Methodology

Two modes: by city, or by state

In city mode, the candidate pool is every US place above 25,000 population — currently 1,910 places (1,618 incorporated cities + 292 Census-Designated Places). In state mode, the candidate pool is all 51 states plus DC, with each state's metric values computed as a population-weighted rollup of its in-scope cities (for crime, disaster risk, unemployment, air quality, and health) or pulled directly from the source for state-native metrics (tax burden, BEA state RPP).

State mode does not score climate. Within-state climate variance — Texas spans humid subtropical to arid desert — makes a single state-level temperature target misleading. The form shows an info note explaining this when state mode is selected.

Per-metric percentile within the candidate pool

For every metric except climate, each city or state's raw value is converted to a 0–100 percentile rank against the same-mode pool — your results are scored against the cities or states they are actually competing with, not against an external reference. Six of the seven scoring metrics are minimize (lower raw value = better); only health outcomes is maximize. County Health Rankings publishes its z-score with lower = better, so we sign-flip it at data ingestion to keep the scoring layer's direction uniform.

Importance weights (0 to 10)

The seven importance sliders multiply each metric's percentile score. An importance of 0 excludes that metric entirely; 5 is the midpoint; 10 is the maximum weight. The total score is a weighted average across the metrics you weighted above 0.

At least one factor has to matter (some importance > 0, or a climate preference set) — otherwise we have no signal to rank with. The form catches this before submitting.

Climate is target-closeness, not percentile

Climate doesn't have a universal "better" direction the way crime or cost of living do. When you pick a climate preference, we score each city by how close its annual mean temperature is to the target — 70°F for warm, 60°F for temperate, 45°F for cold — over a 50°F full range. A city exactly at target scores 100; a city 50°F off scores 0. The implied weight for climate in the total score is 5, the midpoint of the importance scale, when a preference is set; nothing when it's set to "No preference."

Housing budget filter

Before scoring, we filter out cities (or states) whose median housing exceeds your monthly budget under a buffer — 10% in city mode, 25% in state mode. State mode uses a larger buffer because within-state housing prices vary widely. For buy mode, we estimate monthly payments with a 7% annual rate, 20% down, 30-year fixed. For rent mode, we use median rent directly. The candidate pool you see at the top of the results panel reflects how many cities or states actually made it through the filter.

When a metric is missing for a city

Federal data is uneven. Some metrics — like Census population — cover every city. Others — like EPA air quality monitors — only cover a subset of counties. We handle missing data in a three-step fallback so the scoring stays comparable across cities even when data sparseness varies:

1. First choice: the most-precise published source. For most metrics that's a county or metro-area number. (Confidence: high.)

2. If that's missing: a state-level fallback. For air quality, that means averaging the monitored cities in the same state. For cost of living, BEA publishes a state-level price index we substitute directly. (Confidence: medium. A flag appears on the result card explaining what was used.)

(A subtle point: when we compute a state-level average, we only include cities that have real measurements at the more-precise level. Cities that received the state-average fallback themselves don't feed back into the state aggregate — that would be circular. This keeps the fallback's confidence honest: a “medium-confidence” state average means N real measurements averaged, not N + M echoes of itself.)

3. If neither is available: the metric is excluded for that city. The city competes on its other metrics; the result card's Notes section lists which factor was unavailable.

Cities don't get a free pass on missing data — the medium-confidence flag and the Notes section are explicit signals that the score is less precise than a fully-measured city's.

City coverage

The current pool is every US place with population at least 25,000 — 1,910 places (1,618 incorporated cities + 292 Census-Designated Places) across all 51 states plus DC. State-level rankings cover all states uniformly. Smaller places aren't in scope today; expanding to a roughly 5,000-city pool that includes cities down to 10,000 population is on the roadmap.

If your top results cluster geographically, here's why

Several metrics in this tool are published by the federal government at the county or state level, not the city level. Air quality, health outcomes, disaster risk, and unemployment come from county-level sources — every city in a county receives the same number. Tax burden and (when the state-level fallback fires) cost of living come from state-level sources — every city in a state shares those values.

That has a visible consequence in the rankings: when the metrics you weight most happen to favor a particular county or state, you'll see several cities from that area cluster in your top results. A top-5 with three South Dakota cities, for example, isn't a scoring bug — those cities share their state's tax burden and (often) their county's health, disaster, and air quality scores. The scoring is working correctly against the data we have.

To see what actually differs between cities clustered this way, look at the per-metric bars on each card. The metrics where the cities diverge — cost of living, crime, climate, housing cost — are where one city distinguishes itself from its neighbors.

You see this most directly in the “Nearby cities” section under each winner card. In city mode across all states, we show the four geographically-closest other cities within 100 miles of each winner, scored under your same preferences. Their totals often land close to the winner's because they share the same county-level data (health, air quality, unemployment, disaster risk) and sometimes the same state-level data (tax burden, cost of living when the state-fallback fires). Where a winner's per-metric bars diverge from its neighbors shows which metrics distinguish that city locally.

In states with many adjacent CDPs sharing a county, a state-filtered search can return several similarly-scored nearby places — they share the same county-level health, disaster risk, unemployment, and air quality values. The per-metric bars and the housing-cost columns are where these clustered CDPs distinguish themselves. See the Census-Designated Places card below for more on how CDP metrics work.

Census-Designated Places (CDPs) are included by default

About 290 of the ~1,900 places in our pool are Census-Designated Places — unincorporated communities the Census Bureau tracks as distinct places even though they aren't legally incorporated cities or towns. They're common in metropolitan suburbs and in states with few incorporated municipalities, where many people actually live.

Because CDPs aren't legal cities, most data sources don't publish city-specific numbers for them. We inherit their metrics from the surrounding county or metro area — air quality from the county's EPA monitor, health outcomes from the county's CHR rank, cost of living from the metro's BEA index, and so on. Each CDP result card carries an “Unincorporated area” badge near the city name and a sentence in the justification explaining that the metrics reflect the surrounding area rather than the place itself.

If you'd rather see only legally incorporated cities, check “Show incorporated places only” in the form. By default this checkbox is off — we surface CDPs because (a) measuring showed they don't crowd incorporated cities out of all-states results in practice, (b) some state-level searches depend almost entirely on CDPs (states with few or no incorporated municipalities), and (c) the per-card badge and disclosure sentence let you tell at a glance which results are CDPs.

Some scores come from sparser data

Every metric value carries a confidence tier (high / medium / low). High = direct measurement at the city or county geography that matches our pool. Medium = a fallback method, like substituting a state-level average when a city has no county-level monitor data. Low = no data.

Today, when a city uses a fallback method, its result card shows a confidence indicator on the per-metric bar for the affected metric, plus an entry in the “Notes” section explaining what was substituted.

Different "as of" dates across sources

Federal data publishing cadences vary — BLS annual averages lag by a year, NOAA climate normals refresh every decade, FEMA NRI is updated less often than EPA AQS. The per-source dates above reflect each one's most recent publication at acquisition time; we don't hold back fresh data to align dates artificially.

• Per-capita FEMA disaster risk variant, to discriminate better between large urban counties.
• NREL National Solar Radiation Database for the sunshine / clear-days signal that the 1991–2020 NOAA Climate Normals product dropped.
• Expansion to a roughly 5,000-city pool including cities down to 10,000 population.