AreaProfile

Editorial policy

Every claim on AreaProfile has to be traceable to a number in its database.

How statements are written

The sentences that appear on a profile — the quick read, the section summaries — are produced by a fixed set of rules. A rule has a numeric condition and produces a sentence built from the values that satisfied it. If the condition is not met, the sentence does not appear. Nothing is written to fill space.

Every generated statement stores the exact values that triggered it. An administrator can ask why a sentence appeared and get the numbers back.

Where the words come from

Every sentence about a place is produced by a rule with numeric conditions, from figures already stored and checked. A statement appears because the numbers behind it crossed a threshold, and the values that triggered it are kept alongside it. Nothing on a profile is composed freehand, and no figure anywhere on this site is modeled, imputed or guessed to fill a gap. Where a measure is not published at a location’s own geography it is either inherited from the geography containing it or aggregated from the smaller ones inside it — HUD publishes rents by ZIP area, so a city’s rent is the weighted mean of its own ZIP areas, and FEMA publishes risk by county, so a city’s share is allocated by population. About 20% of stored values are derived that way. Each one carries a note on the page saying how it was reached, and is stored marked as derived rather than measured, because a figure that arrives by arithmetic is a different kind of thing from one the agency published for that place.

Comparisons

AreaProfile does not say one place is better than another. It says where each one scores higher, on what measure, and against which cohort. Where a comparison involves a tradeoff, the tradeoff is stated rather than resolved — how much each side of it matters is a personal judgment.

Language

AreaProfile describes rather than promotes. “Building permit activity increased 18% during the latest measured period” is the register; “this city is booming” is not. Physical risk is stated in comparative terms — “above the national median” — rather than as alarm.

Publication threshold

Every place AreaProfile models has an address that loads, but only some are published, and only those are offered to a search engine. The decision is made by ten checks, run on every published candidate on every pipeline pass, and it is not a judgment call:

  1. Canonical geography resolved — the place has coordinates and a state.
  2. At least 40 measured values.
  3. At least 7 separate data categories, of ten.
  4. At least 3 independent federal sources.
  5. At least 40 percentile comparisons, so the numbers have comparative context.
  6. At least 2 data visualizations built from real series.
  7. An Area Score, or at least 4 signals in its place.
  8. A Quick Read of at least 2 observations, meaning the rules found something worth saying.
  9. No open critical data anomaly.
  10. For a state, metro or county, at least one published profile beneath it.

Two filters decide what is published, and it is worth being precise about which does the work. AreaProfile models roughly 70,000 US geographies. Curation selects the candidates for publication. A city needs about ten thousand residents. A ZIP area needs fifteen hundred, and it has to differ from the city containing it by at least ten percent on median household income or median home value — a ZIP that simply restates its city is a second page saying the first page’s numbers again. At most three ZIP areas per county from any one city are published, so a large city contributes its most distinctive few rather than all of them. That leaves 9,702. The ten checks above then run on each one, and a page that fails any of them is not offered to search — it stays reachable for anyone who lands on it, with every link working, rather than being deleted or padded out with prose to reach a word count. Which checks a page failed is recorded against it.

Curation is much the stricter of the two, which is why almost everything that survives it also passes the ten checks — 169 of 9,702 currently fail them. That is the point of the ordering: it is better to decide up front that a place does not warrant publication than to publish it and then dress it up. The count of pages is not the point, and the rule that keeps a page out of search also keeps advertising off it.

Advertising

Google’s advertising script loads on every page, and two advertising positions are mounted on a published profile — one after its jobs, schools and housing sections, and one below the coverage note. They stay empty until Google approves the site and issues the units. Advertising does not appear above the answer a visitor came for, inside search, inside the scoring controls, on a page that is not published or on a ZIP profile, or disguised as profile content: the first three are settled by where the positions are mounted, the fourth by the same rule that governs search, and the fifth by a label that ships with the unit rather than with the page. No advertiser will influence a score, a ranking or a metric, and no figure will be withheld or softened because of one. The privacy policy describes the cookies involved and how to decline them.

Corrections

See the corrections policy.