One Child Moves More Than Eight

Rokia raised twice what the statistics did. Add statistics to Rokia and giving fell. Explain the bias and giving fell again — downward, not up.

In 2007 three researchers handed people five dollars and then asked for some of it back.

Deborah Small, George Loewenstein and Paul Slovic paid participants for filling in a survey, then offered them an envelope and a chance to give to Save the Children. Some read about Rokia, a seven-year-old girl in Mali, with her photograph. Others read the numbers: millions of children short of food in Malawi, millions of Zambians facing hunger, four million Angolans driven from their homes.

Rokia got roughly twice as much money.

The number that should have helped

The obvious reading is that people are bad at arithmetic and good at faces. That reading is too flattering to the numbers. The statistics in that study were not abstract in any way that mattered — they described exactly the same catastrophe Rokia was living through, only larger. Adding scale should have added urgency. It removed it.

The study also ran a third condition, and this is the one worth sitting with. Participants were shown Rokia and the statistics together. Giving fell below what Rokia alone had produced. The numbers did not amplify the girl. They interfered with her.

This is not a new observation, only a well-measured one. Fifteen years earlier a team of economists asked people what they would pay to keep migrating birds from dying in oil ponds. One group was told two thousand birds. Another twenty thousand. A third two hundred thousand. The answers came back at around eighty dollars in all three cases. A hundredfold difference in the harm produced no difference at all in the response.

Then they told people about the bias

Here is where the study stops being a curiosity about donations and becomes something more uncomfortable.

In a further condition, participants were nudged into analytical thinking before they decided — in some versions simply by being told that people tend to give more to identified individuals than to statistical victims. You would expect the correction to run upward: told about the distortion, participants compensate, and giving to the statistics climbs to meet the giving to Rokia.

That is not what happened. Giving to Rokia fell.

The correction ran downhill. Knowing that your sympathy is unreliable did not make you generous toward the millions. It made you stingier toward the one. Whatever the feeling was doing, thinking about it switched it off, and nothing arrived to take its place.

Why a group of eight does worse than one

Kogut and Ritov pushed on the same seam a couple of years earlier. They compared a single identified sick child with a group of eight identified sick children — named, pictured, equally real. The single child drew more.

Eight children is a better reason to act than one child. It is also, apparently, a worse trigger. Somewhere between one and eight the mind stops seeing a person and starts seeing a category, and the category does not move anybody.

What to do with an inconvenient finding

There is a tempting conclusion here that I think is wrong: that since feeling is unreliable, we should give by spreadsheet and ignore the face. The evidence does not support it. When people were pushed toward the spreadsheet, they did not become better calculators of need. They gave less. Deliberation in these experiments functioned less like a corrective lens and more like a dimmer switch.

The honest version is less tidy. The feeling is the engine. It is a badly calibrated engine — it cannot tell two thousand from two hundred thousand, and it stalls when you explain it to itself — but it is the only one that reliably starts. The work is not to shut it off. It is to point it at real people often enough that it keeps running, and to let deliberation choose where the help goes rather than whether to feel anything at all.

Which is roughly the argument for putting individual acts on a map instead of a counter on a homepage. A number that says four hundred people were helped this month is information. It is not, on this evidence, a reason to do anything. One person, named, in a place you recognise, is both.

The Rokia experiment was rerun in 2023 with just over a thousand participants, and the headline effect did not appear. We went through that replication in A thousand people were shown Rokia. The effect wasn't there.

Sources: Small, D. A., Loewenstein, G. & Slovic, P., "Sympathy and callousness: The impact of deliberative thought on donations to identifiable and statistical victims," Organizational Behavior and Human Decision Processes, vol. 102, no. 2 (2007). Kogut, T. & Ritov, I., "The 'identified victim' effect: an identified group, or just a single individual?", Journal of Behavioral Decision Making, vol. 18 (2005). Desvousges, W. H. et al., contingent-valuation work on migratory birds (1992), widely discussed by Kahneman and others as the standard illustration of scope insensitivity. Checked 30 August 2026. The dollar amounts and the bird figures are reported as they appear in the secondary literature; the direction and size of the effects are robust and repeatedly replicated, but exact per-person averages vary between the studies' individual experiments, so treat the specific figures as illustrative rather than as precise constants.