The Broken Compass and the Centennial Catastrophe: A Polling Industry that Keeps Promising Outcomes it Cannot Deliver

America just threw itself a 250th birthday party, and the polling that came with it reads like a physician’s chart for a patient in trouble. A PBS NewsHour/Marist survey released for the semiquincentennial found that 37 percent of Americans — and 58 percent of adults under 30 — now believe violence may be necessary to get the country back on track. A Reuters/Ipsos poll taken weeks earlier found roughly four in ten Americans doubt the United States will hold together for another 250 years.

Those numbers should alarm anyone. But they shouldn’t surprise anyone who has spent time with American history. Lay the last four centuries side by side and a pattern emerges that is hard to unsee — one I’ve come to call the Centennial Catastrophe Theory: every century since before the founding, the American people have divided into two roughly balanced, mutually hostile camps — and every time, a national cataclysm has followed.

And after forty years in the polling and campaign business, I can add a second finding the historians won’t give you: in our century, the fuse that connects division to detonation runs straight through the survey research industry. The most dangerous object in American politics today is not a militia or a manifesto. It is a topline number that sets 60 million people’s expectations — and then misses by twenty points.

The Pattern, Century by Century

The 18th century opened with colonial stress and closed in revolution. The French and Indian War roughly doubled Britain’s debt, and Parliament’s attempt to make the colonies pay split colonial society over a fifteen-year buildup from 1763 to 1775. The Revolution was not merely a war against Britain; it was America’s first civil war, with Patriot mobs targeting Loyalist neighbors and roughly one in six colonists siding with the Crown. The division outlived Yorktown: the War of 1812 passed Congress by the narrowest war-declaration margin in our history, every Federalist voting no, and New England Federalists edged close enough to secession to meet secretly at Hartford to discuss it.

The 19th century delivered the paradigm case. Division metastasized through the 1850s — Bleeding Kansas, Dred Scott, a sitting senator beaten unconscious with a cane on the Senate floor. By 1860 the country had divided so completely that Lincoln won the presidency with roughly 2 percent of the Southern vote. The cataclysm that followed killed approximately 698,000 Americans — about 2.1 percent of the entire population, still the deadliest event in our history.

The 20th century ran the cycle again. The World War I era brought more than 2,000 federal prosecutions of Americans for “disloyal” speech, the racial bloodletting of the 1919 Red Summer, and Prohibition’s cultural civil war of rural “drys” against urban “wets” — while a revived Klan claimed millions of members and, by some estimates, sixteen U.S. senators. What followed was the Depression, a world war, and a Cold War that carried us to the edge of nuclear annihilation in October 1962.

And now the 21st. Trust in the federal government has sat below 30 percent continuously since 2008 — it was 77 percent in 1964. More than nine in ten partisans view the opposing party unfavorably. Congress is more polarized than at any point since Reconstruction. Ten consecutive presidential elections have been decided by single-digit margins, and the current House majority — 220 to 215 — is the narrowest in roughly a century. A 50/50 nation, again.

The Dashboard Is Flashing Red

Consider where the needles sit in August 2026. Capitol Police opened nearly 15,000 threat investigations against members of Congress in 2025 — up 58 percent in a single year. The past two years brought two assassination attempts against President Trump, the assassinations of Minnesota’s Melissa Hortman and of Charlie Kirk, and the arson attack on Governor Josh Shapiro’s residence. The University of Chicago estimates roughly a third of each party — some 60 million Americans — now express support for political violence to achieve their goals. Forty percent of Americans say a civil war within the decade is at least somewhat likely. And the V-Dem Institute cut America’s liberal-democracy score by 24 percent in one year, reclassifying the United States from a “liberal democracy” to a mere “electoral democracy” for the first time in over half a century.

The kindling is stacked. The question is what strikes the match.

The Instrument Is Broken

Here is what we know about the instrument the country uses to form its political expectations. In late July, four reputable national polls put the Democratic edge on the generic congressional ballot at eight points (CNN/SSRS), eleven points (Emerson College), three points (Washington Post-Ipsos), and two points (Denver Gazette). Same weeks, same question, nine-point spread. Somebody’s model of the American electorate is wrong.

Then, on Tuesday, the field test. In Wisconsin, a final-week public poll gave Francesca Hong a two-to-one advantage over David Crowley — 49 to 28 head-to-head. Crowley won by half a point in a photo finish the Associated Press could not call until Wednesday morning. In Michigan, Emerson had Abdul El-Sayed up fifteen; he won by one. In Minnesota, the June SurveyUSA poll had Angie Craig ahead of Peggy Flanagan by five; Flanagan won by twenty. Wrong margin, wrong margin, wrong winner — and not all in the same ideological direction, which means this is not a lean that can be corrected. It is an instrument that no longer measures.

None of this is new. The industry’s own association found that 2020 polling error was the highest in 40 years for the national popular vote and the highest in at least 20 years at the state level, overstating the Biden-Trump margin by 3.9 points nationally and 4.3 points in the states — with the error running overwhelmingly in one direction (AAPOR Task Force). Four years earlier, state polls in the Upper Midwest systematically understated Trump’s support while the forecasting industry converted those polls into headline probabilities of a Clinton victory ranging from 71 percent to over 99 percent (AAPOR 2016 Report).

Expectations Are the Fuse

Why does a bad poll matter more than any other bad data product? Because polls do not merely describe elections; they set the expectations against which the results are judged. Research by Sean Westwood, Solomon Messing, and Yphtach Lelkes found that probabilistic forecasts — the “90 percent chance” horse-race coverage — leave voters substantially more certain of the outcome than the underlying numbers justify, and less likely to perceive the election as competitive at all.

Now follow the logic one step further, into territory the academics are more reluctant to map. A voter who has been told for months that his side is winning — or that the other side’s lead is insurmountable — does not process a confounding result as new information. He processes it as evidence of theft. The expectation does not die on election night. It converts into grievance. And in a country where 60 million people have already told researchers that force can be justified, grievance is the one commodity we cannot afford to manufacture at industrial scale.

That is the theory’s connective tissue: the volatile trigger in the 21st-century cycle is the electoral result that confounds the polling. Not the result itself — the gap between the result and the expectation the polling built.

January 6 Was the Proof of Concept

Reread the last popular uprising with that lens. The 2020 public polls projected a Biden blowout; the reality was a narrow race the polls had missed by their worst margin in four decades. Millions of Trump voters had spent two consecutive cycles being told — correctly, as it turned out — that public polling undercounted them. When the count came in tighter than the toplines and then reversed across election week, they faced a choice between two conclusions: the polls were broken, or the election was. Reuters/Ipsos polling traces how thoroughly that second conclusion had been pre-loaded — a substantial share of Trump’s supporters were predisposed to doubt the legitimacy of the outcome no matter what it was.

The result was a mob at the Capitol trying to stop the certification of a presidential election, an investigation that has charged more than 1,200 defendants across nearly all 50 states, including roughly 444 charged with assaulting or impeding officers. A year later, 71 percent of Republicans still regarded Biden’s victory as illegitimate; three years later, as many as two in three still believed the election was stolen. And the corrosion is not confined to one party: by the first anniversary, 64 percent of all Americans — across both parties — said American democracy was in crisis and at risk of failing.

Be precise about the claim. The polls did not cause January 6; a months-long campaign of fraud claims did the heavy lifting. But the polling failure is what made those claims plausible to ordinary people. When the official instrument of political expectation misses by the largest margin in 40 years, the public does not conclude that surveys are hard. It concludes that somebody is lying — and each side fills in its own liar. That is the mechanism by which a bad poll becomes a national-security problem.

California Is Where the Fuse Meets the Powder

If you want to know where the next American rupture is being rehearsed, don’t look at Washington. Look out the window.

In June 2025, a president federalized thousands of California National Guard troops — backed by 700 active-duty Marines — and put them on the streets of Los Angeles over the loud objection of the state’s governor. Sacramento sued. A federal appeals panel first blessed the deployment; six months later a federal judge ordered it ended with a sharp rebuke of the administration’s justification. Then came the maps: when Texas rammed through a mid-decade gerrymander in August 2025, California answered with a countermeasure redraw of its own, and commentators started calling it a “redistricting arms race”. A federal court ultimately threw out the Texas map, but the precedent stands: the two largest states in the Union now treat congressional lines as ordinance, fired at each other between censuses.

Now place the polling mechanism on that map. Control of the House runs through California — the Proposition 50 map put as many as a half-dozen of the country’s decisive races inside our borders, and the Supreme Court declined to disturb it. California is also the state that counts for weeks: the Supreme Court’s decision this June in Watson v. RNC preserved the counting of late-arriving mail ballots — a legal win for mail voting that even sympathetic observers concede leaves the legitimacy question culturally unsettled. And the state’s voters are already split on the integrity of their own elections before a single ballot is cast: Berkeley IGS finds Democrats overwhelmingly confident that California’s count is secure, and Republicans decidedly less so.

So run the November scenario. The national toplines spend the fall somewhere between D+2 and D+11, because that is where they are today. Whatever happens on election night confounds somebody’s expectation — mathematically, it must, because the polls do not even agree with each other. The decisive seats sit in California, where the margin will materialize over two to three weeks of late-ballot counting, watched by one party that trusts the process and one that does not, in a country where four in ten already expect civil war. If you wanted to engineer conditions for a second expectation-shock crisis — a January 6 with the House majority rather than the presidency at stake — you could not draw them better.

Californians should recognize the script. Troops in the streets over a legislature’s objection, dueling claims of illegitimate authority, each side convinced the other has broken the compact first — that’s Boston in 1768. It’s Kansas in 1855. The venue changes. The play doesn’t.

An Honest Caveat

I’ll concede what the historians would insist on. The clock isn’t a metronome: the gaps between America’s great crises run closer to 80 or 90 years than a clean 100 — roughly once a human lifetime, once living memory of the last catastrophe dies out. That’s the rhythm Peter Turchin found when his models flagged the 2020s for instability back in 2010. And division is not always the trigger: Pearl Harbor united the country overnight. The theory is a diagnostic, not a prophecy. But every gauge on the dashboard now reads the way it read in the 1770s, the 1850s, and the 1920s.

The Question Mark

The last entry in my notes on this theory is a question mark: what is the 21st century’s cataclysm? History suggests it will not look like Gettysburg — each successive American catastrophe has claimed a smaller share of the population than the one before. The likelier shape is the one already forming: sustained constitutional confrontation, decentralized political violence, states and Washington treating each other as adversaries, and 60 million people who have quietly made their peace with the idea of force — all of it primed, cycle after cycle, by an expectations polling industry that keeps promising outcomes it cannot deliver.

The founders whose anniversary we just celebrated understood something we’ve forgotten: a republic this size survives on the consent of people who disagree. Every century, we’ve tested that proposition to destruction. The pattern says the test is here again. The pattern does not say how it ends. That part — for the first time, with the warning lights visible in real time — is still up to us. The clock is striking, as it has struck for four centuries running. The least we can do is stop handing a broken compass to an angry country and telling it which way to march.