Counting patents, miscounting innovation
For those who draft, examine, litigate, or value patents, counting patents to measure innovation can sound like a terrible idea.
Every year, we see rankings of the “most innovative” countries, regions, companies, universities, or even individual inventors. The tables vary, but the logic is often the same: count patents, rank actors, and infer innovativeness. But for those who draft, examine, litigate, or value patents, counting patents can sound like a terrible idea.
Patent professionals know that patents vary widely. Some protect foundational technologies, while others cover narrow improvements. Some are commercially central; others are defensive, speculative, or never used. Some inventions are never patented, and some patented inventions are economically irrelevant. In complex technologies, a single commercial product may be surrounded by hundreds or thousands of patents.
So why do researchers and policymakers keep counting patents? The short answer is that, as imperfect as they are, patent data remain among the best indicators available. As Zvi Griliches put it in his 1990 article, “in this desert of data, patent statistics loom up as a mirage of wonderful plentitude and objectivity.” They are public, structured, dated, classified, and linked to inventors, firms, technologies, and jurisdictions. Few other traces of inventive activity offer such breadth over such long periods of time.
Importantly, this reliance is not merely pragmatic. A substantial body of empirical research has validated the use of patent counts as indicators of innovative output at the country level. Early work by Luc Soete and Sally Wyatt showed a remarkably strong relationship between national business R&D expenditures and per capita foreign patenting in the United States across OECD countries. More recent contributions have addressed long-standing concerns about noise and heterogeneity. Some work I did with Bruno van Pottelsberghe showed that differences in patents per researcher across countries reflect genuine differences in research productivity, not merely administrative or strategic variation. Panel analyses with Jérôme Danguy further confirmed a stable, positive long-term relationship between R&D stock and patent output across industries and countries. Complementary bibliometric studies and work by Zoltan Acs and colleagues reinforce the conclusion that, while imperfect, patent counts provide a reasonably reliable and widely accepted proxy for innovative activity.
Thus, the problem is not that we count patents. The problem is that we too often forget what patent counts actually measure. A familiar saying captures part of the issue: not everything that counts can be counted, and not everything that can be counted counts. In the world of patent statistics, we should add a third proposition: even when something can be counted, the precision of the count is often an illusion.
Patents are not innovation
A common conceptual mistake is to treat patents as innovation itself. They are not. Patents are better understood as intermediate outputs of inventive activity. [1] Spending on research and development (R&D) measures an input: resources devoted to producing knowledge. New products, productivity gains, or firm growth measure outcomes: what happens after ideas are commercialized and diffused. Patents sit somewhere in between: they record the decision to disclose and claim a technical invention through the patent system.
That makes them very useful. A patent document tells us that, at a particular point in time, someone claimed to have solved a technical problem in a way that might merit exclusive rights. It also tells us something about the technological domain, the applicant, the inventors, the geography of inventive activity, and the legal strategy surrounding the invention. But it does not, by itself, tell us whether meaningful innovation occurred.
This distinction matters. A patent count is not a count of inventions, let alone a count of innovations. It is a count of patenting events. Sometimes patenting events are good indicators of inventions. Sometimes they are mostly indicators of legal strategy, institutional incentives, market exposure, or administrative practice.
The missing inventions
The first source of distortion is simple: not all inventions are patented. Some inventions are not patentable. They may fall outside the scope of patentable subject matter, fail to meet the requirements of novelty or inventive step, or be too abstract, too obvious, or too difficult to claim.
Other inventions are patentable but not patented. Firms may rely on secrecy. They may prefer lead time, data advantages, complementary assets, branding, regulatory exclusivity, or speed of execution to capture innovation rents. [2] In some industries, process innovations are difficult to detect, making infringement hard to prove and patent enforcement unrealistic, so firms may not bother to file patents. In others, disclosure itself may be too risky because a patent teaches competitors how to imitate.
The propensity to patent varies dramatically across sectors. Pharmaceuticals and chemicals rely heavily on patents. Software, services, finance, and many forms of organizational innovation may leave a much weaker patent trail. The ongoing generative AI revolution is a case in point: relatively few patents are issued, yet the industry is widely regarded as highly innovative, with billions of dollars in R&D investment.
In short, a low patent count does not necessarily indicate weak innovation. It may simply mean that the relevant actors innovate in ways that are not well captured by patents. The reverse is also true. A high patent count does not necessarily indicate strong innovation. It may mean that firms operate in a technological field where patents are strategically useful.
One invention, many patents
A second problem is that even when inventions are patented, there is no stable one-to-one relationship between inventions and patents. One invention may generate one patent, but it may also generate none, several, or many.
Patent law itself contributes to this. Rules on unity of invention, restriction practice, divisionals, continuations, and claim scope shape how applicants divide technical material across documents. A single technological project may be split into multiple applications because the patent office requires it, because the applicant wants to preserve flexibility, or because portfolio strategy rewards fragmentation.
This aspect is especially important in complex-product industries. A smartphone or telecom standard may embody hundreds (if not thousands) of patented components. By contrast, a drug is often covered by a relatively clear set of patents, such as those covering the molecule, formulation, use, or manufacturing process.
Every patent gets one vote
A third problem is that raw counts treat all patents equally. This assumption is convenient, but it is also very strong.
A foundational patent, a narrow design-around, a defensive publication dressed as a patent, and a low-value incremental claim are all counted the same way: one patent. But patent value is famously skewed. [3,4] Many patents are worth little or nothing, while a small minority accounts for a disproportionate share of technological and economic value. The average patent is not very informative because the distribution is so uneven. It follows that a patent count can appear precise yet be conceptually crude. The number may be exact, but the units being added are radically unequal.
This problem is not limited to patents. Counting scientific publications faces a similar issue: one article in a minor outlet and one pathbreaking paper both count as one publication. But the problem is particularly acute for patents because the reasons for filing are so diverse. As explained in a previous post, a patent may be filed for a variety of reasons, including to exclude competitors, attract investors, signal technical capability, secure freedom to operate, build a bargaining chip, support a standard-essential portfolio, obtain subsidies, impress a board, or simply preserve an option.
One common response is to weight patents using proxies for value. Forward citations are the most familiar example: a patent cited more often by later patents is presumed to have left a larger technological footprint. But patent citations are not like scientific citations. They are heavily shaped by patent-office practices, examiner searches, applicant disclosure duties, and jurisdiction-specific rules. This makes citation-weighted patent counts useful in some settings but difficult to compare across countries or offices. Xuesong Tong and Davidson Frame proposed weighting patents by the number of claims. This approach is appealing because claims define the legal boundaries of the invention. But it also pushes the problem one step further: not all claims are equally broad, important, or enforceable. A claim count is still a count.
Geography makes the problem worse
Patent counts become even more fragile when they are used to compare countries. At first glance, international comparisons seem straightforward. Count patents from each country and rank them. But even this simple idea hides an important ambiguity: what exactly is being counted? Some studies rely on filings in a single national office (such as the USPTO or the EPO), others on domestic filings within each country, and still others on international routes, such as applications filed under the Patent Cooperation Treaty (PCT). Each choice captures a different slice of inventive activity and reflects different economic and strategic decisions by applicants.
The first distortion is home advantage. Domestic applicants are more likely to file in their home office than foreign applicants. A count based on USPTO patents naturally reflects exposure to the U.S. market and legal system. A count based on EPO applications similarly reflects exposure to Europe. By contrast, a count based on national filings reflects local costs, incentives, and expectations about enforcement. In other words, the patent office of reference is not a neutral window onto world innovation. It is itself part of the geography being measured.
The second distortion is legal and administrative variation. Patent offices differ in examination practices, thresholds, backlogs, applicant behavior, fee structures, and grant rates. A patent application in one jurisdiction is not necessarily comparable to one in another. A granted patent in one office is not necessarily equivalent to a granted patent elsewhere.
The third distortion is policy-induced patenting. When governments reward patent filings, applicants respond. China provides a well-documented example. For years, local and national programs offered cash subsidies, tax incentives, and performance targets tied to patent filings, which contributed to a rapid rise in applications, many of which were of limited quality, as documented by Philipp Boeing and Elisabeth Muller, as well as by Zhen Sun and colleagues. [5,6] This is a version of Goodhart’s law: when a measure becomes a target, it ceases to be a good measure.
Applications, grants, and families
Researchers have spent decades trying to address these problems. One obvious refinement is to distinguish applications from grants. Applications measure attempts to obtain rights, while grants measure applications that have survived examination. Neither is a direct measure of innovation, but they provide different information.
Applications are timely (subject to an 18-month delay) and broad. They tell us where applicants seek protection. But they also include speculative filings that are ultimately abandoned or rejected. Grants meet a legal threshold. But that threshold varies across offices, technologies, and over time.
Another refinement is to count patent families rather than individual documents. A patent family groups related applications that protect the same underlying invention across jurisdictions. This brings us closer to the invention level and avoids counting each national phase or foreign equivalent as a separate invention.
This logic underlies indicators such as triadic patent families, which focus on inventions protected at the EPO, the JPO, and the USPTO. [7] The intuition is simple: if an applicant is willing to incur the cost of protecting an invention in several major markets, the invention is more likely to be valuable than a filing that remains purely local. Furthermore, by requiring protection in multiple offices, these indicators also reduce the “home advantage” inherent in single-office counts, since they are less influenced by domestic filing behavior and better reflect inventions with broader international relevance. Other measures follow a similar logic, including PCT-based indicators, IP5 families, and transnational patents. The latter measure, introduced by Rainer Frietsch and Ulrich Schmoch, counts PCT families with an EPO member.
These indicators are often better than raw counts, but they introduce their own selection bias. They privilege inventions that are worth protecting internationally. That may be exactly what we want when studying globally valuable technology, but it may be less appropriate when studying local innovation, incremental improvements, or technologies whose markets are geographically concentrated. In other words, the cure also shapes the diagnosis.
So should we count patents?
Yes. But we should count them with care. Patent statistics remain indispensable because they provide a rare, systematic view of technological activity across dimensions that matter, notably countries, firms, sectors, and time. They allow us to study inventive geography, science-technology linkages, knowledge diffusion, firm strategy, policy shocks, and long-run technological change. The mistake is not to use patent data. The mistake is to forget that patent data measure patenting behavior first and innovation, only indirectly.
A good patent indicator starts with the research question. If the question concerns legal activity, raw applications may be appropriate. If the question concerns economically meaningful inventions, patent families, renewals, citations, claims, or international filings may be better. If the question concerns local inventive activity, worldwide priority counts may be more relevant. [8] If the question concerns firm strategy, the portfolio itself may be the object of interest. Put differently, there is no universally correct patent count. There are only counts that are more or less appropriate for a particular question.
The illusion of precision
Patent statistics are seductive because they produce clean numbers. A country filed 100,000 patent applications. A company obtained 5,000 grants. A university produced 200 patent families. The numbers are precise, but the meaning is not. A country with 100,000 applications is not necessarily twice as inventive as one with 50,000. This is the central paradox of patent statistics: they offer clarity in form while remaining ambiguous in substance.
The right response is not to stop counting patents but to interpret them with discipline—asking what exactly is being counted, why it was counted that way, and what kinds of innovation fall outside the frame.
More broadly, it is good practice to measure innovation using multiple indicators rather than relying on a single metric. A prominent example is WIPO’s Global Innovation Index, which draws on multiple sources of evidence to provide a more balanced and nuanced picture of innovative activity.
If you enjoy evidence-based takes on patents and innovation, join hundreds of readers who receive The Patentist directly in their inbox.
Please cite this post as follows:
de Rassenfosse, G. (2026). Counting patents, miscounting innovation. The Patentist Living Literature Review 14: 1–7. DOI: TBC.