You already know the machinery of Nash equilibrium: a stable arrangement where no player, taking everyone else’s choice as fixed, wishes they’d done something different. The uncomfortable secret that machinery hides is this — plenty of games have more than one such arrangement, and the payoffs alone won’t tell you which one you’ll actually get. Something else picks. That something is what the whole previous three lessons were building toward: beliefs about beliefs. Which equilibrium a group lands on is decided not by the numbers in the payoff table but by what each person thinks everyone else will do — and what they think everyone else thinks they’ll do, all the way up.
A coordination game, refreshed
A coordination game is any game where players are trying to match — they do best when they choose the same convention as everyone else, and it matters less which convention wins than that everybody picks the same one. The cleanest example is which side of the road to drive on. If everyone drives on the right, your best move is to drive on the right too. If everyone drives on the left, drive left. Both “everyone right” and “everyone left” are Nash equilibria: given what everyone else is doing, no single driver gains anything by swerving to the other side (they’d just crash).
Notice what that means. There are (at least) two equilibria, they’re equally fine as destinations, and nothing inside the payoffs says which one a country will adopt. Sweden drove on the left until 5 a.m. on 3 September 1967, then switched the entire nation to the right in one coordinated instant. Both conventions “worked” — the switch was pure equilibrium selection, a whole country hopping from one Nash equilibrium to another on a scheduled morning.
The signature of a coordination game
Multiple equilibria, and the players agree they’d rather all be on the same one than argue about which. The interesting question stops being “what’s the best move?” and becomes “which equilibrium will we end up on — and can we steer it?”
Before you read — take a guess
Take a guess before we dig in. In a coordination game with two equally-good equilibria, what ultimately decides which one the group lands on?
The Stag Hunt: the flagship
Rousseau left us the perfect miniature. Two hunters head into the forest. Together they can bring down a stag — a huge meal — but only if both commit and hold their posts. Alone, either can instead catch a hare: small, but a sure thing you can grab by yourself, no partner required. The catch: if you wait for the stag and your partner wanders off after a hare, you go home with nothing.
Put real numbers on it. Each cell shows (Row’s payoff, Column’s payoff):
| Column hunts Stag | Column hunts Hare | |
|---|---|---|
| Row hunts Stag | 4, 4 | 0, 2 |
| Row hunts Hare | 2, 0 | 2, 2 |
Read the corners. If both hunt Stag: 4, 4 — the best outcome for everyone. If both hunt Hare: 2, 2 — modest but safe. If one hunts Stag while the other grabs a Hare: the hare-catcher still gets 2 (they didn’t need a partner), and the would-be stag hunter gets 0 (they held for a partner who never came).
This game has two pure-strategy Nash equilibria. Check them:
- (Stag, Stag) = 4, 4. If your partner hunts Stag, your best reply is Stag (4 beats the 2 you’d get from defecting to Hare). Neither wants to move. Equilibrium.
- (Hare, Hare) = 2, 2. If your partner hunts Hare, your best reply is Hare (2 beats the 0 you’d get from holding for a stag alone). Neither wants to move. Equilibrium.
(Stag, Hare) and (Hare, Stag) are not equilibria — the abandoned stag hunter is sitting on a 0 and desperately wishes they’d grabbed a hare too. So the Stag Hunt is a coordination game with the same structure as the road: two stable conventions, and the payoffs alone won’t tell you which one two nervous hunters will reach.
In the Stag Hunt table above, why is (Stag, Hare) not a Nash equilibrium?
Payoff-dominant vs risk-dominant
The two equilibria aren’t just different — they’re different kinds of good, and game theory names the two flavours.
A payoff-dominant equilibrium is the one that gives everybody the highest payoff — the outcome you’d unanimously vote for if you could sign a binding contract first. In the Stag Hunt that’s (Stag, Stag) = 4, 4. It is, flatly, the best place to be.
A risk-dominant equilibrium is the one that’s safest under uncertainty about what your partner will do — the choice that protects you best when you’re not sure you can trust the other player. In the Stag Hunt that’s (Hare, Hare). Here’s the reasoning, worked out. Suppose you think your partner is a coin flip: 50/50 to hunt Stag or Hare.
- If you hunt Stag: half the time you get 4, half the time you get 0. Expected payoff = 0.5 × 4 + 0.5 × 0 = 2.
- If you hunt Hare: you get 2 no matter what your partner does. Expected payoff = 2 as well — but with zero risk.
At exactly 50/50 they tie in expectation, but Hare carries none of the downside. Now nudge your belief even slightly toward “my partner might bolt” — say you think there’s only a 40% chance they hold for the stag. Then Stag yields 0.4 × 4 = 1.6, while Hare still yields a rock-solid 2. The moment you’re even a little unsure, Hare wins. That’s what “risk-dominant” captures: it’s the rational pick for a player who can’t count on their partner. Stag is better if you can trust them; Hare is better if you’re guessing.
Lock in the two labels.
Pick the right option for each blank, then check.
The equilibrium giving everyone the highest payoff — here (Stag, Stag) — is the one. The equilibrium that is safest when you're unsure what your partner will do — here (Hare, Hare) — is the one. A cautious player who can't their partner leans toward the safe one.
The key idea: equilibrium selection by higher-order beliefs
Here is the heart of the lesson. You will hunt stag only if you believe your partner will hunt stag — because holding for a stag alone pays 0. But that’s not enough. Your partner faces the identical worry, so they’ll hunt stag only if they believe you will. So you also need to believe that they believe you’ll hunt stag. And they need to believe that you believe that they believe it. The chain doesn’t terminate — reaching the good equilibrium requires not just agreement but common knowledge of intent: everyone intends to hunt stag, everyone knows everyone intends to, everyone knows that, and so on up the ladder from lesson 2.
Being right isn't enough — you need the right beliefs
Both hunters can be completely correct that (Stag, Stag) is the better outcome and still end up on (Hare, Hare). Knowing the stag is worth more doesn’t put the stag on your table. What puts it there is each of you being confident the other will show up — a fact about beliefs, not about payoffs. This is why the payoff table, by itself, can’t predict the outcome.
Work it through. Two hunters, both privately certain the stag is the better deal. Hunter A thinks: “I’d love the stag, but I’m not sure B will hold — and if B bolts, I get zero. Safer to grab the hare.” Hunter B reasons identically. Result: both grab hares, both get 2, and both walk home having knowingly left the 4 on the table. Nobody made a mistake. Given what each believed about the other, hare was the correct move. The good outcome was strangled not by bad payoffs or bad players but by a missing rung of belief — neither could be sure enough of the other.
Now change one thing: the two hunters lock eyes, shake hands, and each watches the other commit to the stag post. Now it’s not just true that both will hunt stag — it’s mutually visible that both will, and visibly visible, and so on. The belief ladder is complete. Both confidently hunt stag; both get 4. Same payoffs, same players, opposite outcome. The only thing that changed was the beliefs about beliefs.
Two startup co-founders both privately believe quitting their jobs to go full-time (the 'stag') would build a far more valuable company than moonlighting on nights and weekends (the 'hare'). Yet both keep their day jobs. What's the most likely reason?
Scaling up: coordination games and critical-mass thresholds
The two-hunter story generalizes to crowds, and there it fuses with the critical-mass idea you already met. In an n-player coordination game, each person has a private threshold — the level of participation they’d need to see before they’re willing to join. A cautious protester might join only if they believe thousands of others will; a bolder one joins at hundreds. Everyone is, in effect, a stag hunter scanning the forest for enough partners to make committing safe.
Now the belief problem multiplies. It’s not enough that enough people privately want to act — each must believe that enough others will act, and believe the others believe it, and so on. This is exactly why the public-signal machine from lesson 3 was so powerful: a signal everyone witnesses together (the child’s shout in the crowd; a packed public square) doesn’t just inform individuals — it manufactures the common knowledge that lets everyone conclude the threshold is cleared at the same instant, and jump together from the low-participation equilibrium to the high-participation one. Private certainty leaves everyone frozen; common knowledge lets the whole group leap as one. (The interactive tipping-point board from the last lesson is the picture of this — the crowd sitting still until a shared signal tips it over the edge.)
Sort each move by whether it can safely be done alone, or whether it only pays off once you believe enough others will do it too (needs common knowledge).
Place each item in the right group.
- Being the first to storm out of an autocrat's rigged election
- Holding your post for the stag
- Grabbing the hare while your partner does whatever they want
- Adopting a brand-new group messaging app nobody you know uses yet
- Keeping your steady day job
- Sticking with the phone platform all your friends already use
Focal points: how groups actually pick
If beliefs about beliefs select the equilibrium, then anything that makes one option obviously expected by everyone can act as a steering wheel. Thomas Schelling called such an option a focal point (now often a “Schelling point”): an outcome that stands out so vividly that everyone expects everyone to expect everyone to pick it, even with no communication. A focal point is a coordination device made of pure salience — it works precisely by being the option each person guesses the others will guess.
Schelling’s classic experiment: you must meet a stranger in New York City tomorrow, but you were never told where or when — and you can’t talk beforehand. Any location in the city is, in payoff terms, a valid meeting-equilibrium. Yet a striking share of people say the same thing: Grand Central Terminal, at noon. Why? Not because it’s payoff-dominant — every spot pays the same if you both show. It wins because it’s salient: the obvious landmark, and noon is the obvious round-number time. You go there not because you love it but because you expect the other person to expect you to go there. The focal point resolves the belief ladder without a single word being exchanged.
Focal points are everywhere once you look: a negotiation snapping to a round number ($100,000, not $97,412), because both sides expect the other to expect the round figure; “let’s meet under the clock at noon”; a 50/50 split when two people must divide a prize and can’t agree on anything finer. In each case a coordination game with countless equilibria gets resolved by the one option beliefs pile onto.
Which statement best captures WHY a focal point works?
The pitfall: coordination failure is a stable equilibrium too
Here’s the trap that catches optimists. It is tempting to think that because the good outcome is better for everyone, groups will naturally drift toward it. They won’t. Coordination failure — being stuck on the worse equilibrium — is itself a perfectly stable Nash equilibrium, and escaping it is nobody’s individual job.
Sit in the (Hare, Hare) outcome and ask who should deviate. Not you: unilaterally switching to Stag drops you from 2 to 0. Not your partner, for the identical reason. Given everyone else’s behaviour, no single person can improve by changing — which is the exact definition of an equilibrium. So the bad outcome isn’t a mistake anyone is making; it’s a trap that holds precisely because each person is responding correctly to their beliefs about the others. This is why payoff-dominance is a promise, not a guarantee: knowing (Stag, Stag) is better does nothing to get you there if the beliefs aren’t in place. Whole industries, teams, and societies sit for years on visibly worse equilibria — an inferior standard, a dysfunctional norm, a market everyone privately dislikes — not from stupidity, but because no one can safely move first.
Payoff-dominance is not a rescue rope
“But the other equilibrium is better for everyone!” is true and useless. Being better doesn’t make a group jump — only the right structure of beliefs does. A better equilibrium sitting one belief-rung out of reach is just as unreachable as one that doesn’t exist. Never assume a group will find the good outcome on its own; ask what would make the leap common knowledge.
Payoff-dominant vs risk-dominant, side by side
| Payoff-dominant | Risk-dominant | |
|---|---|---|
| What it optimises | The highest payoff for everyone if all cooperate | Safety when you’re unsure what others will do |
| Stag Hunt outcome | (Stag, Stag) = 4, 4 | (Hare, Hare) = 2, 2 |
| A player picks it when… | They’re confident others will also pick it | They doubt others — better to hedge |
| What it needs to be reached | Common knowledge of intent | Nothing — it pays regardless |
| Real example | Co-founders both going full-time; a mass protest that succeeds | Keeping the day job; staying home when the outcome’s in doubt |
When to reach for this model
Pull out the coordination-game lens whenever a group is stuck somewhere everyone privately dislikes, or whenever you’re trying to launch anything that only works if enough people join at once — a new standard, a norm change, a strike, a startup, a platform. The diagnostic question is never “is the better outcome better?” (obviously) but “what belief is missing, and what shared, witnessed signal would supply it?”
Match each term to what it means.
Pick a term, then click its definition.
Recap
Big picture
Beliefs about beliefs select the equilibrium
- Coordination games
- Multiple Nash equilibria
- Drive left OR right; both stable
- Payoffs alone can't pick one
- Stag Hunt
- (Stag,Stag)=4,4 payoff-dominant
- (Hare,Hare)=2,2 risk-dominant
- Selection by higher-order beliefs
- Hunt stag only if you trust they will
- Needs common knowledge of intent
- Steering devices
- Focal points (Schelling): salience
- Public signals clear the threshold at once
- The pitfall
- Bad equilibrium is stable too
- Payoff-dominance doesn't guarantee escape
- Multiple Nash equilibria
The lesson in one line: payoffs decide which arrangements are stable, but beliefs about beliefs decide which stable one you get. Keep this cocked and loaded, because next we point it at the wild: bank runs, revolutions, and crowds — where thousands of people play one enormous Stag Hunt, and a single shared signal can tip the whole thing from frozen to unstoppable in an afternoon.