with Ruben Gaetani
We document that the diffusion of new scientific ideas beyond their field of origin has declined over the past four decades. This contraction is linked to increasing specialization in scientific language: research that employs more technical terminology tends to be adopted less broadly. We develop a theory of scientific discovery in which the diffusion of new ideas depends on the degree to which potential adopters can understand and process them. Scientists choose how to communicate their research, trading off technical precision targeted at their immediate peers against accessibility to broader audiences. As knowledge accumulates and frontier research builds on deeper layers of prior work, this tradeoff increasingly favors specialized language, narrowing diffusion. Testing these predictions in a corpus of scientific publications, we find that less accessible language accounts for 15--23 percent of the observed decline, pointing to a novel channel behind the long-run slowdown in research productivity. Policy interventions that align scientists’ incentives with the social value of diffusion can broaden adoption and increase the social returns to scientific research.
with Kristina Manysheva and Martí Mestieri
How much do international knowledge spillovers contribute to innovation and productivity growth around the world? Using a century of global patent data, we document technological waves led by a changing set of countries and, after 1990, a sharp rise in reliance on foreign knowledge, especially in computing, information processing, and medicine. By the 2000s, patents filed outside the United States and Japan cited foreign work more than twice as often as domestic work. Guided by a simple theoretical framework, we leverage the predetermined network of patent citations to construct each country-sector's exposure to foreign knowledge, using a 2000--2014 country-sector panel. Greater exposure raises future value added per worker and TFP growth, and predicts domestic patenting with an elasticity of innovation to foreign knowledge spillovers of roughly one third. Decomposing TFP growth by knowledge source, we find that foreign knowledge accounts for 30 percent of measured TFP growth outside the United States and Japan and for 40 percent in three frontier sectors: computer programming, electronics, and pharmaceuticals. The United States and Japan are the leading sources of foreign knowledge, contributing on average 11 percent of measured TFP growth elsewhere and 21 percent in the frontier sectors. Restrictions on the international flow of knowledge would carry first-order productivity costs, concentrated in the countries and sectors that rely most on foreign sources.
with Aiday Sikhova and Bruce Weinberg
We use unique, linked data on biomedical researchers and an extended AKM-style design to study the role of leaders in engaging and providing opportunities to researchers. Specifically, we study how women principal investigators (PIs) affect gender disparities among the faculty on their projects. We find that research teams with more women PIs are significantly more supportive of female faculty in three major ways. First, teams with more female leaders employ a larger share of women on their teams. Second, using researcher-year and paper fixed effects to account for unobserved characteristics, we find that teams with women leaders give significantly more credit---in terms of authorship---to female faculty who work on those teams, with the largest effect on high impact papers and in heavily-male fields. Third, the impact of female PIs extends beyond the time female faculty are part of the teams, leading to an increase in the number of co-authored papers with female PIs after female faculty leave the teams. Our estimates are strongest for the most productive women, suggesting that, although women leaders tend to narrow the gender gap in authored papers, they seem to perpetuate within-group inequality in authorship for women.
with Davide Coluccia, Gaia Dossi, and Mara P. Squicciarini
We show that U.S. counties hit harder by the 1918 influenza pandemic became both more religious and more science-oriented, a result at odds with the existing literature that documents a negative relationship between religiosity and science. Using a newly assembled name-based measure of religiosity, we find that this coexistence was driven by heterogeneous responses: individuals from more religious backgrounds became more religious, while those from less religious backgrounds were more likely to turn toward science. As a result, the pandemic widened preexisting differences in religiosity and increased polarization within society.
with Mike Andrews and Ruveyda Gozen
Patents do not report inventors' gender, requiring researchers to infer the gender of inventors. To conduct these inferences, researchers must make several choices. We show how these researcher choices can affect conclusions about the role of women inventors in the U.S. from 1845 to 1924. More specifically, we compare two automated methods to determine inventor gender for the universe of U.S. patents: inferring gender from inventors' first names and linking inventors to census data. These methods paint similar pictures about aggregate patterns of patenting by women, but often give different predictions about the gender of particular inventors. Both automated methods identify a larger number of patents by women inventors than have previously been identified in the literature. Using the gender inferred by these two methods, we study how the characteristics of patents and inventors differ by gender.
Patents are commonly used as the main source of data for empirical studies related to innovation and technological change. The large amount of information about the underlying innovative process contained in each patent has certainly contributed to their popularity. Nevertheless, due to the lack of reliable data, historical analysis has focused on relatively small time frames or on specific dimensions of patents data. The goal of this paper is to fill this gap. I build and release a comprehensive time series of the universe of U.S. patents. The data set contains all the variables commonly used in the literature and, importantly, geolocates every inventor and assignee reported in each grant over the period 1836-2016.