Arrow’s Impossibility Theorem: Why Voting Rules Break

Build a clear mental model for why no ranked voting system can satisfy a small set of fairness rules at once. You will see how cycles arise from sincere preferences, why IIA creates the real squeeze, and what designers do instead.

Majority rule feels like it should produce a sensible group ranking if everyone votes honestly. Arrow’s Impossibility Theorem says that hope is misplaced. With three or more candidates, you can demand a few axioms that sound completely reasonable and still end up with a social ranking that is inconsistent or a rule that secretly gives one person total control. The shock is not that people are irrational. The trap shows up even when every voter has a perfectly consistent ranking.

Take a look at how a preference cycle can appear even with sincere voters.

Once you have seen the cycle, the key question becomes what kind of object a voting rule is supposed to output.

The basic object Arrow studies

Arrow is not talking about picking a single winner yet. He starts with a rule that takes everyone’s rankings and returns a full group ranking of all candidates. That rule is a social welfare function (SWF). Each voter supplies a complete, transitive ranking like $A\succ B\succ C$. The SWF takes the whole profile of rankings and outputs one social ranking, also complete and transitive, like $B\succ A\succ C$.

The theorem needs a big domain assumption. Voters are allowed to have essentially any rankings at all, as long as each individual ranking is internally consistent. This is often called unrestricted domain. It matters because cycles are easiest to force when the space of possible profiles is wide.

Explore how an SWF maps a profile of rankings into one social ranking.

If the rule must always output a tidy ranking, the only way to avoid contradictions is to impose constraints somewhere. Arrow’s axioms are the attempted constraints.

Arrow’s axioms as design constraints

Here is the featured snippet version of the core idea.

Arrow’s Impossibility Theorem states that when there are at least three candidates, no social welfare function can simultaneously satisfy Pareto efficiency, Independence of Irrelevant Alternatives (IIA), and non-dictatorship while allowing unrestricted domain. If the rule always produces a complete, transitive social ranking, at least one of those fairness requirements must fail.

What each axiom demands

• Pareto efficiency: if every voter ranks $A$ above $B$, the social ranking must put $A$ above $B$.
• IIA: the social ranking between $A$ and $B$ depends only on how each voter ranks $A$ versus $B$, not on where they place other candidates.
• Non-dictatorship: no single voter’s ranking always becomes the social ranking for every pair of candidates.

See the axioms side by side, including what each one forbids in practice.

Pareto and non-dictatorship usually feel uncontroversial. IIA is the one that sounds clean on paper and becomes brutal when you try to make a full ranking.

Hidden coupling
IIA tries to make every pairwise decision independent, but a full ranking forces those pairwise decisions to fit together globally.

That coupling is why the next step focuses almost entirely on IIA.

Why Independence of Irrelevant Alternatives bites

IIA says the group’s opinion about $A$ versus $B$ should not change if nobody changes their mind about $A$ versus $B$. That is a strong stability promise. It rules out the common experience where adding a spoiler changes who wins, but it also rules out subtler effects like a third candidate changing how a method aggregates information.

The tension is local versus global. Locally, IIA asks for clean pairwise decisions. Globally, the SWF must output a single transitive ranking that stitches all pairwise comparisons together. With three candidates, you can engineer profiles where maintaining IIA forces the rule to make pairwise choices that cannot all be satisfied in one consistent ordering unless some voter’s comparisons dominate.

Simulate what happens when a third candidate is added or removed while $A$ versus $B$ orderings stay fixed.

Notice what makes this hard. Many reasonable methods use information beyond the $A$ versus $B$ comparisons to decide $A$ versus $B$, because that extra context helps avoid cycles or produce a complete ranking. IIA forbids using that context.

Proof idea in one spine

The formal proof is a construction, but the intuition fits on one backbone. Start with the idea of a pivotal voter. For some pair $A,B$, imagine changing voters one by one from preferring $B$ over $A$ to preferring $A$ over $B$. If the social ranking flips at all, there is a first voter whose change flips society from $B\succ A$ to $A\succ B$. That voter is pivotal for that pair under that profile.

Now leverage unrestricted domain to build profiles where the same voter becomes pivotal in a way that, combined with IIA, forces their preference on other pairs too. The move is always the same. You hold fixed how everyone ranks one pair, tweak how they rank a third candidate, and use IIA to argue the social choice on the original pair cannot move. By doing this across carefully chosen profiles, you spread the pivotal voter’s control from one pair to all pairs.

Once one voter’s ranking determines every social pairwise comparison, the social ranking mirrors that voter’s ranking. That is dictatorship, contradicting non-dictatorship. So if you keep Pareto and IIA on the unrestricted domain, dictatorship is not an accident. It is the only way to guarantee a transitive social ranking.

How real voting rules fail the axioms

Most real systems do not even try to be SWFs. They are closer to a social choice function (SCF) that picks a winner, not a full ranking. Still, Arrow is a warning sign. Any time you want a coherent group outcome from ranked preferences, you should expect trade-offs that look a lot like Arrow’s.

A quick way to read common rules is to ask which promise they break when the inputs get adversarial.

See how common methods line up against Arrow-style constraints and typical failure cases.

A few practical anchors help:

• Plurality can ignore broad consensus and is vulnerable to vote splitting. That is not an Arrow axiom failure per se, but it shows how thin information can mislead.
• Instant-runoff voting (IRV) uses elimination rounds, which makes outcomes sensitive to how lower-ranked candidates enter the race. That is a direct clash with IIA-type stability.
• Borda count uses full ranking information, which can reduce cycles but makes $A$ versus $B$ depend on placements of others, again violating IIA.
• Condorcet methods chase majority consistency pair by pair, but can face cycles. To output a complete ranking, they need tie-breaking structure that will offend some fairness criterion.

If you feel the urge to ask which rule is best, Arrow’s punchline is that best means which trade-off you can live with.

Living after Arrow

Designers escape Arrow by relaxing something structural rather than arguing about which axiom is morally correct.

One escape is to narrow the domain. With single-peaked preferences, voters sit on a one-dimensional line of opinion, and each voter’s ranking falls away from their favorite as you move left or right. On such domains, majority rule behaves far better, and Condorcet cycles disappear.

Another escape is to randomize. Randomized social choice can soften impossibilities because the output is a lottery over outcomes, changing what it means to satisfy fairness constraints.

A third escape is to leave pure rankings and use cardinal utility or welfare. That shifts the question from only order to how much, which introduces its own assumptions but gives more structure.

Reveal how restricted domains like single-peakedness remove cycles and how this connects to later impossibility results.

Design lever
Every escape route pays for coherence with an assumption. You either restrict which preference profiles matter, weaken what outcomes must look like, or accept some dependence on context.

With that lever in mind, Arrow becomes less of a paradox and more of a checklist.

Where this matters when you choose rules

Arrow’s Impossibility Theorem is not telling you democracy is doomed. It is telling you to stop treating voting rules as neutral plumbing. Each rule encodes a choice about what kinds of unfairness you are willing to risk.

For elections, it changes how you interpret outcomes. If a candidate wins under a method that violates IIA, the winner may be sensitive to who else ran. That does not mean the outcome is illegitimate. It means you should be careful about making claims like the electorate preferred $A$ to $B$ in any context-free way.

For committees, it suggests a practical move. Write down the axiom you care about most for the decision at hand, then pick the method that protects that axiom and openly accepts the failure mode elsewhere. If you are choosing among policies along one ideological dimension, aim for domain restrictions like single-peakedness. If you are allocating resources, utility based approaches might match the job better than rankings.

You will get further by deciding what you want to be stable. Do you want pairwise stability against spoilers, consensus-respecting behavior, monotonicity, or resistance to strategy. Arrow is the reminder that you cannot demand all of them at once.