In 2015, Anne Case and Angus Deaton noticed something that was not supposed to happen in a rich country: mortality among middle-aged white Americans had stopped falling and begun to rise, driven by suicide, drug overdose, and alcoholic liver disease. They called these deaths of despair, and the phrase stuck because it carried a theory inside it — that the deaths were the visible end state of something psychological, a corrosion of life that statistics on income and unemployment were not catching.
But a death certificate records only the end. If despair is real, it should be measurable upstream, in the living — and for a decade Gallup measured almost everything a living person can report. From 2008 through 2017, the Gallup US Daily poll asked roughly 350,000 Americans a year whether yesterday brought physical pain, sadness, worry; where their life stands on a 0–10 ladder; and where they expect it to stand in five years. Pooled, that is 3.5 million interviews, enough to build reliable portraits of how daily life feels in 1,708 individual counties — 1,299 of which also appear in the county-level deaths-of-despair mortality data assembled from CDC records.
That makes a precise question answerable. Of all the things people can tell a pollster about their inner lives, which self-report actually lives in the same geography as the dying? Case and Deaton's account makes a specific prediction: despair is not generic unhappiness. Their story runs through chronic pain, disability, and the collapse of the futures that working-class life once promised. If they are right, county despair mortality should correlate more tightly with reported physical pain and with foreclosed expectations than with sadness, worry, or how people rate their lives today.
It does — and the ordering of what tracks and what doesn't is the cleanest empirical sketch of what "despair" means that these data can draw.
Three different questions about a life
Well-being researchers are careful to keep three constructs apart, and this analysis depends on the distinction. Experiential measures capture yesterday's lived texture: did you experience physical pain, sadness, or worry during a lot of the day? (In Gallup these are yes/no items; nationally, 24.0% of respondents said pain, 16.6% sadness, 29.0% worry.) Evaluative measures ask you to step back and judge: the Cantril ladder, 0 for the worst possible life, 10 for the best, on which the average American stands at 7.0. And a third, future-evaluative question asks where your life will stand in five years — from which we define a "low future" as a rung of 4 or below, a future in the bottom half of the ladder. Nationally only 10.1% of Americans place their five-year future that low, which is what makes the geography of where many do so striking.
For each of the 1,708 counties with at least 300 interviews, we computed the share reporting each feeling, mean ladder positions, the share ever diagnosed with depression, and a felt-despair composite: the average of the county's standardized scores on pain, sadness, worry, and low-future shares. Then we asked how each lines up, county by county, against deaths of despair per 100,000 residents.
Where despair is died and where it is felt
Bivariate map: each county is classed by its tertile of deaths-of-despair mortality (purple, ↑ vertical in legend) and its tertile of the selected felt measure (teal, → horizontal). The dark diagonal is concordance — places where feeling and mortality agree. The saturated corners are the discordant places. Click a legend cell to isolate those counties; hover any county for values and sample size. Light counties: under 300 interviews or no mortality data.
The diagonal dominates: Appalachia, the Ozarks, the Deep South, swaths of the Mountain West and the Rust Belt run dark in both dimensions, while the urban Northeast corridor, the upper Midwest farm belt, and the big metro suburbs run light in both. But the corners exist, and they are not noise — we will name them below.
Which feeling tracks death?
The headline test is a ranking. For each measure we computed the Spearman rank correlation with county despair mortality across the 1,299 counties that have both, with confidence intervals from 2,000 bootstrap resamples of counties. Rank correlation asks the bluntest possible question — do the counties line up in the same order? — and is immune to outliers and scale choices.
The despair-mortality leaderboard
Spearman rank correlation between each county-level measure and deaths of despair per 100,000, across 1,299 counties. Whiskers: 95% bootstrap CIs (2,000 county resamples). Open circle: average correlation computed within rural–urban (RUCC) bands; open diamond: within county-income terciles — the correlation that survives after comparing only similar kinds of places.
Physical pain wins, at ρ = 0.576 (95% CI 0.536–0.614). The felt-despair composite comes second (0.538), then diagnosed depression (0.515) and the low-future share (0.507). Then a gap. Sadness — the feeling the word "despair" seems to name — manages only 0.395. Worry, the most commonly reported negative feeling in America, is the weakest experiential signal of all at 0.256. The county's average rating of life today, the workhorse measure of the happiness literature, correlates at −0.403 — real, but far from the top. Tellingly, the average rating of life in five years tracks the mortality map more tightly (−0.482) than the rating of life today: where people are dying of despair, the distinctive evaluative signature is not present misery but a missing future. None of the bootstrap intervals for pain overlap those of sadness or worry: the ranking's top and bottom are statistically separated, though pain, composite, depression and low-future cluster together at the top.
The magnitudes are plain. Walk from the bottom quintile of despair mortality (under 41.5 deaths per 100,000) to the top (above 71.2), and the share of residents in physical pain yesterday rises from 22.3% to 29.9% — a third more pain. The share who place their five-year future at 4 or below rises from 9.3% to 13.6%, nearly half again as much foreclosed future. Diagnosed depression climbs from 15.2% to 20.4%. The current ladder, meanwhile, slips only from 7.08 to 6.87 — a fifth of a rung. A pollster armed only with the standard life-satisfaction question would barely see the despair belt at all.
Is it just rurality, or poverty?
High-mortality counties are disproportionately rural and poor, and pain is more common in rural and poor places. So perhaps pain "tracks" mortality only because both ride along with county type. To check, we recomputed every correlation inside rural–urban bands (metro RUCC 1–3, micropolitan 4–6, rural 7–9) and inside county median-income terciles, averaging the within-stratum correlations weighted by county count — comparing metro counties only to metro counties, poor only to poor.
The structure survives. Within RUCC bands, pain still correlates at 0.533, the composite at 0.484, low-future at 0.428, while sadness (0.357), worry (0.242) and the current ladder (−0.343) stay behind. Within income terciles the same: pain 0.468, composite 0.407, low-future 0.375, against sadness 0.249 and worry 0.180. The ordering — pain and foreclosed futures first, generic negative affect last — is not an artifact of comparing Appalachia to Manhattan. If anything the pain–mortality link is sharpest among similar, better-off places: within the top income tercile alone, pain correlates with despair mortality at 0.637, and within metro counties at 0.590.
The same gradient inside similar lives
County correlations are facts about places, not people; high-mortality counties are also older and less educated, and pain rises steeply with age. So we cut the individual data the other way: take the 1,299 analysis counties, sort them into mortality quintiles, and compare individuals within the same age band and education level across those quintiles. Does living in a high-despair county predict feeling worse even among observably similar people?
Feeling worse in high-mortality counties — within age × education cells
Each panel is one age band × education cell. Lines show the share reporting the selected outcome by county despair-mortality quintile (Q1 lowest → Q5 highest). All 45 cells exceed n = 10,000 individuals; hover for exact values and n.
Yes — and the gradient is steepest exactly where Case and Deaton's mortality is. Among 35–54-year-olds with a high-school education or less, the share in physical pain yesterday climbs from 28.6% in the lowest-mortality quintile of counties to 37.6% in the highest — nine percentage points, within a single demographic cell. Their low-future share climbs from 10.4% to 13.4%. Among young college graduates the same walk moves pain only from 11.1% to 12.8%. Pain is higher in the top quintile than the bottom in all nine demographic cells, but the slope concentrates in midlife and at lower education — the precise demographic in which deaths of despair surged. (This is a robustness cut, not a causal claim: people sort into places, and places shape people, and a cross-section cannot split the two.)
The places that break the pattern
A rank correlation of 0.58 leaves room for places where the two maps disagree, and the disagreements are instructive. Define discordance two ways: counties in the top mortality quintile whose felt-despair composite sits in the bottom half (45 counties), and counties in the top felt-despair quintile whose mortality sits below the median (51 counties).
Discordant counties: dying without reporting it, hurting without dying
Each dot is a county (n = 1,299): deaths of despair per 100,000 vs. the felt-despair composite (county z-score average of pain, sadness, worry, low-future). Purple: top-quintile mortality but below-median felt despair. Teal: top-quintile felt despair but below-median mortality. Guides mark the top-quintile mortality cutoff and the felt-despair median. Hover any dot.
The high-mortality, ordinary-feeling counties cluster in the Mountain West and Alaska: Fremont County, Wyoming (157.7 deaths per 100,000 — fourth-highest in the data — yet a felt composite of −0.77, well below the median, across 729 interviews), Anchorage (80.8 and −0.90, on a hefty 4,188 interviews) and Juneau, Alaska, Carson City, Nevada (121.0, −0.51). These are places where the mortality is heavily suicide- and alcohol-shaped — the western pattern of guns, isolation, and altitude — and where the living report unremarkable days. Whether that reflects stoic under-reporting, a frontier sorting of who stays, or despair concentrated in a small subpopulation the survey reaches rarely, the deaths there arrive with little statistical warning from the living.
The reverse corner — places that hurt loudly but die at ordinary rates — is almost entirely the small-city South: Henry County, Virginia (felt composite +1.36, with 32.7% of 1,002 respondents in pain yesterday, but mortality of just 36.5), Greene County, Tennessee (+1.99, 35.6% in pain), Starr County, Texas (+2.62 — a felt-despair score exceeded only by a few Appalachian counties — against mortality of just 40.2), Cherokee County, South Carolina, and Acadia Parish, Louisiana. These are largely poor, often majority-minority or deep-rural textile and farm counties — places long familiar with hardship whose suffering had not, at least by the late 2010s, converted into the suicide-and-overdose mortality signature. Case and Deaton noted that Black and Hispanic Americans were initially spared the mortality surge even where material conditions were worse; Starr County, overwhelmingly Hispanic on the Texas border, in pain and pessimistic but not dying of despair, is that finding drawn on a map.
What the ranking means
Three readings, in increasing order of caution.
First, the result is a validation. "Deaths of despair" was christened from mortality tables, and skeptics have asked whether despair — the psychological state — was ever shown to live in the same places. Here, the subjective data line up with the mortality data precisely along the dimensions the theory specified: chronic pain (Case and Deaton devote chapters to it) and the evaluative collapse of the future, not garden-variety unhappiness. Graham and Pinto, working with the same Gallup files, called lost hope a marker of premature death; Blanchflower and Oswald showed extreme distress rising in exactly these populations. The ranking above is the county-level version of both findings in a single ordering.
Second, it is a measurement lesson. The current-life ladder — the question behind most national well-being rankings — is among the weaker trackers of America's most consequential well-being crisis, and worry, the most prevalent negative feeling, is the weakest of all. Worry is apparently what prosperous, striving places feel; pain is what dying places feel. If governments build dashboards to catch the next despair epidemic, a single life-satisfaction number will not catch it. Yesterday's pain and the five-year ladder will.
Third, the caution. Everything here is ecological: these are correlations between county aggregates, and they license claims about places where people hurt and die, never about whether the individuals who report pain are the individuals who die. The mortality measure is a single ~2018-vintage snapshot joined to all ten survey years, so this is a comparison of a decade of feeling against one late-decade mortality map, not a dynamic account. Counties are assigned from ZIP codes, which blurs some county lines; counties under 300 interviews are suppressed; the mortality file covers only 1,341 larger counties; and Gallup's file carries no survey weights, so every number is an unweighted sample estimate. The ranking of correlations is robust to all of these in the cuts we could run — but the corners of the map are flags for reporting, not verdicts.
Case and Deaton gave the crisis its death map. The living, it turns out, had been drawing the same map all along — not in their sadness, which is everywhere and nowhere, but in their bodies and their abandoned futures. Despair, measured where it can still be asked about, is pain plus the death of expectation. The data have been saying so, three hundred and fifty thousand interviews a year, for a decade.