A one-shot test of how much money people will burn, on purpose, just to punish a split they consider unfair.
Behavioral Economics · Negotiation & Decision ScienceIn the Ultimatum Game, one player (the Proposer) is handed a sum of money to split with a second player (the Responder) they will never meet again. The Proposer names a single, take-it-or-leave-it split. The Responder has exactly one choice: accept it, and both players get the proposed amounts, or reject it, and both players get nothing. There are no counteroffers and no second round. Economists Werner Güth, Rolf Schmittberger, and Bernd Schwarze first ran the experiment in a 1982 paper in the Journal of Economic Behavior & Organization, and it has been replicated more than almost any other experiment in behavioral economics because it pits cold arithmetic against something people clearly care about just as much: fairness.
People will pay real money, in a single round with no do-over, to punish a split they consider unfair — even though punishing it costs them money too.
The purely rational, wealth-maximizing prediction — solved by backward induction to find the subgame-perfect equilibrium — says the Proposer should offer the smallest possible positive amount and the Responder should accept it, since anything beats nothing. Real players almost never behave that way: offers cluster around 40–50% of the pot, and offers below roughly 20–30% are frequently rejected. Anthropologist Joseph Henrich and colleagues ran the game across 15 small-scale societies in 2001 and found the exact fairness threshold differs by culture — but the deviation from "accept anything positive" showed up everywhere they looked.
Use this compass to orient yourself before you reach for the model. It tells you what field it comes from, what it travels with, and where the real tension sits.
Behavioral economics, negotiation theory, and decision science. It's a deliberately stripped-down lab game, built to isolate one question: how much will people sacrifice of their own payoff to punish an offer that feels unfair?
The Dictator Game, its close cousin — same setup, but the Responder has no veto, which isolates "fear of rejection" from genuine generosity. Dictator Game offers run lower on average but are often still positive.
The textbook rational prediction (offer next to nothing, accept anything positive) assumes players only care about their own payoff. The robust finding that unfair offers get rejected anyway shows most people will pay to enforce a fairness norm — even against a stranger they'll never deal with again.
Designing any single, take-it-or-leave-it offer — a severance package, a final licensing term, a one-time settlement, a "final offer" in a negotiation — anywhere a low-but-technically-positive number might get rejected on principle.
Work through these four steps in order, then try sending your own offer in the simulator below.
0 of 4 steps opened
One Proposer, one Responder, a fixed pot, a single sequential move, and no counteroffers. The Proposer names a split; the Responder can only accept (both get exactly what was proposed) or reject (both get zero). That's the entire game — deliberately smaller than any real negotiation, so researchers could isolate one variable.
Solve it by backward induction: a pure wealth-maximizing Responder should accept any positive offer, because rejecting throws away free money. Knowing that, a pure wealth-maximizing Proposer should offer the smallest possible positive amount and keep the rest. That's the subgame-perfect equilibrium under a pure wealth-maximizing assumption — and it is almost never what actually happens at the table.
Across decades of replications and cultures, real offers cluster around 40–50% of the pot, and offers below roughly 20–30% get rejected often enough to matter — even though rejecting leaves the Responder with literally nothing. Henrich's 2001 cross-cultural study found the specific fairness threshold varies widely by society, but "reject unfair-feeling splits" showed up everywhere tested. The likely mechanism is inequity aversion — formalized in Ernst Fehr and Klaus Schmidt's 1999 model — combined with a genuine willingness to pay a personal cost to punish it. Alan Sanfey and colleagues' 2003 neuroimaging study, "The Neural Basis of Economic Decision-Making in the Ultimatum Game" (Science), found unfair offers activate the anterior insula, a region tied to disgust, suggesting rejection is often an emotionally-driven response rather than irrational noise.
Raising the stakes shrinks but does not erase the effect: Lisa Cameron's 1999 study in Indonesia, which scaled the pot to amounts meaningful relative to local income, still found a persistent, fairness-driven rejection rate even at high stakes. The practical lesson: a number that is technically positive, and therefore "rational" to accept by pure wealth-maximizing math, can still get rejected in the real world if it reads as insulting. Designing any take-it-or-leave-it offer means budgeting for the fairness threshold, not just the arithmetic.
You have $100 to split with a Responder you'll never interact with again. Drag the slider to decide how much to offer them, then send it. The simulated Responder follows the real offer–rejection curve described above: near-certain acceptance above roughly 40–50%, rising rejection risk as the offer drops below roughly 20–30%, and near-certain rejection near $0.
Estimated acceptance likelihood at this split: ~98%
No offer sent yet
Pick a split above and click "Send this offer" to see how a simulated Responder reacts.
Same pattern, six different rooms. Click a tab. (Composite illustrations — not real named individuals or companies.)
A candidate receives a final offer letter framed as "this is what we can do." If the number sits clearly below what the candidate believes is fair for the role and market, candidates who could easily use the money anyway sometimes still walk away or push back hard — a real-world echo of Responders rejecting a technically-positive split that still reads as insulting.
A target company receives a single buyout term sheet with no further rounds promised. A split of the deal's value that departs too far from what the sellers consider fair — even if it's technically profitable for them — can collapse the whole deal, because sellers are often willing to walk from real money rather than accept terms that feel exploitative.
Management presents a "final" contract offer to a union membership vote. An offer that is a real improvement over the status quo can still be voted down if it falls well short of what members see as a fair share of the gains they helped produce — rejecting it costs members money too, and it still happens.
A company offers a standardized exit package with no individual negotiation. Packages set right at the legal minimum tend to generate more disputes, slower sign-off, and more reputational fallout than packages set a bit above it, even though both are "better than nothing" for the departing employee.
A supplier presents a single non-negotiable price on a renewal. Buyers who feel the number fairly reflects costs tend to simply pay it; buyers who feel it's an opportunistic squeeze sometimes walk to a worse alternative on principle, costing themselves money to avoid rewarding what felt like an unfair ask.
One country proposes a final tariff or quota arrangement with no further talks before a deadline. Terms that read as lopsided — even when accepting would be the narrowly "rational" economic choice for the smaller country — are sometimes rejected anyway in favor of retaliation, because the political cost of appearing to accept an unfair deal can outweigh the direct economic loss.
Picture a company handing out one-time severance packages during a layoff, with no further negotiation offered — a close real-world cousin of the Ultimatum Game: one side names a number, the other can largely only accept it or walk away (often to a lawyer, which is the real-world version of "both sides end up worse off").
Why the gap runs this direction: nothing about the underlying facts of the layoff changed between the two offers — the same role, the same tenure, the same budget. What changed is whether the number read as a fair share or a minimum squeeze. The same inequity-aversion mechanism that makes a lab Responder reject a low split shows up here as a signed release that doesn't get signed, a formal complaint, or a reputational cost that outlasts the layoff itself — all paid for the same reason a Responder in the lab walks away from free money: because the offer felt unfair, not because the math was wrong.
Pick the situation you're facing, and build a starting prompt you can take to any AI assistant to reason through it using this model.
Keep this model on hand as a reusable skill for yourself or an AI assistant.