Collaborative Research: The Impact of Research Costs on the Direction of Scientific Discovery Frontier Tools & Applications

Project Outcomes Statement

In this project, we have developed a series of techniques, including machine learning and Natural Language Processing tools, to analyze drivers of the direction of innovation and we have applied these to a number of research contexts. We have conducted mainly causal analyses that measure the ways in which innovation evolves in response to shocks in automation, information, and public policies. To do this, we have leveraged advances in computing power and empirical research techniques that enable us to map the evolution of research fields and researchers’ project portfolios in “ideas space” and pharmaceutical firms’ drug development research trajectories.

The empirical analyses in our projects are enabled by our having developed text-based tools that enable researchers to document changes in the direction of innovation. One of these builds upon an unassisted machine learning technique, Hierarchical Dirichlet Process (HDP), which flexibly generates categories from a corpus of text and enables calculations of the distance and movement in ideas space. A second involves using a metric known as a “Tanimoto score” or “Jaccard coefficient,” which reflects the chemical similarity of molecules in order to document biopharmaceutical firms’ research directions and which has been used by firms to inform investment choices.

We use these tools to make a number of contributions to research on the rate and direction of innovation. First, we document that the automation of research tasks that previously required substantial human labor presents opportunities for scientists to expand their research portfolios and for policymakers to accelerate the rate and broaden the direction of knowledge accumulation. Contrary to concerns that automating technology displaces individuals from work tasks, such an automation increased the production of ideas and induced researchers to pursue ideas more diverse than and distant from their original trajectories. This work suggest that automation in the context of jobs that involve substantial autonomy can increase productivity and the diversity of outputs by enabling workers to shift their labor to more productive tasks and explore more distant ideas. Second, we find that research technology costs influence cumulative innovation by altering the composition of expertise in teamwork. Sufficiently large reductions in the cost of research technology leads to greater collaborations across research domains. Third, we demonstrate that risk aversion leads pharmaceutical firms to underinvest in radical innovation. Further, pharmaceutical firms are more likely to terminate clinical trials in response to learning about the failure of rival firms’ drugs and to turn to the market to acquire drug projects from other firms, rather than developing internally.

These research efforts have led to numerous publications in leading journals, including papers that have appeared in the American Economic Journal (Economic Policy), Management Science, Organization Science, Nature Machine Intelligence, the Review of Economics and Statistics, and the Review of Financial Studies.

In addition to having produced academic papers, the project grant has supported the training of multiple doctoral students and master’s degree students. In total, the grant supported the education and research projects of more than a half-dozen graduate students and academic publications. As well, the grant has enabled us to develop and share our research tools. The tools are available online (e.g., via Github) and we have disseminated them to academic audiences via dozens of research presentations. We are proud to report that these tools have already been used by scholars in follow-on research. We believe that, in addition to providing value for academic scholarship, they have application for making managerial decisions, e.g., for R&D management, and for informing public policy choices.