Arrow's Impossibility Theorem: Why Every Election Is Unfair

The Night the Wrong Person Won

On the evening of November 7, 2000, Americans went to bed not knowing who their next president would be. By the time the dust settled — 36 days, a Supreme Court case, and several nervous breakdowns later — George W. Bush had won Florida by 537 votes out of nearly six million cast.1

Here's the number that should haunt you: Ralph Nader, the Green Party candidate, received 97,488 votes in Florida. Exit polls showed that Nader voters preferred Gore over Bush by a margin of roughly two to one.2 If Nader hadn't been on the ballot — or if voters had been able to express that Gore was their second choice — Al Gore would have been president. The Kyoto Protocol. No Iraq War. A different Supreme Court. A different world, probably.

Was this a failure of voters? A failure of Nader's ego? No. It was a failure of mathematics.

The American voting system — mark one name, most marks wins — is called plurality voting. It's the simplest system imaginable, and it has a fatal flaw: it can't handle more than two candidates without potentially electing someone that the majority opposes. Political scientists have a clinical name for this: the spoiler effect. Nader "spoiled" the election for Gore not because Nader did anything wrong, but because the voting system literally cannot process the information "I prefer Nader, but Gore over Bush."

So the natural response is: let's fix it. Let's build a better voting system. Let voters rank their candidates. Let them express their full preferences. Surely, with enough mathematical ingenuity, we can design a system that's actually fair.

In 1951, a 29-year-old economist named Kenneth Arrow proved that we can't.3

Not "we haven't yet." Not "it's really hard." Can't. As in: it is mathematically impossible to design a voting system that satisfies a small list of seemingly obvious fairness requirements. Every system that has ever existed, or ever will exist, must violate at least one of them. The theorem earned Arrow a Nobel Prize and permanently altered how we think about democracy, social choice, and the very concept of "the will of the people."

But before we get to the impossibility, let's understand what "fair" was supposed to mean.

Three Things That Seem Obviously True

Arrow didn't start by looking for impossibility. He started by asking: what are the minimum requirements for a fair voting system? Not a perfect system — just one that isn't insane. He came up with a short list. Embarrassingly short. The kind of list where you read each item and think, "Well, obviously."

Requirement 1: Unanimity (the Pareto Condition)

If every single voter prefers candidate A over candidate B, then the group result should rank A above B.

That's it. That's the requirement. If literally everyone agrees that pizza is better than gravel, the voting system should not declare gravel the winner. You would think this goes without saying, and yet Arrow felt the need to say it, because he was about to prove that even this — combined with two other obvious things — is too much to ask.

Requirement 2: Independence of Irrelevant Alternatives (IIA)

This one sounds technical but the idea is beautifully simple. Say you're choosing between steak and chicken for dinner. You prefer steak. Then someone mentions that the restaurant also serves fish. Should that change whether you prefer steak over chicken? Of course not. Fish is irrelevant to the steak-vs-chicken question.

IIA says: the group's ranking of A versus B should depend only on how individual voters rank A versus B. Introducing or removing some third candidate C shouldn't flip the result between A and B.

This is exactly what went wrong in Florida. The relative ranking of Gore vs. Bush among voters didn't change when Nader entered the race. But the outcome did. Plurality voting violates IIA.

Requirement 3: Non-Dictatorship

There should be no single voter whose preference automatically becomes the group's preference, regardless of how everyone else votes.

Again: obvious. A voting system where one person's ballot determines the outcome no matter what isn't a voting system at all. It's a dictatorship with extra steps.

Let that sink in. Arrow's conditions aren't aspirational. They're not "it would be nice if." They're the bare minimum for a system to not be obviously absurd. And mathematics proves you can't have all of them at once.

Three Voters, Three Candidates, Zero Hope

Let's make this concrete. Forget 130 million American voters. Let's use three: Alice, Bob, and Carol. They're choosing among three candidates: let's call them Red, Blue, and Green.

Here are their honest preferences:

Looks simple enough. Now let's run every voting system we know and watch them all fail in different ways.

Plurality: "Just Count First Choices"

Red, Blue, and Green each get exactly one first-place vote. It's a three-way tie. Plurality has nothing to say. But even if Alice had a friend who also ranked Red first, Red would win 2-1-1 despite the fact that two out of three remaining voters ranked Red last. Plurality ignores everything except your top choice.

Borda Count: "Points for Every Position"

Assign 2 points for first place, 1 for second, 0 for third. Red gets 2+1+0 = 3. Blue gets 0+2+1 = 3. Green gets 1+0+2 = 3. Another tie — but change one voter slightly and Borda will cheerfully violate IIA. Add a fourth candidate that nobody likes, and the winner between the original three can flip completely.

Condorcet: "Who Wins Head-to-Head?"

Compare each pair. Red vs. Blue: Alice prefers Red, Bob prefers Blue, Carol prefers Blue. Blue wins 2-1. Blue vs. Green: Alice prefers Green, Bob prefers Blue, Carol prefers Green. Green wins 2-1. Green vs. Red: Alice prefers Red, Bob prefers Red, Carol prefers Green. Red wins 2-1.

So Blue beats Red. Green beats Blue. Red beats Green. It's a cycle. Rock-paper-scissors. There is no Condorcet winner. The "will of the people" is literally incoherent — the group prefers A to B, B to C, and C to A.4

Ranked Choice (Instant Runoff): "Eliminate the Loser"

With a three-way tie for first place, we'd need a tiebreaker to eliminate someone. But even without ties, IRV has its own spectacular failures. We'll see one shortly — from a real election, not a thought experiment.

The point is: every system, applied to these same ballots, either ties, cycles, or produces a result that violates one of Arrow's criteria. This isn't bad luck. It's the geometry of preferences.

The Parade of Broken Systems

Let's tour the graveyard. Each voting system has an advertising brochure and a rap sheet.

The pattern is clear: every system has a fatal flaw. Not because system designers are stupid, but because Arrow proved that fatal flaws are mandatory. Designing a voting system is like being told to draw a square circle — the specifications themselves are contradictory.

The Election Lab

Enough theory. Below is a ballot box. Set up voter preferences — drag the dropdowns to create any scenario you like — and watch five different voting systems produce potentially five different winners from the exact same ballots.

Try the presets first. Then experiment. The goal is to internalize, in your gut, the central insight of Arrow's theorem: the winner depends on the system, not just the votes.

When the Math Hit the Real World

Arrow's theorem isn't an abstraction. It reaches into voting booths and changes history.

Burlington, Vermont, 2009

Burlington was one of the first American cities to adopt Ranked Choice Voting for its mayoral election. In 2009, three serious candidates ran: Bob Kiss (Progressive), Kurt Wright (Republican), and Andy Montroll (Democrat).

Wright got the most first-choice votes. Montroll was the Condorcet winner — he would have beaten both Kiss and Wright in head-to-head matchups. But IRV eliminated Montroll in an early round (he had the fewest first-choice votes), and Kiss won the runoff against Wright.

The system elected a candidate that the majority of voters liked less than another available candidate. Burlington was so disgusted that it repealed IRV the following year.6

France, 2002

France uses a two-round system: the top two from round one face off in round two. In 2002, the left-wing vote was split among multiple candidates. Jacques Chirac (center-right) and Jean-Marie Le Pen (far-right) made the runoff, squeezing out the Socialist Lionel Jospin. In the runoff, Chirac won with 82% — the most lopsided presidential election in French history — because practically everyone preferred him to Le Pen. Jospin, who polls suggested would have beaten either of them, was eliminated by the system.8

Brexit, 2016

The Brexit referendum was a binary vote — Leave or Remain — which dodges Arrow's theorem entirely. (Arrow's impossibility only kicks in with three or more alternatives.) But it illustrates a related problem: the ballot offered two options when there were really many. "Leave" meant different things to different voters: soft Brexit, hard Brexit, Norway model, Canada model. If the ballot had listed the actual options, no single form of departure would have beaten Remain.9 The voting system collapsed a multidimensional choice into a binary one and declared a "clear" result from muddled preferences.

Why Arrow Doesn't Mean Democracy Is Hopeless

It would be easy to read Arrow's theorem as a counsel of despair. If no voting system is perfect, why bother? Let's just flip a coin. Or install a philosopher-king. Or let an algorithm decide.

But this misreads the theorem. Arrow didn't prove that all systems are equally bad. He proved that all systems involve tradeoffs. And tradeoffs are something humans are very good at navigating, once we know they exist.

The deep lesson isn't "democracy is broken." It's that "the will of the people" is not a coherent mathematical object. There is no objective function waiting to be measured. When we vote, we're not discovering a pre-existing truth — we're constructing a decision through a process, and the process shapes the result.

This should make us more humble about election outcomes, not less committed to elections. Knowing that the system is imperfect is like knowing that a map is not the territory: it doesn't make maps useless, but it stops you from driving into a lake because the GPS said to.

The Impossibility We Live With

Kenneth Arrow died in 2017 at the age of 95.11 In interviews late in his life, he seemed at peace with his theorem. He didn't think it meant democracy was doomed. He thought it meant we should be clear-eyed about what we're doing when we vote.

We are not computing the will of the people. We are negotiating it. Every voting system is a set of rules for that negotiation — rules that favor certain outcomes over others, that amplify some voices and muffle others, that make certain coalitions possible and others impossible. The rules matter as much as the votes.

And here's the part that Ellenberg would love: the math doesn't just tell us something is impossible. It tells us why it's impossible, which means it tells us exactly where the tradeoffs lie. Arrow's theorem is a map of the constraints. It shows us the walls we can't walk through and, by implication, the doors we can.

The next time you hear someone say "the people have spoken," remember Arrow. The people spoke. But the system chose which words to hear.