interface refresh in March, so any walkthrough older than that will not match your screen.

The 1-year estimates are a different story this year. Normally, for anywhere with 65,000 people or more, you would reach for the 1-year file because currency beats precision when the question is what a metro is like now. In its 4 August 2026 advisory, release number CB26-129, the Census Bureau said the release date for the 2025 ACS 1-year estimates is being determined while it assesses a new Department of Commerce administrative order on disclosure avoidance and looks for a compliant way to publish those tables later in the year. The CPS income, poverty and health insurance figures are still scheduled for 15 September 2026, but those are national, not local.

So the practical consequence for anyone shortlisting towns this month: your freshest local numbers are a five-year average spanning calendar 2020 through 2024, with its center of gravity somewhere around 2022. A town whose rents moved hard in 2024 and 2025 will not show it here. Neither will a plant that opened last year.

Which is fine, as long as you know it. Treat the ACS as the structural picture — who lives there, what they do for work, what they own — and pull current prices from sources that publish monthly. Asking rents from live listings. Utility rates from the utility's own tariff sheet. Insurance from an actual quote, which is its own exercise and lives on the Risk & Insurance track.

One tab per town, filled in the same order every time

Order matters more than completeness here. Fill the fourth town's tab in a different sequence from the first and you will end up comparing a tract figure against a place figure without noticing. Mine goes source by source:

• From the geocoder — the place GEOID and the tract GEOID. If the address turns out to sit in no incorporated place, write "unincorporated" and stop pretending you have a city to compare.
• From DP05 — total population at both geographies, median age, share under 18, share 65 and over.
• From DP03 — the top three industries by share of employed civilians, and median household income at both geographies. Income goes in with its margin of error attached, always. It is the figure I have most often caught myself over-reading, and stripped of its MOE it looks far more decisive than it is.
• From DP04 — owner-occupied share and vehicles available per household.

Then one line that belongs to no table: which vintage you were looking at. 2020-2024 ACS 5-year, pulled 2026-08-18, typed into the tab. When the 2021-2025 file lands, everything above it moves a little, and the town will not have moved at all.

Doing this by hand is fine for three towns. By the fifth, script it against api.census.gov/data/2024/acs/acs5. A request without a key returned a "Missing Key" page when I sent one on 18 August 2026, so sign up first at api.census.gov/data/key_signup.html. Free, one minute.

Why the order matters shows up in two cells of one tab. For the downtown address above, the same 2020-2024 file puts Las Vegas city at a median household income of $73,877 ±$1,306 and its Census Tract 7 at $38,229 ±$14,149. One address, one geography apart, $35,648 between them — and this time the intervals do not touch, so the gap is real. Which is why the geography label goes in the cell beside the number, every time.

The twenty-minute version

Take one address you are seriously considering. Run it through the geocoder, copy both GEOIDs, open DP05 and DP04 for the tract, and write down four things: median age, share 65 and over, owner-occupied share, vehicles per household.

That is enough to tell whether the listing's description of the neighborhood and the Census Bureau's measurement of it are describing the same place. Sometimes they are. When they are not, you have just found the question worth spending the rest of the week on.

None of this is advice about where to live — I am not qualified to give any, and the ACS was not built to give it either. It describes places. It does not rank them, and neither do I.