This dissertation contains three essays broadly related to technology adoption in developing countries. The dissertation discusses a job search technology, an agricultural technology, and a monitoring technology in the context of agricultural input and labor markets in developing countries. Chapter 1 studies whether the adoption of a new job search technology reduces search frictions in rural labor markets in Tanzania. I develop an SMS-based technology that connects agricultural workers and employers instantly. In a clustered RCT experiment, I find that treatment villages experience a 16-40 percent reduction in within-village wage dispersion. Consistent with reduced wage dispersion, I find evidence that labor is reallocated within villages. Dispersion in per-acre labor input across farms decreases. Workers divert job applications from lower-paying to higher-paying employers.
In joint work with Shilpa Aggarwal, Brian Giera, Jonathan Robinson, and Alan Spearot, Chapter 2 studies the effect of remoteness on the adoption of chemical fertilizer, one of the most important agricultural technologies to enhance productivities. We quantify and estimate the effect through the lens of a structural spatial model, facilitated by an enormous data collection exercise in Tanzania. We find that villages in remote places face very different input and output prices than villages near a city hub. In our reduced form analysis, we find that one standard deviation increase in travel time is associated with 20-25 percent lower input adoption and output sales. Our simulated structural model predicts that reducing transportation costs to reach input-retailers by 50 percent doubles input adoption.
Chapter 3 (joint with Ajay Shenoy and Laura Zimmermann) studies whether corruption "greases the wheels'' of government in the context where a digital monitoring platform is used for implementing government programs. We test this corruption-as-compensation hypothesis in Indian villages whose presidents, in charge of running a massive public works scheme, claim such behavior is widespread. We link millions of administrative job records from the digital platform to election outcomes, enabling us to measure both the presidents' performance and their self-dealing. We do not find evidence in favor of corruption-as-compensation hypothesis. Self-dealing declines over time in villages with plausibly better monitoring capacity (e.g. the presence of internet cafes), even though there is no decline in performance. In villages without such capacity, self-dealing persists.
