PATH DEPENDENCY IN INDUSTRIES WITH MULTIPLE TECHNOLOGICAL TRAJECTORIES

Transcription

1 PATH DEPENDENCY IN INDUSTRIES WITH MULTIPLE TECHNOLOGICAL TRAJECTORIES Authors Anna Bergek, Department of Management and Engineering, Linköping University, Sweden, Ksenia Onufrey, Department of Management and Engineering, Linköping University, Sweden, Abstract In the literature on path dependency in processes of innovation and technical change, two partly conflicting perspectives are presented. Within the first perspective, it is argued that the cumulative nature of technical change creates persistence in innovative activities: accumulated competencies and learning within a specific field generate new research questions and opportunities for innovation and create entry barriers, which works in favour of incumbent firms and limits the role of new innovators in an industry (Malerba et al., 1997). In contrast, the other perspective emphasises that path dependency gradually decreases the number of available future options (Aminzade, 1992; Araujo and Harrison, 2002) and eventually leads to lock-in to inefficient, inferior or unsustainable technology paths (Cowan and Gunby, 1996; David, 1985; Unruh, 2000). Within both these perspectives, paths tend to be conceptualised as single technological trajectories. However, in some industries multiple trajectories are pursued in parallel and new trajectories are added over time. This raises the questions of whether such industries still can be path dependent and, in that case, where path dependency occurs: within or across trajectories and at the company or industry level. To what extent does the incumbents development of newly added trajectories build on their existing knowledge base? The purpose of this paper is to answer these questions by analysing technological activities of three leading firms in the lighting industry. The paper is based on an analysis of lighting patents granted to General Electric (GE), Osram/Siemens and Philips and their key subsidiaries by the US Patent and Trademark Office (USPTO) over a period of 35 years ( ). Lighting-related patents were identified through a combination of class-based search and title- and abstract-based keyword search. Our analysis shows a common patenting pattern between the three companies: about 70% of all the patents in the dataset belong to seven most frequently used classes and about 50% - to the top three classes. Most of these classes can be described as traditional since companies used them during the whole period of analysis. While some of them are declining both in terms of patent shares and numbers (H01K Incandescent lamps), others are stable or growing (H01J Discharge lamps, F21 Lighting, H05B Electric lighting, C09K Materials for applications). Such long-term stability of traditional classes and similarity of patenting patterns between the three companies indicate technological persistence both at the company and the industry levels. The most recent addition to the companies patent stock is the semiconductors class (H01L). It has been intensively developed since the late 1990s, when industry incumbents joined the LED technology which was pioneered by new entrants. However, about 30-40% of the LEDrelated patents of GE, Osram/Siemens and Philips still belong to traditional lighting classes. Companies have, thus, been able to use their previously accumulated expertise in the development of LED lighting, in spite of its discontinuous character. While technological persistence in terms of LED development can be observed at both industry and company levels, there are some differences among the three companies. 1

2 An analysis of patent references shows that when a patent cites one of the company s own lighting patents, in 60-70% of the cases both patents belong to the same first class, which is a clear sign of path dependency inside trajectories. However, pairwise usage of patent classes indicates not only persistence inside technological trajectories, but also a complex relation between them since patents frequently belong to several classes simultaneously. In particular, H01J (discharge lamps) is the most frequently used secondary class. The main conclusions of the paper are the following: first, we have found signs of path dependency in the lighting industry at the company level in a form of technological persistence. Although persistence inside technological trajectories is especially strong, there is also a complex interconnection between trajectories which indicates that previous association of paths with single trajectories is too simplified. Second, a similarity of companies patenting patterns in almost every aspect of the analysis provides a clear evidence of path dependency at the industry level. Third, the LED example shows, on the one hand, a break with previous activities, and on the other hand, the ability of incumbents to use their accumulated expertise when developing a new, even radically different, technology. Overall, it can be concluded that path dependency can exist in industries with multiple technological trajectories. However, whether this path dependency is productive and efficient or will lead to unsustainable lock-in remains to be seen. Key Words: Path dependency, technological trajectories, technological persistence, industry incumbents, lighting industry, patent study 2

3 1 Introduction The phenomenon of path dependency has received a lot of attention in research about processes of innovation and technical change. However, several issues are still under debate by the path dependency research community. A first unresolved issue is the outcome of path dependency. Some researchers argue that path dependency gradually decreases the number of available future options (Aminzade, 1992; Araujo and Harrison, 2002) and will, eventually, lead to lock-in of firms, technological fields or entire societies to inefficient, inferior or unsustainable technology paths (Cowan and Gunby, 1996; David, 1985; Rycroft and Kash, 2002; Unruh, 2000). In contrast, other researchers argue that the cumulative nature of firm-specific technical change tends to create persistence in innovative activities. Accumulated competencies and learning within a specific field generates new research questions and new opportunities for innovation and creates entry barriers, which work in the favour of incumbent firms and limit the role of new innovators in an industry (Malerba et al., 1997). This implies that a company that innovates once in a technical field is likely to innovate again (Malerba et al., 1997), as long as no new innovations appear that render the knowledge and capabilities accumulated within the path obsolete (cf. Anderson and Tushman, 1990; Henderson and Clark, 1990). A second unresolved issue is the question of path dependency when several paths co-exist in a technical field. In most of the previous literature on path dependency, co-existence of paths is assumed to imply competition between technologies, where only one path eventually survives. In contrast, it has been argued that incumbent firms not only have to master the challenge of displacement of old paths by new ones, but also the challenge of articulation, disappearance, and reappearance (Aminzade, 1992) of many different paths within their main field of operation. The question of where path dependency occurs in such a case, e.g. within or across technological trajectories and at firm or industry level remains largely unanswered. Moreover, in such a setting the question of whether industry incumbents can utilise their accumulated knowledge base in new trajectories or become locked in to old paths and locked out of new trajectories seems even less straight-forward to answer than in single-trajectory industries. The overall purpose of this paper is to answer these questions by means of a case study of three incumbents in the lighting industry: General Electric (GE), Osram/Siemens and Philips. The lighting industry has three important distinguishing features that make it suitable for studying the phenomenon of path dependency in a multi-trajectory setting: (a) several technological trajectories have co-existed for some time, (b) a stable set of industry incumbents dominate the industry and (c) the industry has seen a recent addition of a new technological trajectory, which potentially might challenge industry incumbents. 2 Path dependency and innovation 2.1 Levels and types of path dependency David (1985) suggested the notion of path dependency to describe a change process affected by remote, and sometimes chance, events. Since then, a theory of path dependency has been developed, growing from a simple idea that history matters to a complex concept, constituting an important part of the evolutionary view of technological change. The number of articles mentioning or devoted to path dependency is growing, and includes several recent attempts to summarise and formalise the current understanding of the concept (cf. Garud et al., 2010; Sydow et al., 2009; Vergne and Durand, 2010). In theoretical studies, path dependency is conceptualised on different levels: from the microlevel of company resources and capabilities, through the meso-level of technological field, industry or sector to the macro-level of institutions (Vergne and Durand, 2010). Empirical 3

4 studies, however, tend to focus on the first two of these levels firm (Patel and Pavitt, 1997; Rosenkopf and Nerkar, 2001), technological field/industry/sector (Håkansson and Waluszewski, 2002; Mazzoleni, 1997; Unruh, 2000), or both (Essletzbichler and Winther, 1999; Fai, 2003). Another distinction in the literature is the one between technological path dependency and organisational path dependency. Technological path dependency is related to technological competencies of firms and technological choices in industries, whereas organisational path dependency includes all kinds of organisational processes, routines and institutions. Both these types of path dependency can be studied on different levels of analysis. In this paper, we focus on technological path dependency, which we study on two levels: firm and industry. 2.2 Central features of path dependency One of the central features of path dependency is persistence. Sydow et al (2009) refer to strategic persistence in relation to organisational path dependency: companies experience operational rigidity and maintain a particular pattern of change processes which is not obligatorily the most efficient. The related notion of technological persistence highlights the cumulativeness of technological change, and its consequences in the form of constrains for future technological development (Cantwell and Vertova, 2004; Dosi, 1982; Malerba et al., 1997). More specifically, technological persistence implies that past technological achievements, patterns of problem-solving activities or accumulated competencies constrain current technological activities as well as directions in which new opportunities can be searched for (Dosi, 1982; Malerba et al., 1997; Patel and Pavitt, 1997). The reason for this constraining power of accumulated competencies is that learning is local; new opportunities are likely to be found close to areas of existing expertise (Teece et al., 1997) and therefore competencies can be developed only slowly and incrementally (Fai, 2003). Another important feature of path dependency (or even its heart, according to Sydow et al (2009)), is the existence of self-reinforcing mechanisms, which make the path irreversible and give it a momentum of its own (Dosi, 1982). Self-reinforcing sequences are those which at every step generate consequences that amplify the impact of the outcomes of the earlier rounds in ensuing rounds, moving sequences of events along ever-narrower paths (Araujo and Harrison, 2002, p. 7). 1 Literature mentions different kinds of self-reinforcing mechanisms or increasing returns: technological externalities, learning, uncertainty reduction (Cowan and Gunby, 1996), positive network externalities (Vergne and Durand, 2010). Analysis of both technological persistence and self-reinforcing mechanisms is essential in order to understand technological path dependency (Sydow et al., 2009). Understanding technological persistence is a way to map path dependency, define distinct knowledge areas and relations between them. As a next step, self-reinforcing mechanisms should be analysed in order to comprehend how path dependency affects technological choices and activities of actors. The current study is on the first step of the analysis of path dependency in the lighting industry and therefore focuses on technological persistence, while self-reinforcing mechanisms will be considered in further research. 2.3 Unresolved issues in path dependency research In the discussion of path dependency there are several unresolved issues which require separate consideration. We will here discuss three such issues in more detail: (1) whether lock-in is inevitable, (2) whether triggering events, which have led to path creation, always are of crucial importance, and (3) whether co-existence of several trajectories in one technological field is possible. 1 Self-reinforcing sequences can be contrasted with reactive sequences (or, as Karnoe and Garud (1997) call them, path creation) which can lead to shifting of a path into a new direction (Araujo and Harrison, 2002; Mahoney, 2000). 4

5 2.3.1 Lock-in Lock-in is generally understood as a negative consequence of path dependency. In research on technological path dependency, the starting point is often competition between one or more alternative technologies, which ends in the choice of a sub-optimal technology or set of technologies due to self-reinforcing mechanisms described above (Arthur, 1988; Cowan, 1990; Cowan and Gunby, 1996; David, 1985; Unruh, 2000). More generally, lock-in can be defined as a situation characterised by dominance of a particular pattern of actions and lost flexibility; actors lose their ability to adapt to changes and cannot move to a more optimal alternative even if they prefer to do so (Garud et al., 2010; Sydow et al., 2009; Vergne and Durand, 2010). The question of whether lock-in is an inevitable consequence of path dependency does not have a single answer. In David s early work, path dependency was used as an explanation of why a seemingly inferior technology could become dominant (David, 1985), but lock-in was seen as a possible, not necessary, result of path dependency (cf. David, 1994). However, in some of the latest path dependency literature, lock-in is a central and necessary feature. For example, Vergne and Durand (2010) argue that path dependency will lead to lock-in unless an exogenous shock disturbs the system and Sydow et al (2009, p. 695) describe lock-in as a necessary third and final stage of the process of evolving path dependency: If actors were not locked in, one would not call the process path dependent. In contrast, critics of this perspective argue that even if path dependency constrains actors, they are still creative and able to influence the course of events, change directions of paths that they follow or generate new paths (Antonelli, 2009; Araujo and Harrison, 2002; Garud et al., 2010). Actors iterate past patterns of thought and action, but they are also able to generate future scenarios and evaluate alternative trajectories in response to changing demands and opportunities (Araujo and Harrison, 2002). This agency-oriented perspective resonates well with the so-called Schumpeter Mark II pattern of innovation described by, for example, Malerba et al. (1997). According to this literature, innovation is a result of creative accumulation of technological competencies (Bergek et al., 2011), which is why technological persistence does not inevitably lead to the decrease of the level of innovative activities. Firms develop dynamic capabilities which help them to integrate, build and reconfigure internal and external competencies in line with requirements from their external environment (Teece et al., 1997, p. 516). In this paper, we side with the latter perspective and consider lock-in to be a possible negative consequence of path dependency, but not a necessary one Triggering events There is a general agreement among researchers that a path is affected not only by initial conditions or choices, but also by the entire sequence of events along the path. 2 Path dependent processes are, thus, started by triggering events (Sydow et al., 2009), which give impetus for further action and may cause unintended consequences (Sydow et al., 2009). They are associated with critical junctures, i.e. moments in time when actors range of choice is increasing or decreasing (Gagliardi, 2008). However, the nature and importance of these triggering events for studying path dependency remains an open question. Some researchers pay a lot of attention to the moment of path creation and claim that tracing a path back to one or more triggering events is an essential part of path dependency studies (Mahoney, 2000; Sydow et al., 2009; Vergne and Durand, 2010). However, there is some disagreement among these researchers in terms of whether these events are necessarily contingent by nature (Mahoney, 2000; Vergne and Durand, 2010) or if they can be planned and strategic (Sydow et al., 2009). 2 This is what distinguishes path dependency from past dependency (Antonelli, 2009) 5

6 In contrast, some researchers argue that the starting point of a path is not in any way a given (Garud et al., 2010). This is a consequence primarily of the complexity involved in technological change. The literature on innovation systems emphasises that the development of new technologies takes place within emerging or existing socio-technical systems, in which there are few simple cause and effect relationships (cf. e.g. Carlsson and Stankiewicz, 1995). In addition, different actors within the innovation system might have different starting points, depending on when they joined the path (Garud et al., 2010). This implies that it might not be theoretically possible to trace a path back to its starting point, at least not on a technology field level. From an empirical point of view, finding practical evidence for triggering events in historical cases can be very difficult. In the case of the lighting industry, the original path (the incandescent bulb trajectory) was started more than 100 years ago, which not only makes it difficult to identify a specific triggering event, but also largely irrelevant for the purpose of the current paper: it is not necessary to identify the events triggering the creation of historic paths to study how these paths are manifested in actors technological portfolios today Path dependency and technological trajectories In most studies of technological path dependency, the concepts of path and trajectory are used synonymously (cf., e.g., Augsdorfer, 2005; Cantwell and Vertova, 2004; Mazzoleni, 1997). Articles which refer to several paths or trajectories tend to imply that these are alternative technologies, only one of which will eventually be chosen (Araujo and Harrison, 2002; Arthur, 1988). This means that according to current research, competition among several technologies inevitably results in winning of one of them and disappearance of the rest. Although Dosi (1982) potentially accepts the possibility of multiple trajectories by introducing the distinction between technological paradigm (which can be associated with the path concept) and technological trajectory (which represents a pattern of problem solving activities within a paradigm), the consequences of multiple trajectories for path dependency are not discussed. In our empirical case, the lighting industry, we can identify a number of co-existing technological trajectories, most notably incandescent bulbs (including halogen lamps), discharge lamps, and light-emitting diode (LED) lighting. The most recent addition, LED, is especially interesting, considering that it to a large extent builds on semiconductor technology, which might indicate a convergence between two up till now largely separate industries. The three main incumbent actors, GE, Osram/Siemens and Philips, not only participate in all these trajectories, but are among the companies leading their development, which seems to be an indication of some kind of industry-level persistence and, possibly, path dependency. A number of questions, however, remain unaddressed: a) To what extent do incumbent firms in an industry characterised by multiple technological trajectories show signs of technological path dependency and does path dependency occur within or across trajectories? b) To what extent is technological path dependency common to all incumbent firms in such an industry, i.e. to what extent can we talk about path dependency at the industry level? c) To what extent does the incumbents development of newly added trajectories, in this case LED technology, build on their existing knowledge base? In this paper, these research questions will be addressed by analysing the technological activities of the three leading firms in the lighting industry. Thereby, we hope to contribute to a more thorough understanding of the role of path dependency in industries with multiple technological trajectories 6

7 3 Research design The paper uses patent data of GE, Osram/Siemens and Philips in the lighting technology field. Patents stored in the US Trademark and Patent Office (USPTO) database over a period of 35 years ( ) are analysed. Patent data are traditionally used as an indicator of technological capabilities as well as innovative and R&D activities. For example, Patel and Pavitt (1991, 1997) use patents as a measure of technological activities and technological competency; Fai (2003) applies patents as an indicator of innovative activities; and Bergek and Berggren (2004) use patents to measure R&D activities. The applicability of patents to study the specific issue of path dependency is assessed and confirmed by Fai (2003) who concludes that patents are an appropriate measure for investigating evolutionary and cumulative technological development not only because they provide data over long time periods, but also because they usually require comparatively low degree of novelty and, therefore, are able to reflect incremental change processes. This reasoning is also in line with the idea proposed by Patel and Pavitt (1991) that patents, apart from measuring technology generation, also help to trace technology diffusion and imitation (since these processes include improvements of the original technology). When using patents, most of the researchers acknowledge advantages and disadvantages of this data source as summarised by Patel and Pavitt (1991). On the one hand, patent data are available over long time periods, allows for detailed analysis, and are easy and cheap to access. Additionally, as argued by Fai (2003), patents reflect both codified and tacit knowledge. On the other hand, a number of disadvantages of using patents are usually listed, and several of them should be discussed in relation to this study. First, there can be a difference between countries and firms in terms of their propensity to patent (Patel and Pavitt, 1991). Although three multinational companies studied in this work are expected to have similar requirements in terms of patenting in order to gain or retain industry leadership, their patenting strategies may still differ. In order to mitigate this drawback, we compare companies not only per absolute number of patents in different lighting classes and fields, but also per shares of these fields in the overall number of lighting patents each company has. Second, Patel and Pavitt (1997) mention the difficulty to find all patents granted to large firms since they often file patents under the names of their subsidiaries which requires adding manually all the subsidiaries in the list. In this work, it is possible to verify the key subsidiaries due to the limited number of companies under the analysis. 3.1 Choice of the database: USPTO Analysis is performed for three companies: GE, Osram/Siemens and Philips. Since Osram/Siemens and Philips are originally European and GE is American, we had the choice of using either USPTO or the European Patent Register provided by European Patent Office (EPO). Choosing either of these databases implies some bias in the form of possible overrepresentation of patents of the companies for which it is a domestic database (USPTO for GE and EPO for Philips and Osram/Siemens). We decided to use USPTO for three main reasons. First, in USPTO one can search in patent abstracts, which is not possible in EPO. As will be discussed further below, abstract search is an important tool to identify relevant patents especially in the LED field which is not clearly associated with particular classes. Second, EPO contains both granted patents and patent applications older than 18 months, whereas we base our study only on granted patents. In order to distinguish between granted patents and patent applications in EPO, we need to open and manually check each patent document, which makes this database practically not feasible to use in our study, considering 7

8 a large number of patents under the analysis. Additionally, inclusion of patent applications in EPO makes it impossible for us to analyse patents per issue/publication date, since the day when a patent (or a patent application) becomes available in the EPO is assigned as patent publication date. Third, using USPTO leads to comparatively lower losses in terms of the number of patents taken into analysis. As mentioned above, all companies have more patents in their domestic databases. However, using USPTO leads to lower underrepresentation of Osram/Siemens and Philips compared to the underrepresentation of GE when using EPO. EPO contains over two times fewer GE patents than USPTO while USPTO provides only 8% less patents of Philips and 29% less patents of Osram/Siemens than available in EPO. 3 Although USPTO proved to be more appropriate to be used in this work compared to EPO, we acknowledge possible consequences of this choice such as the risk to underestimate Osram/Siemens and (to a much lower extent) Philips technological capabilities in comparison with GE s. This will be taken into account in the analysis section. 3.2 Dataset We started building our dataset by identifying patent classes that are related to lighting technology. An initial set of classes was distinguished based on patents owned by Osram, since Osram, unlike Philips and GE, is fully specialised on lighting. Class names and descriptions of the most frequently used classes in Osram s patents were checked in International Patent Classification and complemented with classes a) used in previous studies of lighting (Sanderson et al., 2008) and b) identified from International Patent Classification Standard Industrial Classification (IPC-SIC) table by Silverman (2004). In addition to class-based search, we used title and abstract-based keyword search. We applied two groups of keywords. The first group was used to cross-check the results of the class-based search and consisted of the following main keywords: lighting, lamp, light source and light emitting device. The second group of keywords was intended to search for LED-related patents and consisted of different variations of writing LED ( LED, light emitting diode, light-emitting diode etc.). We consider title and abstract-based keyword search to be very important for searching LED-related patents, since they are not strictly associated with any particular class which makes class-based search in this field not applicable. 4 Similar search strategy was also used in Sanderson et al (2008). The resulting search strategy (Appendix A) was applied for GE, Osram/Siemens, Philips and their subsidiaries. Key subsidiaries were identified based on annual reports of GE, Siemens and Philips (for at least 15 years) where major active subsidiaries as well as key joint ventures and acquisitions were mentioned. We assumed that subsidiaries found in the annual reports had strategic importance for the companies and should be included in the analysis. Subsidiaries were included in the analysis from the year of acquisition. When two companies were mentioned in the Assignee patent field (company name), corresponding patent was counted for both companies. 3 For the calculations mentioned above the search was performed only in five patent classes available in both databases: H01J, H05B, F21, H01K, C09K. We did not include keywords search since possibility to search in patent abstracts (apart from titles) makes results biased towards USPTO. The class 26 was not included either as it is not used in EPO. Additionally, since EPO contains both issued patents and patent applications, for USPTO we combined search results from patent database and applications database. Searches for this comparison were performed in April According to class description, the class which is most related to LED lighting is H01L Semiconductor devices, electric solid state devices not otherwise provided for. However, this is a broad class, and all of its subclasses are used not only in lighting. Since we are investigating three multi-technology companies, inclusion of this class into our dataset would introduce bias. However, we include two H01L sub-classes into our extended dataset which we use for comparison with our default dataset (without these sub-classes) when discussing LED technology (the two sub-classes are: H01L 31 and H01L 33 Devices sensitive to, or emitting, radiation ). 8

9 Results from the two search methods were combined, and all duplicates were deleted 5. For each patent, the following information was obtained: patent number, title, issue date, inventors, assignee, application number, filed date, US and international classes, US and foreign references. Totally, 7005 patents were gathered and constituted our default dataset, out of which: 2144 patents belong to GE and subsidiaries patents belong to Osram/Siemens and subsidiaries 3264 patents belong to Philips and subsidiaries It should be noted that comparatively few patents of Osram/Siemens before mid-1990s were found. The number is low both compared to the number of Osram/Siemens patents after mid-1990s and to the number of Philips and GE s patents before mid-1990s. The same problem was observed in both USPTO and EPO. Annual reports or web-based materials of the company did not provide explanation for the lack of patents in this period. This should be researched further. 3.3 Measurement As we explained in the theoretical section, in this paper we focus on technological persistence (and leave another fundamental feature of path dependency self-reinforcing mechanisms for future research). Therefore, measurements performed in this work are aimed to trace technological persistence. Most of the measurements are based on association of a part of patent classes with a particular technological trajectory incandescent, discharge or LED. At the company level, path dependency implies that future technological trajectories of a firm are constrained by its current technological capabilities as well as by its history and past experiences (Essletzbichler and Winther, 1999). Consequently, companies are likely to continue doing the same things as they did before. From the patenting perspective, dynamics of development of different patent classes are used to measure technological persistence. At the industry level, technological persistence can be measured by comparing patenting patterns of the companies (and interpreting similarities as signs of path dependency). The fact that we limit ourselves to the three largest companies when considering the industry level, will affect our analysis since we focus on only one type of actors industry incumbents. Implications of this limitation will be discussed in the end of the paper. We use the following measures of technological persistence: Patent shares of traditional (old) classes Development trends of traditional and new classes Presence of traditional classes in patents related to new products References to patents that belong to the same classes Simultaneous presence of several classes in the same patents Among these measures, some indicate persistence inside technological trajectories (for example, development trends of particular classes or references to patents of the same classes). Meanwhile, other measures indicate interconnections between trajectories (for example, simultaneous presence of several classes in the same patents). 5 Data gathering was performed in March 2012; all patents issued later than 2011 were deleted. 9

10 When analysis of class usage over time is performed, patent issue date is usually used as the time indicator in order to avoid bias related to the fact that up to several years can be needed from patent application to patent issue. All applied but not approved (issued) patents are not added into the dataset and consequently this would be reflected as a declining trend in the number of patents during several latest years if plotted per application date. However, we consider application date when analysing and comparing the start of new technologies (for example, LED) since this date is closer to R&D activities of the companies. One more differentiated approach is used in the analysis of patent classes. In the overall analysis of the distribution of patents per classes we use the first mentioned class in the corresponding field (there, shares of all classes taken together constitute 100%). However, when performing more detailed analysis of particular classes, we take all patents where a particular class is mentioned as first or any of secondary classes (consequently, the sum of all the shares is more than 100% in this case). 4 Evidence of path dependency in the lighting industry In this section, we analyse our dataset using different ways of measuring technological persistence outlined in the methodology part. First, we compare the dynamics of usage of different patent classes 6 at GE, Osram/Siemens and Philips. Next, we analyse patent references and consider simultaneous usage of different classes. 4.1 Most frequently used patent classes Since we primarily used class-based search when gathering data, it is logical that the six classes that were included into the search are rated among the most frequently used for all the three companies. These classes are: H01J, H05B, F21, C09K, 26, H01K. These and other frequently used classes together with class descriptions are presented in Table 1. What is particularly interesting, is that one additional class appeared quite high among the most frequently used classes H01L, Semiconductor devices. As was explained in the methodology section, due to a broad content of this class, we did not include it into the search strategy and, instead, used keywords when searching for LED-related patents. Nevertheless, H01L class is ranked quite high in the resulting dataset for every analysed company: it is the 4th at Osram/Siemens and Philips and the 6th at GE. 6 During data gathering, the search rules were constructed at the level of patent sub-classes (see Appendix A for details). But during data analysis, patents were grouped at the level of classes in order to find higher-level patterns and integrate results from the keyword search into the analysis. That is why the analysed classes can include also other subclasses than described in Appendix A. Nevertheless, all the patents in the dataset are relevant for the lighting technology field. 10

11 Table 1. Top-ten most frequently used patent classes (first class checked) Number of patents Per Cumulative Class description Average Class Osram/ cent GE Philips rank Siemens share share H01J ,0 22% 22% Electric discharge tubes or discharge lamps H05B ,0 17% 39% Electric heating; Electric lighting ( ) F ,0 12% 51% Lighting H01L ,7 7% 58% Semiconductor devices; Electric solid state devices ( ) C09K ,3 5% 63% Materials for applications ( ) ,3 6% 68% International classification for industrial designs: 26 - Lighting apparatus H01K ,0 4% 72% Electric incandescent lamps G02B ,0 2% 74% Optical elements, systems, or apparatus H01R ,3 1% 75% Electrically-conductive connections( ) C03C ,7 1% 76% Chemical composition of glasses ( ) G01N ,7 1% 77% H02M ,3 1% 79% G02F ,0 1% 80% G03B ,3 1% 81% Investigating or analysing materials by determining their chemical or physical properties Apparatus for conversion between ac and ac, between ac and dc, or between dc and dc ( ) Devices or arrangements, the optical operation of which is modified by changing the optical properties of the medium of the devices ( ) Apparatus or arrangements for taking photographs or for projecting or viewing them ( ) The total number of classes in the dataset is about 200, with about 120 unique classes at each company. Therefore, it is noteworthy that the top-three classes are exactly the same for all of the companies and account for 51% of all the analysed patents, and that the top-seven classes constitute over 70% of all the patents in the dataset. Moreover, these shares are very high not only in the combined dataset, which theoretically could be caused by extreme domination of these classes in one of the companies, but are approximately equally highranked for each of the three companies (see Table 2). Table 2. Comparing shares of top-7 and top-3 patent classes Items for comparison GE Osram/Siemens Philips Total number of patents 2, ,264 Number of patents in top-7 classes 1,484 1,122 2,456 Number of patents in top-3 classes 1, ,682 Share of top-7 in total 69% 70% 75% Share of top-3 in total 53% 45% 52% More detailed analysis of the most frequently used patent classes shows a combination of general lighting classes (F21 - Lighting, H05B Electric lighting, 26 industrial design class Lighting apparatus), those that refer to particular technology types or technological 11

12 trajectories (H01K Incandescent lamps, H01J Discharge lamps, H01L Semiconductor devices), and one broader class (C09K Materials for applications). The above evidence shows a common patenting pattern for all three companies. However, in order to talk about industry-level technological persistence, this pattern should be traced over time. Figure 1 shows how the shares of the most frequently used patent classes have developed over time. As we can see, most of the classes (apart from H01L, Semiconductor devices) can be considered as traditional since they have been clearly present throughout the whole period of analysis. In the following, we will first analyse the dynamics of these classes and then consider the newest trajectory (H01L). 7 The dynamics of development of the most frequently used classes (3 years moving average of percent shares) 40.00% 35.00% 30.00% 25.00% 20.00% 15.00% 10.00% 5.00% 0.00% H01J Discharge lamps H01K Incandescent lamps H05B Electric Lighting F21 Lighting C09K Materials for applications H01L Semiconductor devices Figure 1. Shares of most frequently used patent classes first class Traditional lighting classes The cumulative share of the five traditional lighting classes decreased from about 60% in to about 50% in Although the observed trend is declining, traditional classes retained a significant part of all patents in the dataset (see Figure 1). H01J, Discharge lamps, has the largest patents share throughout almost the whole period of analysis. Although over the recent 15 years its share is steadily decreasing, it remains the highest compared to all other classes. Another technology-specific class, H01K, Incandescent lamps, shows a decreasing trend and holds less than one per cent of patents in recent years. As for the general lighting classes, H05B and F21, both have slightly increased share in the end of the studied period compared to the beginning, but they differ in terms of dynamics over the years: H05B, Electric Lighting, has two considerable spikes, while F21, Lighting, can be characterised as more stable. The last traditional class, C09K, Materials for applications, shows not high (below 10%), but stable patent share between 1976 and 2004, with a decrease in the end. Thus, from the patent shares dynamics we can see a downwards trend for several traditional lighting classes since the end of 1990s. However, if we look at the absolute number of patents in these classes for the same period, we will see a different picture. Figure 2 illustrates the dynamics of the absolute numbers of patents in the traditional lighting classes since the end of 1990s. 7 It should be noted that one of these classes, 26, represents lighting design patents which do not constitute a single technological trajectory. Therefore, further analysis will not include the 26 class. 12

13 Compared to the dynamics of patent shares, the absolute numbers show more stable and even growing trends. H01J, Discharge lamps, and two general lighting classes (H05B and F21) are clearly growing; C09K, Materials for applications, is stable, and even the decline of H01K, Incandescent lamps, is not very drastic. The cumulative number of patents submitted in these five classes every year is 1,5 times higher in (on average, 202 patents per year) than in (on average, 131 patents per year). In summary, we can conclude that there is technological persistence at the industry level which is demonstrated by stability of lighting shares of the general lighting classes (H05B and F21) and stable or growing trends in the absolute numbers of patents of the four of five traditional classes (H01J, H05B, F21, C09K). The dynamics of development of traditional classes (3 years moving average of absolute numbers for the past 15 years) H01J Discharge lamps H01K Incandescent lamps H05B Electric Lighting F21 Lighting C09K Materials for applications Figure 2. Development dynamics of traditional lighting classes for the past 15 years In order to see whether technological persistence can be observed at the company level also, Figures 3-7 represent the development dynamics of the traditional lighting classes, plotted separately for each of the companies. In order to obtain a more complete picture of how the traditional classes were used by the companies over time, the figures below include all patents where these classes were used (not only as the first class). Figures 3-7 show that the patterns found at the industry level are also valid at the company level: H01K (Incandescent lamps) is slowly decreasing, with quite similar dynamics over the past 15 years for all the three companies. H01J (Discharge lamps) is quite stable over time. F21 (Lighting) is stable at GE, but at the same time shows a considerable growth at Phlips and Osram/Siemens. H05B (Electric lighting) is steadily growing at Osram/Siemens, quite stable at GE, and shows general growth with a considerable peak in the beginning of 2000s at Philips. C09K (Materials for applications) shows quite uneven dynamics for all three companies, but generally a positive trend can be noticed. Based on these patterns, we can conclude that there is technological persistence at the company level. Additionally, similarity of classes dynamics between the companies strengthens our arguments regarding industry-level persistence. 13

14 H01K (incandescent lamps) as any mentioned class - moving average for 3 years GE Osram Siemens Philips Figure 3. H01K class at GE, Osram/Siemens and Philips 8 60 H01J (dischagre lamps) as any mentioned class - moving average for 3 years GE Osram Siemens Philips Figure 4. H01J class at GE, Osram/Siemens and Philips F21 (lighting) as any mentioned class - moving average for 3 years GE Osram Siemens Philips Figure 5. F21 class at GE, Osram/Siemens and Philips H05B (electric lighting) as any mentioned class - moving average for 3 years GE Osram Siemens Philips Figure 6. H05B class at GE, Osram/Siemens and Philips 15 C09K (materials for applications) as any mentioned class - moving average for 3 years GE Osram Siemens Philips Figure 7. C09K class at GE, Osram/Siemens and Philips 8 As was already mentioned in the methodology part, a very low number of patents were found at Osram/Siemens before the mid-1990s. That is why, all the curves of this company show growth in the second half of the chart. 14

15 4.1.2 H01L, Semiconductors devices, and LED lighting The most recent technological trajectory (LED) and, associated with it, class H01L (Semiconductor devices) require separate attention. This trajectory has been intensively developed since the late 1990s. Development of H01L class reflects the decision of the lighting industry incumbents to join the LED technology which was pioneered by new entrants. The fact that a new major class was added into the companies patent portfolio points to the discontinuous character of the new technology. However, 30-40% of the patents related to LED based on keywords-based search still belong to traditional lighting classes (see Figures 8-10). 40 GE: LED-related patens 80 Osram-Siemens: LED-related patents 100 Philips: LED-related patents Other classes Traditional classes H01L H01L as first class in the extended dataset Other Traditional classes H01L H01L as first class in the extended dataset Other classes Traditional classes H01L H01L as first class in the extended dataset Figure 8. LED patents at GE Figure 9. LED patents at Osram/Siemens Figure 10. LED patents at Philips Such extensive usage of traditional lighting classes in LED-related patents implies that the companies have been able to use their previously accumulated expertise in the development of a new technology, in spite of its discontinuous character. This also points to their technological persistence. As for semiconductor trajectory (H01L) of the three companies, it generally follows the same trend as LED-related patents, although it was started earlier. This could be caused either by some other usage of H01L class in these multi-technology companies, or by naming the patents at the early stage in such a way that they did not show a direct relation to LED. As Figure 11 shows, H01L class is used by the three companies not only in LED lighting. One part of H01L patents does not directly relate to lighting (dark-grey columns in Figure 11), especially at Osram/Siemens. Another part of H01L patents relates to lighting, but not LED (medium-grey columns in Figure 11), which is especially characteristic for Philips and GE, and to a lower extent also for Osram/Siemens. The fact that H01L class is used in different lighting products, and not only LED, is an argument in favour of interdependence between trajectories. 15

16 H01L class, Semiconductors devices, at GE, Osram/Siemens and Philips GE Osram/Siemens Philips H01L as the first class in LED patents H01L as the first class in lighting patents (default dataset) H01L as the first class in any of the company's patents (extended dataset) Figure 11. H01L class at GE, Osram/Siemens and Philips Technological persistence observed at the industry level can also be seen at the company level, although with some differences between the three companies. GE shows the highest technological persistence by having 39% of traditional classes among LED-related patents (see Figure 8). At the same time, LED technology in the company demonstrates downward trend after an initial peak in the beginning of the 2000s. The charts of Philips and Osram/Siemens seem similar in terms of timing and scale of efforts, with Philips being more technologically persistent than Osram/Siemens (36% and 32% of traditional lighting classes among LED-related patents, respectively) (see Figure 9 and Figure 10). Additionally, both companies show upwards trend in the recent years. However, the difference between the two companies becomes more visible if the total number of H01L patents including H01L-class search is taken into account. At Osram/Siemens a significant amount of H01L patents did not get into the default dataset (see Figure 11). A lot of H01L patents of Osram/Siemens do not have direct relation to lighting, even during the recent years when all the H01L patents of Osram/Siemens were filed by Osram (and not other Siemens businesses) Analysis of patent references An analysis of patent references is our next indicator of technological persistence. At GE, Osram/Siemens and Philips, when a patent cites one of the company s own lighting patents, in about 60-70% of the cases both patents belong to the same first class. Moreover, as we can see in Table 3, the same holds even when a patent cites one of the other two incumbents lighting patens (for example, for GE it would be citing patents of Osram/Siemens or Philips). Further, both GE and Philips refer to their own patents much more frequently than to patents of their competitors Since Osram has traditionally been a separate entity at Siemens, we were able to check if H01L patent belonged to Osram or some other subsidiary of Siemens. Before 2004, most of H01L patents of Siemens did not belong to Osram, but starting from 2004, almost all H01L patents of Siemens are filed by Osram. 10 The fact that Osram/Siemens rely more on competitors patents than on their own can be related to the lack of their own patents before the mid-1990s. 16