THEY DON'T INVENT THEM LIKE THEY USED TO: AN EXAMINATION OF ENERGY PATENT CITATIONS OVER TIME

Abstract

This article uses patent citation data to study flows of knowledge across time and across institutions in the field of energy research. Popp [2002, Induced Innovation and Energy Prices. American Economic Review, 92(1), 160–180.] finds that the level of energy-saving research and development (R&D) depends not only on energy prices, but also on the quality of the accumulated knowledge available to inventors. Patent citations are used to represent this quality. This article explores the pattern of citations in these fields more carefully. Evidence for diminishing returns to research inputs, both across time and within a given year is found. To check whether government R&D can help alleviate potential diminishing returns, special attention is paid to citations to government patents. The government patents filed in or after 1981 are more likely to be cited. More importantly, descendants of these government patents are 30% more likely to be cited by subsequent patents. Earlier government research was more applied in nature and is not cited more frequently.

Keywords:

• Patents
• Citations
• Research and development
• Energy
• Diminishing returns

Acknowledgements

The author would like to thank Adam Jaffe, William Nordhaus, Joel Waldfogel, Robert Evenson, Ariel Pakes, and three anonymous referees for helpful comments on earlier versions of this work. Despite their helpful suggestions, however, all mistakes are solely the responsibility of the author. In addition, the assistance of Adam Jaffe, Robert Evenson, Jon Putnam, and Samuel Kortum in locating patent data is gratefully acknowledged. Partial financial support for this article comes from DOE Grant 593A3140217.

Notes

¹ Examples of the first type of study include Griliches and Mairesse (1984), Clark and Griliches (1984), and Scherer (1982, Scherer (1984). See Griliches (1995) for a survey of this work. Examples of the second type of work include Putnam (1996), Lanjouw (1993), Pakes (1986), and Pakes and Schankerman (1984). See Lanjouw et al. (1998) for a review of this literature.

² Note that claims of diminishing returns to research within a field need not be inconsistent with the more general notion that there are increasing returns to research. As new research makes the technologies in a given field obsolete, research efforts should switch to other, more productive areas. Such a general equilibrium analysis is beyond the scope of this article.

³ Examples include Jaffe et al. (1998), Jaffe and Trajtenberg (1996), and Jaffe et al. Trajtenberg (1993). A detailed review of the use of patent citations appears in Jaffe and Trajtenberg (2002).

⁴ Empirical works demonstrating the effect of prices or regulation on innovation include Popp (2002), Newell et al. (1999), Jaffe and Palmer (1997), and Lanjouw and Mody (1996). Theoretical models include Milliman and Prince (1989, 1992) and Jung et al. (1996). Jaffe et al. (2003) provide a review of this literature.

⁵ Outside the realm means that the patent holder cannot file an infringement suit against someone whose invention infringes on qualities of the patented invention that were also included in patents cited by the patent holder.

⁶ Jaffe et al. (1998) examined the relationships between knowledge flows and patent citations. Their research included interviews with scientists, R&D directors, and patent attorneys. They found that, at the level of individual patents, not all citations are indicative of knowledge flows, as other concerns, such as strategically including irrelevant patents to satisfy the patent examiner, affected the citation process. However, on more aggregate levels, such as the patents for an organization or a firm, they found that patent citations are an indicator of knowledge flows, albeit a noisy indicator. Lanjouw and Schankerman (2004) find that forward citations (citations made by future patents to an existing patent) are one of the least noisy indicators of the quality of an existing patent. In this article, the focus is on the quality of knowledge embodied in a patent, rather than a specific knowledge flow. Thus, while a citation may not indicate a direct knowledge flow, the fact that it provides a measure of quality upon which future inventors are building is sufficient.

⁷ Note that since the probability of a patent being cited depends not only on the quality of the patent, but also on the number of patents that follow, it is important to look at probability of citation, rather than raw citation counts.

⁸ Popp (2002) considers the stock of knowledge available to inventors as just one of several factors, such as energy prices and government-sponsored energy R&D, influencing energy research. Since foreign inventors are likely to be influenced by conditions not included in that analysis, that paper focuses on citations made by US assignee patents.

⁹ Several researchers have found that grouping patents by the date of application is a good indicator of R&D activity (Griliches, 1990). Moreover, since, before 2001, information on patents was not made public in the United States until the patent was granted, only successful patent applications are included in the data set.

¹⁰ Note that there are a few cases where a patent appears in more than one technology group. When this occurs, only citations within that group are considered, so that for each patent/citing year pair there are two records (one for each technology group).

¹¹ Whereas negative lags are possible (e.g. a patent granted in 1990, but first filed in 1987, citing a patent granted in 1989), such patents rarely occur in the database. Including the possibility of negative lags greatly expands the set of possibly citing patents set while adding little new information, since few citations occur with negative lags. Thus, such observations are deleted.

¹² Moreover, in each model that follows, hypothesis tests reject the null hypothesis that the Poisson model assuming equal mean and variance is appropriate.

¹³ For this paper, estimation is carried out using the gnbreg command in Stata.

¹⁵ Changes in citing behavior over time must be accounted for because of institutional changes at the patent office that make patents more likely to cite earlier patents than was previously true, even if all other factors are equal. In particular, two changes have played an important role. First, computerization of patent office records has made it easier for both patent examiners and inventors to locate other patents similar to the current invention. Second, increasing legal pressure has made it more important for examiners to be sure that all relevant patents are cited.

¹⁴ I label these patents as children so as to provide a short label for discussion. It need not be the case, however, that child patents are direct descendants of government research, meaning that they need not result from work directly related to the government's research efforts. Citations may result simply because both patents are in similar areas, so that there is an indirect knowledge spillover, but no intentional technology transfer between the government and the private patent.

¹⁶ Interested readers may download a more thorough description of the technologies chosen, as well as additional data on these technology groups, at http://faculty.maxwell.syr.edu/dcpopp/papers/patdata.PDF.

¹⁷ Whereas the cited patents used in the regression only go back to 1975, data extending back to 1900 are used to construct the patent stocks described in Section 2.

¹⁸ In addition to data taken from the NBER data file, I also use additional data on the type of assignee made available to the author. Unlike the other data, this variable is only complete through 1996. I thank Adam Jaffe for making these data available.

¹⁹ Note that in this article, self-citations are included. Self-citations are when the citing and cited patent have the same assignee. Many papers on knowledge flows across space (e.g. across countries or institutions) do not include self-citations. However, in this article, the concern is the usefulness of past research to current inventors. Whether research was done by one firm or by two separate firms should not matter for the question of whether or not there are diminishing returns to research over time. As such, it seems theoretically correct to include self-citations. Nonetheless, the results which follow are essentially the same if self-citations are dropped from the data. Results are available from the author by request.

²⁰ The normalization first divides each continuous variable by its mean, multiplies by 10, and then takes deviations from the mean by subtracting 10, which results in normalized variables that have a mean of 0. The variables are normalized to control for differences in the magnitudes across technologies. The number of potentially citing patents is not normalized because it serves as controls for the number of opportunities for citation that a patent has. Since the dependent variable is a level, rather than a normalized variable, the level of the number of opportunities is what matters.

²¹ The heat exchange technology group includes all of patent class 165, while each of the coal technologies includes only a few subclasses.

²² The table presents results for the individual patent characteristics. Results for the dummy variables are consistent across specifications and are available from the author by request.

²³ Given that the results are not sensitive to how the stock is calculated, all regressions in this section use stocks calculated with the AER decay and diffusion rates. As above, results using alternative specifications are similar.

²⁴ Additional support for the change in the nature of government R&D comes from Popp (2002), which finds that government-sponsored energy R&D substitutes private energy R&D before 1981, but is a complement afterwards.

²⁵ The combined effect equals exp(β_gov+β_interact).

²⁶ Their database differs from the data in this article in two respects. First, it contains both patents assigned directly tothe laboratories and patents assigned to private contractors who collaborated on research at the DOE labs. In this article, such patents are assigned to the private sector. Second, Jaffe and Lerner do not limit their study to a subset of energy technologies, as is the case here. Thus, this article focuses on how the usefulness of R&D within a specific technology changes, whereas in Jaffe and Lerner's work, the types of R&D performed at the laboratories may also be changing. For example, federal R&D spending on renewable energy fell during this period.

Additional information

Notes on contributors

David Popp

Tel.:[phone]; Fax:[phone]; E-mail: [email protected]