The Burden of Knowledge and increasing patent trade: why it is more
apparent in the ICT sector
Master thesis Business Administration – Strategic Innovation Management
Martijn Cornelis Sietses
S3749304
Submission date: 16 June 2020
Supervisor: Pere Arqué-Castells
Co-assessor: M. Weck
Word count: 11,899
ABSTRACT
As the first study to do so, this study sheds light on a possible relationship between two well-documented phenomena; the Burden of Knowledge mechanism and the rise in Markets for Technologies. Furthermore,
this study shows how this relationship is particularly apparent in the ICT sector due to its unique characteristics. The team size and underlying knowledge of each patent are taken as proxy measures for
the Burden of Knowledge and the size of the Markets for Technologies is measured using a dataset including patent transactions. Using extensive USPTO datasets on general patent information, patent
citations and patent transactions, some interesting insights are revealed through stylized facts and regression analyses. First, a steady increase in the Burden of Knowledge is witnessed across fields between 1980 and 2012. Second, some correlation is found between the Burden of Knowledge and the rise
in Markets for Technologies between 1991 and 2001 indicating a possible link. However, this only relates to the measure of the underlying knowledge, no support is found regarding the measure for team size.
Finally, the data shows relatively strong support for the hypothesis that the relationship between the Burden of Knowledge and the Markets for Technologies is especially apparent in the ICT sector due to its
unique characteristics.
Keywords: Patents, patent transactions, Burden of Knowledge, Markets for Technologies, ICT,
1
1. Introduction
The importance of innovation for firms and the national economic growth is well-grounded in the
literature (Stokey, 1995; Feldman, 1999; Audretsch, 1995). Nevertheless, recent streams of research have demonstrated that innovation productivity is slowing down due to the so-called “Burden of Knowledge” (BoK) mechanism (Griliches, 1994; Jones, 2005a; Gordon, 2017; Bloom et al., 2017). This stream of research shows that coming up with inventions is becoming harder over time and because of this,
individuals have 2 choices; specialize or study more and longer. Specializing, means that they will need to work together with others in order to develop groundbreaking inventions. Studying more and longer, will delay their age of first invention. In fact, Jones (2005a) has shown a 17% increase per decade in the average inventor team size between 1975 and 1999, and shows that the average age of ‘great inventors’ and Nobel prize winners has increased steadily with around 6 years over the 20th century (2005b).
The fact that innovators increasingly specialize and cooperate could be a possible explanation for a second, and until now unrelated, phenomenon that has been rapidly growing in the last decades. The so-called Markets for Technology where companies trade or license patents and technologies with one another (Arora et al., 2001). Gaining a better understanding of these phenomena is crucial for policy makers, managers and economists as it can simplify making predictions for the future regarding market transactions, upcoming markets, coordination and teamwork, and it can have profound effects on the economic growth and the future workings of the economy. As the first study to do so, this paper aims, first of all, to shed light on a potential link between the BoK mechanism and the rise in markets for technology and, second, to show that this link is especially profound in the ICT sector due to its unique
characteristics.
The BoK mechanism is not a new phenomenon but gained in popularity in recent years. One of the first to show that the innovation productivity is declining is Griliches (1994). He demonstrated in his paper that the amount of patents per R&D dollar declined rapidly between 1920 and 1990. In 2005(a) Jones
introduces the BoK mechanism as a possible explanation to this problem and shows next to the decline in productivity of R&D also the increase of the age at first innovation, specialization and teamwork.
2
Markets for Technologies have increased rapidly in the last decades, especially in the high-tech industries (Arora and Gambardella, 1994; Linden and Somaya; 1999; Arora et al., 2001). This shows that companies increasingly rely on R&D of other firms and develop less in-house. The literature on markets for
technology is quite mature and mainly focuses on the effects of Markets for Technologies on the industry structure and firm-level strategies (e.g. Arora et al. 2001; Caviggioli and Ughetto, 2013). Nevertheless, the current literature is very limited regarding what caused these Markets for Technologies to rise on an industry wide level. This paper aims to add to the literature by providing insights into the mechanisms that caused these Markets for Technologies to rise.
Finally, this paper argues that due to its unique characteristics, the link between the BoK and the Markets for Technologies is especially apparent in the ICT sector. ICT has the unique features that it is mainly used for sharing and producing knowledge (Powell and Snellman, 2004), can be applied to many different areas, has a very wide underlying knowledge base (Corrocher et al., 2007) and caused a dramatic increase in the demand for skilled workers (Caselli, 1999). This has caused a rapid increase in its popularity in the last decades (Jorgenson et al., 2002; Colecchia and Schreyer, 2002) and likely caused the BoK to increase more rapidly in this technology field compared to other fields. If there is indeed a link between the BoK and the Markets for Technologies, and the BoK increased faster in the ICT sector compared to other sectors, then the rapid rise of the Markets for Technologies in the ICT sector documented in the literature (Marco et al., 2015), could possibly be explained by the rising BoK in this sector.
To shed light on these issues, multiple datasets including information on USPTO patents, patent
transactions and patent citations between 1980 and 2012 are being used. With approximately 4,4 million patents, 1,1 million patent transactions and 65 million patent citations, this study uses one of the most extensive datasets to date to explain these phenomena. Grouped at the aggregate technology class level, the BoK phenomenon is measured by an increase in the inventor team size and the underlying knowledge of each patent. In order to remove bias from differences in the absolute amount of patents, the size of the Markets for Technologies is measured as a ratio of the total patents. After showing the development of these phenomena separately using stylized facts, a possible link is investigated using regression analyses and a comparison between the ICT technology class and other classes is made.
3
knowledge of each patent and no correlation is found regarding the measure for inventor team size. For the other time periods (1980 – 1990 & 2002 – 2012) negative or no correlation is found between the two phenomena.
Finally, fairly strong evidence is found for the hypothesis that the aforementioned link is especially apparent within the ICT sector compared to other sectors. Using interaction terms between a dummy variable marking technology classes related to ICT and the measures for the BoK, this study shows positive and significant correlations for the period 1980 – 2001. Truncation and measurement issues possibly biased the data after 2000 and, therefore, no strong statements can be made after this period.
With these results, this paper brings two important contributions to the literature. First of all, proving the increase in the BoK mechanism between 1980 and 2012 using one of the most extensive databases in terms of size to date, extends the findings of Jones (2005a) who used a patent database on US patents between 1963 to 1999 and Bloom et al. (2017) who used case studies as well as firm-level data from Compustat to find evidence for the BoK mechanism. Second, shedding light on a possible link between these two phenomena and making the assumption about the ICT sector will advance the field and hopefully stimulate future research since it is the first study to do so.
2. Literature review
Since this research makes use of patent data, first, a short overview of the literature on using patents as a measurement for inventive activity will be provided. Patents are one of the best known mechanisms used by firms to protect newly created knowledge against imitation by rivals (Somaya, 2003). Admittedly, some downsides to using patent data exist. For example, not all inventions are suitable for patenting (Katila, 2000), which means that non-technological, managerial or organizational inventions are not measured using patent data. Also, patent data only captures a part of the patentable inventions since some firms choose to use other intellectual property protection mechanisms to protect their inventions.
However, overall patents are well suited for research purposes. The reason for this is that data on them is widely available, easily quantifiable and can be disaggregated into specific technological fields (Haščič and Migotto, 2015), the application and grant criteria are highly objective since patents have to be novel, non-obvious and useful (Katila, 2000) and they measure inventive output as opposed to input measures such as R&D data (Griliches, 1990; Trajtenberg et al., 1997).
4
measures such as R&D expenditure (only available for larger companies), sales data (only available for a limited amount of companies) and surveys (limits to quantity due to time and response rate constraints). Finally, although scientific publications have similar characteristics, patent data seems to be better suited since the purpose of this study is to measure the impact of the BoK on the rise on Markets for
Technologies. Markets for Technologies are by definition trades in patents between firms, scientific publications would not capture this.
3. Hypothesis development
The BoK mechanism is first introduced by jones (2005a) who describes the fact that it gets increasingly difficult for individuals to accumulate all the knowledge in a certain field in order to reach the knowledge frontier and come up with novel ideas. The reason for this is that the overall amount of knowledge available across fields is increasing and new ideas should therefore be based on increasingly larger amounts of prior knowledge in order to be relevant in the current time. This makes it harder to invent new things and, as described in the introduction, gives individuals 2 choices; specialize or study harder and for a longer time. Specialization has caused inventors to increasingly collaborate with others and has been shown to rapidly increase team size (Jones, 2005a). Next to this the average age of inventors has increased quite rapidly in the 20th century (Jones, 2005b), indicating that people indeed study for a longer time. Similar findings of team size growth have been reported in academic co-authorships across fields (e.g. Zuckerman and Merton, 1972; Hudson, 1996; Grossman, 2002) and recently Brendel and Schweizer (2019) have explicitly shown the BoK in academic research with increases in team size, age at first
publication, specialization and the number of backward references in academic research between 1970 and 2012. These increases in team size, age, specialization and the number of backward citations are all indicators that the BoK mechanism is present in academic as well as applied research.
In addition, Bloom et al. (2017), show that in the US the Total-Factor Productivity (TFP) growth rates have been relatively stable between 1930 and 2000 while the number of researchers has been increasing rapidly. This means that the research productivity has been declining rapidly in this period and that
inventing new things has become more difficult. In fact, they show that the research productivity in the US has fallen with an average of 5% a year between 1930 and 2000, indicating that it indeed gets harder to invent new things. These findings are confirmed at the firm level by Png (2019), who shows, that between 1975 and 1998 US firms have a decreasing number of facilities and are operating in a decreasing number of different fields while becoming more intensive in terms of human capital. Thus, although the
5
result, put more effort into each R&D field. If companies become more specialized on a certain field it will make it harder for them to create all the inventions needed for their final product in-house and, hence, incentivizes them to cooperate.
On top of this trend of inventing getting harder, innovation also has increased in speed. Technology life cycles are becoming shorter which makes older technologies obsolete at an increasing rate (Agarwal and Gort, 2001) and there is a growing importance for firms to quickly develop and commercialize inventions (Lynn and Akgün, 1998; Oxley and Sampson, 2004). This increasing rate of development is nicely embodied in Moore’s law, which basically states that computing power doubles every two years and therefore grows exponential. Recently, Bloom et al. (2017) showed that this exponential rate is accompanied by an even greater increase in research effort. In fact, they show that the research effort needed to achieve this doubling of the computing power was more than 18 times higher in 2014, compared to 1971. Hence, on the one hand there is an increasing need to innovate by firms due to rapid technological change and shortening life cycles; on the other hand it becomes harder to invent new things due to the BoK mechanism.
When ideas become harder to find and there is a high pressure to rapidly innovate, firms need to put more effort into R&D to get satisfying results, this leaves companies 2 choices; spend more on R&D, which is often not possible, or specialize on one area and collaborate with others to produce a final product. In the literature it is shown that there is an increasing division and specialization of R&D activities (Arora and Gambardella, 1994; Arora and Gambardella, 2010), which has caused an increase in R&D collaborations between firms (Hagedoorn, 2002; Caminati, 2016).
One way for firms to collaborate are the so called Markets for Technologies. Markets for Technologies can be defined as “transactions for the use, diffusion and creation of technology. This includes
transactions involving full technology packages (patents and other intellectual property and know-how) and patent licensing” (Arora et al., 2001; p. 423-424). For the purpose of this study, only the transactions involving patents are being taken into account due to the characteristics of patent data mentioned earlier. It is well grounded in the literature that Markets for Technologies are growing rapidly in the last decades (Arora et al., 2001; Arora and Gambardella, 2010; Marco et al., 2015).
The current literature mentions a number of different reasons for the rise in Markets for Technologies. However, none of them is likely to be the main cause. A short overview of the most important reasons mentioned in the literature will be discussed. First, a reason that is often mentioned is that defensive patent strategies by firms have created a sort of ‘arms race’ in order to have legal leverage vis-à-vis their
6
Technologies since this strategy is only possible on a large scale when patents can be purchased easily and is therefore more likely a result of, than a cause for, the Markets for Technologies. Also, pursuing this strategy is only feasible to large, resource rich firms due to its high costs.
Second, monetary reasons such as the selling of unwanted or non-core patents are often mentioned (e.g.
Brav et al., 2018). According to Monk (2009), this movement started with IBM announcing in 2003 that it earned $1 Billion solely from licensing. This caused many firms to reconsider their strategies and likely increased the amount of patent sales. The fact that this strategy is relatively recent and that a
well-developed patent market needs to be present to effectively perform this strategy makes it unlikely that this caused the Markets for Technologies to rise rapidly.
Finally, so-called exogenous reasons are mentioned in the literature. These can be divided into so-called legal and regulatory and market pull effects. Legal and regulatory reasons are often mentioned as a reason for a dramatic rises in patenting activity (e.g. Heller and Eisenberg, 1998). Specifically, the establishment of a specialized appeals court for patent cases called the Court of Appeals of the Federal Circuit (CAFS) in 1982 helped to stimulate protection through the use of patents by inventors. Nonetheless, Kortum and Lerner (1997) showed that the establishment of the CAFS was not responsible for the rise in patenting activity and patent trade. Therefore, it doesn’t seem a plausible explanation for the rise in Markets for Technologies. Market pull mainly relates to the mechanism that an increase in Markets for Technologies, in turn, stimulates further patent licensing and trade (Arora and Fosfuri, 2003; Fosfuri, 2006;
Lichtenthaler, 2011). In other words firms imitate one another’s patent strategies to remain the status quo. The market pull effect is therefore likely a result of the rise in Markets for Technologies but is unlikely to be the cause.
Hence, on the one hand there is an increasing need to innovate by firms due to rapid technological change and shortening life cycles; on the other hand it becomes harder to invent new things due to the BoK mechanism. This forces firms to increasingly specialize and collaborate with other firms in order to deal with the increasing difficulty of the invention process and to keep up with the technological change and shortening technology life cycles. This increase in collaboration is possibly the main reason why Markets for Technologies have begun to rise and continued to rise in recent years. On top of this, Markets for Technologies seem to have a self-reinforcing nature due to the ‘arms race’ and market pull mechanisms explained earlier which possibly strengthened the increase in the size of the markets for technology, but are unlikely to be the main cause for the rise of Markets for Technologies. Therefore, this study
7
Hypothesis 1: there is a positive relationship between the BoK mechanisms and the size of markets for
technology between 1980 and 2012
When looking at the changes of ownerships of patents between 1980 and 2012 (figure 8), one technological field clearly stands out. These are patents in the Computers & Communications or ICT category. While other markets of technologies in the US stayed relatively stable in terms of patent transactions after 2000, the change of ownership in the ICT category rose dramatically (Marco et al., 2015). The main purpose of ICT is to store and share information (knowledge) and it is well grounded that knowledge has grown in importance to the economy in the last decades (Powell and Snellman, 2004). In addition, research has shown that ICT makes research collaboration more efficient and, in turn, stimulates the creation of new knowledge (Forman and Zeebroeck, 2012) and patents (Dong and Yang, 2019). For this reason, ICT has been found to positively affect innovation activity in a wide range of different areas (Becchetti et al., 2003; Bertschek et al., 2013) and to stimulate patenting activity in many different technological fields (e.g. Gómez et al., 2017; Ravichandran et al., 2017). In fact, the ICT industry is seen by many authors as a so-called general purpose technology (GPT). This means that this technology can be applied to many different areas, there is a broad array of options to improve the technology and it enables easier invention and production of novel products in other fields (e.g. Bresnahan and Trajtenberg, 1995; Jovanovic and Rousseau, 2005). This makes ICT innovations highly valuable since they are applicable to many different areas, such as banking (Autor et al., 2002) or healthcare (Athey and Stern, 2000), and have therefore, gained dramatically in popularity since the 1990’s. In turn, this has caused heavy ICT
investments by many U.S. firms (Jorgenson et al., 2002; Colecchia and Schreyer, 2002).
This increase in popularity, investments and the fact that it is widely applicable, likely caused many inventors to choose to educate themselves in the field of ICT. As Jones (2005a) describes in his paper, every individual has a choice at the beginning of their career to become an innovator or a production worker. When choosing the innovator path, individuals are likely to choose a path that maximizes lifetime income. The rise of the ICT industry made this career choice increasingly popular and therefore likely caused more people to choose this educational path. Also, contradictory to other GPTs such as electricity or the steam engine, ICT has increased the demand for skilled workers instead of unskilled workers (Caselli, 1999). In fact, ICT patents have been shown to have a highly diverse knowledge base and in terms of actors involved and technological fields (Corrocher et al., 2007).
8
for Technologies in the ICT sector. Hence, based on the literature it is likely that the BoK mechanism is especially apparent within the ICT sector between 1980 and 2012 and consequently has caused the rapid rise in Markets for Technologies in the ICT sector.
Hypothesis 2: the effect described in hypothesis 1 is especially high in the ICT sector.
4. Data
For the purpose of this study, a number of different datasets are being used. First, a dataset including information on roughly 4,4 million patents issued by the USPTO between January 1977 and December 2012 (inclusive) will be used. This dataset includes two main sets of variables, one that comes from a data file including detailed information on the inventors listed on all patents issued by the USPTO between January 1977 and December 2012 (inclusive). From this file, the total number of inventors listed on each patent is extracted. The other data file, named BASIC_BIB_14_class includes more detailed information on more than 5,8 million patents issued by the USPTO between July 1902 and December 2012 (inclusive). However, only the data from January 1977 to December 2012 is complete and, therefore, only patents from this period are included. From this file, the application dates and technology categories at the 36-category level will be extracted. The 36-36-category level is a widely used classification of USPTO patents based on technology class developed by Hall et al. (2001). This record-level patent data will be collapsed on the aggregate 36-category technology class and application year level in order to allow comparison over time and between technology classes.
Second, a dataset on patent reassignments provided by the USPTO (Marco et al., 2015) which includes detailed information on approximately 6 million patent assignments and transactions recorded by the USPTO between 1970 and 2012 has been cleaned by Dr. Arqué-Castells to include only pure patent trades. This dataset includes, among other things, the patent identifier, names of the assignee and assignor, patent application dates, execution dates of patent trades, patent grant dates and the aforementioned 36-category technology classes. This cleaned dataset includes over 1,1 million transactions involving over 800 thousand patents and is collapsed on the 36-category technology class and application year level.
9
2000 the USPTO did not publish patent applications (106th congress, 1999; Quinn and Hernandez, 2000) excluding the patent applications will, therefore, avoid inconsistencies. After this, the dataset is merged with the BASIC_BIB_14_class dataset described earlier in order to include, the patent application year and the 36-category technology classes described by Hall et al. (2001). The final dataset includes nearly 65 million patent citations.
4.1. Operationalization of concepts
Below, the concepts for the BoK and the Markets for Technologies are further operationalized so they are suitable for further analysis.
4.1.1. Burden of Knowledge. The BoK mechanism will be measured using 2 variables that Jones (2005a)
showed to be the most representative proxy measures. These measures are the “team size” and the “tree size” behind each patent. For team size, first the number of inventors listed on each patent is measured and displayed as an integer number for each patent. After, the mean of these numbers is taken grouped by technology class on the 36-category level (Hall et al., 2001) and application year. This allows for comparison between technology classes and over different application years. The larger the number of inventors listed on each patent, the higher the BoK. A problem with this data is that the number of inventors per patent is positively (right) skewed, meaning that the mean is higher than the median. This is due to a large amount of patents with one or two inventors and a few with many inventors (up to 80).
10
R&D resources in this field. This can give a biased view since companies might increase the R&D spending, team size or put more experienced inventors on these projects (increases average age). The citation tree accounts for this bias since it is a backwards measure of the BoK and can therefore be less influenced by managers reacting to the attractiveness of the market.
Following Jones (2005), the logarithm of the number of nodes will be taken as a continuous measure of the ‘tree size’. The reason for taking the logarithm is that taking a cross-section, the raw node count is highly positively skewed, which possibly indicates that outliers in the upper tail dominate the analysis. The natural log of the node count contains these upper tail outliers. The argument Jones uses for using the natural log is knowledge depreciation. He argues that there is a decreasing influence as nodes grow older, making younger nodes more relevant than older nodes. Taking logs loosely captures this decreasing influence of older nodes that dominate the node count.
An issue with the node count is the possible bias due to truncation. The dataset contains solely data on patents issued after 1974. Therefore, only the recent part of the citation tree is visible. This decreases the accuracy of the measure of the underlying knowledge of patents closer to 1974. Furthermore, in 1980 the patenting process at the USPTO got computerized, making it easier to find and add citations by patent examiners (Hall et al. 2001). This could provide a possible bias for patents prior to 1980. This paper will therefore start the measure of the underlying knowledge from 1980 to reduce this inaccuracy.
Finally, the aggregation of the team and tree size at the class level might not capture the phenomenon that traded patents have on average a higher number of inventors than non-traded patents. Therefore the level of aggregation will, next to the class and year level, also be split for traded and non-traded patents. In this way both samples can be compared with one another and the difference between traded and non-traded patents can be shown.
4.1.2. Size of Markets for Technologies. In order to measure the size of technology markets, a similar
approach as Marco et al. (2015) use in their working paper on USPTO patents is being used. The size of the technology markets will be measured using the following formula:
Size tech. market it=No. of traded patents with application year t, in tech class i Total no. of patents with application year t, in tech class i t = patent application year from 1980 to 2012
i = tech class on the 36-category level described by Hall et al. (2001)
11
there are many patents with application year t, in tech class i, it is more likely the absolute amount of these patents traded also increases. The formula stated above will show a more balanced measure of the size of Markets for Technologies. A patent is identified as traded (1) when it changed ownership at least once in its lifetime and non-traded (0) when it did not change ownership.
Opposed to Marco et al. (2015), who use the grant year of the patent to measure the size of technology markets, this study uses the application year of patents. There are three reasons to do so; First, the application date indicates the actual end of the invention process instead of the end of the administrative process by the USPTO. Second, using the application year allows for comparing the BoK mechanism with the size of technology markets since it is a uniform measurement of the year. Finally, trade of patents might be dependent on economic contingencies. Measuring the size of technology markets based on the trade year would, possibly, give a biased view since it might be highly influenced by the state of the economy at the time of trading.
4.1.3. Control variables. In order to remove possible biases and to be able to provide more solid evidence
for the correlation between the BoK and the rise in Markets for Technologies, a number of control
variables will be used. First, certain technology classes have a higher general propensity to cite backwards than others (Hall et al., 2001). Therefore, two controls similar to the normalized measures used by Jones (2005a) will be used for the purpose of this research. In order to control for the differences in backward citations made across classes, the deviation of the technology class from the year mean number of
backward citations across classes, divided by the year standard deviation in the number of citations will be taken. The same will be done for the citation tree size so the deviation of the technology class from the year mean tree size, divided by the year standard deviation of the tree size.
12
4.2. Descriptive statistics
Before moving on to the discussion of the methods of analysis, first some descriptive statistics will be discussed as indicators for the BoK and the rise of Markets for Technologies. Jones (2005a) shows, among other things, an increase in team and tree size between 1963 and 1999. These findings will be extended to verify if this trend continues after 1999 as expected. Since both the team size and the average tree size are highly positively (right) skewed, the median and 75th percentile are looked at in addition to the mean team size. Also, the logarithm will be taken for the tree size for aforementioned reasons. Figure 1 displays the mean, median and 75th percentile of the team size across technological fields between 1980 and 2012. Over the whole period 1980 - 2012 a clear rise in the team size can be observed, which confirms the findings of Jones (2005a), and extents them until 2012. Also, between 1990 and 1995 the 75th percentile of the team size increased from three to four and a rise in the median team size can be observed between 2005 and 2010. When plotting the 95% confidence interval of the team size (figure 2), the line seems to steadily increase between 1980 and 2012, indicating that indeed inventor team size is growing. When looking at the increase in the tree size (figure 3), a clear rise can be observed up until 2000 after which it drops and stabilizes.
In figures 7 and 8 the size of the Markets for Technologies are displayed. The markets seem to rise for patents with application dates until 2000 and rapidly decline afterwards. This rapid decline is probably due to truncation issues which means that younger patents have had less time to be traded compared to older patents. Nevertheless, a rapid rise of the Markets for Technologies can be witnessed until 2000 confirming earlier findings in the literature (Arora et al., 2001; Arora and Gambardella, 2010; Marco et al., 2015). In addition, when looking at each technological field individually, it can be observed that the aforementioned trends are visible within each technological field separately and are not just the result of one technological field. Figures 5, 6 and 8 show the tree size, team size and the size of the Markets for Technologies
between 1980 and 2012 for each technological field at the 6-category level as described by Hall et al. (2001).
For the tree size and the size of the Markets for Technologies it seems that the technological field Computers & Communications or ICT shows overall higher values compared to the other fields, this can be seen more clearly in figures 9 and 10 where non-ICT fields are compared with ICT fields. Furthermore, an independent sample T-test using a dummy variable for the Computers & Communications field
13
field are on average based on a larger amount of prior knowledge and change ownership more often compared to other technological fields, while in the Chemical and Drugs & Medical sectors the average team size is the largest.
5. Methodology
For the purpose of this study, measurements on the aggregate technology class level will be used. The reason for this is that the BoK mechanism is an industry-wide phenomenon which is present within entire markets and is, therefore, hard to measure at the firm-level. The technology-class level of measurement seems to be the most appropriate for such aggregate-level phenomena. This is in line with other studies investigating the BoK phenomenon (Jones, 2005a) and studies researching the rise of market for technologies (Arora, 2010; Marco et al., 2015). Moreover, measuring on technology class level is more representative for the different knowledge frontiers that scientists operate in than firm-level
measurements. Firms might operate in many different industries and employ scientists with varying fields of expertise who are working on projects in many different fields. The firm-level measurement would not allow for testing whether the BoK mechanism, is present in different technology fields since it might group several technological fields together at the firm-level.
5.1. Hypotheses testing strategy
In order to test the first hypothesis, the statistical approach of regression will be used. More specifically, a multiple linear regression will be used to test whether a relationship exists between the two measures of the BoK (team and tree size) and the size of the technology markets. Multiple linear regression seems to be the most appropriate since both measures of the independent variable (BoK mechanism) and the measure for the dependent variable (the size of the Markets for Technologies), are measured on a continuous scale. When the correlations are positive and significant, the null hypothesis of H1 will be rejected.
14
period, are subtracted from the values at the end of the period (e.g. values of 1990 minus values of 1980). By purely measuring the variations during a 10 year period grouped by technology class and not the absolute values, it will be measured how the Markets for Technologies evolve within a certain class as the BoK increases in that same class. This will eliminate the bias brought by variances in absolute values across groups. For the panel data, technology class and year fixed effects are being taken into account in order to address this issue.
A second issue is the aforementioned bias due to truncation. As can be seen in figure 7 and 8, the number of patents transacted decreases rapidly near the end of the measurement period. This is due to truncation issues, since patents with a later application year have less time to change ownership compared to patents that have been around for a longer period. Marco et al. (2015) have reported similar findings and found that pre-grant transfer rates are the main driver for the patent changes in ownership and that post-grant transfer rates are fairly stable across technology fields with the exception of Computers &
Communications, which has experienced an increase. A fairly high percentage of patents have been reported to change in ownership up to 12 years after the grant date, which makes it difficult to adequately account for the time lag since the potential transfer period is fairly long. However, when investigating the data it can be observed that the issue of truncation seems to only appear around 2000 (size of technology markets is rapidly decreasing). Therefore, the final period of the increment data (2002 – 2012) seems to be less representative than the preceding periods. Also, in addition to the full period of panel data (1980 – 2012), a second analysis will be performed only taking into account panel data from the period 1980 – 2000. In this way, the issues regarding truncation will be limited to a minimum.
In order to test whether the results described above are higher for the ICT field as hypothesized in hypothesis 2, a dummy variable will be used in order to allow comparison between the coefficients of the technology fields. More specifically, the technology classes 21 (Communications), 22 (Computer
15
5.2. Empirical strategy
This report is mainly based on stylized facts; correlations from the regression analyses shown in the results section and Appendix I below are not inferring causation but solely correlation. The area of the BoK is very novel and the purpose of this paper is to advance the field, not to show causation. The researcher is well aware that a large amount of alternative explanations are possible to explain the rise Markets for Technologies.
6. Results
Table 5 and 6 (appendix I) discuss the regression analyses performed for the purpose of this study. When examining the Variance Inflation Factors of independent variables none of the values in any of the models exceeded 10 meaning that there are no issues related to multicollinearity. Furthermore, the residuals of the models using increments show an approximately normal distribution and the sample size of the panel dataset is sufficiently large (1.188) to assume normality based on the Central Limit Theorem. Also, approximately linear relations can be observed in the scatterplots between the different variables in all models. When performing the extension of the Breusch–Pagan test developed by Cook and Weisberg (1983), no issues related to heteroscedasticity are found for the sample period 1980 – 1990. However, for the periods 1991 – 2001 and 2002 - 2012 some issues related to heteroscedasticity are found. To account for this, robust standard errors will be used for the regression analysis in these periods. Moreover, the standard errors in the panel dataset will be clustered on the technology class level in order to account for the differences between the classes. In the first three sections of table 5 the increments within three periods are measured. In the regression using panel data, year and class fixed effects have been taken into account in order to control for yearly fluctuations and differences between technology classes.
16
Hypothesis 1 predicted a positive relationship between the BoK mechanism and the size of Markets for Technologies between 1980 and 2012. Tables 5 and 6 partly support this hypothesis. The relationships between the increments of the size of the tech market and the tree size are negative and significant -.143 (P<.01) for the period 1980 – 1990, positive and significant .175 (P<.05) for the period 1991 – 2001 and positive and non-significant .041 (P>.05) for the period 2002 – 2012. The panel data shows a positive but non-significant .013 (P>.05) relationship both for the period 1980 - 2012, as well as for the period 1980 – 2000 .002 (P>.05). The relationship between the increments of the size of the tech market and the team size is negative and non-significant -.109 (P>.05) for the period 1980 – 1990, positive and non-significant .128 (>.0.05) for the period 1991 – 2001 and positive and non-significant .008 (P>.05) for the period 2002 – 2012. The panel data show a negative and significant -.083 (P<.01) relationship for the period 1980 – 2012 and a positive non-significant .012 (P>.05) relationship for the period 1980 – 2000.
This means that using the cross-section increment data for the period 1980 – 1990, a contradictory finding to the one formulated in H1 is found. For the period 1991 – 2001, the hypothesis is supported by the tree size and not supported by the team size. For the period 2002 – 2012, the data shows no support for the relationship between the BoK and the rise in markets for technology. Regarding the panel data, the hypothesis is contradicted by the team size and not supported by the tree size for the period 1980 – 2012. For the period 1980 – 2000 no support is found for both variables.
Hypothesis 2 predicted that the relation between the BoK mechanism and the size of Markets for Technologies was especially strong in the ICT industry. In order to test this hypothesis, an interaction between the dummy variable ICT and the two measures of the BoK mechanism was added. The interaction between the tree size and the ICT dummy is positive and significant for the periods 1980 – 1990 .736 (P<.01) and 1991 – 2001 2.202 (P<.01) but negative and non-significant for the period 2002 – 2012 -.368 (P>.05). The panel data show a positive and significant .187 (P<.01) relationship for the period 1980 – 2012 and positive and non-significant .008 (P>.05) for the period 1980 – 2000. The interaction between the team size and the ICT dummy is positive and significant for the periods 1980 -1990 .749 (P<.01) and 1991 – 2001 7.125 (P<.01), but positive and non-significant .141 (P>.05) for the period 2002 – 2012. Finally, the panel data shows a positive and significant .165 (P<.05) coefficient for the period 1980 – 2012 and a negative and non-significant -.003 (P>.05) coefficient for the period 1980 – 2000.
17
Finally, figures 11 and 12 compare traded patents with non-traded patents on the BoK variables. The backwards citation tree size is larger for traded patents compared to non-traded patents over the full period 1980 – 2012, while the team size was lower for traded patents compared to non-traded patents before 1990 and higher between 1990 and 2012. This indicates that patents which are based on larger amounts of knowledge and have larger average team sizes are more likely to change ownership. In other words, innovation projects for which a larger BoK exists are more likely to collaborate compared to innovation projects with a lower BoK. This sheds some additional light on a possible link between the BoK mechanism and the rise in Markets for Technologies.
7. Discussion
This study is based on two phenomena which have been witnessed in prior literature. The increasing BoK, which makes inventing new things increasingly difficult and forces people to either study longer or to specialize and collaborate and the rise in the Markets for Technologies in the last decades. These
phenomena, and a possible link between them, have been empirically examined in this paper. This section will review the results and elaborate on them based on prior literature. Three conclusions can be drawn from this empirical work. First, the BoK mechanism seems to be increasing over the full analyzed period of 1980 – 2012. This finding extents the current literature which shows this increase only until 1999 (Jones, 2005a). Also, this research has shown the BoK mechanism using the most comprehensive dataset to date. This comprehensive use of data made it possible to draw strong conclusions about the increasing BoK phenomenon in the period 1980 - 2012.
Second, although weak, some correlation is found between the BoK and the rise in Markets for Technologies between 1991 and 2001. Nevertheless, this correlation only exists for the measurement regarding the underlying knowledge (tree size) and is not present in other time periods. Also, the panel data showed a negative correlation between team size and the size of the Markets for Technologies
18
account different factors which might have influenced the team size per patent and, second, adjust the measurement period for the markets for technologies in order to avoid truncation issues.
Third, relatively strong empirical evidence is found that the amount of underlying knowledge is higher in the ICT sector compared to other sectors. No support was found that the team size is higher in the ICT sector compared to other sectors. The largest team sizes were found in the Chemical and Drugs & Medical sectors, which might be explained by the fact that co-patenting is more common in these sectors due to the strong patent protection regimes (Hagedoorn, 2002). Also, the link between the BoK and the rise in Markets for Technologies is particularly apparent in the ICT sector. This indicates that the ICT sector has some unique features which have caused the BoK to increase more rapidly and, consequently, has caused the market for ICT technology to rise dramatically between 1980 and 2012. In the current literature these distinct features have been described (detailed description in section 2.2 hypothesis 2) and the dramatic rise of patent trade in the ICT sector compared to other sectors has been documented (Marco et al., 2015). However, it has never been shown that these distinct features have caused the BoK to rise more rapidly in this sector compared to other sectors and that this has caused the dramatic rise in patent trade. This study provides some preliminary evidence for this link in the ICT sector.
An unusual finding regarding the comparison of ICT with the other technology classes is that using the panel data, support is found for the period 1980 – 2012 but not for the period 1980 – 2000 and using the increment data, no support is found after 2001. This is possibly due to the fact that the Market for ICT rapidly rose prior to 2000 and rapidly declined afterwards (figure 8) due to truncation issues. This same decline can be witnessed in the tree size after 2000 (figure 5), which will be further elucidated below. The panel data regression until 2000 does not capture this while the regression until 2012 does, this might explain the difference in significance. The non-significance of the interaction terms after 2001 using increment data can be explained by the truncation issues with the size of the markets for technologies.
19
backwards citations to not been taken into account and likely caused the sudden drop in the tree size after 2000, especially considering the fact that a large part of the citations occur within the first few years after publication (Hall et al. 2001). For patents after 2000 this means that there is a high probability that they have patent applications in their backward citation lists which were not counted in this study. Future research should account for this and link the patent applications and granted patents so they are counted in a similar way across the full period in order to draw stronger conclusions about the data after 2000. Unfortunately, this was not possible during the course of this research due to time constraints.
Finally, although this study only uses patent data and the sample only includes patents filed in the US, the results generated in this study are reasonably generalizable for the invention process within firms. The reason for this is that a very large sample size is chosen, similar findings regarding the BoK have been found in other disciplines such as the academic world (e.g. Brendel and Schweizer, 2019) and patent data offers a comprehensive and objective measurement (see section 2.1 for a detailed explanation). Therefore, it is reasonable to expect the findings of this study to be generalizable to other settings as well.
8. Conclusion
This study investigates the interplay between the BoK mechanism and the rise of the markets for technologies between 1980 and 2012. Also, it is hypothesized that the relationship between these two phenomena is more profound in the ICT sector compared to other sectors. This is investigated using extensive datasets including information regarding USPTO patents issued between 1980 and 2012. The BoK mechanism is measured using the amount of nodes in the backwards citation tree of each patent and the inventor team size per patent. The markets for technologies are measured by taking the ratio of traded patents compared to the total patents per year and technology class.
Evidence is found for the continuation of the increase of the BoK until 2012. Also, weak evidence is found for a link between the BoK and the rise in the Markets for Technologies. Finally, this study shows that for certain periods a higher correlation exists between these phenomena for the ICT sector compared to other sectors. These findings suggest that a possible link exists between the BoK and the rise in Markets for Technologies and that the unique characteristics of the ICT sector might have influenced this
20
References
Agarwal, R., & Gort, M. (2001). First-mover advantage and the speed of competitive entry, 1887–1986. The Journal of Law and Economics, 44(1), 161-177.
Arora, A., & Fosfuri, A. (2003). Licensing the market for technology. Journal of Economic Behavior & Organization, 52(2), 277-295.
Arora, A., Fosfuri, A., & Gambardella, A. (1998). Licensing in the chemical industry. In Conference paper, Intellectual Property and Industry Competitive Standards, Stanford University.
Arora, A., Fosfuri, A., & Gambardella, A. (2001). Markets for technology and their implications for corporate strategy. Industrial and corporate change, 10(2), 419-451.
Arora, A., & Gambardella, A. (1994). The changing technology of technological change: general and abstract knowledge and the division of innovative labour. Research policy, 23(5), 523-532.
Arora, A., & Gambardella, A. (2010). Ideas for rent: an overview of markets for technology. Industrial and corporate change, 19(3), 775-803.
Athey, S., & Stern, S. (2000). The impact of information technology on emergency health care outcomes (No. w7887). National Bureau of Economic Research.
Audretsch, D. B. (1995). Innovation and industry evolution. Mit Press.
Autor, D. H., Levy, F., & Murnane, R. J. (2002). Upstairs, downstairs: computers and skills on two floors of a large bank. ILR Review, 55(3), 432-447.
Becchetti, L., Londono Bedoya, D. & Paganetto, L. (2003). ICT investment, productivity and efficiency: evidence at firm level using a stochastic frontier approach. Journal of productivity analysis, 20(2), 143-167.
Bertschek, I., Cerquera, D., & Klein, G. J. (2013). More bits–more bucks? Measuring the impact of broadband internet on firm performance. Information Economics and Policy, 25(3), 190-203.
Bloom, N., Jones, C. I., Van Reenen, J., & Webb, M. (2017). Are ideas getting harder to find? (No. w23782). National Bureau of Economic Research.
Brav, A., Jiang, W., Ma, S., & Tian, X. (2018). How does hedge fund activism reshape corporate innovation?. Journal of Financial Economics, 130(2), 237-264.
Brendel, J., & Schweitzer, S. (2019). The Burden of Knowledge in Mathematics. Open Economics, 2(1), 139-149.
21
Caminati, M. (2016). Knowledge specialization and R&D collaboration. Journal of evolutionary economics, 26(2), 247-270.
Caselli, F. (1999). Technological revolutions. American economic review, 89(1), 78-102.
Caviggioli, F. (2016). Technology fusion: Identification and analysis of the drivers of technology convergence using patent data. Technovation, 55, 22-32.
Caviggioli, F., & Ughetto, E. (2013). The drivers of patent transactions: corporate views on the market for patents. R&d Management, 43(4), 318-332.
Colecchia, A., Schreyer, P., 2002. ICT investment and economic growth in the 1990s: is the United States a unique case? A comparative study of nine OECD countries. Review of Economic Dynamics 5, 408–442.
Cook, R. D., & Weisberg, S. (1983). Diagnostics for heteroscedasticity in regression. Biometrika, 70(1), 1-10.
Corrocher, N., Malerba, F., & Montobbio, F. (2007). Schumpeterian patterns of innovative activity in the ICT field. Research policy, 36(3), 418-432.
Dong, J. Q., & Yang, C. H. (2019). Information technology and innovation outcomes: is knowledge recombination the missing link?. European Journal of Information Systems, 28(6), 612-626.
Feldman, M. P. (1999). The new economics of innovation, spillovers and agglomeration: Areview of empirical studies. Economics of innovation and new technology, 8(1-2), 5-25.
Forman, C., & Zeebroeck, N. V. (2012). From wires to partners: How the Internet has fostered R&D collaborations within firms. Management science, 58(8), 1549-1568.
Fosfuri, A. (2006). The licensing dilemma: understanding the determinants of the rate of technology licensing. Strategic Management Journal, 27(12), 1141-1158.
Gambardella, A., Giuri, P., & Luzzi, A. (2007). The market for patents in Europe. Research policy, 36(8), 1163-1183.
Gómez, J., Salazar, I., & Vargas, P. (2017). Does information technology improve open innovation performance? An Examination of Manufacturers in Spain. Information Systems Research, 28(3), 661–675.
Gordon, R. J. (2017). The rise and fall of American growth: The US standard of living since the civil war (Vol. 70). Princeton University Press.
Griliches, Z. (1990). Patent Statistics as Economic Indicators: A Survey. Journal of Economic Literature, 28(4), 1661-1707.
22
Grossman, J. W. (2002). The evolution of the mathematical research collaboration graph. Congressus Numerantium, 201-212.
Hagedoorn, J. (2002). Inter-firm R&D partnerships: an overview of major trends and patterns since 1960. Research policy, 31(4), 477-492.
Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER patent citation data file: Lessons, insights and methodological tools (No. w8498). National Bureau of Economic Research.
Haščič, I., & Migotto, M. (2015). Measuring environmental innovation using patent data. OECD Environment Working Papers, (89), 0_1.
Heller, M. A., & Eisenberg, R. S. (1998). Can patents deter innovation? The anticommons in biomedical research. Science, 280(5364), 698-701.
Hudson, J. (1996). Trends in multi-authored papers in economics. Journal of Economic Perspectives, 10(3), 153-158.
Intellectual Property and Communications Omnibus Reform Act of 1999, H.R. 3194, 106th Cong. (1999).
Jones, B. F. (2005a). The Burden of Knowledge and the “death of the renaissance man”: Is innovation getting harder? The Review of Economic Studies, 76(1), 283-317.
Jones, B. F. (2005b). Age and great invention (No. w11359). National Bureau of Economic Research.
Jorgenson, D. W., Ho, M. S., & Stiroh, K. J. (2002). Information technology, education, and the sources of economic growth across US industries. mimeo.
Jovanovic, B., & Rousseau, P. L. (2005). General purpose technologies. In Handbook of economic growth (Vol. 1, pp. 1181-1224). Elsevier.
Katila, R. (2000). Using patent data to measure innovation performance. International Journal of Business Performance Management, 2(1/2/3), 180-193.
Kortum, S., & Lerner, J. (1997). Stronger protection or technological revolution: what is behind the recent surge in patenting? (No. w6204). National Bureau of Economic Research.
Lichtenthaler, U. (2011). The evolution of technology licensing management: identifying five strategic approaches. R&D Management, 41(2), 173-189.
Linden, G., & Somaya, D. (1999). System-on-a-Chip Integration in the Semiconductor Industry: Industry Structure and Firm Strategies. Draft. University of California, Berkeley.
Lynn, G. S., & Akgün, A. E. (1998). Innovation strategies under uncertainty: a contingency approach for new product development. Engineering Management Journal, 10(3), 11-18.
23
Monk, A. H. (2009). The emerging market for intellectual property: drivers, restrainers, and implications. Journal of Economic Geography, 9(4), 469-491.
Oxley, J. E., & Sampson, R. C. (2004). The scope and governance of international R&D alliances. Strategic Management Journal, 25(8‐9), 723-749.
Png (2019). U.S. R&D, 1975-1998: A new dataset. Strategic Management Journal.
Powell, W. W., & Snellman, K. (2004). The knowledge economy. Annu. Rev. Sociol., 30, 199-220.
Quinn, B., & Hernandez, M. V. (2000). USPTO Will Begin Publishing Patent Applications. Retrieved March 27, 2020, from https://www.uspto.gov/about-us/news-updates/uspto-will-begin-publishing-patent-applications
Ravichandran, T., Han, S., & Mithas, S. (2017). Mitigating diminishing returns to R&D: The role of information technology in innovation. Information Systems Research, 28(4), 812-827.
Somaya, D. (2003). Strategic determinants of decisions not to settle patent litigation. Strategic Management Journal, 24(1), 17-38.
Stokey, N. L. (1995). R&D and economic growth. The Review of Economic Studies, 62(3), 469-489.
Trajtenberg, M., Henderson, R., & Jaffe, A. (1997). University versus corporate patents: A window on the basicness of invention. Economics of Innovation and new technology, 5(1), 19-50.
24
Appendices
Appendix I: Tables
Table 1: correlation matrix variables increments 1980 - 1990
Table 2: correlation matrix variables increments 1991 - 2001
Variables Min Max Mean Std. dev. (1) (2) (3) (4) (5) (6) (7)
(1) tree size .39 1.443 .782 .257
(2) team size -.152 .661 .333 .172 -.100
(3) size tech market -.045 .172 .034 .051 .595*** -.381**
(4) nor back cite -1.905 2.941 0 .888 .448*** .252 .219
(5) nor tree size -1.45 1.927 0 .746 .696*** .031 .248 .452***
(6) first occurrence 1902 1960 1941 15.886 -.157 -.455*** .105 -.171 -.306*
(7) ICT 0 1 - - .345** -.417** .817*** .054 .137 .068
(8) tech market 1991 .096 .206 .151 .028 .212 .088 .170 .194 .045 -.200 .371**
*** p<0.01, ** p<0.05, * p<0.1
Table 3: correlation matrix variables increments 2002 - 2012
Table 4: correlation matrix variables panel data 1980 - 2012
Variables Min Max Mean Std. dev. (1) (2) (3) (4) (5) (6) (7)
(1) tree size .185 1.085 0.755 0.192
(2) team size -.0274 .567 .287 .134 -.578***
(3) size tech market -.021 .085 .028 .028 -.690*** .412**
(4) nor back cite -.927 1.246 1.9e-08 .544 .574*** -.141 -.276*
(5) nor tree size -2.277 1.168 5.8e-09 .756 .536*** -.006 -.259 .511***
(6) first occurrence 1902 1960 1941 15.886 -.106 .011 .050 -.004 .002
(7) ICT 0 1 - - -.472*** .355** .475*** -.027 .030 .068
(8) tech market 1980 .045 .178 .118 .0292 .316* .004 -.578*** .228 .324* -.162 -.081
*** p<0.01, ** p<0.05, * p<0.1
Variables Min Max Mean Std. dev. (1) (2) (3) (4) (5) (6) (7)
(1) tree size -.264 .298 -.042 .122
(2) team size -.187 .564 .199 .149 .556***
(3) size tech market -.18 -.017 -.095 .036 .221 .078
(4) nor back cite -2.475 .996 0 .559 .499*** .259 .099
(5) nor tree size -.559 1.011 0 .286 .356** .239 .318* .245
(6) first occurrence 1902 1960 1941 15.886 -.311* -.423** -.069 -.146 -.217
(7) ICT 0 1 - - -.121 .051 -.776*** -.104 -.359** .068
(8) tech market 2002 .093 .323 .177 .054 .012 .137 -.809*** -.046 -.315* -.026 .854***
*** p<0.01, ** p<0.05, * p<0.1
Variables Min Max Mean Std. dev. (1) (2) (3) (4) (5) (6) (7)
(1) tree size 1.122 7.092 3.442 1.049
(2) team size 1.306 4.232 2.243 .5 .333***
(3) size tech market .031 .371 .15 .051 .241*** .057*
(4) nor back cite -2.004 4.304 0 .986 .218*** -.306*** .041
(5) nor tree size -1.8 4.599 0 .986 .265*** .021 .382*** .570***
(6) first occurrence 1902 1960 1941 15.671 .035 -.145*** -.032 .001 -.009
(7) ICT 0 1 .111 .314 .104*** .079*** .474*** -.101*** .408*** .068**
(8) tech market 1980 .045 .179 .118 .029 .082* -.092* .585*** .327*** .050 -.162*** -.081*
25 Table 5: multiple linear regression using increments and panel data 1980 -2012
Cross-sectional regression using increments Panel data with class & year fixed effects
Variables (Dep. var.: size tech. market it) 1980 - 1990 1991 – 2001 (Robust std. err.) 2002 – 2012 (Robust std. err.) 1980 – 2012 (Robust std. err.)
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4
Tree size -.112*** (.020) -.143*** (.049) .157*** (.042) .175** (.067) .079* (.040) .041 (.065) .007** .013 (.013) Team size .085*** (.027) -.109 (.121) -.134** (.054) .128 (.100) .045 (.029) .008 (.057) -.022*** (.006) -.083*** (.025) Nor. measure back citations -.008 (.009) .008 (.007) -.004 (.008) .009 (.007) .006 (.007) .000 (.007) .014 (.011) .006 (.006) .003 (.009) -.005 (.008) .001 (.007) -.005 (.009) .014** (.006) .013** (.006) .016*** (.005) .016*** (.004) Nor. measure tree size .000 (.006) .008* (.005) -.001 (.006) 0.008 (.007) .018 (.014) -.017 (.017) .011 (.013) -.017* (.010) .005 (.017) -.003 (.012) .001 (.015) -.005 (.015) -.007 (.005) -.008 (.005) -.007 (.006) -.008* (.005) First occurrence .000 (.000) .000 (.000) .000 (.000) 0.000 (.000) .001 (.001) .001 (.000) 0.000 (.001) .000 (.000) .000 (.000) .000 (.000) .000 (.000) .000 (.000) - - - - Size tech market start period -.532*** (.150) -.447*** (.107) -.540*** (.132) -.485*** (.137) .348 (.350) .089 (.321) 0.302 (.299) -.501*** (.147) -.537*** (.053) -.556*** (.494) -.561*** (.074) -.423*** (.127) - - - - ICT yes/no . -.263*** (.060) -2.061*** (.414) -.037 (.125) - - - - ICT x tree size .736*** (.093) 2.202*** (.419) -.368 (1.074) .187*** (.047) ICT x team size .749*** (.189) 7.125*** (1.258) .141 (.452) .165** (.066) Tree size x team size .169 (.151) -0.178 (0.111) .052 (.104) .003 (.006) ICT x tree size x team size -1.915*** (.279) -7.257*** (1.274) 3.006 (4.406) -.062*** (.021) R-squared .357 .688 .516 .781 .146 .430 .504 .844 .666 .712 .691 .717 - - - - N 36 36 36 36 36 36 36 36 36 36 36 36 1,188 1,188 1,188 1,188 R2 within - - - - - - - - - - - - .022 .058 .052 .353 R2 between - - - .158 .111 .136 .468
26
Table 6: Multiple linear regression using panel data 1980 - 2000
Variables (Dep. var.: size
tech market it)
Panel data with class & year fixed effects 1980 – 2000
(Robust std. err.)
Model 1 Model 2 Model 3 Model 4
Tree size .015*** (.003) .002 (.011) Team size .075*** (.014) .012 (.018) Nor. measure back citations .021** (.008) .021*** (.005) .018*** (.006) .011** (.004) Nor. measure tree size -.017** (.008) -.020*** (.005) -.016*** (.006) -.009** (.004)
ICT x tree size .008
(.090)
ICT x team size -.003
(.050) Tree size x team
size
.003 (.005) ICT x tree size
x team size .012 (.035) R2 within .049 .336 .264 .524 R2 between .280 .198 .003 .367 N 756 756 756 756
27
Appendix II: Figures
Figure 1: mean, median and 75th percentile of the team size across technology fields
Figure 2: the 95% confidence interval of the team size between 1980 and 2012
28
Figure 3: the mean logartithm of the backward patent citation tree size between 1980 and 2012
29
Figure 5: the mean log of the backward patent citation tree size between 1980 and 2012 by technological field
30
Figure 7: size of Markets for Technologies between 1980 and 2012
31
Figure 9: markets for technology comparison ICT versus non-ICT technological fields between 1980 and 2010 with 10 year intervals
32
Figure 11: mean logarithm of the node count comparison traded versus non-traded patents between 1980 and 2012