I am grateful to Duncan Green for giving me an opportunity to respond to his comments on The Great Escape. I summarize the key evidence, and try to give a coherent story of how I think aid works, and when it will fail. Like Duncan, I fully recognize (and am motivated by) the moral imperative to help people who are suffering, and I strongly endorse the idea of using aid to promote the international public goods that he describes.

My argument does not rely on evidence about projects; I have no doubt that many of these are good, but project evaluation cannot address the aggregate effects of aid on development. Cross-country regressions can estimate the aggregate effects in principle, though there are many problems, not least that neither writers nor commentators are good at explaining where the results come from, and authors sometimes retreat behind a fog of econometric technique. Yet we cannot ignore these studies; informal, undisciplined stories are not helpful.

Here is my take-away: (a) controlling for the factors that usually appear in growth regressions, aid as a share of recipient GDP is negatively correlated with growth, (b) similarly controlled, the change in growth is positively related with the previous period’s change in the aid to GDP ratio, where periods are several years long, and (c) the share of aid in GDP is larger for small recipients than large recipients, but the latter grow more rapidly than the former.

Point (a) is typically dismissed because well-directed aid will respond to bad growth shocks, which biases down the estimate of aid effectiveness. It is not clear that aid actually behaves like this, but I accept the point for now. On (b), lagging the aid variable and then taking changes flips the sign of the bias (the error term is this period’s shock minus last period’s shock and is this positively correlated with last period’s aid minus aid from two periods ago), so that if (a) is not evidence against aid, then (b) cannot be evidence for aid. I interpret point (c) as evidence against aid, not because country size might not favor growth directly, but because there is no other obviously powerful mechanism, while my account of aid has exactly that implication.

Why might aid fail in aggregate? One of my favorite stories in Duncan’s book is about owners of fishponds being violently dispossessed by more powerful people, and then getting them back through political action. Money and know-how were not the issues; power was the problem, and politics the solution. But this good outcome is unusual. The worst case I know happened in Goma in 1994, when the perpetrators of the genocide in Rwanda fled into the eastern DRC with their wives and families. Perhaps two-thirds of the aid for the humanitarian emergency was diverted for training the murderers to go back to finish off the Tutsi “cockroaches.” Alex de Waal, in Famine Crimes, explains over and over how aid can only reach the victims of war by paying off the warlords, and sometimes extending the war. Such aid saves lives, but at the price of other lives later.

The mechanisms of poverty and power are the same in peacetime, albeit with less catastrophic outcomes. With a government in control, it is impossible to reach those who are powerless without paying the powerful, and paying the President and the government will make them less interested in listening to their people. Instead of having to raise money through taxation and deliver services in return, they can instead use their people to extract money from donors. They can enrich themselves by keeping their population poor; such aid is an instrument of inequality. Some governments may be more benevolent, but large, prolonged amounts of aid ultimately corrupt benevolent rulers, or cause them to be replaced by exploitative rulers. Domestic NGOs or CSOs that monitor the state lose their legitimacy when they accept funds from abroad.

Duncan agrees that such undermining is a possibility, but doubts the evidence. Yet we need only look at the (very large) historical (and recent) literature on the resource curse; rulers who need not raise revenue from their people will often behave badly, sometimes spectacularly so.

There is also an extensive literature on how this happens by observers of aid, from historians, aid workers, journalists, and political scientists. Useful and insightful accounts include those by Michela Wrong (on Zaire, Eritrea, and Kenya), Michael Maren (Somalia), Nina Munk (Millennium Villages), Matthew Connelly (“aid” for population “control”) with broader discussions from Jonathan Glennie, Fiona Terry, Linda Polman, Martin Meredith, Charles Ferguson, and James Scott, as well as the well-known accounts by Bill Easterly and Peter Bauer. Many of these combine close observation with deep political, economic, and historical knowledge.

Histories of long-run development by Eric Jones, David Landes, Ian Morris, or by Daron Acemoglu and Jim Robinson (whose last chapter takes the same view of aid as my own) provide ample evidence on how development was crippled by rulers who had no reason to tax or consult their subjects. Jakob Svensson and Tim Besley and Torsten Persson have developed more formal accounts. Svensson is particularly insightful in noting that donors are responsive to well-meaning but necessarily ill-informed domestic constituents, including domestic interest groups from the aid industry, a process that is only too well-understood by the recipient governments. (Such politics makes “country ownership” impossible, and dooms agreements such as the Paris Declaration.) We are left with what Francisco Toro has aptly called co-dependency, in which the harm is largely borne by the citizens of the recipient countries who have no voice in this political equilibrium.

Duncan says that aid has rather moved on since Mobutu and the Cold War. I wish. The US gives aid to those who support the “war on terror,” or who recognize the state of Israel, and European and American politicians can use aid to burnish sullied images, even when it harms the recipients. Is it really so much better to support regimes that imprison, torture and kill their enemies provided only that child mortality rates are falling? Perhaps. Yet it is well to recall Amartya Sen’s argument that the components of freedom are also instrumental in producing it: we make trade-offs between different freedoms at our practical and ethical peril.

Not sure if this is a good idea, but let’s have a poll (you can choose more than one option). I’m particularly anxious about this one, because Angus is a global survey guru, advising Gallup among others, and I haven’t had time to consult with him over the questions. But interactivity trumps prudence, so over to you to vote. And I suspect that there will be more contributions on this topic over the coming weeks.

And if you have 90 minutes to spare, I recommend Bill Easterly v Owen Barder, covering some of the same ground in last week’s CGD debate.