2.2 What works, and what it costs

Some health programmes are unusually cheap and unusually well tested. They include:

• insecticide-treated bed nets, which protect people from malaria-carrying mosquitoes while they sleep;
• seasonal malaria chemoprevention, which gives young children preventive malaria medicine during the rainy season;
• vitamin A supplements for young children;
• small cash incentives for parents to complete their children's routine vaccinations.

GiveWell is an independent charity evaluator that has spent nearly two decades assessing charities like these. Before you see its figures, make a guess.

People often guess far lower, perhaps the price of a bed net, or far higher, perhaps the cost of an intensive-care stay. The real answer is in between for a simple reason. A bed net costs a few dollars, but most of the people it protects would have survived anyway. You have to distribute many nets to prevent one death. The cost-effectiveness models behind these figures work through steps like that one by one: how many people are reached, how many would otherwise have died, how much the programme reduces that risk, and what it all costs.

Why trust any of this? Partly because of the quality of evidence behind it. Evidence comes in grades. At the top are systematic reviews of randomised controlled trials, where some communities get a programme and similar communities don't, and researchers compare the results. Below that are observational studies, then expert opinion, then individual stories. PlayPumps had plenty of stories and little else. The programmes above have trials.

GiveWell also compares programmes against a simple benchmark: giving people in poverty cash and letting them decide how to spend it. A programme has to do more good per dollar than cash to earn a recommendation.

None of this is precise. The models rest on judgement calls, and the same programme can be much more or less cost-effective in a different country. GiveWell expects the cost of saving a life to rise over time as the cheapest opportunities get funded. The EA Fellowship spends a week taking one of these models apart.