Randomised trials in economics: what about social networks?

01May11

Having just reviewed “Poor Economics“, a new book by Abhijit Banerjee and Esther Duflo (high on my reading list…), The Economist is hosting a debate inviting prominent economists to discuss the question of whether randomised trials can be regarded as the future of economics.

The debate is definitely interesting and the viewpoints expressed quite relevant. Randomised trials have enabled significant progress in microeconomics, but are not an option in macroeconomics. They are highly beneficial but can only be implemented under restrictive conditions -in particular, they must produce results after a short time period. And the enthusiasm for rigour that this method has brought about may take the lead over the fundamental societal questions that the discipline is meant to address.

I would like to add my own point to the debate -which in line with my research interests, concerns social interactions and networks. Randomised trials are hardly an option when interpersonal relationships can spread the effects of treatments in ways that the experimenter cannot control. Charles Manski studied vaccination: even if only part of a population is vaccinated, the risk of illness diminishes for all. The distinction between treatment group and control group becomes blurred and it is no longer possible to precisely measure the effects.

The same may be said of other forms of policy intervention: for example, efforts to educate/inform subjects on some important matter, for instance health-related, or to promote new productive methods or techniques: tin all these cases, treated subjects may spread the information to the control group so that randomised trials would lose the ir much acclaimed rigour and power.

Yet the social networks literature has pointed out that innovation diffusion, behaviour spread, opinion dynamics and other forms of social influence may be more complex than epidemics spreading. For example, adoption of an innovation by an individual may not depend on just one contact with another, but be conditional to a majority of others adopting it in his or her social network (a so-called “threshold” effect). Under these conditions, it may still be possible to keep the treatment and control groups distinguished as long as we remain below the threshold.

In short, further study is needed to illuminate the conditions under which experiments over networks are possible. Some research is already being done on the topic, but more is needed. Networks are now acknowledged to be a highly effective tool to spread the effects of social policies, and a better understanding of how they function may also support further progress in the scientific study of society.

About these ads