Promoting accountability in public projects: donors, audits, and rural electrification

Abstract: International donors and governments require contracting mechanisms that provide accountability in order to mitigate leakage and corruption in public infrastructure projects, especially when construction is outsourced to private sector contractors. This is an important goal because low-income countries spend on average 15% of GDP on public procurement. One mechanism that is less commonly studied is donor conditionality, where international aid donors attach stringent conditions to the financing provided. Does conditionality improve the quality of infrastructure construction, and how does this compare to independent audits? We study this question in the context of one of Kenya’s largest public infrastructure projects: the Last Mile Connectivity Project, which seeks to achieve universal household electricity access by 2022. We exploit two sources of variation. First, to examine the impacts of donor conditionality, we exploit quasi-random donor assignment in the first phases of LMCP construction, in which construction sites across the country were assigned to be funded either by the World Bank or the African Development Bank. Second, to examine the impacts of third-party monitoring, prior to construction we announce construction audits for a randomly selected subset of construction sites. Our novel on-the-ground engineering assessments measure detailed site-level outcomes around the timing of construction progress and the quality of a host of network attributes, including transformers, conductors, poles, and household connections. We find that the World Bank’s relatively stringent conditions, and our randomized independent auditing intervention, both improve construction outcomes by around 0.2 standard deviations on various outcomes. But, importantly, the World Bank’s restrictions cause significant delays of several months on average. Future work will build on these preliminary findings by investigating the channels through which these results arise.

Authors: Catherine D. Wolfram, University of California, Berkeley, Edward Miguel, University of California, Berkeley, Eric Hsu
University of California, Berkeley, Susanna B. Berkouwer, University of Pennsylvania