the ground / stratum · Mechanism
Three friends are choosing a restaurant. Each ranks the three options honestly, top to bottom. Surely there is some fair way to combine three honest rankings into one group decision. In 1951 Kenneth Arrow proved there is not — and the only rules that even come close are dictatorships. This page lets you check that yourself, by hand and then by brute force.
The goal looks innocent. You have a group — voters, friends, a committee — and a set of options, here three: A, B, C. Everyone hands you a ranked ballot, a strict order of preference. Your job is to produce one group ranking out of theirs: a fair, consistent verdict on what the group prefers. A constitution, in the old word for it — a rule that takes everyone's ranking and returns society's.
What should such a rule obey? A few conditions seem beyond argument. It should work for any ballots people hand in (call it universal domain — you don't get to forbid opinions). If everyone ranks A above B, the group should too (unanimity, the Pareto principle). The group's verdict should be a real ranking — complete and transitive, so if society prefers A to B and B to C, it had better prefer A to C. And no single person should be a dictator whose ranking simply becomes society's, everyone else ignored.
That is a low bar. It is also, Arrow proved, an impossible one. No rule meets all four conditions at once — plus one more, the subtle one, that we will come to. Push on any rule that tries, and it either contradicts itself, ignores a unanimous wish, or quietly crowns a king. Let's start with the oldest crack, found two centuries before Arrow.
The most natural rule on Earth is majority: for each pair of options, whichever beats the other in a head-to-head wins. In 1785 the Marquis de Condorcet noticed it can eat itself. Set three voters' ballots below and watch the head-to-head winners. Majorities are real in every pair — and yet they can chase each other in a circle, with no option able to beat both others. There is no winner at all.
Click any option in a ballot to promote it up that voter's ranking. The triangle shows each head-to-head: the arrow points from winner to loser, with the tally. When the three arrows form a loop, majority has no answer.
The cycle isn't a quirk of bad luck. Of the 216 ways three voters can fill out three-way ballots, exactly 12 — one in eighteen — produce a cycle (under the standard impartial-culture assumption that every ballot is equally likely; real electorates cycle far less). And when a cycle exists, the "winner" is decided by whoever sets the agenda: vote A-vs-B first and the survivor against C, and you can install any of the three as champion just by choosing which pair fights first. Majority rule, the fairest-sounding rule there is, hands the outcome to the chair.
You might think this is majority's private failing — that some cleverer rule escapes it. So let's build rules and test them. Arrow's master move was a condition called independence of irrelevant alternatives (IIA): the group's choice between A and B should depend only on how people rank A against B — not on where some third option C sits. (Whether you prefer coffee to tea shouldn't flip because cocoa appeared on the menu.) Under IIA, a rule is nothing but a verdict for each pair, decided pair by pair. Build one.
Below, choose how the group settles each of the three pairs. Each can be decided by majority, or handed to one named voter. The machine then checks every profile of ballots and hunts for one where your rule's verdicts form a cycle — where society prefers A to B to C and back to A, an impossible ranking. The only way to make the cycle impossible, you'll find, is to let the same voter decide all three pairs. A dictator.
① Pick a rule for each pair. The machine searches all profiles for one that breaks transitivity.
② Don't take our word that only dictatorships survive. Enumerate every IIA rule that respects unanimity, and filter to the ones whose verdict is always a consistent ranking.
③ The escape hatch. If the voters share a common axis — A < B < C, say left-to-right — and no one's tastes are "double-peaked," majority rule stops cycling. Restrict to those single-peaked ballots and search again.
So the conditions are not redundant and not negotiable: drop any one and rules reappear. Drop transitivity and plain majority is "fair" — it just sometimes gives a cycle (Instrument I). Drop IIA and the Borda count works beautifully — but its winner can flip when a hopeless third candidate enters or leaves, and, as we'll see, it can be gamed. Restrict the domain to single-peaked tastes and the median voter rules, cleanly and non-dictatorially (Black, 1948). Keep all the conditions at once, on all possible ballots, and the only thing left standing is a throne. That is Arrow's theorem.
One door is still open. We've been combining honest ballots. What if voters lie? Maybe a rule that can't be fair can at least be strategy-proof — safe to vote your true preference, because lying never helps. In the early 1970s Allan Gibbard and Mark Satterthwaite independently closed it. For three or more outcomes, the only rules immune to strategic voting are — again — dictatorships. Pick a real voting rule and catch the lie.
Each rule below is one people actually use or propose. The machine searches for a profile and a single voter who, by submitting a false ballot, changes the winner to one they genuinely prefer. It always finds one — for every rule except the dictatorship, which has nothing to gain by lying because the dictator already gets their way and no one else is heard.
It is easy to over-read this. Arrow's theorem does not say democracy is impossible, that voting is pointless, or that all rules are equally bad. It says something exact: no rule that turns ranked (ordinal) preferences into a group ranking can satisfy universal domain, unanimity, independence of irrelevant alternatives, and non-dictatorship all at once, whenever there are at least three options. Every word is load-bearing, and every word is a door.
Restrict the domain. If preferences are single-peaked along a shared dimension (left–right politics, hot–cold), majority rule is transitive and the median voter wins (Black 1948). The cost: you've assumed the disagreement is one-dimensional, which real ones often aren't.
Use more than rankings. Approval voting and score/range voting ask how much you like each option, not just the order. Because they read cardinal information, Arrow's ordinal theorem simply doesn't apply to them — they step outside its frame rather than beating IIA within it. The cost: they need interpersonal or absolute judgments of intensity, and they remain manipulable — Gibbard's 1973 result still bites.
Accept imperfection. Give up exactly one condition on purpose. Most real systems quietly drop IIA (they let irrelevant alternatives matter) and live with the spoiler effect, or drop a clean group ranking and settle for picking a single winner.
The census in Instrument II is a complete, exact enumeration for 2 and 3 voters over 3 alternatives — every IIA rule that respects unanimity, checked on every possible profile, with the survivors shown to be precisely the dictatorships. That is a real proof of Arrow's theorem for those cases, run in front of you. The full theorem (any number of voters, any number ≥3 of alternatives) is the cited mathematical result, proved by induction down to a small base case — exactly the structure a computer-aided proof exploits (Tang & Lin 2009 reduce it to 2 voters / 3 alternatives and discharge it with a SAT solver). We verify the base; we cite the induction. The same enumeration also confirms the social-indifference case (allowing ties in the group ranking gives no new escape) and reproduces Condorcet's 12 / 216 and the single-peaked rescue. All of it is in /research/arrows-theorem/ — 26 checks, and the page recomputes every number live.
Condorcet, Essai sur l'application de l'analyse à la probabilité des décisions rendues à la pluralité des voix (1785) — the voting paradox.
Kenneth J. Arrow, "A Difficulty in the Concept of Social Welfare," Journal of Political Economy 58(4):328–346 (1950); Social Choice and Individual Values (Wiley, 1951; 2nd ed. 1963, correcting a gap noted by Blau 1957). Nobel Memorial Prize in Economic Sciences, 1972 (with John Hicks) — at 51, the youngest economics laureate until Esther Duflo (46) in 2019.
Duncan Black, "On the Rationale of Group Decision-making," J. Political Economy 56(1):23–34 (1948) — single-peaked preferences, the median voter.
Allan Gibbard, "Manipulation of Voting Schemes: A General Result," Econometrica 41(4):587–601 (1973); Mark A. Satterthwaite, "Strategy-proofness and Arrow's Conditions," J. Economic Theory 10(2):187–217 (1975).
G.-Th. Guilbaud (1952) — the limiting cycle probability for three options, ≈ 8.77% as voters → ∞. Pingzhong Tang & Fangzhen Lin, "Computer-Aided Proofs of Arrow's and Other Impossibility Theorems," Artificial Intelligence 173(11):1041–1053 (2009).
Every quoted figure recomputed in-browser and verified offline first. Historical claims adversarially fact-checked against primary and reference sources before publishing.