Let people take instead of give, and giving falls from 71 to 10 percent. The people did not change. The menu did.

List (2007): 71 percent of students gave to a stranger. Add the option to take, and 10 percent give. Nobody changed; the menu did. What that means for measuring kindness.

In the spring of 2005, students at the University of Maryland sat in two rooms that had no contact with each other. Everyone in both rooms got five dollars. The students in room A got five more, and one instruction: you may pass any part of this extra five dollars, in fifty-cent steps, to a stranger in room B. Nobody would ever know who gave what.

Seventeen of the twenty-four gave something. The average gift was $1.33, about a quarter of the stake. That is the dictator game, and that result is the one in the textbooks: most people share, even when the stranger is anonymous and no one is watching.

John List, the economist who ran the study, then changed one line of the instructions.

One extra option

In the second version, everything was identical except that the menu now ran from minus one dollar to plus five. The dictator could still give up to five dollars. He could also take one dollar from the stranger in room B.

The share of givers dropped from 71 percent to 35 percent. Sixteen of forty-six gave anything at all. The median choice was zero.

In the third version the menu was symmetric: take up to five, give up to five. Five of fifty gave something. Ten percent. The median choice was to take $4.50, which means at least half the room took four and a half dollars or the whole five. The mean outcome for the stranger was minus $2.48.

Nothing about the people had changed. Only the list of allowed actions had.

Not everyone took everything

Even in the symmetric version, where taking the whole five dollars was as easy as ticking a box, the room did not collapse into pure self-interest. Plenty took. Some took nothing. A few still gave. List's own summary: agents do not ubiquitously choose the most selfish outcome, regardless of which actions are on the menu. He also notes, in a footnote, that nearly twice as many subjects took the full five dollars when five was allowed as took the full one dollar when only one was allowed. The moral cost of taking exists, but for most students it was somewhere under five dollars.

The fourth version made the money earned instead of handed out. Before the allocation, everyone in both rooms spent thirty minutes sorting and stuffing envelopes for a charity mailing. Room A was told they had earned ten dollars, room B five. Then the symmetric menu again.

Two thirds of the dictators now chose the neutral point. Neither gave nor took. Only thirteen of forty-seven took anything. Dictators treated money they had worked for differently from windfall money, and the price of taking it went up.

What the game had been measuring

List's paper, published in the Journal of Political Economy in 2007, had a narrower target than human kindness: an interpretation. For twenty years, giving in dictator games had been read as a direct readout of social preferences. People share because they care about fairness, and here is the number.

But if a preference is a preference, adding a choice nobody is forced to make should not move it. It moved it from 71 to 10.

List's explanation runs through social norms. In the standard game, an authority figure has just handed you money and asked whether you would like to share it. A zero looks like pure greed. In the game with a taking option, a zero sits in the middle of the menu. It becomes a decent, neutral choice, and the dictator now has what List calls the moral authority to give nothing. Besides limiting what people can do, the menu tells them which norm applies.

Nicholas Bardsley had found the same pattern independently, with a different design, and the two papers are usually cited together.

The hundred-experiment picture

Christoph Engel's 2011 meta-study pooled more than a hundred dictator game experiments. On average, dictators gave 28 percent of the pie. But the average conceals the shape of it: a third gave nothing, about one in six split exactly in half, and one in twenty gave everything away.

"People share about a quarter" is true on average and useless for any one person. Who gives, and how much, depends on who is watching, whether the money was earned, and, as List showed, which actions the sheet allows.

Why this matters off the map

Good People Map records deeds in a world that looks a lot like List's third room. Nobody hands you a stake and asks you to share. Taking, in its mild everyday forms of not stopping and not noticing, is permanently on the menu and costs nothing. In that world the laboratory's 71 percent is not the baseline. Ten percent is closer. Engel's third who give nothing is closer still.

Which is what makes the recorded deeds worth something. A helped neighbour in a setting where helping was optional and unobserved is a different kind of fact from a dollar passed across a laboratory partition. It is the fact the experiment was trying, and failing, to measure.

For anyone designing a place where people help each other, including us, the options you put in front of people change what they do, far more than any poster about kindness. Make the helpful action the default, the thing that takes no heroism. And for anyone reading the next headline about how generous humans are: ask what the menu looked like.

Sources

• List, J. A. (2007). "On the Interpretation of Giving in Dictator Games." Journal of Political Economy, 115(3), 482–493. Table 1: Baseline (24) 0.71 positive offers, mean $1.33; Take($1) (46) 0.35, median $0, mean $0.33; Take($5) (50) 0.10, median −$4.50, mean −$2.48; Earnings (47) 0.06, 66 percent at the neutral point, 13 of 47 took. Footnote 6 on taking $5 versus $1. University of Maryland, spring 2005. voices.uchicago.edu (PDF)
• Bardsley, N. (2008). "Dictator game giving: altruism or artefact?" Experimental Economics, 11(2), 122–133.
• Engel, C. (2011). "Dictator games: a meta study." Experimental Economics, 14(4), 583–610. Mean gift 28.35 percent; 36.11 percent give nothing; 16.74 percent choose the equal split; 5.44 percent give everything. Max Planck preprint 2010/07
• Kahneman, D., Knetsch, J. L. & Thaler, R. H. (1986). "Fairness as a Constraint on Profit Seeking." American Economic Review, 76(4), 728–741. Three quarters chose the equal split of a hypothetical $20.
• Forsythe, R., Horowitz, J. L., Savin, N. E. & Sefton, M. (1994). "Fairness in Simple Bargaining Experiments." Games and Economic Behavior, 6(3), 347–369.