Two strangers are handed a sum of money and one rule. The first player proposes how to split it. The second player either accepts, in which case both walk away with their shares, or refuses, in which case both walk away with nothing. That is the whole game. The structure was described by John Harsanyi in 1961 and given its first lab version in 1982 by Werner Güth, Rolf Schmittberger and Bernd Schwarze, who recruited students, put them in separate rooms, and watched them divide money under conditions where a crude calculator could have told you the only sensible answer. The sensible answer is that the second player accepts anything above zero, because a coin beats no coin. Knowing that, the first player offers the smallest unit the rules allow. Backward induction gives a single clean prediction for the whole interaction: the proposer takes almost everything, and the responder says thank you.
Nobody plays it that way. Not the students in that first run, not the thousands of subjects since, not the small-scale societies where researchers have carried the game for the last twenty-five years. The ultimatum game is one of the most replicated findings in experimental economics, and what it replicates is a population of people who will pay real money to punish a bad offer.
🧠 What Actually Happens
A 2004 review of 37 ultimatum-game studies put the average proposal at roughly 40 percent of the pot. Offers below 30 percent are frequently turned down, meaning the responder walks away with zero rather than let the proposer keep an amount that feels contemptuous. The rejection is not a bluff and it is not a show for an audience. The game is deliberately one-shot and anonymous, designed so that refusing cannot buy you a reputation or a future favor. You are simply choosing to light your own share on fire.
Two small details in the data make it more interesting. Average offers shrink as the pot gets bigger, and they shrink as players gain experience with the game. Neither pattern is comfortable for anyone who wants fairness to be a fixed human constant. Offer norms are closer to a negotiated local price than a moral absolute.
| What the theory says | What the lab keeps finding |
|---|---|
| Proposer offers the minimum | Proposer offers about 40 percent |
| Responder accepts anything above zero | Offers under 30 percent get refused |
| Refusal is irrational and never happens | Refusal is common, and it is costly to the refuser |
🤔 Two Systems, and Only One of Them Takes the Money
In 2003, Alan Sanfey and colleagues at Princeton put ultimatum-game players in an fMRI scanner and watched them read offers. Unfair proposals lit up two regions at once: the anterior insula, an area tied to disgust and visceral emotion, and the dorsolateral prefrontal cortex, the workhorse of deliberate control. When the researchers separated accepted from rejected unfair offers, the anterior insula activity was significantly higher on the rejections. Something closer to revulsion than arithmetic was running the decision.
Then a 2006 study in Science by Daria Knoch and colleagues took the same game and temporarily switched off the right dorsolateral prefrontal cortex with low-frequency repetitive transcranial magnetic stimulation. Subjects became substantially more willing to accept deliberately unfair offers. Here is the detail that makes the study famous: they still described those offers as very unfair. Their moral judgment was untouched. What the stimulation removed was the capacity to act on the judgment against their own economic interest.
The usual story about that brain region is that it holds the rational adult in check while the emotional toddler wants things. In this game, the roles are reversed. The insula supplies the anger at being treated badly, and the prefrontal cortex supplies the muscle to do something about that anger at a personal cost. Fairness needs both the grievance and the follow-through.
This is also why the rejections do not look like cool strategic teaching. A 2014 study by Carey Morewedge, Sanjeev Krishnamurti and Dan Ariely found that intoxicated players rejected unfair offers more often than sober ones — which fits the account that refusal is a prepotent impulse that self-control normally holds in check rather than a calculated investment in someone else’s future behavior. Both explanations have supporters, and the argument is far from settled.
🔗 The Refusal Is Doing Work Somewhere
If costly punishment were pure waste, evolution should have flattened it long ago. It has not, because the willingness to burn your own share is what makes low offers expensive to propose in the first place. Ernst Fehr and Simon Gächter showed in Nature in 2002 that in a public-goods game, cooperation flourishes when players can punish free riders at their own expense, and collapses to nearly nothing when punishment is not available. The threat works without anyone ever having to collect on it. A seller who knows that a mean split gets rejected offers closer to half. The punishment only needs to be credible.
You can watch the same mechanism in commercial settings, where the currency is loyalty rather than cash. A price that clears the market can still read as an insult, and buyers respond by punishing at their own cost: cancelling an order they wanted, writing the review, churning the subscription, refusing the negotiated discount on principle. Discount structures that cut the newest customers in and leave the earliest ones paying full freight trigger this reliably. So do refund policies that seem to assume the customer is lying, and trial tiers with the useful feature surgically removed.
The design implication is that the number is not the whole offer. Who is making it, whether it looks deliberate, and what the other side visibly gave up all feed into whether the split reads as fair or as a dare. When an offer is refused at a loss to both sides, the deal was never the point.
There is a version of this in any story worth reading. The scene where a character turns down a deal that would clearly help them is not evidence of a badly written character. It is evidence that their price list has something on it that money is not denominated in, and the reader is about to find out what.
🎲 Three Things to Do With This
Run it on a friend. Offer to split ten dollars and take eight for yourself, and watch how fast a rational person becomes willing to leave with nothing. Then make the same offer with a reason attached, and watch the accepting get easier. You have just reproduced the entire literature in one kitchen.
Notice the next refusal you make. If you catch yourself walking away from something you wanted because of how the terms were framed rather than what they were worth, you have located your own 30-percent threshold. Most people can name the number after the fact and cannot name it in advance.
And a piece of field evidence from the cross-cultural version: when researchers took the game to Mongolian herders, the proposers offered even splits even though they knew very unequal offers would almost always be accepted. The norm ran the table instead of the calculation. Whatever rule you think you are following, someone else is playing under a different one, and neither of you is running the numbers.
Sources: Harsanyi’s 1961 game structure and the 1982 first study by Güth, Schmittberger and Schwarze; Camerer’s 2004 review of 37 studies (average offers near 40 percent, offers under 30 percent often refused); Sanfey et al., Science, 2003; Knoch et al., Science, 2006; Fehr and Gächter, Nature, 2002; Morewedge, Krishnamurti and Ariely, 2014; cross-cultural offer variation reported in the literature on small-scale societies.