Arrow's Impossibility Theorem

What the theorem says

Arrow asked a deceptively simple question: can we aggregate individuals' preference rankings into a coherent social ranking? Suppose every voter ranks alternatives like ice cream flavors — chocolate > vanilla > strawberry. Is there a rule that combines those rankings into one "society's preference" that respects basic fairness?

Arrow demanded five conditions:

1. Unrestricted domain. The rule must accept any combination of individual rankings — voters are free to prefer anything.
2. Non-dictatorship. No single voter's preference automatically becomes society's.
3. Pareto efficiency (unanimity). If every voter prefers A to B, society must prefer A to B.
4. Independence of Irrelevant Alternatives (IIA). Society's ranking of A vs B depends only on individuals' rankings of A vs B — not on where C sits.
5. Transitivity. If society ranks A > B and B > C, then it ranks A > C.

Arrow proved: with three or more alternatives, no rule can satisfy all five at once. Either the rule is dictatorial, or it produces cycles, or it sometimes flips the social verdict on A vs B because of where C is ranked, or it ignores a unanimous vote.

The proof is short — modern presentations fit on two pages — but the conclusion was seismic. Welfare economists had spent decades constructing "social welfare functions" that summed individuals' utilities. Arrow showed that without making interpersonal utility comparisons (which positivists rejected as unscientific), no such function could exist.

The Condorcet paradox: Arrow's theorem in miniature

Two centuries before Arrow, the Marquis de Condorcet (1785) noticed something unsettling about majority voting. Consider three voters and three alternatives:

Voter	1st choice	2nd choice	3rd choice
Alice	A	B	C
Bob	B	C	A
Carol	C	A	B

Run pairwise majority votes:

• A vs B: Alice and Carol prefer A; Bob prefers B. A wins, 2-1.
• B vs C: Alice and Bob prefer B; Carol prefers C. B wins, 2-1.
• C vs A: Bob and Carol prefer C; Alice prefers A. C wins, 2-1.

Society prefers A>B and B>C, which by transitivity should give A>C. But the actual majority says C>A. Pairwise majority rule produces a cycle: A > B > C > A > ... There is no Condorcet winner. Whichever alternative you pick, a majority would rather have something else.

Arrow's theorem generalizes this paradox. It says: any ordinal aggregation rule meeting the four other axioms will, on some preference profile, behave at least as badly. There is no clever fix.

Voting systems against Arrow's axioms

System	Pareto	IIA	Non-dictatorial	Transitive	Notes
Plurality (first-past-the-post)	Yes	No	Yes	Yes	Spoiler effect: Nader 2000.
Borda count	Yes	No	Yes	Yes	Ranking shifts when alternatives drop in or out.
Pairwise majority	Yes	Yes	Yes	No	Condorcet cycles possible.
Instant-runoff (IRV)	Yes	No	Yes	Yes	Non-monotonic: more support can hurt a candidate.
Approval voting	Yes	—	Yes	Yes	Cardinal — outside Arrow's setup.
Dictatorship (one voter decides)	Yes	Yes	No	Yes	Trivially satisfies the rest.
Random ballot	No	Yes	Yes	Yes	Pareto fails: can pick an option everyone disliked.

Every row violates at least one axiom. That's Arrow's theorem in tabular form.

Sketch of the modern proof

The cleanest proof, due to John Geanakoplos (2005), uses a "decisive coalition" argument:

1. Call a coalition decisive over (A, B) if its unanimous preference for A over B forces society's. By Pareto, the full electorate is decisive over every pair.
2. Show that if a coalition is decisive over one pair, it is decisive over every pair. (This step uses IIA and transitivity heavily.)
3. Use a contraction argument: take a smallest decisive coalition; show that some strict subset is also decisive. By minimality, the coalition has size 1.
4. That single voter is decisive over every pair — a dictator. Contradicts non-dictatorship.

The proof shows the theorem isn't a quirk of a specific voting rule. It is a structural feature of any setup that demands ordinal aggregation plus IIA on a free domain.

Counterarguments and escape routes

Sen's critique. Amartya Sen argued that Arrow's framework is too thin: it assumes only ordinal preference and forbids interpersonal comparisons. If we allow even modest comparisons ("Alice's gain matters more than Bob's loss because Alice is starving"), the impossibility softens. Sen's own theorem on "the impossibility of a Paretian liberal" (1970) showed a different tension — between Pareto and even minimal individual rights.

Restricted domain. Black's median voter theorem (1948) escapes Arrow by assuming single-peaked preferences along one dimension. Real political space is rarely one-dimensional, but in narrow contexts (committee votes on a budget level) the assumption holds.

Cardinal information. Range voting (score 0-10) and approval voting use cardinal data, sidestepping Arrow's ordinal frame. They face Gibbard-Satterthwaite manipulability instead.

Probabilistic rules. Allowing randomization (e.g., random ballot, where each voter has probability proportional to a count) recovers strategy-proofness in some settings, at the cost of determinism.

Variants and related impossibilities

• Gibbard-Satterthwaite (1973): Every non-dictatorial deterministic voting rule with ≥3 alternatives is strategically manipulable. Arrow says fair aggregation is impossible; Gibbard-Satterthwaite says strategy-proofness is impossible.
• Sen's Paretian liberal (1970): No social choice rule can simultaneously respect Pareto efficiency and grant each individual decisive power over even one personal pair (e.g., what to read in private).
• Muller-Satterthwaite (1977): Replaces IIA with monotonicity; same dictatorship conclusion.
• Arrow's theorem with infinite voters: If we allow infinitely many voters, "ultrafilters" can serve as non-dictatorial decisive coalitions — but they're non-constructive.

Common pitfalls

• Reading "dictator" as a real person. Arrow's dictator is a logical construct: someone whose preference, by the rule's structure, always determines society's. No actual ballot says "Alice decides".
• Concluding democracy is broken. Arrow shows trade-offs, not futility. Most working democracies sacrifice IIA — accepting spoiler effects — and live with it.
• Ignoring the ordinal restriction. Arrow's setup forbids cardinal utility. Many real systems (budgeting, market prices) use cardinal information and aren't bound by the theorem.
• Treating cycles as rare. With many voters and many alternatives, the probability of Condorcet cycles approaches certainty under uniform preferences. Real elections avoid them mostly because preferences are correlated, not because cycles are mathematically rare.
• Conflating Arrow with Condorcet. Condorcet's paradox is a specific failure of pairwise majority. Arrow's theorem proves every ordinal rule has some failure — a vastly stronger claim.