The past year has been a great one for fascinating (though somewhat depressing) research on the demographics of inventorship. In late 2017, The Equality of Opportunity Project (now Opportunity Insights) released the results of their study Who Becomes an Inventor in America? The Importance of Exposure to Innovation. (See my discussion of some of its findings here.) Then, in April, a study on gender differences in the patenting process was published by a group from the Yale School of Management in Nature Biotechnology. The study found that inventors with first names usually given to women – especially those with names easily identified as such – fared worse than inventors with first names usually given to men.
The Yale study was made possible by the U.S. Patent & Trademark Office’s release a few years ago of the Patent Examination Research Dataset (“PatEx”). The original 2015 PatEx release (since updated in 2016 and 2017) included information on over 9 million U.S. patent applications, including bibliographic data (e.g., inventors, examiners, technology area) and prosecution history data (i.e., information on the interactions between patent applicants and the USPTO during the examination process). Certain bibliographic data was reasonably accessible in bulk before, but PatEx’s prosecution history data provides a treasure trove for all kinds of new empirical research about the patenting process. These prosecution histories have been publicly available online for years through the USPTO’s Patent Application Information Retrieval (PAIR) system. But PAIR is designed for individually accessing application records (and since 2007 has had reCAPTCHA to deter data miners), which made large-scale empirical research using PAIR difficult.
In the Yale study, the authors used PatEx data on publicly available nonprovisional utility patent applications filed from 2001 to 2014 (about 3.9 million applications), which they combined with USPTO databases containing the full text of published patents and applications and maintenance fee payment history. The study attempted to assign each inventor a probable binary gender based on a gender disambiguation dataset compiled from U.S. Social Security data and two services that collect name and gender data from online sources. A gender was assigned if more than 95% of individuals in the disambiguation dataset with a particular first name were women, or more than 95% were men. (The authors also checked the robustness of their results by running their analyses using a 90% cutoff, and confirmed that the results were the same.) The 95% cutoff lead to gender assignments for 88.6% of the inventors, such that about 70% of the applications had a gender assigned for every named inventor – 76.3% of applications with inventors all residing in the U.S., and 64.6% of applications having at least one inventor residing outside the U.S.
The resulting dataset allowed the study’s authors to examine characteristics of female inventorship, such as changes in female inventorship over time, rates by technological area, and the extent to which female inventors tend to invent with other female inventors. These findings are interesting, but the Yale study’s particularly significant contribution lies in its analyses related to the patenting process. (Though gender and patents is a relatively underexplored area, past studies and USPTO reports have performed various gender-related analyses of bibliographic patent data – look for example here, here, and here.)
The authors found that even after accounting for technology area, patent applications with all female inventors fared worse than applications with all male inventors. (The effect of an application’s proportion of female inventors on these same outcomes is detailed in the supplementary data and reflects the same trends.)
Patent applications with all female inventors were less likely to result in granted patents than applications with all male inventors. Among the applications that became granted patents, those with all female inventors had a greater decrease in the number of independent claims (the broadest claims of a patent) during prosecution, and their independent claims’ lengths were increased by more words. (Claim length is often used as a proxy for claim scope, with longer claims assumed to be narrower.)
Applications with all female inventors were also less likely to have appeals filed after receiving final rejections. (The study’s authors describe the filing of an appeal as “[t]o some extent . . . captur[ing] the perseverance of the applicant.” That may be somewhat true, but its significance is less than the language might suggest – a “final” rejection is not really “final” at all, and there is another, and more common, way of continuing prosecution after a final rejection that does not require filing an appeal.)
The study also found that once granted, patents with all female inventors were less likely to have fees paid to keep them enforceable, and less likely to be cited by other applicants and examiners in prosecution of other patent applications.
As the authors acknowledge, the study can’t tell us what is causing these differences. There are typically a number of decisionmakers involved over the lifetime of a patent – for example, the decisionmakers could include the inventor(s), in-house counsel, outside counsel, company executives, and the patent examiner. One way that the authors attempted to tease out the mechanism was by looking at the relationship between gender differences and the popularity of inventors’ first names, using popularity as a proxy for the ease of inferring gender from a name. To do this, the authors looked at the subset of applications having a single inventor, to avoid the problems of aggregating first name frequency across multiple inventors. Of this subset, they compared gender differences for applications of inventors with common first names with those of inventors with rare first names. One finding was that the difference in grant rates between applications with female and male inventors was considerably larger when the inventors had common first names. Among the applications of inventors with common first names, the difference was 8.2% – that is, applications with female inventors were 8.2% less likely to result in a granted patent than applications of male inventors, whereas among applications of inventors with rare first names, the difference was only 2.8%. This suggests that female inventors are less disadvantaged (or male inventors are less advantaged) when their genders are harder to infer from their first names. Because company-side decisionmakers are more likely to know (rather than infer) the gender of an inventor, the authors suggest that about two-thirds of the lower female-inventor grant rates can be attributed to the behavior of USPTO examiners.
Given this suggestion that examiner behavior is a substantial contributor to gender differences in grant rates, I found one of the study’s other results somewhat surprising. The authors found that the first inventor’s gender had little effect on grant rate (and little effect on the other results) beyond the effect due simply to his or her contribution to the overall gender make-up of the inventors. I expected that the first inventor’s gender would have a disproportionate effect: Although the order of inventors has no legal significance, the first inventor’s name is commonly used for identification purposes during prosecution. For example, documents submitted by patent applicants to the USPTO, and documents mailed back to applicants from the USPTO, typically list only the first inventor. (Published applications and issued patents similarly display the first inventor more prominently than other inventors, but only by last name.) Clearly, there is much more to explore here about the source of gender differences and the roles of various decisionmakers, and I’m looking forward to what others do with the PatEx data.
Kyle Jensen, Balázs Kovács & Olav Sorenson, Gender Differences in Obtaining and Maintaining Patent Rights, 36 Nature Biotechnology 307 (2018).