Deepak Hegde (NYU Stern), Kyle Herkenhoff (Minn. Econ), and Chenqi Zhu (NYU Stern PhD candidate) have decided to attack the problem from a different angle: using the AIPA (which required patent disclosure at 18 months) as a natural experiment. The paper is on SSRN, and the abstract is here:
How does the disclosure of technical knowledge through patents affect knowledge diffusion, follow-on invention, and patenting? We study this by analyzing the American Inventor's Protection Act (AIPA), which required U.S. patent applications filed after November 28, 2000 to be published 18 months after filing, rather than at grant, and advanced the disclosure of most U.S. patents by about two years. We estimate AIPA’s causal effect by using a counterfactual sample of identical European “twins” (of U.S.patents) which were not affected by the U.S. policy change and find that AIPA (i) increased the rate and magnitude of knowledge diffusion associated with U.S. patents (ii) increased overlap between technologically distant patents and decreased overlap between similar patents. Patent abandonments and scope decreased, while patent clarity improved, after AIPA. The findings are consistent with the predictions of our theoretical framework which models AIPA as provisioning current information about related technologies to inventors. The information, in turn, reduces follow-on inventors’ R&D and patenting costs. Patent disclosure promotes knowledge diffusion and clearer property rights while reducing R&D duplication.This was a clever project. There have been AIPA studies before, but none that try to measure the value of the diffusion, so far as I know. What makes it go is the matching with European patents (which had always been published), which allows for their measurements to be independent of quality of invention.
They find that the earlier disclosure made a difference in diffusion: faster citations, more citations in the same amount of time, changes in technological overlap that implies that patents are being read, fewer abandonments (due to overlap), fewer but longer claims. They did not find these changes in European patents centering on the passage of the AIPA. Furthermore, they do not find these effects for patents that opted out of the AIPA, which implies that it really is the AIPA that is driving these results.
So, it is clear that somebody is reading these disclosures, but the question is who? The draft implies that one conclusion is decrease in duplicative R&D, which should be one of the primary benefits of the patent system and early licensing, as I argue in my essay Licensing Acquired Patents (among others). But there are some hurdles this article faces to get to that conclusion.
First, the European patents were published pre-AIPA, so why wasn't anyone reading and citing those? There are a couple of good responses to this. I believe that most citation study work has shown that citation is stronger more locally (another argument for diffusion effects). If so, then European patents may have been ignored (indeed, Cotropia, Lemley, and Sampat have a paper on this). In addition, the authors correct for this by examining whether European cites turned into US cites and for patent families.
Second, one has to buy into the idea that these are the appropriate measures of diffusion. Are they cites added by examiners or by applicants (the paper has data on this). Are the claims narrowed by the examiner, by the patent attorney, or the inventor? Are there fewer abandoned applications because of less duplicated R&D, or because patent attorneys are not claiming as much? In other words, the diffusion may well be to patent attorneys at the point of filing, and not to inventors at the time of inventing. This, I think is a difficult hurdle. The authors attempt to justify this assumption based on the cost of secrecy and the risk of disclosure for important inventions, but they make clear that it is a key assumption for their definition of knowledge diffusion.
Even if one does not buy into the strong version of their conclusions, however, this study has a really important real world payoff: publishing patents has an effect. They are seen, and they affect the body of prior art in a way that limits future claims (at the very least) and increases citations (which makes searching easier). This alone is an important function; as Lemley, Sichelman, Wagner, and I argued in prior work, one of the costs of not having software patents through the 1970s was that there was no prior art to knock out all the software patents of the 1990's. This study confirms this for us.