The Flip Counties Felt It First

Roughly two hundred American counties voted for Barack Obama in 2012 and for Donald Trump in 2016, and on the strength of that pivot they became the most over-interpreted places in the country. One camp read them as casualties of trade shocks and deindustrialization — the "economic anxiety" thesis, backed by work like Autor, Dorn, Hanson and Majlesi's study tying import competition to political polarization. Another camp, following Diana Mutz's status-threat findings and Sides, Tesler and Vavreck's Identity Crisis, argued the flip was about identity and relative position, not pocketbooks. The argument has never really been settled, partly because almost all the evidence is either aggregate — county unemployment rates, factory closures, mortality statistics — or collected after the fact, when 2016 had already happened and memories had rearranged themselves around it.

There is a rarer kind of evidence. From 2008 through 2017, Gallup's U.S. Daily survey called hundreds of Americans every night and asked them, among other things, to place their life on a ladder from zero to ten, to say where they expected to stand in five years, and whether their standard of living was getting better or worse. The interviews are geocoded to the county. That makes it possible to do something the post-mortems could not: listen to what people in the flip counties said about their own lives, year by year, before anyone knew their counties would flip.

This article does that, with a specific hypothesis on the table: that flip-county residents' assessments of their current lives looked unremarkable — no deeper misery than demographically similar places — but that their trajectory measures, the questions about the future and the direction of travel, sagged through the early 2010s. Not despair; foreclosed expectation. The data mostly cooperate, with one twist worth being honest about: the sag is real, it is concentrated in 2011–2014, and by election year it had already begun to close.

A telescope, not a post-mortem

The raw material is 3,172,536 Gallup interviews conducted from 2008 through 2016 in counties whose presidential votes can be matched to MIT Election Lab returns. Of those, 194,153 interviews — about 6 percent — took place in one of the 223 counties that went Obama in 2012 and Trump in 2016. (The file extends into 2017, but this piece is strictly about the run-up, so post-election interviews are excluded throughout.) One important limitation up front: this extract of the Gallup file ships no survey weight, so every number here is an unweighted sample estimate. Patterns and contrasts between groups are the point; exact national levels should not be quoted from this page. And a second limitation that the whole article should be read through: this is descriptive journalism, not a causal study. Nothing below says the flip was caused by anything.

The geography explains why the comparison group matters. Flip counties cluster along the upper Mississippi, the Great Lakes, and the rural Northeast — Iowa alone has dozens — and they are disproportionately small-metro and rural, middling in income, and far from the coastal metros that dominate any "rest of America" average. Comparing Howard County, Iowa to the United States at large would confound place with everything else.

Finding the right mirror

So the flip counties are compared here to non-flip counties that look like them on three structural dimensions, none of which involve anyone's survey answers: urbanicity (USDA Rural-Urban Continuum codes, banded 1–3, 4–6, 7–9), census region, and tercile of county median household income. Crossing those produces 33 cells that contain at least one flip county. The comparison pool is every interview from a non-flip county in one of those cells — 2,943,451 interviews across 2,788 counties — reweighted so the pool's distribution across cells exactly matches the flip group's. In plain terms: when a flip respondent lives in a lower-income rural Midwestern county, she is compared to people in lower-income rural Midwestern counties that didn't flip. This is coarse, deliberate, and descriptive — a mirror, not an identification strategy. A toggle on the next chart swaps in the unmatched all-non-flip comparison, which tells a similar story with slightly different levels.

Read the top panel first, because its dullness is the finding. In 2008 flip-county residents put their lives at 6.53 on the ladder; the matched comparison said 6.58. Both groups slumped with the financial crisis, recovered, and drifted upward together for eight years — by 2016 the flip counties stood at 7.05 and the matched counties at 7.12. The gap wobbles between roughly 0.02 and 0.08 ladder steps and never grows. If you were hunting for unusually miserable places, you would scroll right past these counties.

The bottom panel is where the groups come apart. Take the hope gap — each respondent's expected ladder in five years minus their ladder today, a person-level measure of how much better people think life is about to get. In 2008 it was identical in the two groups: 0.710 in flip counties, 0.712 in the matched comparison. (Both numbers are large because in a crisis the present scores badly while five-year expectations hold; hope gaps are widest at the bottom.) Then, as the recovery took hold, hope compressed everywhere — but it compressed faster in the flip counties. By 2014 the matched comparison still expected to climb 0.47 ladder steps; flip-county residents expected 0.37. A 0.11-step shortfall in expected progress had opened in places that started with none.

How big, and how sure

Differences in annual means can hide composition: maybe flip counties just have more of the kinds of people whose hope was fading everywhere. The chart below adjusts for that. It plots, for each year, the flip-vs-comparison difference relative to its 2008 baseline, from a regression with year fixed effects and controls for age, race, education, and income, with confidence intervals clustered by county (223 flip + 2,788 comparison counties). It is drawn like an event study, and that resemblance should be resisted: the elections are markers on the axis, not treatments, and nothing here is an effect.

On the current-life ladder the adjusted series is a flat line: relative to 2008, the flip-county coefficient in 2016 is +0.008 ladder steps with a confidence interval from −0.046 to +0.062 — statistically and substantively nothing, in any year. On the hope gap, the same machinery finds a real mid-decade sag: −0.073 in 2011, −0.070 in 2013, and −0.098 in 2014 (95% CI −0.151 to −0.046), each distinguishable from zero. The five-year-expectations ladder alone shows the same 2014 trough (−0.084, CI −0.147 to −0.022). Then it fades: by 2016 the hope-gap coefficient is −0.033 and its interval (−0.089 to +0.022) covers zero. The honest headline is not "flip counties despaired into the election." It is that they stopped expecting improvement years earlier — the divergence peaks two years before anyone voted, then partially closes.

Keep the magnitudes in view. A tenth of a ladder step is small against the 0–10 scale, and against, say, the two-step gap between rich and poor respondents. But it is measured with enormous samples — roughly 10,000 to 23,000 flip-county ladder interviews per year — on a question about expected change, where the entire nine-year movement of these series spans about half a step. As a share of expected progress, a 0.10-step shortfall on a 0.47-step baseline is a quarter of all the improvement people thought was coming.

Five measures, one pattern — almost

The standard-of-living question — is yours getting better, the same, or worse, scored +1/0/−1 — complicates the tidy story, usefully. Flip-county residents were persistently gloomier on it: −0.151 versus −0.101 in crisis-bottom 2008, +0.245 versus +0.307 in 2016, a gap near 0.05 points in most years. But the gap does not widen. Once demographics are controlled, the year-by-year coefficients never stray beyond ±0.034 from the 2008 baseline, and in 2016 the coefficient is −0.003. What remains is a small, stable baseline deficit of −0.028 (CI −0.054 to −0.003) that predates the decade's politics. On this measure the flip counties didn't deteriorate; they entered the period already a touch more pessimistic about material progress and never caught up — even as both groups swung from deeply negative to solidly positive with the recovery.

And the misery measure stays quiet. The share who experienced worry for much of the previous day moved from 28.9 percent (2008) to 26.1 percent (2016) in flip counties, and 28.8 to 27.1 percent in the comparison — if anything, slightly less worry in the flip counties by the end, with an adjusted 2016 difference of −1.1 percentage points that does not clear conventional significance. Whatever was happening in these places, it did not register as day-to-day distress. That is exactly the signature the "foreclosed expectation" hypothesis predicts and the "economic misery" caricature does not.

One more check matters. County averages can drift because counties' populations change — young people leave, the composition shifts — rather than because anyone's outlook changed. The sharpest available cut is to hold the demographic constant: restrict both groups to white respondents without a four-year degree, the demographic at the center of every 2016 explanation, with roughly 5,300 flip-county ladder interviews a year even in the half-sample years. The toggle on the main chart applies it. The same shape survives: current-life ladders move in lockstep (6.85 vs. 6.92 in 2016), while the hope gap runs below the comparison through the middle of the decade — 0.23 vs. 0.29 in 2014 — despite having started above it in 2008. The divergence is not just the flip counties holding more white non-college residents; it shows up inside that group.

What this settles, and what it can't

Where does this land in the economic-anxiety-versus-status-threat fight? Awkwardly for both slogans, which is probably the point. The misery version of economic anxiety finds no support: by their own account, flip-county residents' lives were fine and getting better, their daily worry unexceptional. But the pure status-threat account, in which material experience is beside the point, has to reckon with the fact that the measures that did diverge are precisely the material-trajectory ones — expected progress and standard-of-living direction — and that Carol Graham and Sergio Pinto have shown lost hope of exactly this kind concentrates among white, less-educated Americans in struggling places, while Herrin and colleagues found county well-being tracked the 2016 vote better than economics alone. What the flip counties uniquely add is timing: in places about to pivot politically, expectation gave out mid-decade, quietly, while life-as-lived stayed ordinary.

The limits are real. Two hundred twenty-three counties is few, and they are smaller and more rural than America at large; per-county samples range from a handful of interviews to tens of thousands, which is why the map suppresses thin cells. The matching is coarse and the estimates unweighted. People move, so a county's respondents in 2014 are not its respondents in 2008 — the white non-college cut narrows but cannot eliminate that. The hope gap partially recovered by 2016, which fits no simple morality tale. And nothing in a survey can say why these particular counties translated a mid-decade sag in expectations into a changed vote while similar-looking counties did not. What the data can say is narrower and still worth saying: years before the op-eds, the people of the flip counties were telling a nightly telephone survey that their lives were fine — and that they had stopped believing the next five years would be better. They felt it first. Whether "it" was the economy or something the economy stands for is the part no survey settles.