Economic data aggregation bias: Empirical evidence from the energy sector

This project generates population estimates of the local economic impacts of energy development, specifically focusing on wind energy generation facilities. We compare results from two prominent methodologies in the literature: (1) an “outward-propagating†model that aggregates earnings and employment outcomes for workers and establishments located in increasing radii around the locations of a wind generation facilities and control sites, and (2) a “spatial lag†model (which has become the dominant approach in the literature during this project) that uses the individual workers as the unit of observation and aggregates their exposure to wind energy generation facilities at increasing radii around their georeferenced residence locations. We further explore how data aggregation impacts results by repeating these analyses using data that has first been aggregated to the county level before further aggregating to an “outward-propagating†or “spatial lag†framework. We have two main findings. First, we find that the spatial lag approach gives much more reliable results than the outward propagating model, with the latter model likely overstating the aggregate local impacts of a given economic shock and producing less stable estimates. Second, we find that aggregating underlying individual data to the county level before implementing either of the models severely dampens economic impact estimates in most cases, highlighting the importance of either gaining access to representative georeferenced samples or finding another geographic aggregation at which to produce publicly available data products in order to generate reliable estimates of local impacts of economic shocks.