|Figure 1 from Abrams et al. (note that data is normalized)|
Abrams et al. develop a model to explain this inverted-U as the result of two different types of innovation: productive innovation, when the most valuable inventions are built on by more subsequent inventions (leading to the traditional increasing relationship between value and citations), and strategic innovation, which focuses on producing fencing patents to prevent others from entering the field. Because the most valuable strategic patents are the ones that create the largest barriers to entry, there is a negative relationship between value and citations for these patents. The coexistence of productive and strategic patenting results in the inverted-U.
While there may be other ways to explain the results, this paper makes a valuable contribution to a tremendously important problem. Many of the questions at CELS focused on how generalizable the patent valuations in this dataset are—e.g., do patents that get sold to an NPE tell us much about the value of patents that operating companies choose to keep? But even having this NPE valuation data available for the first time is terrific, and these results certainly complicate the literature that has relied on citation-weighted patent counts. In future work, the authors plan to provide guidance on how to better proxy for patent value, and I look forward to these results.
But as useful as this work is, I also think it is worth emphasizing what is perhaps obvious: even if scholars like Abrams and his coauthors lead us to much better patent value estimates, patents will remain an incomplete innovation metric because many valuable innovations occur outside the patent system. See, e.g., Petra Moser on innovation without patents in the 19th century, Kal Raustiala & Chris Sprigman on the role of informal norms and market incentives in producing innovation where IP protection is unavailable, and Amy Kapczynski & Talha Syed on health innovations like ICU checklists that fall outside the patent system. And patent-based metrics are particularly ineffective when scholars are trying to study the effect of patent laws themselves. That doesn't mean that patent citations can't be a useful proxy in some contexts—but innovation scholars should remain wary of the streetlight effect of only searching where it is easiest to see the data.