In 1982, a small group of German economists published a paper that would quietly reshape how psychologists and economists measure fairness. The paper introduced the ultimatum game, a two-player bargain where one player proposes a split of money and the other can accept or reject. A rejection leaves both with nothing. That veto gave the responder power, and the game seemed to capture something real about human fairness.

Then, almost as an afterthought, the same authors described a stripped-down version. The responder lost the veto. The proposer simply dictated the split. That version became the dictator game, and it has been run in hundreds of labs ever since. But the change from bargaining to giving was not a neutral simplification. It quietly introduced a stakes confound that has shadowed the field for four decades.

This is a story about a single word, a design choice, and the slow, unglamorous work of untangling what an experiment actually measures. It is also a reminder that the history of a method is not a smooth line from crude to refined. Sometimes the field takes a step sideways, and it takes a generation to notice.

A Single Word Change That Reshaped a Field

The shift from ultimatum to dictator was not a discovery. It was a methodological convenience. The ultimatum game required two active players, a proposer and a responder, and the interaction between them. The dictator game required only one. It was cheaper, faster, and easier to run in a classroom or a lab.

Researchers wanted a clean measure of pure altruism, untainted by strategic fear of rejection. By removing the veto, they thought they could see what people would give when there were no consequences. The word 'dictator' captured the asymmetry: the proposer had all the power, the recipient had none.

But the change did more than remove a strategic layer. It changed the stakes for both roles. In the ultimatum game, the responder's rejection could cost the proposer real money. In the dictator game, the recipient had no way to punish selfishness. The proposer's stake was trivial: giving away a few dollars cost little, and keeping them was easy.

The hidden confound was not that stakes were low. It was that stakes were unequal across roles. The proposer faced a small loss either way. The recipient faced a passive gain or loss of the same magnitude, but with no agency. That asymmetry, baked into the design, made the dictator game a measure of something narrower than 'fairness'.

For a field that prided itself on clean controlled comparisons, the confound was invisible for years. The dictator game became a standard tool, cited in textbooks and used in hundreds of studies, without anyone asking whether the stakes structure was doing some of the work.

The Ultimatum Game's Bargaining Baseline

The 1982 paper by Werner Guth, Rolf Schmittberger, and Bernd Schwarze is the origin point. They ran the ultimatum game with real money, not hypotheticals. Proposers offered a split of a fixed sum, and responders could reject. Rejection meant both got zero.

Their results were striking. Many proposers offered roughly 40–50% of the pie. Responders rejected offers they judged too low, even though rejection cost them money. That behavior contradicted the standard economic prediction: a rational responder should accept any positive amount, and a rational proposer should offer the smallest possible nonzero sum.

The fairness that appeared in the ultimatum game was strategic. A proposer offered a decent split partly because they feared rejection. A responder rejected a low offer to punish the proposer, even at personal cost. The game captured fairness in a context of interdependence.

That interdependence is exactly what the dictator game removed. In the dictator game, there is no punishment, no strategic reason to be fair. The only reason to give is intrinsic. So the dictator game was meant to isolate intrinsic fairness from strategic fairness. But the isolation came at a price.

How the Dictator Game Dropped the Teeth

The dictator game procedure is deceptively simple. A proposer receives an endowment, say five dollars, and decides how much to pass to an anonymous recipient. The recipient cannot respond. The proposer keeps the rest. The game is one-shot, anonymous, and there is no reputation at stake.

In early dictator games, proposers gave away a surprising share, often around 20–30% of the endowment. That seemed to suggest that people were intrinsically generous, even without any threat of punishment. The finding was widely celebrated and became a cornerstone of behavioral economics.

But the design had a problem. The proposer's stake was small, typically a few dollars. The recipient's stake was equally small but entirely passive. The proposer faced no risk, no social pressure, and no consequence for keeping everything. The only cost of giving was the small amount of money forgone.

That cost was trivial. For most participants, five dollars is not a meaningful sacrifice. The absence of a veto meant that the proposer could act selfishly without any downside. The game measured what people would do when being selfish was nearly free.

The 'teeth' that the ultimatum game had, the responder's veto, were not just a strategic detail. They were what made the stakes matter. Without them, the game became a test of generosity under almost no pressure. And that is a very different thing from fairness under constraint.

The Stakes Confound: Unequal Risk on Both Sides

The confound is not that stakes were low. It is that stakes were asymmetric. In the ultimatum game, both players faced a real risk: the proposer could lose the whole pie if the responder rejected; the responder could lose the offered amount by rejecting. Both had something to lose.

In the dictator game, the proposer had nothing to lose. They could keep the entire endowment and walk away. The recipient had nothing to gain or lose actively; they just received whatever was sent. The asymmetry meant that the proposer's decision was not about fairness under threat, but about fairness when giving was almost free.

Meta-analyses of dictator game studies have shown that giving rates vary widely, from near zero to over 50%, depending on small procedural details. One of the most influential variables is the size of the stake. When the stake is a few dollars, giving is modest. When the stake is larger, giving tends to drop, though the effect is not always linear.

Some estimates put the stake effect at roughly 1–2 percentage points of giving per additional dollar of endowment. That may sound small, but it compounds across the range of stakes used in the literature. A study using one-dollar stakes might show 30% giving, while a study using one-hundred-dollar stakes might show only 10%.

The confound matters because it makes the dictator game a test of how people behave when money is nearly meaningless. That is not the same as how they behave when money matters. The field built a large literature on a measure that was, in effect, calibrated to a very specific and unusual stakes level.

Replication and Critique: The Stakes Debate

The critique gained traction in the early 2000s. A 2002 study by Todd Cherry, Peter Frykblom, and Jason Shogren found that when proposers had to earn their endowment, giving dropped dramatically. Participants who worked for the money kept almost all of it, giving only a few percent on average.

That result suggested that the endowment's origin, not just its size, mattered. Money that felt earned was guarded more fiercely than windfall money. The finding was a blow to the idea that the dictator game measured a stable trait of generosity.

John List, a prominent experimental economist, showed in 2007 that demand effects could explain a large part of dictator giving. When participants could choose to take money from a box instead of being handed it, giving fell sharply. The act of being handed money created an implicit expectation to share.

Replication efforts have been mixed. Some studies replicate the basic finding of modest giving, while others show near-zero giving under stricter conditions. The stakes confound is not fully resolved because it is not a single variable but a bundle of related design choices: stake size, stake origin, stake asymmetry, and the social context of the experiment.

A 2012 review by Barmettler, Fehr, and Zehnder concluded that the dictator game's sensitivity to design details makes it a poor tool for measuring intrinsic social preferences. They argued that the game is better understood as a measure of situational generosity, not a stable trait.

Practical Lessons for Designing Giving Experiments

The history of the dictator game offers concrete lessons for anyone designing a study of giving or fairness. The first is to match stakes across roles. If one player faces a real loss and the other faces a passive gain, the comparison is confounded.

The second is to use real money, not hypotheticals. Hypothetical stakes produce different behavior, often more generous, because the cost of being generous is imaginary. Real money, even small amounts, forces participants to weigh an actual trade-off.

The third is to control for the origin of the endowment. If participants earn the money, giving drops. That is not a bug; it is a feature of how people value earned versus windfall resources. A study that ignores this will overestimate generosity.

The fourth is to report effect sizes with confidence intervals. The dictator game's giving rates vary so much that a point estimate without an interval is almost meaningless. Readers need to see the range of plausible effects.

Finally, pre-register the analysis plan. The dictator game's flexibility creates many researcher degrees of freedom. Pre-registration forces decisions about stake size, randomization, and analysis to be made before data collection, reducing the risk of post-hoc selection.

None of these lessons is new, but they are often ignored in practice. The dictator game remains popular because it is easy to run, not because it is well understood. A careful researcher should treat it as a fragile instrument that needs calibration, not as a fixed tool.

What the Confound Teaches About Methodology

The dictator game's history is a cautionary tale about the power of small procedural choices. A single word change, from 'ultimatum' to 'dictator', altered the stakes structure and, with it, the meaning of the result. The field did not notice for years because the change seemed innocuous.

The lesson is not that the dictator game is useless. It is that experimental measures are not neutral windows into human nature. They are constructed instruments that embed assumptions about what matters. The stakes confound is one such assumption, hidden in plain sight.

Replication alone cannot fix a confound. Replicating a flawed design produces the same flawed result. The field needed theory to see the problem, and it took critics like Cherry, List, and the 2012 reviewers to articulate why the design was problematic.

The slow path from the 1982 paper to the current understanding mirrors the broader arc of behavioral science. A promising result, a wave of enthusiasm, a period of replication, and then a hard look at the method. The dictator game is still used, but it is used with more caution, and its results are interpreted with more humility.

The confound also teaches that stakes are a design variable, not noise. Every experiment sets a stakes level, and that level shapes behavior. Researchers should report it, vary it, and think about what it means. The dictator game's failure to do so is a reminder that methodology is not a set of rules but a set of choices, each with consequences.

As the field moves forward, there is a growing effort to design giving experiments that control for stakes asymmetry, such as using endowments that are earned by both players, or using larger stakes, or adding a real cost to giving. These improvements are welcome, but they will not erase the confound from the historical record. The 1982 shift remains a lesson in how a tiny procedural change can reshape a field, for better and for worse.

Beyond the Lab: Field Experiments and the Stakes Question

The stakes confound is not confined to the laboratory. Field experiments on giving and fairness have inherited similar issues, often with real-world consequences. Consider a study that measures charitable donations in a natural setting, such as a door-to-door fundraising campaign. The 'endowment' is the money in the donor's pocket, but the stakes are asymmetric: the donor faces a small loss if they give, while the recipient organization gains a small amount. The same confound appears, though it is rarely named.

One notable field experiment by List and his colleagues examined giving in a market setting, where participants could buy and sell goods. They found that when participants had to part with their own money, giving rates were lower than in lab dictator games with windfall endowments. This suggests that the stakes asymmetry is not an artifact of the lab but a fundamental feature of how people value resources they have earned.

Another field experiment, conducted in a developing country, varied the size of the endowment in a dictator game played with real money. The researchers found that giving rates declined as the stake increased, but the decline was modest. This echoes the lab findings and reinforces the idea that the stakes level is a critical design parameter.

These field results have practical implications. If policymakers use dictator game results to predict charitable giving or to design tax policies, they must account for the stakes asymmetry. A policy that assumes people will give generously based on lab experiments with small stakes may be disappointed when real-world stakes are higher.

Moreover, the stakes confound extends to other experimental paradigms, such as public goods games and trust games. In public goods games, contributions often decline when the marginal return on contribution is low, which is a form of stakes asymmetry. In trust games, the trustee faces a passive recipient role, similar to the dictator game's recipient. Understanding the stakes confound in the dictator game can inform the design of these related games.

For instance, a recent study on trust games varied the stake size and found that trust and trustworthiness declined with higher stakes. This is consistent with the idea that the stakes confound is a general phenomenon, not unique to the dictator game. Researchers who study trust should be aware of this pattern and design their experiments accordingly.

In summary, the stakes confound is a pervasive issue in experimental economics. It affects not only the dictator game but also many other paradigms. By acknowledging and addressing this confound, researchers can improve the validity of their findings and make more accurate predictions about real-world behavior.