(Mis)Measuring Innovation around the World

Research on corporate innovation often focuses on firms with positive (USPTO) patent activity and reported R&D, thereby excluding over 90% of the firms in Compustat. Exploiting data from 30 global patent offices, we investigate the nature of missing innovation data. We find systematic and predictable patterns in firms with missing patents and R&D suggesting such missingness is not completely random. Based on these findings, we compare the empirical efficacy of excluding firms without USPTO patents or reported R&D, to simple replacement methods, and to various econometric solutions for missing patent and R&D data. We then show how excluding or deleting firms without USPTO patents or reported R&D provides biased coefficient estimates and standard errors in studies of innovation; leading to specific distortions in tests related to corporate growth and country level innovation capacity. Further analysis indicates a variety of problems with using partial sample methods to handle missing innovation data.