Effective altruism's most distinctive claim is an empirical one. Charities and programs differ enormously in how much good they do per dollar, and the differences are so large that choosing well can matter more than giving more.
Differences of thousands
In a 2013 essay, Toby Ord drew on cost-effectiveness estimates from a large World Bank-backed study, Disease Control Priorities in Developing Countries. Among interventions against HIV and AIDS alone, the most cost-effective was education for high-risk groups, estimated to be about 1,400 times as cost-effective as the least, treatment of Kaposi's sarcoma, a cancer common among people with AIDS. Across the full range of health interventions the spread was wider still. Ord argued that differences in cost-effectiveness between health interventions can run to factors of thousands, so a donor who picks the best does vastly more good than one who picks at random.
Estimates like these are rough, and later researchers warned that the most dramatic figures often shrink when checked. Still, even if the true spread were ten times smaller than Ord's figures suggested, it would remain large enough to matter.
Randomized trials
To know what a program achieves, you need to know what would have happened without it. The movement leaned on a generation of development economists who used experiments. A randomized controlled trial assigns people or villages by chance to receive a program or not, so that differences in outcomes can be credited to the program. Because chance decides who is treated, the two groups should be alike in every other way on average. Three pioneers of this approach to fighting poverty, Abhijit Banerjee, Esther Duflo and Michael Kremer, shared the Nobel prize in economics in 2019.
Trials have limits. A result in one region may not carry over to another, and many important questions, such as whether a country should reform its courts, cannot be randomized at all.
The counterfactual question
A second habit of mind is counterfactual thinking. Counterfactual impact is the difference your action makes compared with what would have happened anyway. If a charity would have received the same money from someone else, your gift adds little. If a government was about to fund a program, private money may simply replace public money. GiveWell therefore asks not only whether a charity works but whether it could put more money to good use right now.
The deworming dispute
Evidence can be fought over. In the late 1990s, a study in western Kenya by Edward Miguel and Michael Kremer found that treating schoolchildren for intestinal worms sharply reduced absence from school. Deworming is cheap, and the findings made it one of the movement's favorite causes. In 2015 a team of epidemiologists reanalyzed the data and reported weaker effects, setting off a public quarrel that journalists called the worm wars. Later follow-ups reported higher adult earnings among those treated, while critics questioned how far the findings could be generalized. GiveWell long kept deworming charities among its recommendations while stating its doubts openly. The deworming dispute showed that even well-studied programs can rest on contested evidence.
Weighing unlike goods
How do you compare a death prevented with an income doubled? Any ranking requires such moral weights. Rather than rely only on its own staff, GiveWell commissioned a survey asking people in low-income communities how they would weigh such outcomes. The respondents, in Kenya and Ghana, put a higher value on saving children's lives, relative to cash, than GiveWell's staff had.
The measurability worry
Critics argue that the stress on evidence carries a bias. The measurability objection holds that favoring what can be counted steers money away from valuable work whose effects are hard to count, such as political organizing, legal reform or basic research.
Defenders reply that the movement increasingly funds hard-to-measure work, judged by expected value. Expected value multiplies each possible outcome by its probability and adds up the results. A project with a one in ten chance of a very large success can beat a sure but small one. That tool brings troubles of its own, as debates about the far future would show.
Worked example: the guide dog
Singer has compared two ways of spending money on blindness. Training a guide dog for a blind person in the United States can cost around 40,000 dollars. Surgery for trachoma, an eye infection that can cause blindness in poor countries, has been quoted at somewhere between 20 and 100 dollars per patient. Taken at face value, the money for one dog could protect the sight of hundreds of people.
Now apply this level's tools. GiveWell and others questioned the surgery figure: the operation treats a late stage of the disease, it does not always succeed, and it mostly prevents further loss of sight rather than restoring it. The counterfactual matters too: would those patients have been treated anyway by a government program? Once those corrections are made, the ratio shrinks, perhaps a great deal.
The example also raises a moral question. Does comparing a guide dog with eye surgery treat the blind American unfairly, setting needy people against one another? Critics say yes. Defenders answer that such trade-offs happen whenever money is spent, and that making them visible is more honest than pretending they do not exist.
