Quiet Corners

The quiet corners of America

Quiet Corners scores rural U.S. counties on stability, remoteness, sprawl risk and affordability, and surfaces the ones that are staying quiet and are likely to keep staying quiet. The methodology below sets out how stability, remoteness and sprawl risk are computed and how they roll up into the overall score, and the data trust model names the source behind each input. The dashboard below is public and free to use in your browser — no account, no signup.

Every data point comes from the U.S. Census Bureau or USDA, with one disclosed exception: interstate proximity (30% of the sprawl-risk composite, below) is measured against a hand-maintained waypoint list, not a government dataset. No commercial APIs, no scraped listicles, no vibes.

Open Explore, at the top of the panel column beside the map (under the map on a phone), to filter counties by stability, remoteness, sprawl risk or overall score. The same panel holds the results table, so the counties that match are right there below the filters.

Methodology

Independent scores, each 0-100:

Stability — Coefficient of variation (Bessel-corrected) plus linear regression slope across 5 years of population estimates. Low CV + flat slope = high stability. Weighted 60/40 CV-to-slope.

Remoteness — Great-circle distance from county centroid to nearest city with population over 25,000. Uses the full Census place gazetteer — capturing regional hubs, college towns, and county seats that rural residents actually drive to. The pipeline exports this as raw miles (distance_to_metro_mi), not a normalized 0-100 score; the dashboard’s Remoteness filter slides across that raw distance, up to 250 miles, and slides all the way down to Any, which switches it off: a county ten miles from a small city can be as quiet as one ninety miles out, and a county whose distance was never computed is not evidence that it is close to anything.

Sprawl Risk — A five-factor composite, all five live on this run: interstate proximity (30%), building permit trend (25%), state migration acceleration (20%), fast-growing metro proximity (15%), and recreation/retirement county typing (10%). Lower is better. Permits and metro proximity used to be dead — the Census API retired both endpoints — and are now read from the keyless bulk files on www2.census.gov instead (Building Permits Survey county annual files; PEP subcounty estimates).

A county missing an input is scored over the factors it does have, with the remaining weights renormalized, so a gap cannot quietly deflate a score toward “no sprawl risk”. A factor that scores 0 from real data still counts at full weight.

The trust panel on the dashboard lists, per factor, how many counties it scored above zero on the run that produced this page — read from the data, so it stays true across regenerations. Those counts are not coverage rates. A factor reads 0 both when its input is missing and when the county genuinely has no pressure from it: most rural counties carry no USDA recreation or retirement type at all, migration scores 0 for every state whose inward flow is flat or falling, and a county more than 100 miles from an interstate scores 0 there by construction. Permits are the one factor where a low count is partly a real gap — many rural counties have no usable multi-year permit series in the survey’s county files.

Overall — Quietness 55% and Affordability 45%. Quietness is itself stability 40%, remoteness 30%, and inverted sprawl risk 30%. Both levels renormalize over the inputs a county actually has, and a county with nothing scoreable gets no score rather than a zero. Livability was removed: it returned the affordability score verbatim, which double-weighted affordability while advertising a dimension that did not exist.

Counties must have a USDA Rural-Urban Continuum Code of 6 or higher (nonmetro with urban population under 20,000) to appear in results. Within qualifying counties, candidate towns (incorporated places with population 1,000-3,000) are surfaced with scores inherited from their parent county.

Reading the map

The county shading and the map key show one measure at a time, and the Color the map by control chooses which: Stability, Overall, Quietness, Affordability, or plain miles to the nearest city. The key relabels itself in that measure’s own units — a distance key reads in miles, not on a 0–100 ramp — and outlines and dot sizes follow the same bands, so the map is never colour alone. It is separate from how the tables sort (Overall for counties, Stability for candidate towns); changing the colour changes no filter and drops no row. A county whose value was never computed is drawn in its own grey and listed as Not measured, rather than shaded into the bottom band where a gap in the data looked like a low result.

The map names two kinds of place. The towns it scores come from the candidate export, and they are small by definition — so the country-wide view also names the large places you would orient by, from a separate list of 347 cities that exists only to say where the small towns are. Those names are capped so they frame the map rather than fill it, and they carry the everyday form of the name rather than the Census one.

Each table says what it is ranked by, in words — Sorted by Overall, highest first — and the heading it names carries a caret and an underline, not just a colour. The county table and the candidate towns below it sort independently, and neither follows the map’s colour. One line above the tables says all three at once — Map color: Stability · County sort: Overall · Town sort: Stability — because Stability is a component of Overall rather than a rival ranking of it, so a county can shade dark on the map and still sit low in the table. It is written from the live selection and follows every change you make to it.

The page sets figures and words in different faces on purpose. The score columns stay monospace, because seven columns of digits only line up when every glyph is the same width. Everything you read rather than compare — filter labels, column headings, the map key, the sentences above the tables and the provenance line under the title — is in the ordinary interface face, and the county’s own name is the largest thing in its row.

On a phone each row becomes a card instead of a table line, and every measure comes with it. The tables used to drop affordability, remoteness and sprawl below 1024px while the filters that set them stayed — so the figure you had just filtered on was the one off screen. Each value now carries the column heading beside it, read from the table’s own header, so a card can never label a number with the wrong measure.

Share view copies a link to exactly what the map is showing now — center, zoom, base layer, surface and any active overlays — as a #map= fragment in the URL, so pasting it elsewhere reopens the same view. The address bar keeps that fragment in sync as you pan and zoom once a shared link has been opened, and the browser’s back and forward buttons move through those views. Reset view clears the map back to its default and forgets your saved base-layer choice; it never touches a print recipe, a saved print project, a chosen area, or your saved places — those belong to the Field Maps builder or live independently of any view reset.

Saved places lets you bookmark a named place from its details card and keep it in a list you can revisit, right in your browser. It sits with tonight’s sky and forecast below the map, and carries its own Download, Import and Clear controls; it never affects a shared link, a print recipe, or a saved print project. A road or park camera’s card has a Save camera button too: it keeps only the camera’s name, the operator and its position, never a picture or a stream address. Show on a saved camera turns that camera’s layer on and opens its card; if the operator’s current list no longer carries it, the row says so and opens nothing. A saved camera with no position (an imported list can carry one) stays in the list but cannot be shown on the map or checked for conditions, and says why. Cameras have no Share button.

Conditions answers what the sky and the forecast do at the exact place you last picked — a county, a town, or a named summit, gap, ridge or populated place. Asking is explicit: opening a named place’s details card never starts a request on its own, and its card’s own Before you go button — also offered on a saved place’s row — is what asks. A Conditions for header above tonight’s sky, the forecast and the alerts names the place and states what each of the three sources actually covers: the Sun and Moon are calculated for that exact point; the forecast is the National Weather Service’s value for the grid cell that contains the point, not a reading at the point itself, with a further note on a summit, ridge or gap that the cell’s one value is not a measurement on the feature; and the alerts are the Weather Service products whose area covers the point. Opening a different place’s card while an earlier place’s conditions are still on screen asks nothing new — the header instead says whose conditions those still are, so you never read one place’s weather under another’s name.

Under the alerts, Cameras nearby lists the road or park cameras within 25 km, straight-line, of the place you asked about — but only cameras from a layer you have already turned on in the Field Maps panel. It never turns a layer on and never asks a camera operator for anything by itself; pressing a camera’s name is what opens its card on the map. A camera 25 km away is not the same weather across a ridge, and the list says so.

The same map carries the Field Maps builder’s Layers panel — base map choice, a names-and-search layer, a Places in this view list, display controls, and a set of starting presets, one of which, Compare counties, turns on this page’s own county-ranking wash (off by default) through the same switch the Layers panel’s Land, sky & light context group already uses for every other context surface: there is no separate toggle for it. Undo and Reset sit beside the presets for backing a shortcut out again. See the Field Maps page for the full list and what each one moves.

On a wider screen the four panels — Explore, Conditions, Layers and Print — sit in one column to the left of the map, one open at a time, so none of them ever covers the map; the column is the same width whichever panel is open, so the map never shifts under you, and Explore’s county and town tables show each place as a small card with every figure labelled instead of a wide table. On a phone, opening any panel — Explore, Conditions, Layers or Print — opens it in a sheet under the map instead of over it, so the map stays visible and never blank. Drag the sheet’s handle up or down, or use its Expand/Shrink button, to see more of the panel or more of the map; Close or Escape returns you to whichever control you opened it from.

Use my location takes one click and asks your browser for your position exactly once — it never watches your location in the background. Nothing about that fix is stored: not in the address bar, not in your browser’s saved preferences. If the fix is inside the area this map covers, the map pans and zooms to it and drops a marker with an accuracy circle, both held only in memory for that visit. A location-derived view is left out of a shared link unless you explicitly tick “Include my location in the link” first; Share view’s status line says whether it was included. Finding your location still means asking the map’s tile provider for the area now on screen, the same request any pan or zoom already makes.

Search places matches as you type, against the same town export the map scores. Typing a coordinate pair instead jumps straight to it. Neither reaches outside your browser — both search a list already on the page. The box, its suggestions, its notes and its Search button share one solid panel, so no map label shows through the text. Type a street address (a house number and street, a lone apartment or unit, or a ZIP code) and nothing is sent while you type: the box says to press Search. Pressing Search (or Enter) then sends what you typed, through this site, to the U.S. Census Bureau’s address geocoder, and lists up to five street matches under “Street addresses (estimated)”, each marked “estimated along the street, not a doorstep”: the point is interpolated along the street’s address range, never a household. Picking one moves the map there and the conditions heading names the matched address followed by “(street estimate)”; like picking a county or a town, it asks the National Weather Service for the forecast and alerts at that estimated point, rounded to four decimal places. When no street matches, it says so, asks you to check the spelling, and offers a town or named place, coordinates as latitude, longitude, or dragging the map to the spot. When the lookup is unavailable (the service is down, busy, slow past ten seconds, or you are offline) it says address lookup is unavailable right now and offers the same alternatives. A place already in the town list wins, and then nothing is sent. A newer search cancels an older one and a late answer is dropped. The typed address is never written to the page’s link, a shared link, saved views, browser storage, the page title or analytics. A name that only looks like an address (“Lake Unit 4”, “Highway 441”) is searched like any other name, and an address never reports a missing town or swaps in a nearby one. Search is the second, explicit search you ask for with its own button or by pressing Enter with no suggestion highlighted, and it asks the U.S. Geological Survey’s own gazetteer (GNIS) for named summits, gaps, ridges and populated places by name — up to 80 characters, optionally narrowed to a state the way “Ashland, IL” narrows a town search. A result too broad to list in full says so and asks you to narrow the name or add a state, rather than claiming there were no matches. A geographic name in the results carries a badge naming its kind — Summit, Gap, Ridge/range and the rest — alongside its state and county, right in the merged list with the towns and coordinates, so a peak reads as a peak before you open it. Picking one moves the map to a view sized for what it is and opens the same details card a map pin opens, naming it, its kind, its state and county, its USGS GNIS source and its coordinates, its approximate terrain elevation fetched from USGS 3DEP once you open the card, with buttons to save the place, copy the coordinates, or copy a shareable link to that place. The map stops where that card can be read: clear of the buttons laid over the map (and, on a phone, the panels) where there is room, and where there is not, with as much of its name clear as the space allows, then as much of the rest of the card. Show on a saved named place aims the same way, but the panel you pressed it in stays open, so the card has less room.

Tab or Shift+Tab out of the open search closes it, as its button does, so the search never sits over the control you moved on to. Dragging the map leaves it open with what you typed.

Data trust model

Every figure on this page is regenerated from source by a pipeline that refuses to invent a number it does not have — but it discloses gaps two different ways depending on the dimension. Stability, remoteness and affordability report a missing county’s field as absent (null) rather than defaulted; the coverage panel states the real scored percentage for each. The sprawl sub-factors instead report a per-factor non-zero count in the sprawl-factors panel, and a factor with no input for a county is dropped from that county’s composite with the remaining weights renormalized — never defaulted to a convenient number. The generation timestamp is stamped into the exported data rather than typed by hand.

Sources the pipeline draws on:

• County populations: Census PEP Vintage 2024 (2020-2024)
• County coordinates: Census Gazetteer 2024
• Nearest cities: Census Place Gazetteer 2024 (all places >25K)
• Rural classification: USDA Rural-Urban Continuum Codes 2023
• County typing: USDA County Typology Codes 2015 (recreation/retirement flags)
• Interstate proximity: Static, hand-maintained waypoint list along major US interstates (422 points across 30 route chains, no duplicates), with distance measured to the nearest route segment rather than the nearest point — not a government dataset
• County geometry: us-atlas counties-10m.json (ISC License, © the us-atlas project) — a TopoJSON build derived from Census TIGER/Line cartographic boundaries, served from this domain rather than a third-party CDN
• Building permits: Census Building Permits Survey, county annual bulk files (www2.census.gov/econ/bps/County/) — the API timeseries endpoint is retired; these keyless files replace it
• Fast-growing metro proximity: Census Subcounty Population Estimates bulk file (SUB-EST2024) — the PEP place API endpoint is retired; this keyless file replaces it
• State migration: Census ACS 1-Year B07001 (Geographical Mobility)
• Affordability: Census ACS 5-Year (median home value, gross rent, household income)
• Candidate towns: Census Subcounty Population Estimates 2024 (SUB-EST2024)

No data is sampled, simulated, or interpolated beyond filling the 2001-2009 intercensal gap with Census-published CSV data. What you see is what those sources reported.

The county map, its geometry and the scores it shades by are served from this domain, with no third-party scripts and no analytics on it. This page shares its Layers panel with Field Maps: turning on a base map or a context layer there requests tiles or data directly from that layer’s own service, described on the privacy page.