← Civic Value Atlas How it works · Sources · Limits

How it works

The question this map answers

Every county in America spends public money — on schools, health, public safety, and more — and every county gets results residents can feel. This map puts those two things side by side and asks a simple question: is a county doing better or worse than you'd expect for its circumstances?

Some places have every advantage and still fall short. Some face long odds and punch above their weight. The map's main view — "Performance vs Expectations" — is built to surface that difference: the counties doing better (or worse) than you'd expect once you account for the hand they were dealt and what they spend.

The two ways to color the map

• Outcomes view colors counties by how well they're doing on the things you've weighted with the sliders, from 0 to 100. Brighter = better outcomes. This is the straightforward "who's doing well" picture.
• Performance vs Expectations view colors counties by whether their results beat or fall short of what we'd expect given their circumstances and spending. Deep teal = doing clearly better than expected; deep rust = clearly worse; cream = about as expected.

How each county gets an outcome score

For each topic — education, health, safety, economic mobility — we line up all 50 states plus DC and rank every county against the others, from 0 (worst) to 100 (best). When a topic is measured by more than one number (health, for example, blends length of life and quality of life), we average those pieces. A county missing one piece still gets a score from the rest.

Safety is the one weighted blend: it's three-quarters homicide, one-quarter drug overdose. That mix reflects what most people mean by "safe" while still capturing the overdose crisis.

How we decide what to "expect" from a county

To judge performance fairly, we first need a fair expectation. Two counties aren't really comparable if one is a dense, wealthy suburb and the other is a small rural county with an older population — they face different costs and different starting points.

So for each topic we build a statistical model that learns the typical relationship, across all counties, between a county's results and a handful of things it can't change overnight:

• how much it spends on that topic (adjusted for local cost of living)
• how many people live there, and how densely
• how rural it is
• how old or young its population is
• its income level

The model uses those inputs to predict what a county "should" score. The gap between a county's actual score and its predicted score is the heart of the Performance vs Expectations view. A county well above its prediction shows up teal; well below, rust.

How good is the prediction?

No model is perfect, and we don't pretend otherwise. The table below shows how much of the difference between counties each topic's model actually explains. The rest — leadership, local institutions, history, plain luck — is what shows up as the gap.

Health is the most predictable: how long people live tracks closely with income and circumstances. Education is the least: how much a place spends on schools turns out to explain relatively little about test results once you account for income and demographics — a finding well established in education research. Lower numbers here aren't a flaw; they mean more of the story is left in the gap, which is exactly what the map is trying to show.

One honest caution: the gap is a comparison, not a verdict. A teal county isn't proof that its government runs better — it means its results are higher than the model predicted, for reasons the model can't see. Read it as "this place does better than similar places," not "this place is well governed."

Which spending we match to which result

We don't just throw a county's whole budget at every topic. We match each result to the spending most likely to affect it, and we lock those pairings in using published research before looking at the data — so we're choosing them on principle, not cherry-picking whatever happens to fit.

What we hold steady (the hand a county was dealt)

When we set each county's expectation, we hold steady the things its leaders can't change inside a budget cycle: population and density, how rural it is, the age mix of residents, and income level. We also factor in child poverty for health, safety, and mobility, because a county's starting poverty shapes those results in ways spending alone can't undo.

We deliberately don't hold poverty steady when judging schools — for education, poverty is too tangled up with the very thing we're measuring, so controlling for it would hide real differences. And we leave out things like residents' own education levels — too close to the outcome itself.

Does this tool use race?

No. The model never sees a county's racial makeup. It works only from outcomes, spending, and the structural circumstances described above — population, density, how rural a place is, its age mix, income, and child poverty.

But race shows up in the results anyway, and it's worth being straight about that. The plain Outcomes map tracks the country's racial lines: counties with larger Black, Hispanic, or Native populations tend to score lower on the things we measure. We didn't build that in — it falls out of the actual results. That isn't a flaw in the map. It's a measure of how real, and how deep, these gaps still are.

The Performance vs. Expectations view exists partly to separate the conditions a place faces from the place itself. Once we account for the circumstances a county can't change quickly — income, poverty, and the rest — most of that racial gap closes. That tells you how much of it is tied to those conditions. Some of it still remains. We read what's left not as a verdict on the people who live somewhere, but as the fingerprint of forces this map can't see — things like segregation and long-standing differences in resources. A county isn't doing worse because of who lives in it. It's carrying history the numbers don't fully capture.

Why tiny counties get smoothed

In a county of 3,000 people, two tragic deaths can swing a "rate" wildly — not because anything really changed, but because the numbers are so small. If we let those swings stand, a few tiny counties would dominate the brightest and darkest colors on the map.

So we gently pull small counties' gaps back toward their predicted value, more so the smaller they are. A county of 100,000 is left essentially untouched; a county of 10,000 lands about halfway between its raw figure and its prediction; a county of 1,000 is pulled most of the way back. Counties under 5,000 people also get a "treat with caution" flag in the detail panel.

Comparing a county to its peers

The "vs. similar counties" block in the detail panel finds the 20 counties most like the one you picked — closest in size, density, rurality, age mix, and income — and ranks your county against them. It's a fairer mirror than the national average: it answers "how do we stack up against places genuinely like us?"

Comparing within a single state

Flip "compare across" to your state only and each county is re-ranked against its in-state neighbors: 50 is the state middle, 0 is worst in the state, 100 is best. Why offer this? Because voters can change their own state's laws and school-funding rules, but not their neighbors' — so it's useful to see who's ahead under the same state policies.

How we count spending

"Local public spending" here means the money actually spent by every local government operating in a county — the county itself, plus cities, towns, school districts, and special districts. We add up what they spend directly on services and construction, and we're careful not to count the same dollar twice when one government simply passes money to another.

We show three flows separately, never mashed together:

• What gets spent — total spending per resident
• Paid by residents — local taxes and fees per resident
• Paid from outside — state and federal money received per resident

The difference between what comes in and what goes out is money going to reserves or paying down debt — or, if it's negative, a shortfall covered by dipping into savings or borrowing.

Two adjustments make counties comparable. First, we adjust every dollar for local cost of living, so a dollar spent in rural Mississippi and a dollar in coastal California count as the same real amount. Second, we average spending over five years (2018–2022), because one-time projects — a new school, a road rebuild — would otherwise make a single year look misleadingly high or low.

When one big taxpayer skews the numbers

A few small counties host a single giant taxpayer — a nuclear plant in Coffey County, Kansas; the oil fields of the North Slope, Alaska; a power plant in Beaver County, Utah. There, the per-resident spending figure is inflated by one company's tax bill and doesn't reflect what ordinary residents pay. We mark those places with a ★ "concentrated tax base" note so you can read their numbers with that in mind.

Counties whose boundaries changed

A handful of places redrew their boundaries recently, which would otherwise leave them blank on the map. We fill them in from the most closely matching older area, and note that it's an approximation:

• Connecticut replaced its 8 counties with 9 planning regions in 2022. Most data still reports under the old counties, so we carry each new region's values from the old county it most overlaps.
• Alaska's Valdez-Cordova area split into two in 2019; both halves inherit the old area's figures where newer data hasn't caught up.
• New York City's five boroughs share one city government, which reports its finances as a single unit. The boroughs share that citywide spending figure, while their results (schools, health, and so on) are still reported separately for each borough.

The colors

The palettes are chosen to be readable for colorblind viewers and free of political coding. The outcomes view runs cream → teal → navy (light to dark as results improve). The Performance vs Expectations view runs rust ↔ teal with a cream middle. We deliberately avoid red-green (hard for many people to tell apart) and red-blue (reads as partisan in a civic tool). The legend uses gentle language — "needs improvement / performing well," "below expected / above expected" — rather than blunt "worse / better" labels for places where people live.

Where the data comes from

Every number on the map traces back to one of these public datasets.

Census TIGER/Line 2025 — county shapes

tl_2025_us_county.zip
County boundaries for all 50 states, DC, and the territories, simplified for fast loading in the browser.

Stanford Education Data Archive (SEDA) 2025.1 — school achievement

Educational Opportunity Project at Stanford
School-level student achievement on a common national scale, pooled over the 2022–2025 school years. We place each school in its county and average up, weighting by the number of students tested. A small number of rural counties with no recent school data fall back to the 2009–2019 pool.

County Health Rankings 2025 — health (length of life)

countyhealthrankings.org
Life expectancy and early-death rates, plus county population and demographic figures used throughout the site, with a trend file back to 1997.

CDC PLACES 2025 — health (quality of life)

data.cdc.gov / swc5-untb
County-level rates of people reporting fair or poor health and frequent mental or physical distress. These are modeled estimates — the overall level is solid; small year-to-year moves should be read cautiously.

NCHS Drug Poisoning Mortality — safety (overdose)

data.cdc.gov / rpvx-m2md
County-level drug-overdose death rates. Modeled so that every county gets an estimate even where raw counts are too small to publish.

CDC Injury & Violence Mapping — safety (homicide)

data.cdc.gov / psx4-wq38
County-level homicide rates, 2019 onward, smoothed so that nearly every county has a value for every year.

Opportunity Atlas (Census Bureau + Opportunity Insights, 2024 update) — economic mobility

census.gov / Opportunity Atlas data tables
Measures how kids from lower-income families turn out as adults — specifically, the adult income rank of children whose parents were near the bottom of the income ladder. We use the 1992 birth cohort: kids who grew up in the 2000s, measured as young adults around 2019.

Census of Governments — local spending

census.gov/programs-surveys/gov-finances
Detailed finances, 2018–2022, for every county, city, town, school district, and special district. This is what we add up — carefully avoiding double-counting — to measure each county's local public spending.

BEA Regional Price Parities — cost-of-living adjustment

bea.gov
State-level price levels used to convert spending into comparable real dollars across the country.

Natural Earth — map water

naciscdn.org/naturalearth/50m/physical
Oceans and lakes drawn behind the county colors. Public domain.

What it can't tell you

A few things this tool doesn't solve, that a careful reader should keep in mind:

It describes; it doesn't prove cause

A teal county isn't proof that its decisions caused better results. It means its results are higher than expected for a place with its spending and circumstances. Anything the model can't see — a strong local institution, a particular history, a state program run from the capital — lands in the gap and can push a county's color either way.

Spending years and result years don't line up perfectly

Spending is a five-year average (2018–2022), while the results come from different years depending on the source. That's a deliberate trade-off — a steady spending baseline against moving results — but it does mean a 2023 result is being judged against a 2018–2022 spending average, not against 2023 spending.

Money the state spends directly isn't counted

When a state runs a service itself — state courts, state highways, some social services — that money never shows up in any local government's books, so it's missing from our county totals. States do this to very different degrees, so the spending figure understates total public investment more in some states than others. Be cautious comparing spending across state lines.

Federal payments to individuals aren't counted

Social Security, Medicare, food assistance, and the like are left out on purpose. They're income support flowing to people, not county services — including them would muddy the "is local government getting results" question this tool is built around.

Old debts and pensions can look like waste

Interest on debt, and contributions to underfunded pension systems, count as current spending even though they don't buy current services. A county weighed down by past obligations can look "inefficient" when it's really paying for decisions made long ago. These payments aren't separated out from spending on current services.

Counties dominated by one big taxpayer

Where a single plant, mine, or oil field dominates the local tax base, per-resident spending doesn't reflect what ordinary residents pay or get. We flag the clearest cases, but the flag is a rough rule of thumb, not a precise fix.

School districts that cross county lines

When a school district serves several counties, the source books all its spending to the county where the district is headquartered. In rural areas this can inflate one small county's school spending while shortchanging its neighbors.

The mobility data is a window into the past

Economic mobility tracks people born in 1992, measured as young adults around 2019. It's a historical signal, not a current readout — it reflects how a county launched kids who grew up there in the 2000s, not how it's doing for today's children. The detail panel makes this clear.

Tiny counties are smoothed, which hides detail

Because small counties' numbers are so jumpy, we pull their figures toward what's expected — which is honest, but means a genuinely excellent or struggling tiny county can look unremarkable on the map. The detail panel shows both the smoothed and the raw figures so you can judge for yourself.

Redrawn boundaries are approximated

Connecticut's new planning regions and Alaska's split areas inherit their numbers from the older areas they most overlap, rather than a precise town-by-town blend, so their figures are approximate.