Terence Chau

Research

My research combines causal inference and big historical data to learn about technological innovation. My main work discusses how specific government investments in research and development can lead to innovations with wide ranging applications across the economy. Empirically, I leverage data on the universe of US patents to study the creation of NASA during the 1960s Space Race using difference-in-differences and event study models. Relative to comparable fields, I find that the Space Race increased patenting in spaceflight-related technology fields, that these inventions were highly influential, and that their influence extended beyond to non-spaceflight technologies. I also link NASA affiliated inventors to their patents, and find that most did not hold patents prior to joining NASA or obtaining a NASA contract.

As part of a team with Richard Hornbeck, Anders Humlum, Martin Rotemberg, and Shanon Hsuan-Ming Hsu, I have applied supervised learning techniques to longitudinally link recently digitized 19th century Census manuscripts containing the near-universe of manufacturing firms. In exploratory work using this novel manufacturing dataset and high resolution National Hydrography GIS data, we characterized available waterpower for each individual river across the entire US to study how 19th century firms that depended on waterpower chose their locations.

Historical data allows us to tackle questions that cannot be answered with modern data, but it also comes with its challenges. Archival records rarely have unique identifiers like Social Security numbers, so a major part of my research focuses on linking large datasets using machine learning—training models to find people and firms across different censuses, like a genealogist at Ancestry would do, but at scale. Such probabilistically linked datasets can lead to biased estimates, so I have worked on proposing bias-corrected estimators in such cases with Lucas Mation.

_{Image: National Bureau of Economic Research}