Predicting Mortality Outcomes in Randomized Trials with Proxies [Preliminary draft]
Many children under the age of five die of preventable causes in low-income countries. A number of cost-effective interventions in global health focus on reducing mortality among this group. However, estimating the impact of these interventions on mortality in a randomized controlled trial can be challenging due to the large sample size required to measure mortality outcomes accurately. At the same time, these studies often measure intermediate outcomes that lead to the final outcome of mortality reduction, for instance, increased utilization of health facilities or adoption of health products like bed nets. Even though the intermediate outcomes can be measured with higher accuracy, using them as proxies in place of mortality outcomes may lead to biased estimates. We apply a recently developed technique which increases the accuracy of predicting the true outcome of interest to this problem. The method combines a large amount of proxy data and a small amount of true data in a two-step estimator with L1 penalty. We demonstrate the effectiveness of this approach using data from randomized controlled trials on a community health intervention and a health information intervention aiming at reducing child mortality in Sub-Saharan Africa.
Measuring the Preferences of Development Program Recipients: Results from Ghana and Kenya (with Alice Redfern and Daniel Stein) [Slides]
Understanding how people trade-off between different types of outcomes, e.g. increasing income versus reducing mortality risk, is important for policy making and resource allocation. Despite many studies addressing this question in high-income countries, e.g. on the value of statistical life (VSL), there is a lack of evidence among low-income countries where much of international aid is directed. We conducted a stated preference survey on people living in extreme poverty in Kenya and Ghana to measure their VSL as well as their moral preferences over increasing income versus saving lives in their community and over saving lives of different age groups. We find that their VSL is higher than the result of a linear extrapolation using the relationship between VSL and income from high income countries. In addition, when asked about resource allocation from a community perspective, many people have a much stronger preference for saving lives over increasing income compared to when the question is framed from an individual perspective (i.e. their VSL). We also find that people’s preference over saving lives of different age groups is decreasing in age, which broadly in line with findings from high-income countries.
Information Sharing in Trade Credit Markets: Evidence from Kenyan Retail Shops [PDF]
In developing countries financial frictions hinder firm growth. Credit constraints result from poor contract enforcement and asymmetric information in the credit market. One solution is to provide infrastructure for lenders to share information on borrowers’ credit history, which can mitigate adverse selection and improve repayment incentives, reduce resource misallocation and accelerate firm growth. Information flow facilitates informal enforcement which may be particularly important in an environment where formal (legal) enforcement is weak. I investigate the barriers to and impact of introducing an information sharing service for small and medium enterprises (SMEs) and their trade credit providers (suppliers) in the retail sector in Kenya, by means of randomized information intervention and subsidy of take-up. I focus on borrowers and lenders’ decisions to adopt and share information, as well as the impact of the service in reducing information asymmetry, increasing borrowers’ repayment incentives, buyer-supplier relationships and spillover among retail shops. I find that offering free credit reports to retail shops increases credit report ownership and knowledge, as well as shops’ likelihood of applying for supplier credit, but not access to supplier credit. Lack of response from the supplier side seems driven by their unwillingness to rely on information in the credit reports as well as some suppliers’ lack of ability to provide credit.
Democracy, Devolution, and Local Spending: Evidence from Kenya’s Constituency Development Fund (with Pascaline Dupas) [PDF]
Kenya's Constituency Development Fund (CDF), introduced in 2003, was designed to better address local needs through locally-driven selection and implementation of development projects, with greater budget allocation to poorer areas. However, until 2013, Members of Parliament (MPs) of each constituency could appoint members of the local CDF committee, de facto controlling the CDF budget. As MPs have incentive to select projects in a way that maximizes reelection gains, this may compromise efficiency of project allocation. We present evidence on project allocation, including targeting to local needs and timing with elections, as well as project performance, to support the theory that MPs indeed allocated projects for reelection gains at the expense of efficiency.
Farmer Credit and Firm Profits: Experimental Evidence from a Monopsony Buyer in Mozambique (with Ignacio del Busto, Veronica Polin, and Daniel Stein) [PDF]
Low usage of productive inputs leads to lower levels of smallholder farmer production, which can also affect the profitability of purchasers and processors of agricultural goods. But whether firms have incentive to directly intervene in farmer investments depends on its impact on production. We conduct an experiment with a large cotton company that has monopsony purchasing power in Mozambique. Among relatively productive farmers in the region, the firm randomly allocated farmers additional extension services, or additional extension services combined with drastically increased access to credit. We find that providing additional extension services increases the number of farmers who cultivate cotton, resulting in increased cotton purchases by the firm. However, the increases are modest and not profitable for the firm. Providing farmers credit has much stronger effects. Farmers offered credit plus extension are 67.7 percentage points more likely to grow cotton and increase cotton yields by 39.1 percent, resulting in drastically increased cotton production. However, overall repayment rate on cash credit is low (80.5%), and the intervention may not be profitable for the firm. Assuming similar impact on cotton growing and loan repayment rates, a hypothetical intervention with the same amount of cash credit and no additional extension service would be more profitable.