Research

Abstract: In frontier sectors such as biotechnology and artificial intelligence, firms increasingly compete to hire PhD-trained scientists to drive innovation. At the time of hire, firms face substantial uncertainty about how individuals will perform in industry, yet they can observe measures of academic productivity. How informative is academic performance about individuals' productivity in industry, and how does this impact hiring outcomes? Using confidential administrative data on 40 cohorts of PhD graduates across all fields linked to wage and publication records, I estimate individuals' productivity in both academia and industry, regardless of their actual sector of employment. I do so using a Marginal Treatment Effects framework that accounts for endogenous sorting into sectors and exploits variation in industry demand for PhDs at graduation. I find that academic productivity provides only limited information about individuals’ productivity in industry, with an average correlation of 0.20--0.29. This informativeness has declined over time and varies markedly across majors. In majors where academic productivity is a weaker proxy for industrial productivity, firms' hires are significantly more likely to be laid off, consistent with greater difficulty identifying individuals whose skills translate effectively to industry. Together, the results suggest that limited alignment between academic and industrial productivity may create systematic challenges for firms when hiring scientific talent.

Abstract: Many inputs to scientific production are scarce. As a result, researchers often receive fewer inputs than their research agendas would require. Which projects do scientists prioritize when they cannot pursue all the projects in their pipeline? One possibility is that they preserve work with more dependable paths to publication and recognition; another is that they concentrate scarce inputs on riskier projects with greater upside if successful. We study this question in the context of high-performance computing using XSEDE, an NSF-funded centralized allocation system for supercomputing resources. Exploiting variation in the extent to which projects receive fewer computing resources than expert reviewers judged they required, we examine how constraints on high-performance computing relate to subsequent research output. We find that tighter constraints reduce the number of papers produced and narrow topical scope. They also shift output toward work that is more conventional, less recent, less novel, and less exploratory. The share of papers in both tails of the citation distribution declines, while the share in the middle increases, consistent with a shift away from riskier projects. Taken together, these findings suggest that constraints push scientists toward projects with more dependable paths to recognition. This tendency appears stronger for established scientists, who retreat more sharply from exploratory activity. Overall, the results indicate that constraints on research inputs do not simply reduce scientific output; they also redirect the trajectory of scientific progress toward safer research trajectories.

Abstract: Startups vary in the extent to which their technologies build on their inventors’ own prior research, rather than on the work of others. Yet, we know little about how such differences shape outcomes in exit markets. In this paper, we examine: (1) how the extent to which a startup’s technology relies on its inventors’ own scientific knowledge relates to exit outcomes and (2) how this relationship depends on whether the inventor of the technology is also the founder of the startup. Using data on 1,035 bio-medicine startups founded between 2005 and 2012, we find that greater reliance on inventors' own scientific knowledge is associated with poorer exit outcomes: a 10 percentage-point increase in reliance on inventor-origin science is associated with a 31% decrease in the likelihood of acquisition or IPO relative to the mean. Additional analyses suggest that this negative relationship may reflect three challenges for external stakeholders. Startups that rely more heavily on inventor-origin science appear to build on knowledge that is more tacit, less redeployable, and associated with lower option value. These patterns are not readily explained by differences in technological quality or commercial potential. Finally, we find that this negative relationship is substantially attenuated when the inventor is also a founder, suggesting that an inventor-founder mitigates some of the challenges associated with tacit, inventor-dependent knowledge. These findings highlight that knowledge-based advantages in entrepreneurial contexts may be contingent on aligning knowledge characteristics with founder identity, and underscore the importance of distinguishing between the source of knowledge and the individual who embodies it.