In a recent talk he gave a forceful explanation of why economists should be 'thinking small' like this when assessing what works:
What we can say now is that this attempt to find the determinants of growth has failed so decisively, so comprehensively, that anyone today who makes any policy recommendation based on a growth regression has zero credibility.
Instead, he argues, RCTs should be used to measure the effectiveness of small-scale interventions. Those shown to work can then be rolled-out more widely.
Some critics have argued that a positive RCT result in one context doesn't necessarily mean the intervention will be successful elsewhere.
I think this line of argument will lose its power over time, as more and more trials are conducted. Once hundreds or thousands of trials have been done, people will have the information to adapt interventions to new contexts.
I'm in no doubt that RCTs offer an exciting new way to approach aid effectiveness. But there's one remaining concern, which few commentators have written about.
RCTs try and make development evaluation a scientific discipline. I firmly believe it is wrong to present 'science' as the only route to knowledge - especially when considering people's lives.
Let's not forget that 'scientific thinking' was at the heart of many fascist and socialist movements. Science has a self-legitimising power which can be very difficult to keep in check.
How will we cope with RCT results that suggest unpalatable or unethical interventions?
What will we do if a community rejects a project that we know will be good for them?