Optical Science as a General Purpose Technology: A Patent Analysis

Transcription

1 Eindhoven, Optical Science as a General Purpose Technology: A Patent Analysis by R.P.G.R (Roger) Füchs identity number in partial fulfilment of the requirements for the degree of Master of Science in Innovation Sciences Supervisors: Dr. Önder Nomaler Eindhoven University of Technology, Dept. of Innovation Sciences Dr. Bert Sadowski Eindhoven University of Technology, Dept. of Innovation Sciences

2 Table of Content I. List of Figures... 2 II. Summary Introduction GPT s: a phenomenon or an ideal? Empirical Literature on GPT s & Patents Patents Empirical Literature Optical Science Literature Methodology Pervasiveness Innovational Complementarities Scope for Improvement Results Pervasiveness Unrelated variety in the co-occurences The number of IPC codes that each IPC code shows up toghether with Innovational Complementarities Scope for improvement Growth in patent applications Growth in patent citations Conclusion Discussion Limitations Indications for Future Research References Appendix 1: Comparison of G2B, G2C and G2F on the 3 and 8 digit level Appendix 2: Comparison of citations of G2B, G2C and G2F on the 4 and 3 digit level Appendix 3: Ranking of IPC codes based upon entropy... 6 Appendix 4: Distribution of the entropy of the top 5 IPC codes Appendix 5: Summary of Empirical Literature Appendix 6: Table of IPC-codes used by Feldman & Lendel

3 I. List of Figures Figure 1 Summary of Theoretical Literature Figure 2 Tree structure of IPC Codes Figure 4 Difference in relatedness of IPC codes Figure 6 A world of 13 patents and 1 IPC Codes Figure 7 Counting the Co-occurrences and calculating the shares Figure 8 Calculating the entropy Figure 9 Calculating the unrelated variety on the 4-digit level Figure 1 Example where self-occurrence is not allowed Figure 11 Calculating the unrelated variety without self-occurrence on the 4-digit level Figure 12 G2B, G2C and G2F description Figure 13 Unrelated variety of G2B and the Averages of the Rest on the 8 digit level Figure 14 Unrelated variety of G2B and the Averages of the Rest on the 4-digit level Figure 15 Unrelated variety of G2B and the Averages on the 3-digit level Figure 16 Unrelated variety of G2C and the Averages of the Rest on the 8-digit level Figure 17 Unrelated variety of G2C and the Averages of the Rest on the 4-digit level Figure 18 Unrelated variety of G2C and the Averages of the Rest on the 3-digit level Figure 19 Unrelated variety of G2F and the Averages of the Rest on the 8-digit level Figure 2 Unrelated variety of G2F and the Averages of the Rest on the 4-digit level Figure 21 Unrelated variety of G2F and the Averages of the Rest on the 3-digit level Figure 22 Comparison 4 Digit Level Figure 23 Estimation of the number of IPC codes that G2 show up together with Figure 24 Estimation of the number of IPC codes that G2B,C & F show up together with Figure 25 Unrelated Variety of the Citations on the 8-digit level Figure 26 Unrelated Variety of the Citations on the 4-digit level Figure 27 Unrelated Variety of the Citations on the 3-digit level Figure 28 Unrelated Variety of the Citations on the 8-digit level... 4 Figure 29 Unrelated Variety of the Citations on the 4-digit level... 4 Figure 3 Unrelated Variety of the Citations on the 3-digit level Figure 31 Unrelated Variety of the Citations of the 8-digit level Figure 32 Unrelated Variety of the Citations to 4-Digit Figure 33 Unrelated Variety of the Citations of the 3-digit level Figure 34 Citations Comparison 8 Digit Figure 35 Total Count of G2 in patent applications Figure 36 Fractional Count of G2 in patent applications Figure 37 Total Count of G2B, G2C & G2F Figure 38 Fractional Count of G2B, G2C & G2F Figure 39 Amount of Patents that received at least one citation Figure 4 Amount of Patents Citations over Time Figure 41 Amount of Patents that received at least one citation Figure 42 Amount of Patent Citations over Time Figure 43 Average Number of Citations Received by Patents that have received at least one citation48 Figure 44 Average Number of Citations Received Figure 45 Average Number of Citations Received by Patents that have received at least one citation49 Figure 46 Average number of Citations Received Figure 47 Summary of Results Figure 48 Comparison 3 Digit Level Figure 49 Comparison 8 Digit Level

4 Figure 5 Citations Comparison 4 Digit Figure 51 Citations Comparison 3 Digit Figure 52 Ranking of IPC codes based upon entropy. Top 5 out of digit IPC codes including A&B & 385 without A&B... 6 Figure 53 Distribution of 4-digit, exlcuding itself, without classes A&B entropy of the top 5 IPC codes Figure 54 Distribution of 4-digit, exlcuding itself, with classes A&B entropy of the top 5 IPC codes. 61 Figure 55 Distribution of 8-digit, exlcuding itself, without classes A&B entropy of the top 5 IPC codes Figure 56 Distribution of 8-digit, exlcuding itself, with classes A&B entropy of the top 5 IPC codes. 62 Figure 57 Summary of Empirical Literature Figure 58 Table of IPC-codes and description

5 II. Summary The central theme in this research is General Purpose Technology (GPT). There is no consensus in the literature discussed in this research on how a GPT is defined. Breshnahan and Trajtenberg (1995) identify three key characteristics of a GPT which are used in this research in order to define a GPT. The first characteristics is Pervasiveness, the second characteristic is Innovational Complementarities and the last characteristic is Scope for Improvement. Pervasiveness means that a GPT is visible throughout the economy and is used in many different sectors. Innovational complementarities relates to other technologies depending on the GPT. If there is an innovation in the technology of the GPT, the application sectors where the GPT is used will benefit from this innovation as well. The last characteristic is scope for improvement and means that the GPT improves over time. In order to measure whether a technology can be considered a GPT, first of all a definition of what a technology is, is needed. Next a definition and metrics are needed in order to test whether or not a technology can be considered as a GPT. This research defines a technology by means of patent IPC codes. In order to measure whether or not a technology can be considered as a GPT, a new concept of GPT ness is introduced. GPT ness does not classify a technology as a GPT binary. GPT ness allows for comparisons to be made between technologies in order to investigate whether or not a technology is more of a GPT than another technology. This research has analyzed the domain of optical science as a possible candidate for being a GPT. Optical science relates to technologies that manipulate light and how light interacts with matter. Therefore the research question of this research is: Why can optical science be classified as a general purpose technology In order to analyze whether or not optical science can be considered as a GPT, this research carried out a patent analysis. Patents contain valuable information about technologies are this information was used in this research in order to test to what extent optical science can be considered as a GPT. Patents are classified by means of IPC codes which have a tree like structure. In this research the domain of optical science is at first defined by all the technologies belonging to the IPC class of G2. The unit of analysis of this research is IPC codes and not individual patents. This means it is possible to identify several subgroups of the G2 class that do meet the characteristics of a GPT whereas other subclasses might not meet these criteria. This research provides a new type of methodology in order to measure the characteristics of pervasiveness and innovational complementarities. This new type of methodology addresses the problem of the old methodology that was used in the past in order to determine these characteristics. This new type of methodology is the main contribution of this research to the domain of general purpose technologies. In order to measure scope for improvement, this research makes use of existing methodologies. 4

6 Pervasiveness is measured by means of the variety of the co-occurrences of IPC codes. This is done by making a distinction between related variety which represents variety within IPC code groups and unrelated variety which represents variety between IPC code groups. The unrelated variety of the cooccurrences of the IPC codes is used in order to measure pervasiveness. Pervasiveness is calculated for the IPC groups of G2B, G2C and G2F and is compared to average pervasiveness of all the other IPC codes. Innovational complementarities is measured by a similar methodology as the methodology of pervasiveness. Instead of looking at the variety in the co-occurrences of the IPC codes, this research now looks at the variety in the patent citations. The unrelated variety in the citations of G2B, G2C and G2F is again compared to the average unrelated variety of all the other IPC codes in order to benchmark the outcomes of the G2 class. Scope for improvement is measured by the existing metrics namely the growth in patent applications over time and the growth of patent citations over time. The results show that the subclass G2B is the most promising candidate of being classified as a GPT. G2C and G2F do not fulfil all of the three characteristics. G2B scores until 197 on all three characteristics above average. After 197 the unrelated variety in the citations tend to get more closer to the average unrelated variety of all the other IPC codes. The pervasiveness of G2B does not drop in more recent years as it stays largely above average. This research has concluded that defining optical science by means of the IPC class G2 is too broad as the results indicate there is a large difference in variety between the subgroups of G2B, G2C and G2F. Therefore the definition was adjusted to G2B as being the technologies that relate to optical science. The conclusion of this result is that optical science is a GPT of the past and not an emerging GPT as the evidence of being a GPT is stronger in the past than in more recent years. The results are not in line with what could be expected with an emerging GPT as one would expect an emergent GPT to score largely above average on all three characteristics in more recent years which is not the case. Finally is must be noted that the average results of all the other IPC codes was used as a benchmark in order to test whether or not optical science can be considered as a GPT. Therefore all the results and conclusions should be interpreted as relative results. for example: G2B is considered more of a GPT than the average 4-digit IPC code instead of G2B is a GPT. This is the essence of GPT ness. 5

7 1. Introduction Throughout history, many technological advancements have been made. Some of these technological breakthroughs lead to radical changes in the economy and opened up new possibilities. Bresnahan and Trajtenberg (1995) introduced the notion of general purpose technology (GPT) in context of economic growth. GPT s are technologies that can be used in many sectors and can increase the productivity levels in these sectors drastically by opening up new opportunities. Thus a GPT can be considered as a drastic innovation that is widely used in a broad domain. Examples include the steam engine, the electro motor and semiconductors. GPT s are highly pervasive and often result in innovational complementarities. This means that an innovation in the GPT s sector can result in to greater returns of each application where the GPT is used. This causes incentives to innovate which then leads to more innovation in the GPT s sector (Bresnahan & Trajtenberg, 1995). A GPT is characterized by pervasive use and by technological dynamism. Since the article of Bresnahan and Trajtenberg (1995) other research has also been carried out with respect to the notion of GPT. This notion has been further developed. In order to measure GPT s, one first has to define what a technology is and second, how to test whether or not a technology can be considered as a GPT. The unit of analysis differs across the literature. This means that there is not much consensus on what exactly a GPT is. There are many different classifications and metrics in order to measure whether something is a GPT or not. Furthermore the theoretical literature shows there is also not much consensus on defining a technology. This results into different units of analysis in the empirical literature. There is also not much quantitative work done on GPT s (Bashir & Sadowski, 214). Therefore there is a need for a standard methodology or concept in order to compare and measure technologies. Instead of a classifying technologies as a GPT or not this research introduces the concept of GPT ness which measures to what extent a certain technology is more of a GPT than the average. This research will look at optical science as a GPT. Optical science can be considered an enabling technology that is used in a wide range of sectors, for example telecommunications, medical instruments, semiconductors etc. (Feldman & Lendel, 21). This research consists of two parts: a theoretical and an empirical part. In the theoretical part of this research GPT s in general based upon literature by Bresnahan and Trajtenberg (1995) and other literature that has referred to these two authors will be discussed. First of all the notion of GPT will be examined and how it has developed over time. This will set the stage in order to introduce the new concept of GPT ness. Next literature that combines empirics with this theory will be analyzed. The empirical literature can be used in order to determine what metrics can be used in order to measure GPT ness of a technology. The limitations of the current metrics will be discussed which provide an argument for introducing a new type of methodology in order to measure GPT ness. In the empirical part of this research these metrics will be tested using patent data related to optical science. This research will measure variety in optical science patents by means of cooccurrence of IPC codes. Furthermore co-occurrence across patent citation links will be analyzed. The variety in these co-occurrences will be measured by introducing a new type of methodology based 6

8 upon unrelated and related variety by Frenken et al. (27). Using this type of methodology a problem of analyzing the pervasiveness of GPT s by means of a patent analysis will be addressed. This methodology is the main contribution of this research to field of GPT s. The variety in the cooccurrences of IPC codes and the variety in the citations of optical science patent applications will give an overview of how optical science patents relates to other patents. Based upon the outcome of this empirical analysis and the definition of GPT it is possible to answer the research question: (Why) can optical science be classified as a general purpose technology? This report is structured as follows: First of all the literature regarding GPT s will be analyzed, both theoretical and empirical in section 2. Also literature related to optical science and how to define it by means of patents will be discussed. Section 3 contains the methodology of the patent analysis. Section 4 covers the results. Finally conclusions will be drawn and the research question will be answered in section 5. This reports end with a discussion and recommendations for further research in chapter 6. 7

9 2. GPT s: a phenomenon or an ideal? The aim of this first section is to analyze the notion of General Purpose Technology (GPT). What exactly is a GPT, what defines it? In this section both theoretical as well as empirical literature regarding GPT s will be discussed. When reading this first section it is important to lay emphasis on how the different authors define a GPT, what their unit of analysis is and what metrics they use in order to measure whether something can be classified as a GPT or not. From this first section it will be clear that there is no single common understanding of what exactly a GPT is and that the concept has changed over time. Also there is no consensus on a workable definition of a technology that empirical studies can use in order to define a technology. These differences and the lack of a consensus on a workable definition of a technology will serve as an argument for introducing a new methodology in order to measure and define GPT s which is described in section Theoretical Literature on GPT s Bresnahan and Trajtenberg (1995) published their work: General purpose technologies: Engines of growth. It is important to note that growth is calculated on the aggregate level and thus Bresnahan and Trajtenberg are looking at GPT s at the macro level. According to Bresnahan and Trajtenberg (1995) a GPT is defined by three characteristics (Bresnahan & Trajtenberg, 1995). The first characteristic is pervasiveness and means that the GPT is visible and being used in many different sectors. The second characteristic is called scope for improvement which means that a GPT is able to improve over time. The third characteristic is that GPT s lead to innovational complementarities (Bresnahan & Trajtenberg, 1995). This means that an innovation in the GPT s sector can result in to greater returns of each application where the GPT is used. This results in incentives to innovate which then leads to more innovation in the GPT s sector. These three characteristics are the criteria in order to classify something as a GPT. The unit of analysis are sectors. The authors make a distinction between the GPT s sector and the application sectors. The GPT s sector contains the technology itself. The application sectors are sectors where the GPT is being used. For example in the GPT sector of semiconductors, there exist multiple application sectors such as hearing aids, radio & TV, computers, CT scanners and cars (Bresnahan & Trajtenberg, 1995). Development in the GPT s sector can lead to innovation in the applications sectors. It can also create new application sectors. Innovation in these application sectors is often very diverse. The relation between the GPT s sector and the application sectors leads to two externalities, a vertical externality (between GPT s sector & application sectors) and a horizontal externality (within GPT s sector and application sectors). The vertical externality relates to the difficulties of reaping the rewards of innovation in the GPT s sector. The payoffs of the GPT s sector and the applications sector are linked and therefore if either one innovates, the other will have the benefits as well. Regulations and institutional conditions often prevent the GPT inventor from full appropriation of the rewards. Therefore the GPT is undersupplied compared to the social optimum. 8

10 The horizontal externality relates to the amount of application sectors and their coordination. The individual application sectors do not collaborate jointly in order to improve the GPT as a whole. They rather seek for innovation within their own field. The GPT s sector is likely to innovate more if there is more demand from the application sectors. This means that the more application sectors there are, the more incentive there is for the GPT sector to innovate. Furthermore, coordination by the different application sectors results in to encouragement for improvement of the GPT. A better GPT benefits them all, and therefore it is more efficient to collaborate. However Bresnahan and Trajtenberg show by means of their model that in equilibrium state there are less application sectors than the amount that would be optimal for innovation in the GPT (Bresnahan & Trajtenberg, 1995). Breshnahan and Trajtenberg assume a monopolist exists who owns the GPT. This monopolist decides how fast the GPT will improve and the users of the GPT decide how fast a specific application of the GPT improves. Following up on the article by Breshnahan and Trajtenberg (1995) on economic growth, Helpman and Trajtenberg (1998) extend the notion of GPT. They use the three characteristics of Breshnahan and Trajtenberg as a baseline. Helpman and Trajtenberg (1998) introduce a mathematical model that aims to explain historically observed fluctuations of economic growth and include the concept of GPT in this model. They refer to GPT s as the encouragement of the development of so called compatible components. For example if a computer is seen as GPT, software can be seen as a compatible component. They assume that there can only exist one GPT at the same time. An R&D sector is responsible for supplying the GPT s sector with compatible components to improve. They assume that each GPT arrives at a predetermined time interval and this time interval is the same for each GPT. This time interval is called a cycle and this cycle exists out of two or three phases. Whether there are two or three phases depends on whether or not R&D was still going on while a new GPT is being introduced. When a new GPT arrives, it is immediately recognized as a GPT and all R&D in the previous GPT stops and is diverted to the new GPT. The first phase is the time where the new GPT is still being developed while the old GPT is still responsible for manufacturing. During this phase the innovator does not make profit but innovates anyway because the innovator expects that the rewards will be there in the future. During the second phase, the old GPT is completely replaced by the new GPT and both R&D and manufacturing are carried out by the new GPT. The authors also define phase three where no R&D exists and all resources are spend on manufacturing. Phase three only exist if R&D of the old GPT has stopped before the new GPT was introduced. When a new GPT is being introduced, labor is redistributed from output manufacturing of the old GPT into R&D of the new GPT. This causes, what Trajtenberg & Helpman (1998) call, an output slowdown (Helpman & Trajtenberg, 1998). Aghion & Howitt (1998) acknowledge this output slowdown of Trajtenberg & Helpman (1998) but do not agree with the timing of when these output slowdowns occur. The reason for this is that a new 9

11 GPT needs a significant amount of time to develop. Therefore it is unlikely that immediately upon arrival of a new GPT, the old GPT is replaced and the output slowdown occurs. Instead they propose that there is a period where the old GPT is still in use while the new GPT is being developed. Aghion and Howitt (1998) identify three reasons for the existence of such a period. The first two reasons are measurement errors and complementarities. The third reason is social learning. They argue that a company cannot instantly start with creating components for the new GPT. Instead these companies first have to produce an intermediate good. Companies create a specialized workforce that engage in the creating of a so called template. By trial and error, probability of acquiring a successful template increases and as more and more companies acquire their template for the GPT, the GPT can gain momentum and this is the moment when the shift from the old GPT to the new GPT takes place (Aghion & Howitt, 1998). Using these three theories about GPT s, Lipsey et al. (1998) identify several GPT s such as ICT, several power delivery systems (waterwheel, steam, electricity) and means of transportation (railways, motor vehicles). These GPT s are all in the past and it s clear to see now what their impact was on daily life and the economy. However in order to identify GPT s that are currently being developed or do not exist yet, a clear definition of GPT s is needed. Using this definition one can evaluates whether or not a certain technology matches that definition. In order to come up with a such a definition it is important to make sure that it includes at least all the GPT s of the past. In order to do this Lipsey et al. (1998) try to find common characteristics through all of these GPT s (Lipsey et al., 1998). Lipsey et al. (1998) argue that the definition of a GPT by Breshnahan & Trajtenberg (1995) as well as the model by Helpman and Trajtenberg (1998) are too generic and broad. Therefore they do not look at sectors as unit of analysis, instead they look at technologies as unit of analysis. Their procedure in order to define a technology is to search important technologies according to their observed economic effects They searched for technologies which are widely used and caused changes that pervaded the entire economy. (Lipsey et al., 1998, p. 21). Lipsey et al. (1998) also do not agree with the criteria proposed by Breshnahan & Trajtenberg (1995). They argue that by using the three criteria by Breshnahan and Trajtenberg (1995) as the definition of a GPT too many technologies will be classified as a GPT (Lipsey et al., 1998). Instead Lipsey et al. (1998) identify four characteristics of which all four have to be met in order for a technology to qualify as a GPT. The first characteristic is again scope for improvement. This means that a GPT has gone through a visible evolutionary track and thus has developed over time. Over time the GPT is being developed, costs drop and the range of use widens. This means when a technology is being introduced it is far from perfect and still has to go through an evolutionary development path in order to become a GPT. Scope for improvement is thus a necessary condition in order for a technology to qualify as a GPT (Lipsey et al., 1998). 1

12 The second characteristic that Lipsey et al. (1998) identify is called wide variety of use (Lipsey et al., 1998). This means for a technology to qualify as a GPT, it has to be used in many different applications. As the GPT evolves, more and more applications of the GPT become apparent. The further the GPT is in its development stage, the wider its variety in use is. In order to have a wide variety of use a GPT must have the capability to adapt to where it is being used. A GPT must be modifiable so it can be used in many different applications. The third characteristic is wide range of use (Lipseyet al., 1998). This means that a GPT must be visible across the entire economy in many different sectors. It is important to stress the difference between wide range of use and wide variety of use. In order to stress this difference Lipsey et al. (1998) use the example of the lightbulb. The lightbulb is visible across the economy in many different sectors and thus meets the requirement of wide range of use. However the lightbulb has only one type of application: producing light. Therefore it does not meet the requirement of wide variety of use. In order for a technology to qualify as a GPT it must fulfill both (in fact all four) of these characteristics. The last characteristic of a GPT is that a GPT has to have complementarities (Lipsey et al., 1998). Two types of complementarities are identified: Hicksian complementarities and technological complementarities. A technology A has Hicksian complementarities if the demand of another technology B is affected by the price of A. This means that a GPT is highly competitive with other technologies and that changes in price & demand of the GPT affect other technologies as well. The main difference between Hicksian complementarities and technological complementarities is causal direction of the complementarities. Hicksian complementarities are complementarities that are caused by improvements in a certain technology so that other related technologies will benefit from this improvement as well. Technological complementarities are complementarities that occur when the benefits of a certain new technology are so great, that the existing capital and existing technologies will be altered in order to benefit from this new type of technology. The effects of technological complementarities cannot be modeled as changes in price as is opposed to Hicksian complementarities. Lipsey et al. (1998) use the introduction of electricity as an example: Power plants were only redesigned after the introduction of the unit drive. The unit drive is not compatible with the steam era. Even if the price of steam was zero, the plants would not be remodeled until after the introduction of the unit drive. In other words the introduction of electricity did not depend on a change in price but on changing existing capital and related technologies. In order for a technology to qualify as a GPT, it must have both Hicksian complementarities as well as technological complementarities (Lipsey et al., 1998). 11

13 Using these four (or five if you count the complementarities separately) characteristics, Lipsey et al. (1998) come up with the following definition for GPT s: A GPT is a technology that initially has much scope for improvement and eventually comes to be widely used, to have many uses and to have many Hicksian and technological complementarities. (Lipsey et al., 1998, p. 43) It is important to note that all four characteristics have to be met in order to qualify as GPT. Lipsey, et al. (1998) give numerous examples of technologies that meet some of these characteristics but not all and therefore cannot be considered as a GPT. Examples include: the light bulb, television and shipping containers. From the literature above it is important to notice that there is no common interpretation of what exactly a GPT is. Breshnahan and Trajtenberg (1995) look at sectors as unit of analysis and came up with three characteristics to define a GPT. Helpman and Trajtenberg (1998) use the concept of GPT in a mathematical model in order to explain fluctuations in economic growth, this model is then further adjusted by Aghion & Howitt (1998). Lipsey et al. (1998) do not look at sectors as a unit of analysis but instead they look at technologies. They also adjust the criteria of a GPT to four characteristics. These differences are depicted in the figure below. So how can GPT s be measured and by what metrics? In the next paragraph 2.2, empirical literature will be discussed where different kinds of GPT s are analyzed by different metrics and methodologies. This empirical literature provides answers to the questions raised above. While reading paragraph 2.2 it will be clear that there are differences in how to measure GPT s and therefore a new concept is introduced at the end of paragraph 2.2 in order to overcome this problem. Theoretical Literature Author Unit of Analysis Criteria for being a GPT Breshnahan & Trajtenberg (1995) Sectors Scope for Improvement Pervasiveness Innovational Complementarities Helpman & Trajtenberg (1998) and Aghion & Howitt (1998) Technologies or "Compatible Components" Mathematical Model Lipsey, Bekar & Carlaw (1998) Figure 1 Summary of Theoretical Literature Technologies that are widely used and caused changes that pervaded the entire economy Scope for Improvement Wide Range of Use Wide Variety of Use Hicksian & Technological Complementarities 12

14 2.2. Empirical Literature on GPT s & Patents So far, no consensus on a workable definition for a technology has been reached which resulted in to different units of analysis. Therefore empirical literature regarding GPT s will now be discussed in order to examine the different units of analysis that the authors used. Some of the empirical literature on GPT s that is depicted below makes use of patents in order to define a technology. Therefore, before looking at the empirical literature on GPT s, brief information about patents is discussed in order to make the next piece of literature more comprehensible Patents According to the World Intellectual Property Organization (WIPO) patents can be described as follows: A patent is an exclusive right granted for an invention. Generally speaking, a patent provides the patent owner with the right to decide how - or whether - the invention can be used by others. In exchange for this right, the patent owner makes technical information about the invention publicly available in the published patent document. (WIPO, 216) Patents are assigned an International Patent Classification (IPC) code based upon the different areas of technologies (WIPO, 216). One patent can have multiple IPC codes assigned to it. IPC codes always start with a letter (A to H) assigning the patent to the broadest category possible. For example IPC codes starting with a G belong the field of Physics. Within these categories, further categories are defined allowing for more detail within a category. This means that IPC codes have a tree like structure. Below an illustrative example is given of the tree like structure of the IPC classification: Figure 2 Tree structure of IPC Codes 13

15 There are also other classification systems in order to classify patents. For example the United States Patent Classification (USPC) is also used by some of the authors in the literature below (USPTO, 216). Patents are also able to cite other patents. Patent citations provide information about how certain technologies are linked with each other. The information published in patent documents is used by some of the authors below in order to identify GPT s. Below several examples of empirical literature regarding GPT s are discussed in chronological order Empirical Literature Rosenberg & Trajtenberg (21) identify the Corliss steam engine in the late nineteenth century as a GPT. What is worth noting is that the unit of analysis is thus a certain type of steam engine instead of a technology or even a whole sector. The data on the Corliss steam engine that was used originated from a denied petition that contained a list of buyers of Corliss steam engines. In their research, among many other findings, they conclude that the Corliss steam engine can be considered as a GPT. They try to identify innovation complementarities by relating the improvements in Corliss steam engine to the role of water (wheel) energy. Next they look at the impact of the Corliss steam engine on large scale production. Lastly they look at the importance of the Corliss steam engine on rolling mills. The data used on the Corliss steam engine is based upon a denied petition by George Corliss that detailed a list of 257 buyers of Corliss engines. They tried to classify these buyers into different sectors and based upon the results they concluded that the Corliss steam engine did indeed fulfilled the requirement of having a wide range of applications across sectors. Furthermore they compared the role of waterpower to the role of steam power in general and to the role of Corliss steam engine specifically. They concluded that it was not steam power in general but it was a new type of steam engines, starting with the Corliss, that caused a shift in the usage of waterpower (Rosenberg & Trajtenberg, 21). The research by Rosenberg and Trajtenberg (21) is one of the first that uses empirical data when looking at GPT s. The unit of analysis is a specific type of steam engine, the criteria for being a GPT are scope for improvement, pervasiveness & innovational complementarities. The metrics that used are both historical as well as empirical. Moser & Nicholas (24) measure to what extent electricity can be considered as a GPT. Therefore their unit of analysis is a technology. Moser & Nicholas measure GPT s on three new levels in order to test to what extent electricity can be considered as a GPT. These levels are originality, longevity and generality (Moser & Nicholas, 24). Their metrics are based upon patent data, for example patent citations.. They use U.S. patents dating from 192 and onwards. First of all they divide their patents in to four different industry groups in order to measure the general purpose characteristics for different industries in order to compare electricity to for example the chemical industry. Next they 14

16 define the characteristics originality, longevity and generality. Originality is measured by means of patent citations. Longevity is measured by time lags between patents and generality is in terms of the variety in the USPTO patent classes found in the citing patents. One way of doing this would be to count the unique number of USPTO classes found in all citing patents. However, some patent classes occur more often in the citing patents than the others. Accordingly the authors used an alternative metric, the Herfindahl-Hirschman index (HHI) in order to define variety. HHI, which is essentially the squared sum of the shares of occurrence, clearly takes the relative frequency of occurrence into consideration.. They calculate for each of the four industry groups the originality, the longevity and the generality. Comparing the outcomes for each characteristic for each different industry sector they conclude that electricity based upon the definitions used cannot be considered as a GPT (Moser & Nicholas, 24). The research by Moser & Nicholas introduces three new criteria in order to empirically measure if electricity can be considered as a GPT. The unit of analysis is a broadly defined technology: electricity. This technology is measured by U.S. patents granted in 192, 1922, 1924, 1926 and 1928 that were cited 34 times by patents between 1976 and 22. (Moser & Nicholas, 24) Only patents assigned to publicly traded firms were included. Comparing this approach to the approach of the Corliss engine, it is clear that there is a difference in unit of analysis as well as in the criteria in order to measure a whether or not a technology can be considered as a GPT. Youtie et al. (27) investigated to what extent nanotechnology can be considered as a GPT. The criteria for being a GPT are scope for improvement, pervasiveness & innovational complementarities. The level of analysis is a technology. This technology is defined by nanotechnology patents. Nanotechnology patents were obtained using the USPTO classification system as well as the keyword system built by the CNS-ASU team. (Youtie et al., 27). Only pervasiveness is tested empirically. For the other two criteria they indicate possible measurements that can be obtained from patent data. In order to measure the characteristic of pervasiveness, they make use of the HHI Index of the patent classes assigned to forward citations of the focal patent. This is similar to the research by Moser & Nicholas (24). They mention that nanotechnology is a new type of technology and therefore the patent data set will not be as large due to the developing phase that nanotechnology is in. By means of benchmarking nanotechnology to ICT (presumed a GPT) and Drugs (presumed not a GPT) they concluded that nanotechnology does show the characteristics of being a GPT (Youtie et al., 27). Shea et al. (211) follow up on the research of Youtie et al. (27). The preliminarily evidence that Youtie et al. (27) found for nanotechnology being a GPT was continued. The criteria for being a GPT are again scope for improvement, pervasiveness & innovational complementarities. The level of analysis is a technology: nanotechnology. They used patents in order to define nanotechnology. By 15

17 means of a keyword search on the USPTO database, patents were extracted. The patents were then assigned a technological sector according to IPC codes. The innovational characteristic of nanotechnology is measured by the amount of granted nano patent applications over time as well as the amount of nano patent applications as a percentage of all patent applications. They also briefly look at patent citations. The second characteristic, pervasiveness, is again measured by the HHI Index of the patent classes assigned to forward citations of the focal patent.. Based upon their results and the results of previous related studies they concluded that nanotechnology also fulfills the second requirement of being a GPT. The last characteristic, scope for improvement is not measured by means of patents. Instead the development of the nanotube is used in order to provide an example that nanotechnology is an evolving and improving technology. This development is measured by improvements in the production efficiency of nano-tubes. The authors acknowledge that it is difficult to draw a general conclusion from example about the improvement of the entire nanotechnology sector. However they conclude that the nanotube provides preliminary support that nanotechnology is improving over time. Based upon their findings they conclude that nanotechnology does show signs of being a GPT and implications of nanotechnology as a GPT are being discussed (Shea et al., 211). Kreuchauff & Teichert (214) follow up on Shea et al. (211). The criteria for being a GPT are again scope for improvement, pervasiveness & innovational complementarities. The level of analysis is a technology: nanotechnology. Nanotechnology is defined by means of patents. Patents were extracted from the PATSTAT database based upon a keyword search. Also publications listed in the Thomson- IsI WoS database were included based upon a keyword search. Only patents and publications between the years 198 and 28 were included. For each characteristic they have multiple metrics. For pervasiveness they use three metrics. The first metric is diffusion rates of patents, which is calculated by calculating the share of nano patents in an innovative company patent portfolio. The second metric is the HHI index of 3 technological fields on the share of citations received by patents. These technological fields were based upon the International Standard Industrial Classification (ISIC). In order to link IPC codes to ISIC, a concordance table was used. The last metric measures the relatedness of patent citations and is called technological coherence (Kreuchauff & Teichert, 214). This last indicator is of particular interest as a similar metric will be used in this research as well. More information on why this last metric is important will be discussed in paragraph Scope for improvement has two indicators: growth in the amount of nano-patents & patent (forward) citations. Innovation spawning capability is measured by two indicators: the share of nanotechnology patents in overall patenting activity over time & growth in nano-citing technology classes. The results are benchmarked to ICT (presumed a GPT) and the combustion engine (presumed not a GPT). Their results show that nanotechnology does show the characteristics of a GPT. However they note that nanotechnology is a new technology still undergoing development and therefore the data is limited. (Kreuchauff & Teichert, 214). 16

18 As is visible from the empirical literature, there is no consensus of what the unit of analysis of a GPT should be although it should be noted that the latest researches define a technology by means of a set of patents. Also in the empirical literature there are different criteria for being a GPT as well as different metrics. These differences do not provide consensus on what methodology to use in order to determine if something is a GPT or is not a GPT. So is a GPT a phenomenon? Can some technologies or sectors or even specific types of technologies be classified as a GPT whereas others cannot? Instead of looking at GPT s as a binary value (either something is a GPT or isn t a GPT) this research introduces the concept of GPT ness. This means that some technologies can be considered more of a GPT than others. This way it is possible to make comparisons between technologies for each of the three characteristics. The GPT ness of technologies can be compared throughout time to the weighted average GPT ness of all the other technologies. It is also possible to make an absolute ranking at a certain point in time to see what technologies are scoring the highest when it comes to GPT ness. This way it will be possible to identify whether or not a technology stood out or not instead of marking it either a GPT or not a GPT. In order to measure GPT ness the existing criteria for being a GPT: scope for improvement, pervasiveness and innovational complementarities can still be used. How these three criteria can be measured was already briefly discussed in the literature above. In the next three paragraphs 2.2.1, and this research will zoom in into these metrics in order to understand the advantages and disadvantages of the metrics. These metrics can also be found in appendix 5 where they are summarized in a table. After these metrics have been examined, this research will present its metrics in order to measure GPT ness Scope for Improvement Kreuchauff & Teichert (214) measure scope for improvement by two indicators. The first one is looking at the growth of the amount of patents and comparing it to the growth of the amount of patents of the benchmark set in the same period (Kreuchauff & Teichert, 214). Kreuchauff & Teichert (214) base this indicator upon research by Palmberg & Nikulainen (26) who also investigated to what extent nanotechnology meets the criteria of a GPT (Palmberg & Nikulainen, 26). The second indicator of Kreuchauff & Teichert (214) is based upon research by Schultz & Joutz (21) who investigated how to identify a GPT based upon a patent analysis (by means of case study on nanotechnology) (Schultz & Joutz, 21). This indicator is based upon patent citations creating a technological trajectory. If a later patent cites the original invention, improvement is indicated (Kreuchauff & Teichert, 214). This is also known as patent forward citations. By computing the average amount of forward citations for the optical science database and the benchmark it is possible to compare the results with each other and see to what extent optical science has been improving more compared to the benchmark during the same time period. 17

19 Hall and Trajtenberg (24) suggest based upon the characteristics of a GPT that it is expected that GPT patents will have many citations within their technological area indicating a technological trajectory. (Hall & Trajtenberg, 24) Pervasiveness Many of the earlier discussed authors use a generality index in order to measure pervasiveness. To give some examples, this index is used in the research on electricity by Moser & Nicholas (24), by Youtie et al. (27) in their research on nanotechnology, by Shea et al. (211) in their research on nanotechnology and by Kreuchauff & Teichert (214) also in their research on nanotechnology. This index is based upon Hall & Trajtenberg (24) and is denoted as follows: n i 2 Generality = G i = 1 s ij (Hall & Trajtenberg, 24, p. 9) This index is based upon the Herfindahl-Hirschman Index (HHI) which us used in order to measure the concentration (market share) in a particular industry. In this formula s ij stands for the percentage of citations that are received by patent i which belongs to patent class j out of a total of n i patent classes. This means that if a patent has many citations from many different technological fields (patent classes) the generality index will be close to 1 which indicates a highly pervasive technology (Hall & Trajtenberg, 24). It is noted that the generality index cannot be calculated when a patent receives zero citations. The advantage of using this generality index is that it is not simply counting patent citations but it takes into account the share (relative weight) of each patent class However, this method of calculating pervasiveness has one major drawback which is mentioned by Hall & Trajtenberg (24) as an area for future research. The generality index makes no distinction between technologies that are very closely related but not in the same patent class and technologies that are very far apart. (Hall & Trajtenberg, 24, p. 2). Kreuchauff & Teichert (214) solve this problem by means of technological coherence (Kreuchauff & Teichert, 214). This research will use a different methodology in order to solve this problem by calculating the unrelated variety for each IPC class. This methodology is one of the main contributions of this research. An illustrative example of the problem is given below. j 18

20 Figure 3 Difference in relatedness of IPC codes As is clear from the figure above, there is a difference between the relatedness of these patent classes. In order to overcome this problem this research does not make use of the generality index on patent citations. Instead this research measures pervasiveness by means of co-occurrences of patent classes, or more specific of IPC codes. If a technology can be classified by means of IPC codes it will be possible to determine how these IPC codes are related to other IPC codes by means of co-occurrences. Recall that one patent can have multiple IPC codes. If a certain IPC code has a lot of different IPC codes showing up together with that IPC code, the variety in the co-occurrences of that focal IPC will high. If the variety in the co-occurrences of that focal IPC is high it means that technologies that are classified by means of the focal IPC are pervasive. By comparing the variety to the average variety it will be possible to measure the relative pervasiveness which is exactly what is needed in order to measure the previously introduced concept of GPT ness Innovational Complementarities Hall & Trajtenberg (24) suggest that the time between the grant date of a patent and the arrival of the patent s forward citations tend to be longer in the case of a GPT. They argue this is because it takes time for a GPT to pervade the economy (Hall & Trajtenberg, 24). This time is referred to as the patent citation lag. Hall & Trajtenberg (24) note that when looking at a fixed period of time, the patent data is subject to truncation and therefore they look at mean citation lags that are large relative to the average citation lag for patents applied for in the same year. Furthermore Shea et al. (211) look at the sheer amount of patents over time as well as the relative amount of patents of a technology as a percentage of the total amount of patents. Based upon Hall & Trajtenberg (24) who argue that GPT s have patents that are highly cited within their technological area (Hall & Trajtenberg, 24), they look at the amount of citations as a measurement of a technology being a GPT (Shea et al., 211). 19