S9 Ep31: How well does patent screening work?

Someone once held a patent on the swing. A piece of wood. Two ropes. The US Patent Office granted it. How often does that actually happen, and what does it cost when the system gets it wrong? Or, how often is a valid patent claim rejected?Until now, no one knew. Tim Phillips talks to Mark Schankerman of LSE and CEPR, who with co-authors William Matcham spent eight years building the tools to find out. Using natural language processing across a dataset of around one million patent applications, twenty million claims, and fifty-five million examiner decisions, they measure how similar each incoming claim is to the hundred million claims that preceded it, going back to 1976. They find that 81% of initial patent claims fall below the patentability threshold; examiners must negotiate that figure down round by round. And they do a pretty good job. But around a third of all abandoned applications contain at least one valid claim the system failed to protect. You don’t see patents that aren’t awarded, so those errors have, until now, been invisible.The research behind this episode:Matcham, William, and Mark Schankerman. Forthcoming. "Screening Property Rights for Innovation." Econometrica. Available as CEPR Discussion Paper DP18334 (gated). Current version dated January 2026.To cite this episode:Phillips, Tim, and Mark Schankerman. 2026. “How “well does patent screening work? VoxTalks Economics (podcast). Assign this as extra listening. The citation above is formatted and ready for a reading list or VLE.About the guestMark Schankerman is Professor of Economics at the London School of Economics, where his research spans innovation, intellectual property, and the economics of technology. His work has examined how patent rights shape R&D incentives, the market for technology, and the behaviour of innovative firms, with particular attention to the institutions that govern how property rights are allocated and enforced.Research cited in this episodePrior art. In patent law, prior art is any publicly available knowledge that predates a patent application. Examiners are required to search prior art and reject claims insufficiently distinct from it. The concept defines the outer boundary of what can be granted protection; the closer a claim is to prior art, the weaker the case for granting it.Type I and Type II errors in patent screening. A Type I error occurs when an examiner grants a claim that should have been rejected, typically because it is too similar to prior art. This allows the holder to charge royalties and, in the US context especially, to bring litigation. A Type II error occurs when a valid claim is refused or abandoned, depriving the applicant of protection they deserve and reducing future incentives to innovate. Schankerman argues that Type II error is systematically under-discussed in public debate: you can point to a patent that should not have been granted; you cannot point to the invention that was never protected.Structural model. The paper uses a dynamic structural model, meaning it models the actual institutional rules, incentives, and decision sequences that govern patent prosecution at the USPTO. Structural models allow researchers to run counterfactual experiments, asking what would happen if specific rules or incentives were changed, without running those experiments for real. This is the methodological basis for the paper's policy analysis.Patent distance measure. The paper's key methodological innovation is a quantitative measure of how similar a patent claim is to existing claims, constructed using natural language processing. The algorithm is trained on existing patent documents and compares the textual content of each incoming claim against all prior claims, covering roughly a hundred million filings going back to 1976. This produces a scalar distance figure that can be compared against an estimated patentability threshold.Deadweight loss. The standard economic term for the welfare cost created when prices are raised above competitive levels. In the patent context, a wrongly granted claim allows its holder to charge higher licensing fees than the market would otherwise bear, generating a cost for users without a corresponding social benefit.Request for Continued Examination (RCE). A procedural mechanism in the US patent system that allows applicants to re-open a finally rejected application in exchange for a fee. Unlike the European Patent Office or China's patent system, the USPTO places no hard limit on how many times an applicant can return. Schankerman's counterfactual analysis fi