Variation in infrastructure delivery and management quality among Ghana’s local governments

Many infrastructure projects in developing countries are started but never finished. Despite the vast resources invested in infrastructure and the obvious inefficiency of abandoning projects mid-construction, there is little research on this problem. I have collected, digitized, and coded administrative records of over 14,000 local government infrastructure projects in Ghana, covering multiple project types and delivery mechanisms. Approximately one-third are never completed, consuming onefifth of total infrastructure expenditure. There is a bifurcation of outcomes: projects tend to be completed either promptly or not at all. There are large differences across districts in completion rates, but there is also significant variation within districts across different delivery mechanisms, each of which has varying degrees of involvement from local government, central government, and donors. These differences are extremely robust and remain statistically and economically significant even after controlling for project characteristics and district, community, and contractor fixed effects. I discuss issues of causality, and argue that it is implausible that endogenous project sorting can account for the magnitude of the observed differences. These results are consistent with generalized hold-up problems as a primary mechanism of non-completion, but are not well explained by common explanations for poor service delivery such as low bureaucratic capacity, technical complexity, political favoritism, or simple corruption. The findings complement micro-level experimental studies as well as macro-historical research on state capacity and abstract institutions, by providing rigorous observational evidence that the design and implementation of government institutions and delivery mechanisms can affect project outcomes in political and bureaucratic equilibrium. I discuss implications for theories of policy implementation, state capacity, and distributive politics.

The data can be downloaded from the dataverse as in Stata.do and dta files. Do file replicates multiple figures with escriptions in do file for figures replicated.