When you’re lost and running late, it is frustrating to stop and figure out the lie of the land. Nevertheless, that has to be better than speeding off in the wrong direction, however fleetingly satisfying the illusion of activity may be.

Few problems are more urgent or important than that of desperate poverty.

Aid can feed the starving, heal the sick, educate young children before they grow too old for school. But Abhijit Banerjee, an economist at the Massachusetts Institute of Technology (MIT), is one of a growing band who believe that the development industry nevertheless needs to stop and work out whether it is moving in the right direction.

In Making Aid Work, a slim new book that deserves an audience, Banerjee cites a recent World Bank report as recommending a cornucopia of initiatives, including but not limited to ”computer kiosks for villages; cell phones for rural areas; scholarships for girls attending secondary schools; school-voucher programs for poor children; joint forest-management programs; water-users’ groups... ” The problem is that while this stuff sounds sensible, we don’t really know how much of it works. Whether and when aid works at all remains a hotly debated subject, which tells you something about the quality of the research devoted to working out whether it does.

In fairness, that is not an easy job. Economists are well aware that correlation isn’t the same thing as causation - emergency aid goes hand in hand with earthquakes but does not cause them - and it is hard to figure out whether some improvement happened because of an aid programme or despite the aid programme.

Banerjee argues that there is a solution: aid agencies should copy medical researchers and run randomised trials. In one famous trial in western Kenya, economists used the alphabet to decide randomly which of 178 Kenyan schools would receive flip charts. Because the allocation was random rather than based on need, enthusiasm or political connections, improvements in test scores or attendance were almost certain to be due to the aid programme. Previous evidence suggested that children learnt more with flip charts, but the randomised trial proved otherwise.

I would count the flip chart trial a big success. Even though the children learnt nothing, the aid community learnt a lot.

There are, of course, limits to randomised trials; good luck using one to evaluate the relative merits of fixed and floating exchange rates. But Banerjee’s back-of-the-envelope calculations suggest that we have now sufficient randomised evaluations of specific projects to spend more than $10bn of aid a year on programmes with a proven record of success. That is about 15 per cent of official development aid, and more results are coming in each month. Successful projects included giving deworming tablets to children, which increased school attendance.

A more serious objection is that aid projects may cause wider problems that a randomised trial could not detect. For example, a trial may prove that hiring trained nurses pays dividends for people’s health. Yet spending a lot of money on nurses may suck skilled people out of other important activities, such as teaching or even - dare I say it? - setting up and running businesses. Or the aid dollars will tend to push the currency up and make it hard to be a successful exporter. These problems may be a price worth paying, but they are certainly beyond the scope of the original trial.

Nevertheless, the ”randomistas” deserve to win over a few converts in their evangelic zeal for proper controlled trials. Many aid programmes are introduced gradually for want of funds; it does not take much to turn a gradual introduction into randomised delay for the purposes of finding out what works. These operations are too important to carry out with our eyes closed.

More Tim Harford

Tim Harford’s book, ”The Undercover Economist”, is now out in paperback.