Sachs Not Vindicated

Dani Rodrik, in a blog posted here, has announced that the new paper by Cohen & Dupas presented at CGD last week "vindicates Jeff Sachs." If by this, Dani means that the Cohen-Dupas paper lends support to the view that bed nets should always be free in poor countries and social marketing programs that depend upon small payments should be abolished, I believe that he is reading much more into the Cohen-Dupas results than is justified. April Harding and I have already blogged about why the supply side might be an important omitted consideration. But we have not directly examined the ability to generalize the Cohen-Dupas findings. So perhaps there is room for another blog on this paper. Cohen & Dupas established with great precision that, in communities which had recently absorbed half-a-million white bed nets at a subsidized social marketing price (1.4 per household), pregnant women visiting antenatal clinics were willing to accept more bed nets at lower prices. This shows that in this particular Western Kenyan context, when everyone has received information on the value of a bed net but not everyone has one yet, a highly targeted distribution campaign can achieve greater penetration when the bed nets are even less expensive. It says nothing about what total coverage would have been if the free distribution had been the only mechanism for delivering nets throughout the community for the past few years, as Sachs seems to propose.Cohen & Dupas also find, through visits to the women’s homes, that in 95% of the homes where a bed net has been purchased there was an "apparently new" white bed net hanging over a bed and this did not depend on how much the woman had paid for the net. At first glance, this finding is quite surprising, since it contradicts the finding from other malaria researchers that up to 40% of acquired bed nets are unused and that purchased bed nets are more likely to be used than free ones. However, the authors offer three possible explanations to which I will add another. First, thanks in part to the demand promotion activities associated with the social marketing campaign which preceded and formed the context for the experiment, bed nets are apparently an appreciated and valued commodity in this region of Kenya. Second, women who have just given birth and received a free bed net to be used for the baby may be more likely to use nets than most population groups. Third, the small amounts by which the price of the net differed across experimental clinics may not have been sufficient to trigger the "sunk cost fallacy" or "cognitive dissonance" which has previously been observed to reduce the use of free nets. And fourth, I add, the surveyors who visited the households may not have been able to distinguish a new white bed net received from the ante-natal clinic from a new white bed net which the household purchased through a social marketing channel in the same community. So Cohen and Dupas have confirmed that demand curves slope downward. More than that, they have found that there is residual demand for bed nets from mothers with new babies in a community that has benefited from extensive demand promotion activities in conjunction with a social marketing campaign for bed nets, which are very highly subsidized, but not free. But the more fundamental issue has to do with the hypothesis that Cohen and Dupas did not test. What does a free distribution policy do to the supply of a commodity? This is the issue I raise in my blog on the CGD website. I argue there that we need to have experiments on the supply side not on the demand side. To my comments previously posted on the CGD website, I add one more point. First, Cohen and Dupas report that they observed corruption in 4 of the 11 clinics which were asked to deliver nets at a price and in none of the five clinics that were asked to deliver them for free. They do not report the statistical significance of this finding. In fact, a simple test of the equality of the two proportions (zero out of 5 and 4 out of 11) does not reject the hypothesis that they are equal at the 95 or even the 90 percent confidence level (the p-value is 0.1195). Another way to look at this data is that, despite expensive monitoring and incentives provided to 16 clinics, corruption occurred in 4 of them, a proportion that is significantly different from zero. Thus, the authors have NOT demonstrated that government distribution will be less corrupt under free distribution, as Dani's blog implies, but they HAVE demonstrated that corruption is extremely difficult to eradicate in government clinics - even when one spends substantially more resources on monitoring the clinics than is typically affordable in these contexts. The presence of this corruption in the government clinics is an argument in favor of the more private sector oriented social marketing approach to bed net distribution, where corruption is easier to control. I do not mean to denigrate the work of Cohen & Dupas. We need more randomized trials on health related topics in developing countries, which are as carefully designed as this one. But it does no service to these researchers or to the topic they have examined to exaggerate their findings. The authors went to great lengths to hold constant the supply of bed nets at the four prices, so they could obtain scientifically accurate information about the demand side of the market. They did not set out to study the supply side, so it's not surprising that are unable to detect a statistically significant impact of the price charged on corruption. As I concluded in my discussant comments, there is an urgent need for similarly rigorous studies of the supply side of the market for health commodities and services, so that we can better understand how governments and donors can design incentive mechanisms that elicit high quality service delivery and high coverage of these merit goods among the poor in developing countries.