Generative AI and the Reconfiguration of Innovation Validating Exposure Measures and Reassessing Corporate R&D after ChatGPT
Description
The public analysis panel (Compustat raw fields stripped, firm IDs anonymized), the derived exposure files, a data dictionary, the replication code (re-pathed to run from the package), the figures, a README, and a manifest. No licensed fields leaked. Now the data dictionary, README, and manifest.
Files
Steps to reproduce
The public panel dataset is provided in the replication package and contains 4,172 firm-year observations for 235 firms. Running the main estimation script on the public panel reproduces the headline result exactly (β = −1.32, firm-clustered t = −2.63), as well as the corresponding O*NET estimates (β = 1.08, firm-clustered t = 2.48). The replication package includes the public panel dataset, exposure files, re-pathed code, figure-generation scripts, a data dictionary, README documentation, and a manifest. All scripts have been tested to ensure that they run directly from the package root directory using the public data. The complete workflow reproduces the reported coefficient (β = −1.32, firm-clustered t = −2.63) and the associated figures from the public panel dataset.