Research

Working Papers


Job Market Paper

I exploit historical natural experiments to study how establishing a new college affects local invention. Throughout the nineteenth to the mid-twentieth century, many
new colleges were established in the U.S. I use data on the site selection decisions for a subset of these colleges to identify “runner-up” locations that were strongly considered to become the site of a new college but were ultimately not chosen for reasons that are as good as random assignment. The runner-up counties are similar to the winning college counties along observable dimensions. Using the runner-up counties as counterfactuals, I find that the establishment of a new college caused 43% more patents per year in college counties relative to the runners-up. To determine the channels by which colleges increase patenting, I use a novel dataset of college yearbooks and individual-level census data to learn who the additional patents in college counties come from. A college’s alumni account for about 3% of the additional patents, while faculty account for roughly 1%. Knowledge spillovers to individuals unaffiliated with the college account about 1.4% of the additional patents. Migration is the most important channel by which colleges affect local invention, as controlling for county population accounts for 15-85% of the increase in patenting in college counties relative to the runner-up counties. I provide suggestive evidence that, rather than simply shifting where invention occurs, establishing a college causes a net global increase in patenting. On the other hand, colleges are no better at promoting invention than other policies that lead to similar increases in population.

Historical Appendix

Featured on: Marginal RevolutionMother JonesBrad DeLong's Blog



To understand the importance of informal social interactions for invention, I exploit a historical policy that restricts one channel through which individuals interact. More specifically, I examine the effects of alcohol prohibition in the U.S. Prior to the enactment of state-wide or nationwide alcohol laws, each county determined its own alcohol policies. Thus, prohibition differentially treated counties depending on whether they were wet or dry prior to prohibition. I analyze three different prohibition episodes: the enactment of national prohibition, the imposition of prohibition at the state-level in the decades before national prohibition, and the removal of prohibition barriers in the 1930s. I consider several sample selection criteria, including utilizing data on county-level voting during prohibition referendums, to ensure that sample counties had consistent views on alcohol, in turn ensuring that changes are not driven by unobservable characteristics that also affect patenting behavior. Following national prohibition, previously wet counties had approximately 10% fewer patents per year. After prohibition at the state level, the estimated effect is even larger: previously wet counties have about 20% fewer patents per year relative to the dry counties. In both cases, the effect is largest in the first three years after the imposition of prohibition and rebounds thereafter. Consistent with the observed decrease in patenting being driven by a disruption of informal social interactions, the fraction of patents with multiple inventors falls, the diversity of patented ideas declines, and first-time inventors decrease their patenting more than serial inventors following prohibition. The patenting rate for men decreases more than that for women in previously wet counties. Removing prohibition appears to lead to a small increase in patenting, although the results are more mixed.


with Sarada and Nicolas Ziebarth

Who invents? This is a central question to understanding possible barriers to entry in the innovation process. To address it, we match the Annual Report of the Commissioner of Patents from 1870 to 1940 to the corresponding U.S. Federal Population Censuses. This matching procedure provides a rich set of demographic information on a comprehensive set of inventors. We first document that patentees over this seventy year period are more likely to be older, white, male and to be living in a state other than the one in which they were born. These patterns are very persistent over space and time. We then attempt to identify correlates of the demographics of patentees focusing on county-level economic and demographic characteristics. Beyond the most obvious, such as the fraction of a particular demographic group in that county, very little explains differences in the demographics of inventors across counties. We then examine two historical institutions that differentially affected particular demographic groups. For blacks, we consider historically black colleges and universities (HBCUs) and for women, state-level extension of the franchise. We find some evidence that HBCUs differentially increased black patenting rates while the extension of the franchise did not seem to have an effect for the representation of women amongst inventors.



I compare the strengths and weaknesses of four historical patent datasets and compare the suitability of each for use in economic research. I show first that a number of historical sources exist that are nearly as complete as are data on contemporary patenting. Second, I describe in detail differences across the datasets in terms of patent inventor information included, reliability of provided information, and potential sample selection issues. Third, I show that while there are some differences across datasets, overall they paint a remarkably consistent picture of invention in the historical United States.



I build a duopoly model in which firms compete on a quality ladder. The only payoff-relevant state variable is the number of steps, or gap, between firms on the ladder. At each stage, each firm decides how much R&D effort to exert as well as which of two "innovation technologies" to use. A "gap-dependent innovation technology" has a high probability of success, but the cost for the follower increases as it falls more steps behind the leader. A "gap-independent innovation technology" has a lower probability of success, but the cost does not depend on the gap between firms. Thus, when the gap becomes sufficiently large, the follower switches to use the gap-independent innovation technology. If the follower has a single success using either innovation technology, it leapfrogs the leader. When only the gap-dependent innovation technology is available, the equilibrium exhibits familiar escape competition and discouragement effects: both firms exert the highest effort when the gap is small and lower effort when the gap is large. When both innovation technologies are available, changing costs in one state trickle down and affect firms' choices in other states in nontrivial ways. For instance, reducing the cost of the gap-independent innovation technology increases the follower's effort when the gap is large but decreases effort when the gap is small because it is less costly for followers to fall far behind and less profitable for leaders to move far ahead. I show that this decrease in effort when the gap is small can be sufficiently large to lower the overall expected arrival rate of successes. Other comparative static results are illustrated with numerical examples.