The objective of this dissertation is to study African smallholder farmers’ transient technology use. Transient use refers to the situation where farmers switch back and forth between two or more technologies. To better understand transient technology use, I first specify a dynamic theoretical model to investigate farmers’ optimal decision rules given the availability of modern and traditional technologies under a range of productivity and market scenarios. The model is then calibrated and solved using a dynamic programming algorithm. Numerical results show that expected profitability and costs of switching between technologies are the two main driving forces influencing the patterns of transient technology use. Next I turn to econometric insights about transient technology use in Africa. The sample data utilized in this dissertation is an irregularly spaced four-wave panel data set, making all existing traditional discrete choice estimators inconsistent for dynamic panel estimation. Therefore, before conducting the empirical analysis, I develop and evaluate the performance of three possible estimators (gap-dummy approach, linear probability model, and indirect inference method) for discrete choice dynamic panel data models with irregular spacing. Monte Carlo simulations reveal that traditional estimators generate downward bias in estimates for the state dependence parameter. Adding gap dummies indicating if the panel period is irregularly spaced could potentially reduce the bias. The other two estimators, linear probability model and indirect inference, fail to reduce the bias of irregular spacing effectively in our simulations. The final task is to undertake an empirical analysis of Kenyan smallholder farmers’ decision to use hybrid maize seed. The gap dummy approach is applied to reduce the bias from the irregular spacing problem. Our findings provide empirical evidence that hybrid maize seed use is a dynamic process with a high degree of state dependence. However transient use does occur regularly and switching is influenced by the expected relative profitability between hybrid and traditional varieties, but that the choice is also highly state dependent consistent with the existence of switching costs and/or learning by doing effects.
Read
