Skip to content
Back to search results

Patent Citation Analysis

Introduced by
Adam Jaffe, Manuel Trajtenberg & Rebecca Henderson; Bronwyn Hall, Adam Jaffe & Manuel Trajtenberg
Year
1993
Type
Bibliometric and network analysis of patent citation data
Family
Strategic management / innovation measurement
Font size

What does it do?

Patent citation analysis uses the references that patents make to earlier patents as quantitative traces of innovation value and the flow of technological knowledge.

The approach was given its empirical foundation by Adam Jaffe, Manuel Trajtenberg, and Rebecca Henderson, whose 1993 Quarterly Journal of Economics study used patent citations to show that knowledge spillovers are geographically localized - inventors disproportionately build on nearby prior art. Bronwyn Hall, Adam Jaffe, and Manuel Trajtenberg's 2001 NBER work then assembled the large-scale patent-citations data file and the methodological toolkit - forward-citation counts, generality and originality indices, citation lags, and self-citation measures - that made citation analysis a standard instrument in the economics and strategy of innovation. By treating the citation network as data, researchers can measure how important an invention is, where its knowledge came from, and where it flowed.

The intuition behind the method

A patent is required to cite the prior art it builds on, and examiners add further references, so the web of citations records, however imperfectly, who learned from whom in the history of technology. Two simple ideas make this powerful. First, if many later patents cite a given patent, that invention probably mattered - forward citations behave like a vote of technological importance, and they correlate with the economic value of the underlying invention far better than a raw patent count. Second, a citation from patent A to patent B is a visible footprint of knowledge moving from B's inventors to A's, so the pattern of citations maps how ideas diffuse across firms, technologies, and places. Jaffe and colleagues' classic finding - that inventors cite nearby prior art more than a matched control would predict - shows the method's power: knowledge, supposedly intangible and frictionless, leaves geographic fingerprints in the citation record. Building on this, analysts construct indices that distinguish broadly influential inventions from narrow ones and self-built knowledge from external knowledge, turning patent databases into a quantitative laboratory for studying innovation.

History

Patent citation analysis grew out of bibliometrics and the economics of technological change as researchers sought objective measures of innovation beyond simple patent counts. The pivotal empirical demonstration came in 1993, when Adam Jaffe, Manuel Trajtenberg, and Rebecca Henderson used patent citations and a matched-control design to show that knowledge spillovers are geographically localized, establishing citations as credible evidence of knowledge flows. The field was transformed again in 2001 when Bronwyn Hall, Adam Jaffe, and Manuel Trajtenberg released the NBER patent-citations data file together with a toolkit of measures - forward and backward citation counts and lags, generality and originality indices, and self-citation shares - and documented their properties. These contributions made citation-based analysis a standard methodology across innovation economics, technology strategy, and science-and-technology policy, and underpin much subsequent work on patent networks and technological trajectories.

When to use it

Use patent citation analysis when you need quantitative, large-scale measures of innovation in technology-intensive contexts - assessing the value and impact of inventions, mapping knowledge flows between firms, regions, or technologies, tracking the trajectory and convergence of technological fields, or benchmarking firms' innovative output for strategy and policy research. It is well suited to econometric studies of R&D, technology strategy, and competitive intelligence where patents are a meaningful output of innovation. It is less appropriate in industries where firms protect innovation through secrecy or complementary assets rather than patents (so patenting understates innovation), for very recent patents whose forward citations are truncated, and as a sole measure of value, since many patents are never commercialized. Citations should be interpreted as noisy proxies, and analyses should always apply truncation and field corrections and triangulate with other indicators.

Limitations and considerations

  • Many citations are added by examiners or for legal reasons rather than reflecting genuine knowledge use, adding noise.
  • Recent patents are truncated - too new to have accumulated forward citations - biasing comparisons unless corrected.
  • Citation propensities differ sharply across technology fields and time, so uncorrected counts are not comparable.
  • Patents capture only patented innovation, missing secrecy-protected or non-patentable advances and uncommercialized inventions.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Provides objective, large-scale, and reproducible measures of innovation from comprehensive public patent data.
  • Forward-citation counts correlate with invention value and firm value far better than raw patent counts.
  • Citation links offer a rare observable trace of knowledge flows across firms, technologies, and geography.
  • Standardized indices (generality, originality, citation lags) enable comparable measurement across studies.

Applications

  • Measuring the value and technological impact of individual inventions and firm patent portfolios.
  • Mapping knowledge flows and spillovers across firms, regions, universities, and technology fields.
  • Tracing technological trajectories and the convergence or emergence of technology areas over time.
  • Benchmarking corporate innovation performance for competitive intelligence and technology strategy.
  • Informing science-and-technology policy by linking R&D, patenting, and downstream knowledge diffusion.

How it works

How it works

1

The analysis begins by assembling patent records and constructing the directed citation network, in which each patent is a node and a directed link runs from a citing patent to the cited (earlier) patent. Hall, Jaffe, and Trajtenberg's data file made this feasible at scale by cleaning and linking millions of patents and their citations along with assignee, classification, and date information. Each patent thus has 'backward' citations (the prior art it references, indicating where its knowledge came from) and 'forward' citations (later patents that cite it, indicating its influence), and this directed graph is the substrate for every subsequent measure.

Formula
G = (V, A),\quad (i \to j) \in A \iff \text{patent } i \text{ cites patent } j
2

The most widely used measure is the forward-citation count: the number of subsequent patents that cite a focal patent. Hall, Jaffe, and Trajtenberg document that forward citations are a strong indicator of a patent's technological importance and correlate with the economic value of the invention and even with the market value of the firm holding it, far better than simple patent counts that treat all patents alike. Counting forward citations therefore turns the binary fact of 'having a patent' into a graded measure of how much an invention actually influenced later technology, which is why it anchors most innovation metrics built on patents.

Formula
c_j = |\{ i : (i \to j) \in A \}|,\quad \text{forward citations as value/impact proxy}
3

Beyond raw counts, the citation pattern is summarized into structural indices, principally generality and originality. Generality measures how widely spread, across technology classes, the patents that cite a focal patent are: a high value means the invention influenced many different fields and is computed as one minus the concentration (a Herfindahl-type sum of squared shares) of citing-patent classes. Originality applies the same logic to a patent's backward citations, capturing how diverse the knowledge it drew on was. Hall, Jaffe, and Trajtenberg formalized these indices so analysts can distinguish broad, general-purpose inventions from narrow ones and synthetic inventions from incremental ones.

Formula
\text{Generality}_j = 1 - \sum_{k} s_{jk}^{2},\quad s_{jk} = \text{share of citing patents in class } k
4

Because each citation is a footprint of knowledge flowing from the cited to the citing inventor, citations are used to trace knowledge spillovers across firms, technologies, and geography. Jaffe, Trajtenberg, and Henderson's landmark design compared the geographic origins of patents citing a focal patent against a carefully matched control sample of patents with the same technology and timing but no citation link, and found that citing patents are significantly more likely to originate in the same region. This matched-control logic is the methodological heart of spillover studies, isolating genuine localized knowledge flows from the background tendency of similar inventions to cluster anyway.

Formula
\Pr(\text{cite} \mid \text{same region}) > \Pr(\text{cite} \mid \text{control}),\ \text{localized spillovers}
5

Finally, the raw measures must be corrected before comparison, because citation data are systematically distorted. Recent patents have had little time to accumulate forward citations (truncation), and citation intensity varies by technology field and over time. Hall, Jaffe, and Trajtenberg provide methodological tools to address this - scaling citation counts by expected citations for a patent's age and field, or using fixed effects - so that a patent's citations are judged relative to its cohort rather than in absolute terms. Applying these adjustments is essential; comparing raw counts across eras or fields produces badly biased conclusions about which inventions or firms are most innovative.

Formula
\tilde{c}_j = c_j / w(\text{age, tech field}),\quad \text{correct for truncation and citation-rate differences}

Common pitfalls

  • Comparing raw forward-citation counts across different ages or technology fields without truncation and field normalization.
  • Treating every citation as evidence of real knowledge transfer, ignoring examiner-added and strategic citations.
  • Equating patent or citation counts with innovation outright, when many patents are defensive or never used.
  • Studying spillovers without a matched control sample, so ordinary clustering of similar inventions is mistaken for localized knowledge flow.

Frequently asked questions

Why are forward citations used as a measure of patent value?

A patent that many later patents cite has demonstrably influenced subsequent technology, so the count of forward citations acts as a graded signal of importance. Hall, Jaffe, and Trajtenberg show that forward-citation counts correlate with the economic value of the underlying invention and even with the stock-market value of the patent-holding firm, and discriminate far better than treating all patents as equal. Citations remain a noisy proxy - many patents are never commercialized and some citations are routine - but among readily available patent measures they are among the most informative about value.

How do citations reveal knowledge spillovers between places or firms?

Since a patent must cite the prior art it builds on, a citation is an observable footprint of knowledge moving from the cited inventor to the citing one. Jaffe, Trajtenberg, and Henderson exploited this by comparing where citing patents originate against a matched control of similar patents that do not cite the focal patent, and found citing patents disproportionately come from the same region - evidence that knowledge spillovers are geographically localized. The matched-control comparison is essential, because similar inventions cluster geographically anyway; only the excess citation from nearby inventors signals a genuine spillover.

What corrections does patent citation analysis require to be valid?

Two distortions must be addressed. Truncation: recent patents have had little time to accrue forward citations, so they look less influential simply because they are young. And field and time effects: patenting and citation intensity differ markedly across technology areas and decades. Hall, Jaffe, and Trajtenberg provide tools to correct these, typically by scaling a patent's citations relative to others of the same age and field or by including age-by-field fixed effects, so that comparisons reflect real differences rather than measurement artifacts.

Related methods

Sources

  1. Jaffe, A. B., Trajtenberg, M., & Henderson, R. (1993). Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations. The Quarterly Journal of Economics, 108(3), 577-598.

    Open DOI reference
    Copied.
  2. Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER Patent Citation Data File: Lessons, Insights and Methodological Tools. NBER Working Paper 8498.

    Open DOI reference
    Copied.