This dissertation examines several topics in the economics of innovation in historical settings. The first chapter examines how government investment in technology creates spillover effects in the production of knowledge, using the 1960s Space Race as an empirical setting. Combining the known universe of patent records with information on their federal reliance, along with a difference in differences design, I estimate that NASA-exposed fields increased their patenting relative to non-exposed fields, and that these patents were more impactful by citation metrics. To study the degree to which these results are driven by the reallocation of scientists and engineers, I use the inventor information in the patent documents to show that NASA-affiliated inventors obtained their first ever patents after joining NASA or obtaining a NASA contract, and not before. The second chapter contributes to data methods in the economics of innovation by comparing traditional rules-based data linkage to machine learning methods. I apply a supervised learning strategy to link patents issued between 1840 and 1900 to individual establishment microdata in the 1870 Census of Manufactures, and conclude that a simple rules-based approach combined with manual verification plausibly yields higher confidence links. I contribute a novel dataset for future researchers by performing this higher confidence linkage. The final chapter provides an overview of the historical patent datasets used throughout the dissertation, with a focus on their accuracy, coverage, and overlap.