Patent data have long enticed varied audiences. At the national level, patent filings and grants by nationality of inventor(s) have long been used as indicators of the technological competitiveness of a country. Economists have attempted to enhance the system by studying optimal patent duration, scope and the interaction of the two (early work by Nordhaus, 1969; Klemperer, 1990). Others have attempted to assess aspects of equity or opportunity from patent files, such as the proportions of individual or small firms filing or domestic versus foreign applicants. Moreover, renewal data have been used to value these patents (Lanjouw et al., 1998).

Early in the exploration of the data, it was apparent that, given the range of inventive significance represented in the patent files, treating a patent as a homogeneous unit was meaningless. Certainly the patents issued for major inventions, such as the light bulb, nylon and transistor, dwarf in importance the technological advances and practicality of the average patent, a point reinforced by findings that only approximately 15% of US patents are ever commercialized, and patent values are highly skewed. Of course, non-commercialized patents can none the less have significance, if only to block the 'inventive space' around a key patent from a competitor. Studies of patent value typically show highly skewed distributions; for example, the 1970 cohort of patent applications in France and the UK showed an average value of only DM11,250, but the top 1% was valued at DM112,500 (Schankerman and Pakes, 1986).

As a means of identifying 'weights' to reflect the relative significance of patents, economists have used counts of both claims and citations. Patent claim counts have the clear limitation that the claims are structured to be non-interdependent, meaning that some very broad claims may be added although they add little practicality to the scope. For example, many animal model patents (e.g. 6727405) in the first claim read to 'a non-human mammal', even though the applications are to rodents, or mice. Because claims are not interdependent, broad claims can be (and sometimes are) halved off with little significance.

The analysis of patent citations has been more fruitful. Hall et al. (2000) found that the size of a firm's patent portfolio weighted by patent citations could explain the firm value better than patent numbers could. The analysis also considered other patents referenced by the patent (out-citations) and subsequent patents that reference it (in-citations). Alternatively, Lanjouw and Schankerman (1997, 1999) used citation numbers as one component of 'patent quality' that could explain which of the patents were renewed to full term and which were litigated. Yet, there are limitations to the use of citations. As Hall et al. (2000, p. 4) note, many citations may be added by an attorney or examiner, and hence do not correspond to the inventors' concept of relevant prior art.

At a broader level, a number of charges have been made in recent years regarding problems at the US Patent and Trademark Office (PTO). Barton (2000) notes that only one-fourth of the challenged patents remain unchanged on review, and approximately 10% cancelled altogether (Graham et al., 2003). The remedy proposed by Barton (2000, p. 1934) is to 'issue better decisions in the first place'. Contributing to the debate of the quality of PTO examination, Quillen and Webster (2001) estimate that up to 9 7% of the applications eventually matures into grants. Yet, in an examination of recent grants, Kortum and Lerner (1999) conclude that the doubling of applications and grants since the 1980s is not traceable to changes in the PTO. Rather, 'a more plausible explanation for the rise in patenting is that either technological opportunities or the process of doing research has improved' (see also Lesser and Lybbert, 2004).

One of the limitations with those analyses is the limited number of patents frequently mentioned. The issue is not 'does the PTO make mistakes?', to which the answer is clearly 'yes'. Many examples of grants, like Animal Ear Protection (to keep a dog's ears out of its food, 4,233,942 - additional examples available at http://colitz. com/site/wacky.htm), which are at best dubious even on peripheral consideration, are evidence of sporadic errors whose practical effect is negligible. Of greater concern are the systemic problems that can be very expensive. The analysis of patent litigation in itself does not reveal general trends, for litigated patents do not represent a random sample of all patents. Rather, patents are typically not litigated unless they are either valuable or represent a new technological area for which a court appeal is required before the true scope of the patents in that class can be determined.

Understanding the system-wide effects of the PTO actions requires an examination of all recent grants, either in total or by class. Here, the analytical power of network analysis to patent files is applied.


The objective of this chapter is the initial application of network analysis to patent files. It provides some illustrative examples of the approach to model patents as a networked structure. The overall goal is to analyse the statistical properties of the network of patents (considering different link structures corresponding to different relationships between patents) to uncover hidden structures, which may provide us with interesting insights in terms of patenting behaviours and practices.

Because network analysis delineates relationships, a full assessment must await a time when: (i) those relationships are better understood; and (ii) a methodology evolves for analysing what are often visual images. Hence, what is presented here are preliminary results intended more as an indication of the potential of network analysis than conclusions with any clear validity in their own right. Although we are still in the early phase of our analysis, several interesting structural properties can be identified.

The analysis is divided into two levels: all patents, and a small number of patents of known technical and commercial relevance - those by Cohen and Boyer. From the all-patent analysis it is possible to make initial judgements about the appropriateness of the patent classification system, as well as getting a broad understanding of the spillovers among the patent classes. At the individual-patent level, it is possible to identify confidently the most important among several patents, as well as making a tentative operational definition of a 'pioneering' patent. Many of these developments are facilitated by using multiple levels of citations (direct and indirect) rather than restricting the analysis to direct citations.

0 0

Post a comment