A thousand people were shown Rokia. The effect wasn't there.

Fundraising's most quoted rule says one named child beats a statistic. A preregistered replication with 1004 participants found no effect at all.

Every fundraiser learns the same lesson early. Don't say nine hundred thousand people are hungry. Say Rokia is seven, she lives in Mali, and your money buys her food.

The lesson has a name — the identifiable victim effect — and a paper trail going back to Karen Jenni and George Loewenstein in 1997. It got its most quoted demonstration in 2007, when Deborah Small, Loewenstein and Paul Slovic ran a set of experiments with real money and found something stranger than the effect itself. Teaching people about the bias didn't fix it. People who were told that a single named child pulls harder on the heart than a statistic simply gave less to Rokia. They did not start giving more to the statistics. Explaining the bias produced less charity overall, not better-aimed charity.

That finding is properly uncomfortable, which is part of why it travelled so well. It has been cited in fundraising handbooks, effective-altruism arguments, journalism ethics seminars. I've used it myself in conversations about how this site should describe the people on it.

In 2023 three researchers went back and checked.

What the replication did

Maximilian Maier at University College London, with Yik Chun Wong and Gilad Feldman at the University of Hong Kong, published a preregistered replication and extension in Collabra: Psychology. They recruited 1004 US participants, folded Studies 1 and 3 of the original into one design, and crossed three ways of presenting the victim — identified, statistical, or both together — against being told about the bias or not.

Their result, in their notation: η² = .000, with a 95% confidence interval of [.000, .003]. In plain language, no identifiable victim effect. Not a smaller one than 2007. Not one that appears in some conditions. Nothing measurable at a sample size powered to catch an effect of 0.012.

The debiasing finding didn't show up either, which follows — you can't remove something that isn't there.

They had gone in suspicious. Before running anything they reanalysed the existing meta-analysis on the effect, by Lee and Feeley from 2016, using robust Bayesian methods that account for publication bias. The reanalysis suggested the literature was inflated by studies that found the effect getting published while studies that didn't stayed in drawers.

The part most people will skip

The authors put a limitation in their own abstract, and it is a serious one. Their participants made hypothetical donations. Small and colleagues in 2007 paid people for a task and then asked them to give real earnings away.

That is not a small difference. Money you actually lose behaves differently from money on a screen. The 2023 team says so directly and calls for a high-powered replication with real donations, run as a registered report, ideally in collaboration with the people who proposed the original effect. That last clause is the opposite of the tone replication fights usually take.

So the honest summary lands short of "the identifiable victim effect is fake": a claim taught as settled for a quarter of a century now has one large, well-run test against it, and a good reason that test may not be the last word.

Scope insensitivity, which is worse

Maier and colleagues offer a different reading of what's going on. Perhaps people were never giving more to the one named child. Perhaps they were failing to give more to the many.

Reframed that way, a face doesn't beat a number; our response barely moves whether the number is one or nine hundred thousand. Flip the effect around and a curiosity about storytelling turns into something bleaker, about the shape of human attention.

I don't know which reading is right. Neither, at the moment, does anyone.

What we do with this

Not much, immediately, and that's the point I'd make to anyone running a small charity and now wondering whether their whole appeal strategy rests on nothing.

Telling the story of one person is still the right thing to do, for reasons that have nothing to do with donation psychology. It is more accurate. A number describes a condition; a person describes a life, and the people we ask you to help on this map are individuals whether or not that raises more money. What changes is the confidence with which anyone should say that the story works better. That sentence needs a "probably" in front of it now.

There's a habit worth borrowing from Maier's group, separate from anything they found. They preregistered what they expected, published data and code, invited the original authors in, and stated the weakness of their own design in the abstract rather than the fourth paragraph of the discussion. Whatever the identifiable victim effect turns out to be, that is how you'd want it settled.

Sources: Maier, M., Wong, Y. C., & Feldman, G., "Revisiting and Rethinking the Identifiable Victim Effect: Replication and Extension of Small, Loewenstein, and Slovic (2007)", Collabra: Psychology 9(1), 2023, doi 10.1525/collabra.90203, data and code at osf.io/n4jkh. Original: Small, D. A., Loewenstein, G., & Slovic, P., "Sympathy and callousness", Organizational Behavior and Human Decision Processes, 2007. Meta-analysis reanalysed: Lee & Feeley, 2016. Read 9 September 2026.

The effect as it is usually told, with the figures that circulate in the secondary literature, is in our earlier piece One Child Moves More Than Eight.