What is the Impact of Academic Science on Industrial R&D in Japan? New Evidence From Co-publication and Patent Citation Data

I will present the preliminary results of a research project on the linkage between academic science and industrial innovation in Japan. University scientists and industrial engineers interact in a number of ways, referred to as "sangaku renkei," which involve different kinds of relationships and generate a range of outcomes.

There are four different categories of interaction, one of which my research focuses on. For the first category of interaction, basic scientific discoveries can open new areas of applied research and development. For at least the past 15 years, there has been an increasingly direct and strong connection between university research and the scientific frontier, as well as the development of new drugs and medical devices. This could be referred to as "industrial researchers using science." My main interest is in this kind of sangaku renkei and it is this that I will seek to measure in my research.

This, however, is only one kind of interaction. Even when academic science has little direct impact on industrial technology, firms can find it useful to obtain guidance from university professors. Many engineering professors in the U.S. and Japan engage in this kind of consulting and this may be referred to not as "using science" but "using scientists."

Between the two extremes of using revolutionary new science and using scientists to help firms apply well-known engineering principles, there is an intermediate kind of university-industry interaction we could call "collaborative implementation." Being influenced by a revolutionary scientific breakthrough produced at one university, firm engineers may seek to collaborate with scientists at another university to apply the new science for the development of a particular product. Because the science is new to the firm, the firm seeks to engage the services of academic scientists, who understand it better. But, these academic scientists may not be the same ones who made the revolutionary basic science discoveries.

Finally, there are cases when university scientists actually produce an invention, patent the invention, and then seek to license the technology to a company. This differs from collaborative implementation, in which both university scientists and firms contribute to the research stage.

All of these interactions are important and they have a place in any modern industrial society. Yet, given that these kinds of interactions are fundamentally quite different, it stands to reason that different methodologies will be required to measure these different kinds of interactions.

Much analysis has been based on a statistical analysis of university patents and statistical analysis of university technology licensing. Yet, in a Japanese context, our understanding of sangaku renkei with these two methods is limited. Given that limitation, many scholars in Japan and the U.S. have turned to surveys and case study analysis. I believe this is an extremely important and useful method of analysis but it has its limitations. If we want to quantify sangaku renkei and changes over time, then we need to go beyond this methodology.

Given the lack of large numbers of university patents or licenses, many researchers have attempted to use scientific articles that are co-authored by firm researchers and university researchers as a way of analyzing the impact of university-science on industrial research and development (R&D). The citations to scientific articles that appear in patented documents are an important source of information for this analysis.

The academic promotion system gives university researchers a strong incentive to publish all results of scientific merit. The U.S. patent system creates an incentive for firms to protect their most important technology with patents, and an obligation under U.S. Patent Law to cite in their patents the prior technology on which they have built. These patent citations can take place even when the firm did not find out about the science through an academic article. This combination of the professors' incentives and the firms' requirements generate a paper trail linking a large number of scientific discoveries to individual inventions.

We collected data from all the patents granted in the U.S. to more than 300 large Japanese manufacturing firms from 1983 to 1999 and also acquired patent data on the citations made to academic scientific articles. We have also collected information for some of the firms on the academic publications produced by the firms' own scientists and engineers. Finally, we did a limited amount of interviews with R&D managers in the electronic and pharmaceutical industries.

We were interested in looking at the patent citations in the U.S. patents of Japanese firms because U.S. patent law legally requires inventors to make "appropriate citations to the prior art." This helps us because there are penalties that U.S. inventors will have to pay if they fail to include relevant citations. On the other hand, it hurts us in that because of the legal function that patent citations play in the U.S., patent citations get added to patent documents by parties other than the inventor. So, there is probably going to be a lot of noise in the U.S. patent citation data and we need to be aware of that. Because of Japan's lack of legal obligations for patent citation, we might be concerned that the citations are incomplete. I am not sure that is true, so perhaps it makes sense to use both U.S. patents and Japanese patents to get the most comprehensive picture.

Another reason for looking at the U.S. patent documents of Japanese firms is that most patents taken out by firms in any country turn out to have relatively little value. We want to focus on the impact of science on Japan's more successful and important innovations. Hence, as it is costly to patent outside of Japan, by focusing on the U.S. patents of Japanese firms, we get a subset of patents that have been judged by firm themselves to be more valuable.

Using the data, we wanted to conduct econometric analysis. While time is limited, I would like to share with you some regression results at two levels of aggregation. First, we presented results at the level of citation in patents. The dependent variable is the total number of citations to science made by a particular patent. We regressed this on the characteristics of the patent and the firm that generated it. With a logit regression, we aimed to predict whether a given patent made zero citations to science or some number greater than zero. A second was the use of Poisson and negative binomial regression models, which are designed to predict the number of citations to science made by a particular patent based on the characteristics of the patent and the firm. This allows us to examine changes in citation trends while controlling for changes in the number of patents and in the distribution across firms and industries.

With respect to our findings, the dummy variable associated with the electrical machinery industry is actually negative. Although the electrical machinery industry generates a large number of patent citations to science, this occurs simply because that industry generates more patents. Yet, even controlling for the number of patents generated by pharmaceutical firms in Japan, the estimated tendency of these patents to cite science is particularly high. As is the case with American investors, Japanese patent citations to science are disproportionately concentrated in areas like drugs, medicine and biotechnology. We also see that the number of patent citations to science is positively related to the size of the firm even after controlling for the number of patents. Finally, the more research alliances a Japanese firm has with U.S. universities and firms, the more patent citations it makes to science.

There has been a sharp increase since the mid 1980s in the degree to which Japanese firms cite academic science, beginning at the time when many Japanese firms started to make serious investments in basic R&D. Having reached the technological frontier in the 1980s and under increasing competitive pressure from lower cost Asian producers outside Japan, Japanese firms increasingly felt they had to innovate in order to survive. Interestingly, this tendency to cite academic science was increasing even before policies to promote sangaku renkei were put into place.

In respect to Poisson and negative binomial regressions at the patent level, patents were grouped into categories corresponding to their grant year. Looking at these coefficients, we again see a clear tendency for the propensity of Japanese patents to cite science to go up strongly over this period. As in the early regressions, it seems that this impact is particularly strong in the pharmaceutical industry, and as in the other regressions, there is a positive impact of firm size on knowledge spillovers from universities and also a positive impact on international technology alliances.

On the firm level, if patent citations to science really reflect knowledge spillovers and trace out the application of useful new science to firm R&D, then controlling for the level of R&D spending, firms that make more patent citations to science should systematically deliver a higher level of R&D output. As such, our data does indicate a positive relationship between the intensity of patent citation to science and a firm's inventive productivity.

If firms learn useful lessons from recent academic science, they should make higher quality inventions. The quality of an invention can be measured by the frequency of its citation by subsequently granted patents. Hence, in our analysis, patents are weighted by the number of citations they receive from patents granted over the next four years. When we regress this quality adjusted measure of patent output on various lags of our measure of the intensity with which firms cite science in their patents, we find a positive and statistically significant relationship for almost all lags.

If this learning effect is real, it should increase not only the quality of the firm's patents but also the measured total factor productivity of the firm. We estimated the production function by regressing real sales on a measure of the real capital stock, employment, R&D spending and various lags of our science citation measure. Lags are especially important as it takes time for a patented product to be developed, marketed and sold, and only after significant sales are made is it that the new product will contribute to measured productivity. And again, we find that there is a positive and statistically significant relationship between increases in science citations and firm productivity growth.

The lags between the generation of a technology and the realization of sales from that technology are pretty long for the pharmaceutical industry, longer than this lag structure can measure. So, we think we are probably underestimating the effect of knowledge spillovers in the pharmaceutical industry. But, even given that, we get a positive and statistically significant coefficient on our science citation measure. Now, the coefficients do not look that large, but the output elasticity of patent citation to science implied by these coefficients is 30-60% as large as the firm's R&D spending. Based on our regression coefficients, an increase of one standard deviation above the mean in terms of patent citation to science implies an annual increase in invention quality of up to 32% and factor productivity growth of up to 14% per year.

There is a rapid increase in the extent to which academic science has spillover effects on Japanese R&D. We also find evidence that this is concentrated in pharmaceutical science, and it is especially American science that is most frequently cited in the U.S. patent documents of Japanese firms. Knowledge spillovers traced out by patent citations to science in our data actually impact the research productivity of the citing firms, and the effect is statistically robust and reasonably strong.

University-industry co-authorships to measure sangaku renkei seem to be strong, and almost all the co-authorship is done with Japanese universities. We doubt the extent to which publications perfectly reflect the strategic technological activities of firms. Many co-publications reflect the training of corporate researchers more than they reflect the strategic technological activities of the firm. But they do not capture perfectly the use of recent science. From our study of four firms, we found that their co-authorship patterns are overwhelmingly with other Japanese scientists and engineers, but in respect to their patent citation to science, the authors cited in their patents tend to be disproportionately foreign and American-based authors tend to play a particularly strong role.

Firm R&D managers tend to use foreign universities and foreign research institutes as the sources of the most innovative research applied in their programs. However, consultation with university faculties about well-established engineering principles tends to be with Japanese professors at Japanese universities. Collaborative implementation is conducted with both foreign and Japanese universities, but because the costs of collaboration with local universities are clearly lower and a lot of barriers to collaboration with foreign universities do not exist when interacting with domestic universities, the number of firms interested in collaborating with local universities has increased quite substantially recently.

Let me conclude by offering just a few potential policy implications of this research. First, Japanese firms have become increasingly connected to academic science, which, as in America, this change has been most dramatic in the pharmaceutical industry. Because much of the path-breaking science occurs outside Japan, the strategic use of this science requires firms to forge close connections to foreign science centers. But the good news is that Japanese firms have been successful in doing so. However, formal patenting and formal patent licensing are not the top methods. Patent citations to science are a useful way of measuring this use of most recent science in industrial R&D, but it is not a good way to measure using scientists and only partially reflects collaborative implementation. If we really want to understand this phenomenon, we need to have many different kinds of approaches, have many different kinds of data, and bring them all together. Many Japanese scholars are working on this, and when it is complete, we will have a deeper understanding of this phenomenon.

Q: (question in Japanese)

A: It is true that long before formal sangaku renkei programs had been established in a large number of Japanese universities, Japanese firms were participating in these programs in the U.S. It is possible that some of the early increases in patent citations to science may reflect this early interaction with U.S. universities. It is certainly true that Japanese firms' interest in interacting with local universities along all these four dimensions of sangaku renkei has increased over the last seven or eight years and that government policy has played a role. I do not want to suggest that government changes have been ineffective, unnecessary, or an inappropriate application of government efforts. I do want to point out here that it is not just Japanese science, but global science, that has an impact on the research activities of Japanese firms. I think this tells us just how entrepreneurial and effective Japanese firms are that they are actually taking steps to exploit and use science generated outside of Japan where it is useful to do so.

Q: Scientific paper citations in patents do not necessarily mean that there is actual collaboration. Rather, it is patent dependence on scientific discovery. What do you think about this difference?

A: I agree. When I spoke of the different kinds of sangaku renkei, I meant a new scientific discovery opens up a new area of applied research and development and in a sense provides a new foundation that firms can build on in terms of making useful new technology. This kind of learning does not require collaboration. When the basic discovery takes place, it can influence the R&D of firms all over the world even without direct contact. This is a knowledge spillover in its purest form. It is important to track the extent to which university science has enabled the creation of new kinds of technology. Of course, there are other kinds of sangaku renkei that are important such as collaborative implementation. I would admit, however, that patent citations to science may not perfectly reflect this since new global science breakthroughs can affect R&D all around the world.

Q: (question in Japanese)

A: One can find examples of new high-tech start-ups in the U.S. and Japan that have been important in converting university scientific discovery into a product. Yet, even in the U.S., a great deal of industrial application of new science has been undertaken by established firms since the vast majority of these start-up firms fail. The creation of a system of formal university patenting and patent licensing probably will be disproportionately important for start-up firms. But I do not think we should over-exaggerate the role of these small venture firms. Even the recent round of bankruptcies of venture firms has not resulted in a massive decline of patent citations to science in the U.S. or a massive slowdown in the rate of productivity growth in the U.S. Hence, in the U.S. and Japan, established firms will play a very important role in this process of converting university science to industrial science.

Dr. TamadA: (question in Japanese)

A: It is often important to involve the inventor in the development of his invention, and certainly a strong intellectual property system and formal procedure for university inventions and licensing them to firms will be important in this context. But what we found in the U.S. is that as the number of university patents has increased quite rapidly, the marginal quality of the average patent goes down. We have the same experience with technology licensing agreements. Many university scientists discover that they are not prepared for the switch from the role of university professor to entrepreneur.

Q: Does the noise in U.S. patent citations result in an overcounting of the degree of linkages or does it distort the balance of the linkages? Might we expect more citations that are not germane?

A: Dr. Tamada has undertaken a study of patent citations to scientific articles in Japanese patent documents. What he and his collaborators have found is that like us, science citations are concentrated in biotechnology, and in biotechnology the overwhelming majority of patent citations to science are made to U.S. research institutions. Hence, it appears that U.S. patent data are biasing our results. Yet, Dr. Tamada and his collaborators find in other technological areas, Japanese scientific articles are cited more frequently. This is a discrepancy with our results, which find a strong American influence across all technology categories. We simply have to admit that there is a possibility that when we use U.S. patent data, not only is there noise in the data but there is at least the possibility that the noisy citations will be disproportionately American. On the other hand, when we talk to Japanese R&D managers, they suggest that breakthrough science still comes disproportionately from foreign universities and particularly American universities. Hence, we should use both kinds of patent citation data in order to get more at the truth.

Dr. Tamada: At a certain point, the U.S. Patent Office's legal enforcement was strengthened. That may have affected the increase in the science linkage in the timeline.

A: U.S. inventors became more careful about their citations over the course of the 1990s, and there is an increasing number of total citations in all patent categories. U.S. patent attorneys often add citations to prior patents that the inventor did not know about for legal and procedural reasons, but they generally do not know the scientific literature and thus rely on the inventor to specify the academic references. In looking at patent citations to academic articles, we are getting a measurement that is less contaminated by these biases than patent citations to prior patents.

Q: Do you have specific variables of sangaku renkei to explain the result of higher productivity? Is this R&D productivity or TFP as whole?

A: We used two measures of inventive outputs. The interpretation of the coefficient on the science citation measure is the impact of changes in this variable on our outcome variable, controlling for the level of patenting. That is positively and statistically significant. In the production function case, we looked at total output as measured by sales. In both cases, you might expect that this learning comes through with a lag. Firms start citing science, and the benefit of learning affects the quality of their invention but with a lag. So we experiment with simple different lags but we consistently get a positive and statistically significant result. If there is a significant scientific contribution that this paper makes, it is here. Many of us who have used patent citation data for a while continue to harbor doubts about the extent that it really reflects knowledge spillovers. But if it did not reflect knowledge spillovers at all, then it would not be connected in any systematic way to measures of R&D productivity, suggesting that at least some of these patent citations really do reflect knowledge spillover from university science.

Q: What will be the exact variable for sangaku renkei? What are variables representing collaboration?

A: We are not seeking to measure the impact of collaboration. We are trying to measure the pure knowledge spillovers from that kind of science.

Q: Is it possible for you to do so, including the sangaku renkei and all the variables?

A: Yes, in principle, if we had firm level data.

Q: Is this only for biotech? Could it be applied to other industries?

A: The impact is strongest in biotech and pharmaceuticals. But R&D managers and electronics industries have suggested that there are parts of their R&D portfolios where university research really does matter. Yet, these results presented are based on a cross section of which pharmaceutical firms are only a minority. So there seems that there is a statistically significant relationship outside biotech and pharmaceutics.