The Math of Voting Systems: Arrow's Impossibility Theorem

You and two friends are picking where to eat. Pizza, sushi, or tacos.

Everyone ranks their preferences. You tally the votes. Simple, right?

Here's the unsettling part: no matter which tallying method you pick, you can construct a set of honest preferences that makes the result look broken. Not "broken" as in you made a mistake. Broken as in mathematically guaranteed, for every method anyone has ever proposed.

That's what Kenneth Arrow proved in 1951. And once you see it, you can't unsee it in any election, any product ranking, any "best of" list ever compiled by combining multiple opinions.

Two options: no problem at all

If it's just pizza vs. sushi, majority rule is bulletproof. Count the votes for each, more votes wins, done. There's no way for this to produce a weird or contradictory result. Every voter has a clear preference, and the group's preference just adds them up.

The trouble starts the moment you add a third option.

Three options: majority rule can contradict itself

Let's bring in tacos. Now imagine three friends with three different rankings:

• Friend 1: Pizza > Sushi > Tacos
• Friend 2: Sushi > Tacos > Pizza
• Friend 3: Tacos > Pizza > Sushi

Look closely. Pizza beats Sushi (2 friends prefer pizza to sushi). Sushi beats Tacos (2 friends prefer sushi to tacos). So pizza should beat tacos too, right? ...Tacos beats Pizza. Also 2 out of 3.

Pizza beats Sushi. Sushi beats Tacos. Tacos beats Pizza. Majority rule just told you A is better than B, B is better than C, and C is better than A. That's not a group of three confused friends, that's the math itself going in a circle.

Notice something: this only happens for some voter splits, not all of them. Push the slider toward one dominant group and the circle breaks, a clear winner emerges. The paradox lives in the balance point, where no single preference dominates. This is called the Condorcet paradox, named after the 18th-century mathematician who first noticed that pairwise majority voting isn't guaranteed to be transitive.

We write "X is preferred to Y" as $X \succ Y$. Transitivity means if $X \succ Y$ and $Y \succ Z$, then $X \succ Z$ should follow. The Condorcet paradox shows group preferences don't have to obey this rule, even when every individual voter's preferences do.

Adding a spoiler shouldn't change who wins... but it does

Here's a separate, sneakier problem. Suppose pizza and sushi are the only two options, and slightly more people prefer sushi. Sushi wins.

Now a third place opens, tacos, and it's genuinely irrelevant, nobody would ever pick it over their favorite. Adding it to the ballot shouldn't change the pizza-vs-sushi outcome. But it can.

Drag tacos close to pizza on the spectrum. Some pizza-leaning voters peel off to tacos instead, splitting that side of the vote. Suddenly sushi wins, not because anyone changed their mind about pizza vs. sushi, but because an irrelevant third option carved up the vote unevenly.

This is the spoiler effect, and it's the same math behind every real election where a third-party candidate gets blamed for "stealing" votes. The formal name is independence of irrelevant alternatives (IIA): a fair system shouldn't let the group's preference between X and Y depend on whether some unrelated Z is on the ballot. As you just saw, plurality voting violates this constantly.

Fine, let's just try a different counting method

Maybe plurality voting (just count first choices) is the problem. There are other methods. Instant-runoff eliminates the last-place candidate and redistributes their votes. Borda count gives points for 1st, 2nd, and 3rd place and adds them up.

Surely one of these is the "correct" one...

Same ballots. Same voters. Just a different rule for turning rankings into a winner. And look, the winner changes depending on which rule you use. Plurality likes one candidate, instant-runoff might pick another, Borda count a third. There's no tiebreaker that tells you which method is "really" fair. Each one satisfies some intuitive notion of fairness and fails another.

This is the deeper problem Arrow was chasing: it's not that we haven't found the right voting method yet. It's that reasonable fairness conditions can't all be satisfied at once, by any method.

So what conditions, exactly?

Arrow formalized "fair" with a short list of conditions any reasonable voting rule should satisfy:

1. Unanimity: if every single voter prefers X to Y, the group result should too.
2. Independence of irrelevant alternatives (IIA): the group's X-vs-Y preference shouldn't depend on some other candidate Z being on the ballot.
3. Non-dictatorship: no single voter's preferences should automatically become the group's preferences, ignoring everyone else.

Each one sounds like a bare minimum for "fair," not a high bar. Arrow's theorem says:

Read that carefully. It doesn't say fair voting is hard. It says that if a voting rule always respects unanimity and IIA for three or more options, that rule is mathematically forced to be a dictatorship, one voter's ranking overrides everyone else's, every time.

Watching the dictatorship emerge

Let's see this directly instead of just stating it. Take five voters with genuinely different rankings, and a dial that controls how much weight Voter 1 gets relative to everyone else.

At low weight, Voter 1 is just one voice among five, and the group ranking is a genuine blend, transitive, sure, but not obviously anyone's personal opinion. Push the dial up. Watch the group ranking drift, then lock, exactly onto Voter 1's personal order.

That's the only way to get a group ranking that's always consistent and always immune to spoilers, for every possible combination of ballots. Give one voter all the power. Anything less than full dictatorship, and you can construct some set of preferences, like our three friends and their three restaurants, that breaks unanimity, IIA, or transitivity.

So is voting pointless?

No, and this is the part that actually matters. Arrow's theorem is a statement about guarantees, not about what happens in practice. It says no method can be perfect on every possible set of ballots. It doesn't say every method is equally bad on the ballots you'll actually see.

Real elections rarely produce perfect Condorcet cycles. Some methods (instant-runoff, Borda count, Condorcet-consistent methods) resist paradoxes and spoilers far better than plain plurality voting, even though none of them can promise perfection. Choosing a voting system is really choosing which failure modes you're willing to risk, not choosing between "fair" and "unfair."

The short version

With two options, majority rule is perfectly fair. With three or more, things break down: pairwise majority votes can cycle (Condorcet's paradox), irrelevant candidates can flip the outcome (the spoiler effect), and different vote-counting rules crown different winners from identical ballots. Arrow's impossibility theorem proves this isn't a design flaw we can patch, any voting rule that always satisfies unanimity and independence of irrelevant alternatives for three or more options must be a dictatorship. Every real voting system is a set of tradeoffs among these failure modes, not an escape from them.

The next time a friend group can't agree on where to eat, you'll know it's not their fault. It's baked into the mathematics of combining opinions.

All visualizations are interactive React components running entirely in your browser, computing every pairwise comparison, plurality count, Borda score, and instant-runoff elimination live from the slider or drag position. No libraries beyond React.