How much does the effect of industrial policy depend on the capacity of the bureaucrats implementing it? We exploit the rotation schedule of managers of South Korea's export promotion offices in 87 countries between 1965 and 2000 to show that a one standard deviation increase in bureaucrat ability boosts product-level exports by 37%, while the policy increases exports by 38% on average. Together, this implies the export promotion policy has no effect when implemented by a bureaucrat one standard deviation below average. Under higher-ability bureaucrats, South Korean exports respond more strongly to destination import demand and expand more for products prone to information frictions, suggesting more effective transmission of market information. We find that performance in the first appointment predicts whether a bureaucrat sees subsequent appointments, highlighting performance-based screening of bureaucrats as a mechanism that increases the policy's effect.

We show that the first nationwide mass vaccination campaign against measles increased educational attainment in the United States. Our empirical strategy exploits variation in exposure to the childhood disease across states right before the Measles Eradication Campaign of 1967–68, which reduced reported measles incidence by 90 percent within two years. Our results suggest that mass vaccination against measles increased the years of education on average by about 0.1 years in the affected cohorts. We also find tentative evidence that the college graduation rate of men increased.

Firm-directed policies are central to governments’ economic agendas. But under low state capacity convincing firms to take up such policies is difficult because (i) lack of knowledge of firms among implementing bureaucrats may render policies less effective or harmful, (ii) firms may mistrust the policies’ and bureaucrats’ motives. We are running an at-scale RCT studying a government-run labor market matching program in Addis Ababa to test a potential solution, increased bureaucrat-firm embeddedness.