The Cost of the Gap
What the patent gender gap costs, measured in inventors, in growth, and in the inventions that never get made.
The gender gap in patenting is usually described as unfair. It is also expensive. Economists have spent two decades putting numbers on what a country forgoes when most of its potential inventors never reach the patent system.
The findings below measure different things. Some are observed disparities; some are model-based counterfactuals; the GDP figure is an illustrative extrapolation built on 2003 survey data. They are not additive, and none is a forecast of what any single policy or product would achieve. What is consistent across the credible studies is direction and scale: unequal access to invention has large consequences, and they show up at more than one point in the pipeline. For each estimate below we give the population, data years, and evidence type.
This article presents published research data for educational purposes. It is not legal, policy, or investment advice. Patent gender statistics depend on name-based inference and methodology choices: see How we measure.
Where things stand (current baseline)
In 2024, women were 18% of inventors named in international PCT applications, and 37% of PCT applications named at least one woman inventor (up from 23.1% in 2010). These are global international-application measures, not the share of all patents or U.S. grants (WIPO PCT Yearly Review 2025). The most-cited detailed U.S. figure, 12.8% women inventors, is a 2019 benchmark, not a current rate.
Lost Einsteins
The most-cited work on who becomes an inventor is Bell, Chetty, Jaravel, Petkova and Van Reenen, "Who Becomes an Inventor in America?" (Quarterly Journal of Economics, 2019). Linking 1.2 million inventors to anonymized tax records, the authors found that children from families in the top 1% of the income distribution are about ten times as likely to become inventors as children from below-median-income families.
The paper introduced a phrase that has since organized the whole field: "lost Einsteins." These are the women, minorities, and lower-income children who would have produced high-impact inventions had they had the same childhood exposure to innovation as their advantaged peers. Much innovation is cumulative and team-based, so the point is not that rare genius is missing; it is that a large share of ordinary inventive capacity never enters the system. The talent exists. The path to expressing it does not.
The "quadruple" figure deserves precise wording, because it is often loosened in retelling. The benchmark group is white men from top-income-quintile families, and the estimate is about the number of inventors entering the patent system, not patent quality, invention output, GDP, or products. The demographic categories overlap; they are not separate effects to be summed. The authors themselves caution against reading it as proof that welfare would quadruple.
An illustrative economic estimate
A widely cited calculation by Hunt, Garant, Herman and Munroe, "Why Don't Women Patent?" (Research Policy, 2013), is that U.S. GDP per capita could be 2.7% higher if women with science and engineering degrees commercialized patents at the same rate as comparable men.
We state the scope plainly. The 2.7% is a modeled difference in the level of GDP per capita, not an annual growth rate, a current figure, or a direct causal estimate. It combines a patenting difference measured in 2003 survey data with a cross-country estimate of the relationship between patent stocks and GDP, and the authors themselves called it a "crude calculation." Read it as an indication of possible scale, not a forecast.
The more durable finding may be the decomposition beneath it. In the authors' sample, women's lower probability of holding an S&E degree accounted for only about 7% of the commercialized-patent gap they measured. Much more was associated with what happened among degree holders: field, occupation, and development and design work. (This is a decomposition of one commercialized-patent outcome in 2003 survey data, not a general statement that "education explains 7% of the patent gender gap.")
A separate, preliminary estimate from Lisa Cook and colleagues puts the range at 0.6% to 4.4% higher GDP per capita if women and African Americans participated fully in the innovation economy. Because its population and method differ from Hunt's, we treat it as context, not direct corroboration of a gender-only number, and its underlying working paper is not peer-reviewed.
The inventions that never get made
The GDP figures measure lost output. They do not capture what is arguably the more concrete loss: the specific inventions that are never made because the people who would have made them are not in the system.
Koning, Samila and Ferguson, "Who Do We Invent For?" (Science, 2021), analyzed 441,504 U.S. biomedical patents from 1976 to 2010. Patents from all-women inventor teams were 35% more likely than those from all-men teams to focus on women's health. Under an equal-representation counterfactual over that study period, the authors estimated roughly 6,500 additional women-focused patented biomedical inventions would have appeared in the data.
Two cautions on that number. It concerns patented inventions over 1976-2010, not products, treatments, or approvals available today, and it does not estimate health outcomes or commercial value. What it does support is a careful, consequential conclusion: patent data cannot tell us how many products were never built, but they do show that who participates in invention can shape which problems receive patent attention. (See the field-by-field breakdown.)
The gap is a pipeline, not a single leak
No single study locates the entire gap at one stage, and the evidence points to losses at several. Childhood exposure shapes who enters an inventive career (Bell). Education, field, and job tasks shape who works on patent-intensive problems (Hunt). Recognition and internal invention processes shape whose contributions get written up and forwarded to counsel. And post-filing decisions matter too: a 2025 randomized trial at the USPTO found that examiner assistance substantially raised women pro se applicants' likelihood of obtaining a patent, and a separate study of nearly a million applications attributes more than half of the issued-patent gap to differential abandonment after an early rejection. Our own prosecution-funnel page walks those post-filing stages in detail.
So the honest framing is not "the loss happens at disclosure." It is that the loss compounds across the pipeline, and internal invention harvesting is one actionable upstream stage, where reducing reliance on self-nomination is comparatively cheap and within an organization's control. It is one intervention point, not the whole explanation, and no single intervention is a demonstrated complete fix.
One institutional case study is often cited here: at Washington University, patent filings on behalf of women faculty were nearly 129% higher in the three years after a targeted tech-transfer initiative than in the three years before. It is a promising single-institution before/after comparison, not a randomized evaluation, and it does not isolate which component drove the change. (See closing strategies that work.)
Where ObviouslyNot fits, and where it does not
ObviouslyNot's scanner reads a codebase and surfaces candidate technical concepts, each with traceable references back to the code, for human review. It offers an additional route into an internal invention-harvesting process, one that does not require an engineer to first recognize and nominate a fully formed "invention."
That role is deliberately limited, and the limits matter:
- It does not determine inventorship. Under U.S. law, inventorship turns on a human contribution to the conception of a claimed invention. Writing code, making commits, or submitting a disclosure does not by itself make someone a legal inventor, and only natural persons can be inventors. Patent counsel assesses inventorship claim by claim.
- Code is one record among many. A scan can miss system design, hardware, data work, security decisions, and user research documented outside the repository. "Strategic concept" is a workflow term, not a legal category; the scanner does not assess patentability, novelty, ownership, or freedom to operate.
- Identity-blind is not automatically neutral. Not filtering by who wrote the code removes one bias, but results still reflect who was assigned to which repositories and whose non-code contributions are visible.
- Keep it confidential. Handle scan outputs and source code under confidentiality controls, and consult counsel before any public release; prefiling public disclosure can forfeit rights in many jurisdictions.
The honest claim is a design hypothesis, not a demonstrated outcome: systematic, code-based concept discovery may reduce dependence on self-nomination at one upstream stage. Whether it narrows a gender gap in any given organization is an empirical question that needs outcome data, which we do not claim to have here.