Three essays on the adoption and impacts of improved maize varieties in Ethiopia

Abstract

Public agricultural research has been conducted in Africa for decades and has generated numerous crop technologies, while little is understood on how agricultural research affects the poor and vulnerable groups such as children, and how farmers' perceptions affect their adoption decisions. This dissertation helps fill this gap with three essays on adoption and impacts of improved maize varieties in rural Ethiopia. The first essay estimates poverty impacts. Field-level treatment effects on yield and cost changes with adoption are estimated using instrumental variable techniques, with treatment effect heterogeneity fully accounted for in marginal treatment effect estimation. A backward derivation procedure is then developed within an economic surplus framework to identify the counterfactual income distribution without improved maize varieties. Poverty impacts are estimated by exploiting the differences between the observed and counterfactual income distributions. Improved maize varieties have led to 0.8-1.3 percentage drop in poverty headcount ratio and relative reductions in poverty depth and severity. However, poor producers benefit the least from adoption due to their small land holdings. The second paper assesses the impacts on child nutrition outcomes. The conceptual linkage between maize adoption and child nutrition is first established using an agricultural household model. Instrumental variable (IV) estimation suggests the overall impacts to be positive and significant. Quantile IV regressions further reveal that such impacts are largest among the most severely malnourished. By combining a decomposition procedure with estimates from a system of equations, it is found that the increase in own-produced maize consumption is the major channel such impacts occur. The third paper explores how farmers' perceptions of crop traits affects their willingness to adopt improved maize varieties. Under a random utility framework, a mixed logit procedure is implemented to model farmer's adoption intention, where perceptions of key varietal traits are first identified, and then instrumented using a control function approach to account for potential endogeneity. Perceived yield is found to be the most important trait affecting farmers' adoption intention. Further, yield perceptions among previous adopters appear to be affected by within-village peer effects rather than the real crop performance.