Clustered into control: Heterogeneous causal impacts of water infrastructure failure

Abstract

We estimate economic impacts from decaying water infrastructure in the United States. Using water main breaks in Washington, DC, and a yearlong panel of hourly traffic speeds, we estimate causal effects of water main failures on traffic congestion. We use k‐means clustering to create clusters of streets that are similar to each other: treated observations are compared to other units in the cluster. We identify heterogeneous treatment effects algorithmically while retaining straightforward standard error calculations. We find strong evidence of heterogeneous treatment effects across clusters but small welfare impacts of water main breaks on traffic patterns overall.