Arrow's result

In 1951, the economist Kenneth Arrow published a proof so counterintuitive that it is still routinely misread. He was not asking whether any particular voting system was good or bad. He was asking whether any ranked voting method — one that asks voters to order candidates from most to least preferred — could satisfy a short list of conditions that almost everyone would consider the bare minimum for a fair procedure. The answer was no.

The conditions Arrow set out are worth stating carefully, because the theorem's force depends entirely on what they actually say. The first is unrestricted domain: the method must produce a result for every possible combination of voter preferences, however contradictory or unusual. The second is Pareto efficiency: if every single voter prefers candidate A to candidate B, then the collective outcome must rank A above B. The third is independence of irrelevant alternatives (IIA): the collective ranking between A and B must depend only on how voters rank A against B, not on where they place some third candidate C. The fourth is transitivity: the social ordering must be internally consistent — if A is ranked above B and B above C, then A must be ranked above C. The fifth is non-dictatorship: no single voter's preference order should automatically become the group's preference order regardless of how everyone else votes.

Each condition, taken alone, seems not just reasonable but almost definitionally required. Together, Arrow proved, they are mathematically incompatible. Any ranked aggregation method that satisfies the first four must violate the fifth — it must, in effect, be a dictatorship.

The proof is often cited to mean that voting is arbitrary, or that no system is better than any other. Neither conclusion follows from it.

Arrow's theorem applies specifically to ranked methods that produce a full social ordering — a complete ranking of all options. It does not apply to cardinal methods, in which voters assign numerical scores rather than ordinal ranks. Whether cardinal systems escape the spirit of the result while technically escaping its letter is a matter of ongoing theoretical debate, but the escape is real. Score-based and approval-based methods operate outside Arrow's framework.

The theorem also does not say that the five conditions conflict in practice with equal frequency or severity. IIA is the condition that bites hardest. It rules out precisely the kind of sensitivity to a third candidate that produces the Condorcet cycle: situations in which a majority prefers A to B, B to C, and C to A simultaneously, making the collective preference circular and intransitive. Real elections with three or more candidates regularly produce preference distributions where some condition must yield, but which condition yields, and how badly, depends heavily on the method in use.

Nor does the theorem prove that all methods perform equally when conditions are violated. First past the post, for example, violates IIA in a particularly stark way — a minor candidate who cannot win can still determine who does — while other methods distribute the compromise differently. The theorem shows that compromise is unavoidable; it says nothing about where to place it.

Arrow's result is best understood as a boundary-marker. It closes off a certain kind of argument: the claim that there exists some correct procedure that simply aggregates individual preferences into a coherent group preference without loss or distortion. No such procedure exists for ranked methods. Every designer of a voting system is therefore not discovering the right answer but choosing which desirable property to sacrifice, and under what conditions.

That is a genuinely useful thing to know. It shifts the design question from "which method is fair?" — as if fairness were a fixed target — to "which trade-off is most tolerable for this electorate, this decision, this institution?" Practical systems are compared on how they behave when the conditions are violated, not on whether they violate them, because violation is guaranteed. The theorem does not end the conversation about which voting method to use; it clarifies what that conversation is actually about.