The Condorcet Paradox: When Majority Voting Goes in Circles

The setup: three voters, three options, one impossible loop

The paradox needs only three of everything. Take options A, B, C and three voters (or three equal-sized blocs) whose rankings are each perfectly rational — no voter contradicts themselves:

• Voter 1: A > B > C
• Voter 2: B > C > A
• Voter 3: C > A > B

Now run the three head-to-head majority contests, counting how many voters put one option above the other:

• A vs. B: Voters 1 and 3 rank A over B → A wins 2–1
• B vs. C: Voters 1 and 2 rank B over C → B wins 2–1
• C vs. A: Voters 2 and 3 rank C over A → C wins 2–1

Read that chain back: society prefers A to B, and B to C, so a rational agent would conclude A beats C. Instead, C beats A. The group's preference is A → B → C → A, a closed loop. Whatever you crown as the winner, a majority strictly prefers something else. Notice the elegant symmetry of the construction: each voter's ranking is just the previous one rotated by one position. That rotational symmetry is exactly what makes the tie impossible to break — no option is anyone's universal favorite or universal reject.

Why it happens: rationality does not survive aggregation

The deep lesson is not about voting per se — it is about aggregation. Each voter possesses a transitive ordering, a property so basic we treat it as the definition of rationality. Transitivity says: if you like coffee over tea and tea over water, you like coffee over water. Majority rule takes these three tidy orderings and produces a group ranking that violates the very property every input satisfied.

Mechanically, the failure occurs because majority rule throws away intensity and position. When A beats B, we learn only that a majority ranks A higher — not by how much, and not who the dissenters are. Each pairwise vote is decided by a different majority coalition: {1,3} carries A over B, {1,2} carries B over C, {2,3} carries C over A. No single, stable majority governs; a rotating cast of coalitions each pushes in a different direction. Formally, let N(X,Y) be the number of voters preferring X to Y. Majority rule declares X > Y whenever N(X,Y) > N(Y,X). Transitivity of this relation is not guaranteed by transitivity of the individual orderings — and the Condorcet profile is a constructive proof that the guarantee fails.

This is why the paradox is genuinely a paradox and not merely a rare glitch: it shows that ‘the majority prefers’ is not always a well-defined ordering at all. The phrase ‘what the people want’ can be, quite literally, incoherent.

The killer consequence: whoever sets the agenda picks the winner

A cycle is not just a curiosity for logicians — it hands enormous, hidden power to whoever controls the order of votes. Because pairwise contests are usually run as a sequence (a bracket), the agenda-setter can engineer any outcome they like by choosing which two options fight first, with the survivor advancing to face the third.

Using our profile:

• Vote A vs. B first (A wins), then winner vs. C → C wins. Final winner: C.
• Vote B vs. C first (B wins), then winner vs. A → A wins. Final winner: A.
• Vote C vs. A first (C wins), then winner vs. B → B wins. Final winner: B.

Every option can be made to win — the process is entirely determined by the sequence, not by any objective ‘people's choice.’ A chair who understands the profile can guarantee their favored result while appearing to run a fair, neutral, majority-respecting procedure. This is the theoretical foundation of agenda manipulation, and it is why legislative procedure (which amendment is voted on against which) is so fiercely contested. The final option in the sequence has a structural advantage — the classic reason skilled parliamentarians fight to be the ‘last-standing’ motion.

How often does it actually happen? Real numbers

If cycles were ubiquitous, democracy would be paralyzed; if impossible, the paradox would be a party trick. The truth sits in between and depends heavily on how you model voter preferences.

The worst case — impartial culture. Assume every voter's ranking is drawn uniformly at random (all 6 orderings of 3 options equally likely). Then the probability of a Condorcet cycle grows with the electorate and converges to about 8.8% (roughly 1 in 11) as the number of voters → ∞. It gets much worse with more options: with 4 candidates it rises past 17%, and it climbs toward certainty as the number of options grows. Importantly, mathematicians have proved that impartial culture maximizes the cycle probability — 8.8% is an upper bound, not a typical value.

The real world — far rarer. Actual electorates are not random. Voters cluster ideologically along recognizable dimensions (left–right, more–less spending). A survey of 37 empirical studies from 1955 to 2009 concluded that Condorcet's paradox ‘might be observed, but... probably is not a widespread phenomenon,’ and that a Condorcet winner (an option that beats all others head-to-head) exists in the overwhelming majority of real elections. Suspected real cycles are debated — some analyses of multi-candidate races and legislative votes on wage/spending bills — but clean, undisputed examples are surprisingly hard to pin down, precisely because human preferences are structured, not uniform.

The escape hatch: single-peaked preferences and the median voter

Why is the paradox so rare in practice? Duncan Black's answer (1948) is one of the most elegant results in political economy: if preferences are single-peaked, cycles are impossible and a Condorcet winner always exists.

Single-peaked means you can arrange all options along one line — say, government spending from low to high — such that every voter has one favorite point and likes options less the farther they are from that peak in either direction. No voter has ‘two humps’ (loving both very low and very high spending while hating the middle). When this holds, the option preferred by the median voter beats every other option in a majority contest. This is Black's Median Voter Theorem, and it is the reason two-party competition tends to converge on the center: the candidate closest to the median voter's ideal point wins.

The Condorcet profile violates single-peakedness by construction. Look again: Voter 3 ranks C > A > B — for this voter, no matter how you order A, B, C on a line, their preferences have a valley, not a single peak, relative to the others. The paradox lives precisely in preference structures that cannot be reduced to one dimension. So the practical takeaway is sharp: the more genuinely multidimensional and non-ideological the choice, the more likely a cycle.

The bigger picture: Condorcet as the ancestor of Arrow's theorem

The paradox slept for 150 years until Duncan Black (1948) and Kenneth Arrow (1950–51) revived it. Arrow's genius was to ask: is majority rule just unlucky, or is EVERY reasonable voting method doomed? His Impossibility Theorem — which won him the 1972 Nobel Prize — answers the latter. With three or more options, no voting rule can simultaneously satisfy a short list of mild fairness conditions:

1. Unrestricted domain: works for any profile of rational voter preferences.
2. Pareto efficiency: if everyone prefers A to B, so does society.
3. Independence of irrelevant alternatives (IIA): society's A-vs-B ranking depends only on how voters rank A vs. B.
4. Non-dictatorship: no single voter's preference always determines the outcome.

The Condorcet Paradox is the concrete failure that Arrow generalized: it is exactly what goes wrong when you demand transitive collective preferences over an unrestricted domain. Arrow proved the only escape from cycling-or-chaos is to abandon one of the conditions — and the least-bad thing to give up is often IIA (which is what rank-scoring methods like the Borda count do, at the cost of vulnerability to strategic and clone candidates). The paradox is therefore not a bug to be patched but a fundamental limit on collective decision-making itself: perfectly fair, perfectly rational democratic aggregation is, in general, mathematically impossible.

Individual rationality vs. collective rationality under majority rule

Property	An individual voter	Majority rule over a group
Completeness (ranks every pair)	Yes, by assumption	Yes — a majority winner exists for each pair
Transitivity (A>B, B>C ⟹ A>C)	Assumed / usually holds	Can FAIL — this is the paradox
A single best option (a 'top')	Always exists	May not exist (a cycle has no top)
Outcome depends on voting order?	No	Yes — agenda-setter can pick the winner
Escape route	Not needed	Single-peaked preferences restore transitivity