Shocks, Reallocation, and the Direction of Innovation: The Case of COVID-19

• In Event: Innovation Under Pressure: Policies, Shocks, and Spillovers

Abstract

This paper examines how sudden shifts in salience affect the allocation of scientific effort across biomedical fields, using the COVID-19 pandemic as a natural experiment. While crisis-driven reallocation can accelerate innovation in urgent domains, it may also divert attention from fields with high long-term social returns. To quantify these trade-offs, I construct a novel panel dataset of over 15 million PubMed-indexed publications (1999–2023), enriched with NIH disease-category metadata and structured using a hybrid natural language processing pipeline. Specifically, I use dimension-reduced Sentence-BERT embeddings and K-nearest neighbors to cluster 330 NIH disease categories into 35 coherent research domains, and fine-tune a BERT classifier to assign probabilistic relevance scores to each publication. The resulting domain-level effort measure—based on monthly, probability-weighted publication counts—captures both field-specific intensity and cross-field complementarities.

To estimate the causal impact of the pandemic, I employ an interrupted time series (ITS) design with field and calendar-month fixed effects, focusing on both discrete level shifts and slope changes post-March 2020. Preliminary estimates reveal a sharp reallocation of scientific attention: within six months of the first U.S. COVID-19 case, aggregate non-COVID biomedical research effort declined by approximately 14 percent. Twenty-three of thirty-five research domains experienced statistically significant losses, with the largest reductions observed in Cognitive Neuroscience (−548 publications/month) and HIV/AIDS (−500). In contrast, a subset of domains—including Cancer Biology, Genomics & Precision Medicine, and Tissue Engineering—exhibited strong gains, in some months exceeding the increase observed in COVID-related research itself.

To interpret these heterogeneous responses, I develop a dynamic model of effort allocation in which a planner reallocates fixed scientific labor across fields subject to convex adjustment frictions and network spillovers. The model’s first-order condition implies that optimal reallocations should be gradual and should spare fields with high systemic importance. I show that matching the observed post-shock dynamics requires introducing a salience distortion parameter that inflates the perceived value of crisis-relevant output relative to its social value. Structural calibration suggests that such distortions explain a significant portion of the crowd-out observed in foundational domains.

Although results are preliminary, they document a strong level change in research output and suggest that real-time policy responses may have unintentionally disrupted the long-term knowledge frontier. This research contributes to the literature on innovation under uncertainty by providing a tractable empirical framework for evaluating shocks to scientific salience. The methods developed—particularly the soft field-level effort metric and clustering pipeline—are generalizable and can be used to study other events such as funding interruptions, high-profile discoveries, or exogenous changes in media attention. Future work will integrate more granular funding and citation data to assess persistence, lagged spillovers, and implications for research equity.

Author

• Mehrdad Baratian, Tulane University
  Presenting Author