The big push back

IMAGINE you run a dirt-poor agricultural economy. Your farmers want to improve their soil but have no fertilisers, so you must either build a fertiliser factory or import the stuff. You will therefore need a port, dock workers, accountants and lawyers, roads and trucks to transport it, bankers to extend credit, and plastic bags to put the fertiliser in, necessitating plastics and packaging factories, too. Your farmers then need seeds and livestock, so you must have vets, market traders, construction companies to build barns and refrigeration systems. And you will need all this before your farmers put a seed in the ground.

In development, it seems, you cannot do anything until you can do everything. That is the idea behind the “big push” theory. Outlined by Paul Rosenstein-Rodan in 1943, this says that even the simplest activity requires a network of other activities and that individual firms cannot organise such a large network, so the state or some other giant agency must step in.

The big push came to grief in the 1970s and 1980s as evidence accumulated that, in Africa at least, public investment and foreign aid had produced no perceptible change in productivity, not least because so much of it was stolen. Recently, though, the idea has come back into vogue. The UN talks constantly about its millennium development goals (eight goals, 21 targets). Jeffrey Sachs of Columbia University argues that if public investment and foreign aid are big enough, they will boost household incomes, spurring savings and boosting local investment. They should also “crowd in” external investment by improving infrastructure.

Unlike most economists, Mr Sachs can put other people's money where his mouth is. He set up the “millennium village project”, taking 14 places in rural Africa with about 500,000 people and, since 2006, making them the subjects of a $150m project run by his university and African governments.

The project—motto: “no single intervention is enough…we must improve them all”—carries huge hopes. Touring a village in Malawi, the UN's secretary-general says he saw the potential of technologies such as smartphones and drip irrigation “to advance human well-being in ways that simply were not feasible even a few years ago”. George Soros, a financier, gave the project $47m and predicted that it would transform entire regions.

But the money has brought into the open simmering disputes about the merits of the project, how properly to measure them and what happens when the money runs out. Now the first independent assessment of one of the villages—Sauri, in Kenya—has suggested that the critics are having the better of the exchange.

The projects' backers claim extraordinary results: a 700% increase in the use of antimalarial bednets; a 350% increase in access to safer water; a 368% increase in primary-school meal programmes. On closer inspection, though, these numbers turn out to be less dramatic. True, the proportion of young children sleeping under mosquito-repelling nets in Sauri has risen sharply. So has the share of households getting clean water in Bonsaaso village in Ghana (see charts). But these improvements might have happened anyway, without the programme. The use of treated bednets in Kenya's Nyanza province (the region of which Sauri is part) also increased, albeit over a longer period. Ditto with clean-water availability in Ghana's Ashanti province.

Michael Clemens of the Centre for Global Development, a think-tank, and Gabriel Demombynes of the World Bank says that a randomised trial is needed to disentangle what the millennium programme is doing from what is happening anyway. In such a trial, each village would be paired with a similar one not getting the same help—and the results compared.

This stirred up a hornets' nest. In a vitriolic letter to another critic, Mr Sachs calls the idea “that one can randomise villages like one randomises individuals…extraordinarily misguided”. Randomised trials cannot work in villages, he insists, because they are too complex and dynamic. Comparing a millennium village with a randomly chosen one “will add surprisingly little”; the proper comparison is with a region or a country as a whole.

David McKenzie of the World Bank then took up the cudgels. He pointed out that if the impact of the project were as great as its backers claim, it should be discernible even against a shifting background; that, in practice, randomised trials can be used to evaluate complex, dynamic processes, not just simple, static ones (though they have to be designed properly); and that comparing a favoured village against another after the intervention has started—which is being done—isn't a randomised trial in the proper sense (properly, one should select pairs of villages, then choose one of the pair randomly as the subject of the programme).

If it works, prove it

Now a Kenyan economist, Bernadette Wanjala of Tilburg University in the Netherlands, has raised further doubts about the project. She interviewed 236 randomly selected households in Sauri who had been offered the benefits and 175 randomly selected ones who had not. In a study with Roldan Muradian of Radboud University, she concluded the first group had raised their agricultural productivity by an impressive 70%. Yet she found that the impact on household income was “insignificant”, and that there had been little extra saving or investment. The villagers had grown more food—and eaten it. They became better nourished, but this did not affect the wider economy.

Better nutrition is important, of course. But the aim of the project is to boost income, investment and economic diversification. It is not clear that these goals are being achieved. For $60 per person per year (which increases the income of the poorest villagers by well over 25%) the project has improved village life only a little more than it would have improved anyway. So far, the project provides little evidence that “big push” development—advancing on all fronts, flags flying—is better than the alternative: gradual, step-by-step changes to remove specific barriers to growth.