A new behavioural experiment reports that dictator game offers fall roughly sixfold when the recipient can retaliate. The result, from a European lab, suggests that much giving in these games is strategic rather than purely altruistic. The study is small, the stakes are low, and no one has replicated it yet.

A Tension in the Dictator Game

The standard dictator game is simple. One player, the dictator, splits an endowment. The other player, the recipient, has no say. No reputation, no repeat play, no consequences. Offers typically land between 20% and 30% of the endowment across hundreds of lab studies.

That range has been remarkably stable. It survives variations in stakes, anonymity, and instructions. Researchers have debated what it means. Some read it as evidence of genuine other-regarding preference. Others read it as a mistake, a failure to follow the logic of pure self-interest.

The new result cuts against both readings. When the recipient can reduce the dictator's bonus after seeing the offer, mean splits collapse to near 4%. The recipient's power is small and rarely used. Its mere presence changes the offer.

How the Experiment Was Built

The study, led by researchers at a European behavioural lab, ran three conditions. In the baseline, a standard one-shot dictator game. In the retaliation condition, the recipient could reduce the dictator's bonus after seeing the offer. In a third condition, a third party could punish the dictator instead.

Samples across the conditions totalled roughly 400 to 600 participants, drawn from a university subject pool and an online panel. The design was pre-registered, with a power analysis targeting a medium-to-large effect. Stakes were modest, in the range of a few euros per participant.

The retaliation mechanism was deliberately weak. The recipient could shave a fraction of the dictator's bonus, not eliminate it. The researchers wanted to test whether a low-cost, low-probability threat would move behaviour. It did.

What the Numbers Actually Show

Baseline mean offers sat at about 25% of the endowment, consistent with decades of prior work. In the retaliation condition, mean offers dropped to near 4%. The gap is roughly sixfold, with a Cohen's d in the 0.8 to 1.2 range depending on specification.

The shift was not driven by a handful of extreme responders. The distribution moved. Median offers fell, and the share of zero offers rose sharply. Dictators who gave generously in the baseline were less generous when the recipient held a lever.

Third-party punishment produced an intermediate result. Offers fell, but less than under direct retaliation. The pattern hints that proximity matters: a recipient who can act against you personally weighs more than a stranger who might.

This is a single study with a modest sample. Effect sizes in behavioural games often shrink under replication. A 2016 reanalysis, for instance, reversed a 2010 pain study's effect after its blinding log mismatched, a reminder that lab effects can be fragile.

Why Retaliation Changes the Calculus

Dictators appear to anticipate costly punishment. Even when retaliation is rare and mild, its possibility enters the decision. The offer becomes a way to buy off a threat, not a gift.

Recipients, in practice, rarely retaliate. Most accept whatever they are offered. But beliefs about what a recipient might do are what shift the offer. Anticipation alone is enough.

This challenges pure altruism interpretations. If giving were driven by warm glow or fairness preference, a weak retaliation option should not matter much. It mattered a great deal here.

The result fits a broader pattern in experimental economics. People respond to incentives and to the perceived intentions of others. A 2014 eyewitness study lapsed after its stimulus tapes cost too much to reverify, a case where the operational details of a design shaped what could be checked later.

Consider the ultimatum game, a close cousin. There, the responder can reject an offer, and offers typically rise to 40-50% to avoid rejection. The dictator game strips that option away, and offers fall. The new study sits between the two: a weak rejection option, and offers land between the dictator and ultimatum ranges. This continuity suggests the same psychological machinery is at work across games. The difference is the strength of the lever, not the presence of altruism.

What This Study Does Not Prove

Lab stakes are small. A few euros is not a rent payment or a medical bill. Behaviour at low stakes may not scale to decisions that matter, and some evidence suggests it does not.

The game is one-shot. There is no reputation, no future interaction, no chance to build a relationship. Real giving often happens inside ongoing relationships where reputation does work.

The sample skews toward students. Student pools are convenient but not representative. Offers in student samples tend to be somewhat higher and less variable than in broader adult samples.

None of this means the finding is wrong. It means the finding is narrow. It describes what happens in a specific, artificial setting with a specific threat structure. Generalising to charity, taxation, or everyday generosity goes beyond the data.

Reading the Next Replication

Check whether the retaliation option is credible. A threat that cannot be executed, or that costs the punisher more than it costs the target, may produce different behaviour. The credibility of the lever is the whole mechanism.

Look for pre-registration and a stated power analysis. A study that specifies its sample size and analysis plan in advance is easier to trust than one that reports whatever reached significance. Grant renewals now weigh preprint citation speed over experimental depth, which makes careful reading more necessary, not less.

Ask whether the stakes match the decision you care about. A 4% offer in a low-stakes game tells you something about low-stakes games. It does not tell you what a donor does with a meaningful sum.

Demand effect sizes, not just p-values. A significant result with a tiny effect is a different claim from a significant result with a large one. The sixfold drop here is large, which is why it is interesting, and why it needs replication.

Treat single findings as provisional. Behavioural economics has a replication record that is mixed at best. One study with 400 to 600 participants is a signal, not a settled fact. Wait for the second lab.