D.M.M. Meyer S2914182
Faculty of Economics and Business University of Groningen
2 ABSTRACT
The environment of invention has been experiencing an increasing amount of Burden of Knowledge. The rapid increase of knowledge stock led to a shift from solitary contributions towards more collaborative team effort for invention over the past 30 years. This shift led to an increase in expenses for companies to assimilate and recombine knowledge for inventions internally, thus leading to increased vertical separation in the innovation process. The
innovation process has therefore become one of transactions between entities to invent,
develop and commercialise an end product. Simultaneous uprising of Information Technology produces an interesting factor in this relationship, namely ‘E-collaboration’, which could be able to mitigate the need for larger teams. This research used various databases involving patents to estimate the effects of the Burden of Knowledge on patent transactions and the potential usage of E-collaboration to mitigate this relationship with a cross-sectional perspective using panel data. The results show that the Burden of Knowledge does affect patent transactions, and E-collaboration, contrasting the hypothesis, have a marginal aggravating effect on this relationship. Thus, showing the importance for policy makers, companies, and inventors for future studies towards knowledge generation and the potential influence of Information Technology.
Keywords
3
1. INTRODUCTION
Significant solitary contributions have been of all time: Albert Einstein, Isaac Newton, Alexander Graham Bell, Nikola Tesla; all of them made contributions to the advancement of our understanding of the world through inventions and ideas that thrived innovation. However, with this increase in knowledge it has become more difficult to assimilate and recombine knowledge as an individual. Knowledge stock becomes exorbitant creating a burden figuratively speaking. Such burden can be found in the rapid increase of knowledge which, as a consequence, increases the rate of complexity and scale for inventors to overcome. To state it metaphorically: ‘one must first climb the back of giants to be able to stand on the shoulders of giants’, implying one must first attain the knowledge stock of these giants before having the ability to expand this same knowledge stock (Jones, 2009). An influencing factor can be found in the increased rate of specialisation by inventors due to the difficulties with acquiring all knowledge in certain fields (Guimera et al., 2005). Inventors attempt to overcome this burden and specialisation gap by teaming up, making the process of invention a team-based affair in contrast to solitary inventions. Thereby, the number of inventors that have been appointed to a granted patent has increased drastically over approximately the past thirty years (Wuchty et al. 2007).
4
products exist of many different patented components, such as the iPhone which was introduced in 2007 and takes into account 57 patents to construct (Purcher, 2018). Consequently, a multitude of patents is needed to generate and commercialise a product with such high level of complexity (Wuchty et al. , 2007). Thereby, trading patents to accomplish commercialisation, also known as patent transactions, has rapidly increased in previous decades displaying a growth of markets for technology.
5
In addition to the research on patent transactions and increasing team size for inventions, researchers have found that cross-industry variation is another factor to take into consideration when looking at knowledge generation. Jones (2009) examined cross-sectional patterns to discover that teamwork and specialisation is greater in fields with deeper knowledge by means of creating the factor knowledge depth through the (sub-)categories of Hall et al. (2001). When looking at the visualisation of results by Jones (2009), it can be observed that large differences across technological categories with regard to team size exist, with high average team sizes for categories such as Chemical, Computers & Communications, Drugs & Medical, and Electrical & Electronic. Lower average team sizes can be observed in other categories. Similarly, Wuchty et al. (2007) found results pointing towards the same direction with research fields that displayed technological complexity experiencing a higher increase in team size than fields with lower technological complexity. These researches offer insight into the effects of knowledge on industries which have greater technological dependence, thus arguably display more complex knowledge stock. Therefore, it should be considered when researching the Burden of Knowledge whether industrial factors such as this technological dependence display more substantial effects.
6
notions on technological complexity and knowledge depth, which has not been further researched in the light of team size and the influence on patent transactions.
Therefore, the research objective proposed is to investigate whether the Burden of Knowledge does affect patent transactions, and how usage of IT for e-collaborative purpose in the invention stages could mitigate the size of this patent market. Additionally, a cross-sectional measure of patent classes will be investigated functioning as fixed effect for these relationships to understand better if industry differences have potential influence.
In order to research these relationships between the various concepts, the following research question will be handled as leading thread throughout this paper:
‘To what extent does patent team size influence the number of patent transactions, and how does E-collaboration mitigate this relationship?’.
The dataset used to explore the possible answer to this research question was constructed by merging various transformed databases from the United States Patents and Trademark Office (USPTO), forming an end product of the key variables which will later on be discussed. The dataset can be identified as panel data due to the inclusion of patent classes and a time component through the usage of application year for patents. Therefore, to analyse the research question a fixed effects approach for panel data was chosen which yielded four model specifications.
Results of these models show that the burden of knowledge does influence the number of patent transactions in a given classj and yeart, however, could not display statistical
7
The remainder of this research paper will follow the outline hereafter described. Firstly, the subsequent chapter will attempt to deliver a more elaborate view on the theoretical background on the previously mentioned topics. Thereafter, the hypothesis development section will include the various relationships between the concepts which will be explored. The data used and transformation needed to prepare for testing will be discussed in chapter four. The potential methods applied will be discussed in the chapter following that. Thenceforth, the result section will display the various regression analyses. Lastly, I will end this research paper by means of summarisation of key findings, display implications for theoretical and practical usage, and derive possible limitations as well as directions for future research. To conclude, the last chapter will bring the paper towards a conclusion.
2. LITERATURE REVIEW 2.1 Review on past literature
In the mid-20th century, Kuhn (1962) described in his essay on ‘the structure of scientific
8
Till now, no research has been done on the effects on technology markets caused by this shift towards teamwork. Thus, a contribution towards this will be attempted in this paper, showing that an increase in the Burden of Knowledge affects patent transactions in the technology market.
Previous papers on patent transactions have used a multitude of perspectives and definitions on markets. Whereas neoclassical economists see the market as “the meeting of
supplier of goods in an abstract place which determines the price of such good” (Marshall,
1961; p. 44), sociologists have accounted more for the human factor in markets. The market is perceived by sociologists as a mechanism for social structures in which structured exchange of market goods between actors takes place (Aspers, 2008). The latter definition will be adopted for this research due to the human factor which is needed to understand patent transactions on the premise of decision making. The market for patents has been researched before under various names, namely market for technology (Arora et al., 2001), market for ideas (Gans & Stern, 2003), or market for inventions (Conti et al., 2013). In addition, a definition for patent transactions has been proposed by Commons (1936, p. 12): ‘a patent transaction constitutes the transfer of ownership of proprietary knowledge embedded in such patent from one entity to another’. This definition will be used for purpose of research and will contribute towards the application in the field of knowledge generation.
Lastly, an increase in the amount of research papers directed towards the usage of information technology has skyrocketed in previous years. An entire new field of research has emerged, which caused digital collaboration or also denoted as E-collaboration to come into play. For purpose of this research the following definition of E-collaboration is used, as drawn up by Kock & D’Arcy (2002):
“Collaboration amongst several individuals whose goal is to accomplish a task together using
9
A number of studies show that E-collaboration has increased prominently, by allowing for higher frequency of technological collaboration across regional and national borders (Guellec & Van, 2001), virtualisation of the workplace (Pallot et al., 2005), and showing as well that increasing amount of patent owners are geographically dispersed (Cantwell, 1989). As seen by Lei et al. (2013) a shift occurs from domestic to international production of knowledge, allowing for trans-local knowledge recombination (Amin & Cohendet, 1999; 2004) and causing to break the favoured proximity rule for innovation (Lee et al., 2010). This research will contribute to the body of knowledge on E-collaboration by looking at the direct effects it has on the Burden of Knowledge.
In order to understand the multiple relationships between the three concepts ‘Burden of Knowledge’, ‘Patent Transactions’, and ‘E-collaboration’, I will expand in the following section on the focus of this research. I will propose various hypotheses on the basis of available literature on these concepts. I will first elaborate on the relationship between the Burden of Knowledge and patent transactions and will follow with the mitigating effects of E-collaboration on this relationship.
3. HYPOTHESIS DEVELOPMENT 3.1 The Burden of Knowledge and Patent transactions
10
Similarly, companies have found themselves in a similar position in which knowledge generation in-house is causing higher internal development costs. Transaction cost economics, developed by Williamson (1975, 1981, 1989), shows that when internal development costs accumulate too much, companies favour the exploitation of external development versus internal development to reduce these costs. This becomes more common through the increase of contracting versus integrating technology to avoid high internal development costs (Teece, 1986; Ghoshal & Moran, 1996).
This can be observed by the increase in research that claims that vertical separation of the innovation process takes place; in other words: the stages of invention, development, and commercialisation become more distant from each other. Such vertical separation has been seen in external knowledge commercialisation (Lichtenhaler, 2005; Arikan, 2009), the globalisation of knowledge production increasing the international efforts in the innovation process (Chaminade & Vang, 2008), and the increasingly dominant strategy of licensing-in by firms for innovation (Laursen et al., 2010).
The influence of this vertical separation has been monitored cross-sectionally in order to understand which industry incurred the highest vertical separation. Researchers found an increase in technology heavy industries, showing a similar trend on the Burden of Knowledge in these same technology heavy industries (Hall & Ham, 1999; Linden & Somaya, 1999; Arora & Fosfuri, 2000; Caviggioli & Ughetto, 2013). This similarity between the Burden of Knowledge for individuals and increased vertical separation shows influence on the process of knowledge generation and commercialisation for companies. As a consequence, the need for patent transactions have risen due to the increased difficulty for firms to create and develop inventions internally, creating a market for technology (Arora et al., 2001).
11
opt for externalising the invention and development stages of the innovation process. As a consequence, this vertical separation of innovation activities causes patent transactions in order to complete the commercialisation stage of said innovation process by companies. Therefore, the relationship between these two concepts is proposed:
H1: The Burden of Knowledge is positively related to the number of patent transactions 3.2 Mitigating effect of E-collaboration
Whereas previously the proximity rule for innovation (Lee et al., 2010) was the main direction to stimulate the innovation process, it is observed that the adverse effects of it causes detrimental knowledge production (Bathelt et al., 2004). Amin & Roberts (2008) noted that cultivation of firms occurs which offers firms a common language, however, also stimulates the Burden of Knowledge due to the lack of novel knowledge sources. This stimulation creates a barrier for entry of novel knowledge due to the continued usage of the same contacts within proximity. This in terms increases the specialisation gap due to constraints on the type of knowledge which is available.
12
geographical dispersion show highest impact on their respective research field (Jones et al., 2008). The introduction of information technology in the world of collaboration therefore has a reducing effect on the knowledge burden.
E-collaboration reduces the transaction costs ensued with knowledge generation thus mitigating the effect of the Burden of Knowledge on patent transactions. Therefore, the aforementioned can be formulated as the following hypothesis:
H2: E-collaboration mitigates the Burden of Knowledge, and consequently reduces the effect of the Burden of Knowledge on the number of patent transactions
A visualisation of the relations stated in both hypothesis 1 and hypothesis 2 can be found in the conceptual model in appendix A.
4. DATA 4.1 Data sources
13
Secondly, for the variable Burden of Knowledge a dataset considering inventor details, constructed by the USPTO, has been used. The data concists of 12,649,388 observations which were recorded between 1st of January 1975 till the 31st of December 2014. This dataset shows
patent ID’s connected to inventor data such as full name, street address, city, state/country code, and zip code.
Thirdly, the disambiguated database created by Morrison et al. (2017) was used to pinpoint more precise location details of inventors. The database shows a total of 25.3 million inventors, considering multiple patent agencies (European Patent Office, Patent Cooperation Treaty) next to the USPTO. This database was required since the USPTO database on inventors shows details on location in alphabetical format which is least favourable to use when calculating distance between these locations. Thus, this dataset provided coordinates per inventor, including a variable showing the accuracy of the inventor’s respective location.
Lastly, the Hall et al. (2001) NBER classification of patent classes into technological categories and sub-categories was used to control for differences between various technological classes. Additionally, a common control variable for patent researches can be found in backward citations, showing the amount of listed prior art per patent. In order to integrate this control variable, the data from the USPTO on patent backward citations was used showing citing patents and cited patents as variables.
4.2 Data transformation
In order to work with the various datasets and create an unified sample, transformative actions needed to be undertaken. The statistical programs Stata and R-studio have been used to perform these actions.
14
to operationalise the Burden of Knowledge in accordance with the previously mentioned literature, a measure was made which showed total team size per patent ID by generation of a counting variable named ‘Team_size’. Additionally, the NBER patent classification was added to the database to show the application date per patent and technological classes as well for the purpose of merging the final dataset.
Secondly, the disambiguated patent dataset by Morrison et al. (2017) was used to gather more precise data on geographical location of inventors through coordinates. These coordinates were needed to operationalise the variable E-collaboration by means of distance as a proxy for the usage of Information Technology. This operationalisation was derived from Tillema et al. (2010) showing that the higher the distance (measured in km’s) the greater the collaborative usage of information technology for knowledge transfer. Additionally, Pallot et al. (2010) showed the importance of spatial distance for E-collaboration to have an effect on creativity and thus knowledge generation.
The dataset was initially opened and transformed using Stata. The initial 25.3 million observations included were reduced to 11,796,560 observations through removal of patents granted by the European Patent Office and by the Patent Cooperation Treaty since these could not be matched with inventor data on team size and thus falling outside of the focus of this research. The coordinates provided in the database were allocated in one variable, thus transformation towards two separate variables was needed in order to calculate distance. A split of the variable ‘loc’ was performed which created the variables latitude and longitude. Thereafter, the variables considering ID, inventor names, and location quality were dropped in order to save memory for calculations.
15
export in csv format was produced and opened in the statistical code R to perform these calculations to circumvent the memory issues. The obtained variable from calculations shows the mean distance (in km’s) between all inventors included in one patent, using pairwise distance calculations (e.g. two inventors on one patent shows mean distance from A-B and distance from B-A). Thereafter, the dataset was collapsed to show the mean distance per patent.
The patent citation data was transformed to show the number of backward citations by means of generating a counting variable per observation and thereafter collapsing per patent identifier to show the sum of cited patents.
Thereafter, the three datasets containing patent team size, mean distance per patent, and backward citations were merged by using the common patent identifier in all. The database was collapsed to show the mean distance, team size, number of patent applications, and backward citations per technological classi and yeart.
The variable patent transactions was operationalised by means of the following formula: 𝑇𝑒𝑐ℎ𝑛𝑜𝑙𝑜𝑔𝑦 𝑀𝑎𝑟𝑘𝑒𝑡 =#𝑡𝑟𝑎𝑑𝑒𝑑 𝑝𝑎𝑡𝑒𝑛𝑡𝑠 𝑖𝑛 𝑐𝑙𝑎𝑠𝑠𝑖 𝑖𝑛 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛 𝑦𝑒𝑎𝑟𝑡
#𝑝𝑎𝑡𝑒𝑛𝑡 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠 𝑖𝑛 𝑐𝑙𝑎𝑠𝑠𝑖 𝑖𝑛 𝑦𝑒𝑎𝑟𝑡
In order to obtain this variable, the dataset on patent transactions was first collapsed based on the sum of traded patents in the equivalent classj and recorded yeart providing the
16 4.3 Sample statistics
All descriptive statistics and the correlation matrix for all variables can be found in table 1 and table 2, respectively. For the purpose of analysis, a sample of 3122 observations was used covering a period of 30 years (1980-2010) and 111 classes. The skewness of each variable indicates that the normality distribution is not present, which is common to see when using panel data.
Observations Mean Standard deviation
Min Max Skewness
Technology Market 3122 3.13 22.45 0.0018 502 12.68 Team size 3122 2.72 0.84 1 17 4.82 Distance 3122 274.89 245.00 0 7105.03 10.23 Applications 3122 561.58 751.29 1 5937 2.92 Transactions 3122 124.28 223.53 1 3224 4.45 Citations 3121 13.08 12.24 1 160.79 5.05 Year 3122 1995.67 8.47 1980 2010 - Class 3122 433.47 176.19 60 850 - Aclass 3122 34.03 10.50 21 49 - Aaclass 3122 3.08 0.92 2 4 -
Table 1: Descriptive statistics
17 Technology
Market
Team size
Distance Applications Transactions Citations Year
Technology Market 1.000 Team size 0.008 1.000 Distance 0.098 0.167 1.000 Applications -0.096 0.145 0.152 1.000 Transactions 0.300 0.114 0.219 0.448 1.000 Citations 0.218 0.064 0.271 0.047 0.362 1.000 Year 0.222 0.248 0.380 0.206 0.475 0.512 1.000 Table 2: Correlations 5. METHODS 5.1 Data analysis
18
Therefore, regression analysis with fixed effects was used in order to test for the relationship between variables in the dataset using the following regression formulae:
[1] 𝑇𝑒𝑐ℎ𝑛𝑜𝑙𝑜𝑔𝑦𝑚𝑎𝑟𝑘𝑒𝑡𝑡 = 𝛽1𝑇𝑒𝑎𝑚𝑠𝑖𝑧𝑒𝑡+ 𝛾𝑋𝑡+ 𝛿𝑡+ 𝑢𝑡
[2] 𝑇𝑒𝑐ℎ𝑛𝑜𝑙𝑜𝑔𝑦𝑚𝑎𝑟𝑘𝑒𝑡𝑖𝑡 = 𝛽1𝑇𝑒𝑎𝑚𝑠𝑖𝑧𝑒𝑖𝑡+ 𝛾𝑋𝑖𝑡+ ∝𝑖+ 𝛿𝑡+ 𝑢𝑖𝑡
[3] 𝑇𝑒𝑎𝑚𝑠𝑖𝑧𝑒𝑡 = 𝛽1𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑡+ 𝛾𝑋𝑡+ 𝛿𝑡+ 𝑢𝑡
[4] 𝑇𝑒𝑎𝑚𝑠𝑖𝑧𝑒𝑖𝑡 = 𝛽1𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖𝑡+ 𝛾𝑋𝑖𝑡+ ∝𝑖+ 𝛿𝑡+ 𝑢𝑖𝑡
In which i = patent class, t = time, α = class specific intercept, δ = time specific intercept, u = error term, and 𝑋 is a matrix that embeds time varying controls including the number of backward citations and the number of patent applications. The outcome of such analysis would yield eight models: the first model with the singular effect of team size, as a proxy of the Burden of Knowledge, on the size of the technology markets through fixed effects of time. The second model would add the control variables to the first model. A third model will be constructed again with the singular effect of team size on the size of the technology markets with addition of time and class fixed effects. The fourth model will consist of model three with control variables added. Similar rationale of model construction will be followed for the other four models in which model five displays a singular effect of distance on team size by means of time fixed effects, model six shows an addition of control variables to model five. Again, model seven will be the singular effect of distance on team size with time and class fixed effects, and lastly, model eight will enrich model seven with control variables.
19
Assumption under regression analysis is that the data used is homoscedastic and thus regression coefficients displays no skew towards one end of the regression. To test for this, I have used a modified Wald test for groupwise heteroscedasticity which is appropriate when considering time and entity fixed effects models (Baum, 2001). The results (see appendix D) displayed that all models with time and class fixed effects were subject to heteroscedasticity. Additionally, I performed a Breusch-Pagan test for heteroscedasticity for models that use time fixed effects only. Results yielded from this test (see appendix E) displayed that models one and two were subject to heteroscedasticity and model five and six show marginal heteroscedastic issues. Resolving this common issue for panel data can be done by using robust standard errors, which replaces the normal standard errors in each model with robust standard errors.
Additionally, a common issue for macro panel data (approximately 20 to 30 years) can be found in serial correlation. This causes the errors terms in yeart to be correlated with the error
term in yeart-1, thus causing inefficiency in the calculated models. To check for serial
correlation, I used a Wooldridge test for autocorrelation in panel data (Wooldridge, 2002). The results (see appendix F) show that model three and four display serial correlation through strong rejection of the null hypothesis. Model seven and eight do not display serial correlation, hence no action is needed for these models. Again, a common solution for overcoming serial correlation is to make use of robust standard errors to increase the efficiency of the models.
20 6. RESULTS
6.1 Hypothesis 1 testing: Fixed effects regression team size on technology market
As mentioned before, to test hypothesis 1 several models have been made using a fixed effects regression analysis considering class- and year-fixed effects. Table 3 shows the results with regard to hypothesis 1 which includes models 1 to 4. Models 1 and 2 have been performed with year fixed and without class fixed effects to monitor the possible between-class variation over time. Models 3 and 4 show a fixed effects regression analysis with both fixed effects included to test for within-class variations over time.
Model 1 displays an OLS-regression with merely the relationship between team size and technology market included. The beta parameter shows to be negative and does not display statistical significance (β= -0.287, p > 0.1). The second model displays the same relationship with inclusion of the control variables. A similar result is yielded for the beta parameter of team size, namely negative direction and it did not approximate statistical significance (β= -0.164, p > 0.1).
Model 3 displays a fixed effects regression with solely the effect of team size on technology market. In this model team size shows a positive, though insignificant effect on technology market (β= 0.598, p > 0.1). Again, in model 4 an addition of control variables is made by providing the variables citations and applications. The beta parameter for team size does change in the fourth model, however, still does not display significance (β= 0.0861, p > 0.1). In both models control variables do prove to be significant. The variable citations displays a positive beta parameter (β= 0.493, p < 0.01) which indicates that one extra citation in classj
and yeart adds 0.493 to the variable technology market. Thus, citations increase the size of the
technology market, ceteris paribus. Inverse, the variable applications displays a negative beta parameter (β= -0.00708, p <0.01) which means that by one extra applicationin classj and yeart
21
of the variable technology market includes the variable applications as denominator which consequently affects the equation. The models thus show that no statistical support can be found to support hypothesis 1 due to insignificance of the beta parameter for team size.
Table 3: Regression results model 1 - 4
(1) (2) (3) (4)
VARIABLES Model 1 Model 2 Model 3 Model 4
Team size -0.287 -0.164 0.598 0.0861 (0.573) (0.587) (1.004) (0.928) Citations 0.114 0.493*** (0.071) (0.102) No. Applications -0.00128 0.00708*** (0.0006) (0.00128) Constant 0.643 -0.245 1.509 0.408 (1.270) (1.476) (2.736) (2.595) Observations 3,122 3,121 3,122 3,121 R-squared 0.502 0.505 0.000 0.083 Adj, R-squared 0.497 0.500 0.000 0.082 F-statistic 12.64*** 7.95*** 0.36 15.86*** Number of class 111 111 111 111
Class FE NO NO YES YES
Year FE YES YES YES YES
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Dependent variable: Technology market
22
model 2 (F=7.95, p<0.01), inadequate fit for model 3 (F= 0.36, p>0.01) and good fit for model 4 (F= 15.86, p<0.01).
6.2 Hypothesis 2 testing: Mitigating effect of E-collaboration
Hypothesis 2 portrays that the usage of E-collaboration could have a mitigating effect on the Burden of Knowledge and consequentially reduces the effect of the Burden of Knowledge on patent transactions. Again, to test this, four models have been constructed which can be found in table 4, with the dependent variable team size. Again, model 5 and 6 are without class fixed effects to observe between-class variation over time and models 7 and 8 include these class fixed effects.
23 Table 4: Regression results model 5 - 8
(1) (2) (3) (4)
VARIABLES Model 5 Model 6 Model 7 Model 8
Distance 0.000299*** 0.000299** 0.000520*** 0.000392** (0.000116) (0.000118) (0.000178) (0.000158) Citations -0.00548*** 0.00730*** (0.00150) (0.00171) No. Applications 0.000092*** 0.000106*** (0.0002) (2.70e-05) Constant 2.171*** 2.170*** 2.582*** 2.462*** (0.1186) (0.1154) (0.0489) (0.0390) Observations 3,122 3,121 3,122 3,121 R-squared 0.080 0.090 0.028 0.046 Adj. R-squared 0.071 0.081 0.027 0.045 F-statistic 14.13*** 16.63*** 8.56*** 21.26*** Number of class 111 111 111 111
Class FE NO NO YES YES
Year FE YES YES YES YES
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Dependent variable: Team size
Again, all models needed the imputation of robust standard errors due to the heteroscedasticity present. With regard to the comparison of models we can observe that all models display adequate model fit with respectively model 5 (F= 14.13, p<0.01), model 6(F= 16.63, p<0.01), model 7 (F= 8.56, p<0.01), and model 8 (F= 21.26, p<0.01). However, the models do show differences in the amount of explained variance. When looking at the between-class variation we can observe that model 5 explains 7.1% of the variation whereas model 6 is able to explain 8.1% of the variation in team size. With regard to within-class variation Model 7 shows and adjusted R-squared of 2.7% whilst model 8 shows a higher 4.5%.
6.3 Post estimates testing and visualisation
24
tests in which the error term was estimated. The summarised results of this estimation can be found in table 5 which displays the estimated fixed effects (uit) under each model.
Model 3 Model 4 Model 7 Model 8
Mean 3.28^-9 -5.72^-9 -8.26^-10 9.22^-10
Standard dev. 3.66 5.76 0.47 0.47
Min -3.74 -11.81 -0.84 -0.84
Max 19.75 15.98 2.30 2.42
Table 5: Estimated fixed effects per patent class
One can observe that in the third model and fourth model large variations exist with regard to the estimated effects, thus showing that the variable technology market is able to vary considerably when taking into account the class and time fixed effects. Model seven and eight display less variation due to the estimated class and time fixed effects.
25
7. DISCUSSION 7.1 Key findings
The objective of this study was to clarify the effects of the Burden of Knowledge on patent transactions, and to study the mitigating factor Information Technology. On the one hand, these concepts were tested through usage of time fixed effects to discover between-class variation. On the other hand, addition of class fixed effects through insertion of patent classes to the formulae were used to test for within-class variation with regard to the tested concepts. In table 6 an overview of regression results is displayed.
Hypotheses Results
H1 The Burden of Knowledge is positively related to the number of patent transactions
Not supported
H2 E-collaboration mitigates the Burden of Knowledge, and consequently reduces the effect of the Burden of Knowledge on the number of patent transactions
Not supported
Table 6: overview regression results
26
year of recording. Therefore, it follows the trend of other research papers (Wuchty et al., 2007; Fiore, 2008; Jones, 2009; Bloom et al., 2019;) displaying the Burden of Knowledge for invention to hold true on the basis of increased team size, however not being able to infer any causal effects towards the firm level.
27
of Information Technology, however is contradicted by the need for breaking local language for novel knowledge by teams (Bathelt et al., 2004; Amin & Roberts, 2008), thus increasing the team size marginally instead of decreasing.
7.2 Theoretical & Managerial implications
Based on the key findings multiple implications directed towards team size, markets for technology and potential E-collaboration can be made. Firstly, I will discuss theoretical implications based on the hypotheses. Found in hypothesis 1 was a similar trend towards the increasing Burden of Knowledge, adding to the existing body of literature describing this increase. The addition of patent transactions to this concept has extended the literature by showing the possible influence of team size on the number of transacted patents in a given year and class. This influence, although found insignificant, shows potential for further research directed towards interactive effects between the knowledge generation rationale and the extent to which companies outsource their development of knowledge for end products.
Similarly, the extension of E-collaboration to the concept of the Burden of Knowledge shows that increased distance does not decrease, but contrarily increases team size for inventors. It therefore adds to the body of literature on E-collaboration by showing that potentially the availability of knowledge through IT could have the inverse effect of making it more difficult for local teams to assimilate and recombine the knowledge. Thus, it invokes the increase of teams by adding in trans-local knowledge to generate novel knowledge which is oftentimes needed for invention.
28
locating and resolving these bottlenecks by looking at the highest rates of patent transactions or team size.
Lastly, from a managerial perspective it is important to understand the manner these concepts influence the innovation process and thereby could impact a firm. It is widely known that outsourcing for innovation has been found to have an increasingly dominant place on company level, however the underlying theoretical knowledge on the potential reasons for this outsourcing to happen have been under investigated by managers. Possibly increasing costs of knowledge generation, by means of increased coordinative costs for teams, causes to increase transactional costs for commercialisation, thus inducing less potential profit after commercialisation.
7.3 Limitations and future research directions
Even though this research does propose some new insights for the knowledge generation and collaboration literature field, it needs to be considered that unequivocally some limitations are present. Firstly, a limitation can be found in the distribution of data. The data used was drawn from various databases which aided at the construction and operationalisation of the variables needed to perform the research. However, the data was not evenly distributed as is needed to increase the validity of the results obtained from the regression analyses. The data for example displayed heteroskedasticity and serial correlation in almost all models, which was solved by using robust standard errors. Nonetheless, robust standard errors are known to affect models by allowing for overestimation of the beta parameters and potential model misspecification (King & Roberts, 2015). Thus, a delimiting factor could be found in the usage of these and could be overcome when adding in better construct validity for variables.
E-29
collaboration, however more factors could be included to the construct to determine whether collaboration occurs, such as noted by Pallot et al. (2010) in the framework for collaborative distance. Spatial distance could in retrospect only capture a small portion of E-collaboration within patents and could also include other non-intended variations towards other constructs.
Similarly, the concept called the Burden of Knowledge was extensively discussed in this research, however, does not have any specified measurements and operationalisations. Suggested operationalisation in this research was by attempting to look at increasing team size as has been found in other papers. Contrary, one could also construct variables for such concept which take into account the expenditures on research and development, the time needed from idea to patent filing, or the number of different specialisations to account for the various skills included in a group of inventors for example. The first alternative would imply the usage of company data for example to impute for transaction costs made, whereas the second and third alternative could be matched with team size to improve the operationalisation of the Burden of Knowledge.
Thirdly, a limitation towards the contextual factors of patent transactions should be considered. The number of patent transactions has increased over time, however why such patents were traded needs to be further examined since it could be that transactions merely happen for purpose of protection of intellectual property towards competitors and thus is able to skew the numbers by doing so (Merges, 1999; Wang, 2010).
30
Additionally, the introduction of other patent categories, which have been left out of the scope in this research, could be of added value to the data since it could offer a better oversight of the potential differences amongst these various categories and enrich the data by adding possibility of variation. Thus, the suggestion is to add such categories when performing future research for the matter of data validity.
Next to that, the mitigating relationship which was observed in model five to eight surprisingly did not hold true, even though marginally, thus causes reason for additional research towards why the direction of the parameter for distance contributes positively to the Burden of Knowledge. The scope for future research should therefore be of adding to the construct validity of both concepts as mentioned before and potential confirmation of these primary results towards the aggravating effect of E-collaboration on the Burden of Knowledge.
31
8. CONCLUSION
Knowledge generation has become more difficult over time, for both individual inventors and companies that experience a burden. The need for collaboration to establish a new patent has increased in multiple industries which are coping with high knowledge depth. Simultaneously, the number of patents that are transacted by companies to be able to commercialise full products has fostered a vertical separation in the process of innovation. This research attempted to contribute to this research gap by looking at the relationship between the Burden of Knowledge and the number of patent transactions recorded in a given year. Additionally, a light was shed on the various effects per patent class, showing that variation between industries exists. It has been hypothesised that the Burden of Knowledge has a causal positive relationship with patent transactions. It was found that considering the direction of team size in the models the number of patent transactions do increase, however do not reach sufficient significance to infer causal relationship. Furthermore, it has been investigated if E-collaboration had a mitigating effect on patent team size and consequently could reduce the effect of the Burden of Knowledge on patent transactions. It was found that instead of a mitigating effect the inverse was true, and E-collaboration aggravates the Burden of Knowledge marginally. Therefore, to answer the research question posited before: ‘To what extent does
patent team size influence the number of patent transactions, and how does E-collaboration mitigate this relationship?’. It can be said that the number of patent transactions increases by
32
9. BIBLIOGRAPHY
Amin, A., & Cohendet, P. (1999). Learning and adaptation in decentralised business networks. Environment and Planning D: Society and Space, 17(1), 87-104. Amin, A., & Cohendet, P. (2004). Architectures of knowledge: Firms, capabilities, and
communities. OUP Oxford.
Amin, A., & Roberts, J. (2008). Knowing in action: Beyond communities of practice. Research policy, 37(2), 353-369.
Arikan, A. T. (2009). Interfirm knowledge exchanges and the knowledge creation capability of clusters. Academy of management review, 34(4), 658-676.
Arora, A., & Fosfuri, A. (2000). The market for technology in the chemical industry: causes and consequences.
Arora, A., Fosfuri, A., & Gambardella, A. (2001). Markets for technology and their
implications for corporate strategy. Industrial and corporate change, 10(2), 419-451. Aspers, P. (2008). Analyzing order: social structure and value in the economic sphere.
International Review of Sociology–Revue Internationale de Sociologie, 18(2), 301-316
Audia, P. G., & Goncalo, J. A. (2007). Past success and creativity over time: A study of inventors in the hard disk drive industry. Management science, 53(1), 1-15.
Bathelt, H., Malmberg, A., & Maskell, P. (2004). Clusters and knowledge: local buzz, global pipelines and the process of knowledge creation. Progress in human geography, 28(1), 31-56.
Bathelt, H., & Turi, P. (2011). Local, global and virtual buzz: The importance of face-to-face contact in economic interaction and possibilities to go beyond. Geoforum, 42(5), 520-529.
Baum, C. (2001). XTTEST3: Stata module to compute Modified Wald statistic for groupwise heteroskedasticity.
Cantwell, J. (1989). Technological innovation and multinational corporations. B. Blackwell. Caviggioli, F., & Ughetto, E. (2013). The drivers of patent transactions: corporate views on
33
Chaminade, C., & Vang, J. (2008). Globalisation of knowledge production and regional innovation policy: Supporting specialized hubs in the Bangalore software industry. Research policy, 37(10), 1684-1696.
Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative science quarterly, 128-152.
Commons, John R., 1936: Institutional Economics. In: American Economic Review 26, 237-249.
Conti, R., Gambardella, A., & Novelli, E. (2013). Research on markets for inventions and implications for R&D allocation strategies. Academy of management Annals, 7(1), 717-774.
Cummings, J., & Kiesler, S. (2003). Coordination and success in multidisciplinary scientific collaborations. ICIS 2003 Proceedings, 25.
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.
Gans, J.S., & Stern, S. (2003). The product market and the market for ideas:
Commercialization strategies for technology entrepreneurs. Research Policy, 32(2), 333–350.
Gans, J.S., Hsu, D., & Stern, S. (2002). When does start–up innovation spur the gale of creative destruction?. RAND Journal of Economics, 33(4), 571–586.
Gertler, M. S. (2003). Tacit knowledge and the economic geography of context, or the undefinable tacitness of being (there). Journal of economic geography, 3(1), 75-99. Ghoshal, S., & Moran, P. (1996). Bad for practice: A critique of the transaction cost theory.
Academy of management Review, 21(1), 13-47.
Glückler, J. (2007). Geography of reputation: The city as the locus of business opportunity. Regional Studies, 41(7), 949-961.
34
Guimera, R., Uzzi, B., Spiro, J., & Nunes Amaral, L. (2005). Team Assembly Mechanisms Determine Collaboration Network Structure and Team
Performance. Science, 308(5722), 697-702. https://doi.org/10.1126/science.1106340
Guellec, D., & de la Potterie, B. V. P. (2001). The internationalisation of technology analysed with patent data. Research Policy, 30(8), 1253-1266.
Hall, B. H., & Ham, R. M. (1999). The patent paradox revisited: determinants of patenting in the US semiconductor industry, 1980-94 (No. w7062). National Bureau of Economic Research.
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.
Hausman, J. A. (1978). Specification tests in econometrics. Econometrica: Journal of the
econometric society, 1251-1271.
Hofstede, G., & Bond, M. H. (1984). Hofstede's culture dimensions: An independent
validation using Rokeach's value survey. Journal of cross-cultural psychology, 15(4), 417-433.
Jones, B. (2009). The Burden of Knowledge and the “Death of the Renaissance Man”: Is Innovation Getting Harder?. Review Of Economic Studies, 76(1), 283-317.
https://doi.org/10.1111/j.1467-937x.2008.00531.x
Jones, B., Wuchty, S., & Uzzi, B. (2008). Multi-University Research Teams: Shifting Impact, Geography, and Stratification in Science. Science, 322(5905), 1259-1262.
https://doi.org/10.1126/science.1158357
King, G., & Roberts, M. E. (2015). How robust standard errors expose methodological problems they do not fix, and what to do about it. Political Analysis, 23(2), 159-179. Kock, N., & D Arcy, J. (2002). Resolving the E-collaboration paradox: The competing
influences of media naturalness and compensatory adaptation. IM-MUNCHEN-, 17(4), 72-77.
35
Landry, R., & Amara, N. (1998). The impact of transaction costs on the institutional structuration of collaborative academic research. Research policy, 27(9), 901-913. Laursen, K., Leone, M. I., & Torrisi, S. (2010). Technological exploration through licensing:
new insights from the licensee’s point of view. Industrial and Corporate Change, 19(3), 871-897.
Lee, K., Brownstein, J., Mills, R., & Kohane, I. (2010). Does Collocation Inform the Impact of Collaboration?. Plos ONE, 5(12), e14279.
https://doi.org/10.1371/journal.pone.0014279
Lei, X. P., Zhao, Z. Y., Zhang, X., Chen, D. Z., Huang, M. H., Zheng, J., ... & Zhao, Y. H. (2013). Technological collaboration patterns in solar cell industry based on patent inventors and assignees analysis. Scientometrics, 96(2), 427-441.
Lichtenthaler, U. (2005). External commercialization of knowledge: Review and research agenda. International Journal of Management Reviews, 7(4), 231-255.
Linden, G., & Somaya, D. (1999). System-on-a-Chip Integration in the Semiconductor Industry: Industry Structure and Firm Strategies. Draft. University of California, Berkeley.
Marshall, A. (1961). Principles of Economics: Text (Vol. 1). London; New York: Macmillan for the Royal Economic Society.
Merges, R. P. (1999). Institutions for intellectual property transactions: the case of patent pools. University of California at Berkeley Working Paper, 1-74.
Molinari, G., Sangin, M., Dillenbourg, P., & Nüssli, M. A. (2009). Knowledge
interdependence with the partner, accuracy of mutual knowledge model and computer-supported collaborative learning. European journal of psychology of education, 24(2), 129.
Morrison, G., Riccaboni, M., & Pammolli, F. (2017). Disambiguation of patent inventors and assignees using high-resolution geolocation data. Scientific data, 4, 170064.
36
Pallot, M., Prinz, W., & Schaffers, H. (2005, June). Future Workplaces, towards the"
Collaborative Web. In Proceedings of the 1st AMI@ Work–Communities Forum Day, Nottingham.
Pallot, M., Martinez-Carreras, M. A., & Prinz, W. (2010). Collaborative Distance: A Framework for Distance Factors.
Parker, H., & Brey, Z., Collaboration costs and new product development performance, Journal of Business Research (2015), http://dx.doi.org/10.1016/j.jbusres.2015.02.013 Purcher, J. (2020). Apple Won 57 Patents Today covering a Design Patent for the Original
2007 iPhone, Liquid Metal used in iPhones & more. [online] Patently Apple.
Available at: https://www.patentlyapple.com/patently-apple/2018/11/apple-won-57- patents-today-covering-a-design-patent-for-the-original-2007-iphone-liquid-metal-used-in-iphones-more.html.
Sampat, B. (2005). Determinants of patent quality: an empirical analysis. Columbia Univ.,
New York.
Stark, D. (2011). The sense of dissonance: Accounts of worth in economic life. Princeton University Press.
Teece, D. J. (1986). Profiting from technological innovation: Implications for integration, collaboration, licensing and public policy.
Tillema, T., Dijst, M., & Schwanen, T. (2010). Face-to-face and electronic communications in maintaining social networks: the influence of geographical and relational distance and of information content. New media & society, 12(6), 965-983.#
Wang, A. W. (2010). Rise of the patent intermediaries. Berkeley Tech. LJ, 25, 159.
Williamson, O. E. (1975). 1975 Markets and hierarchies: analysis and antitrust implications New York: Free Press.
Williamson, O. E. (1981). The economics of organization: The transaction cost approach. American journal of sociology, 87(3), 548-577.
37
Wilson, J. M., O’Leary, M. B., Metiu, A., & Jett, Q. (2005). Subjective distance in teams. In INSEAD Working Paper Series 2005/38/OB.
Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data MIT press. Cambridge, MA, 108.
Wuchty, S., Jones, B., & Uzzi, B. (2007). The Increasing Dominance of Teams in Production of Knowledge. Science, 316(5827), 1036-1039.
38
39 Appendix B
VARIABLE OPERATIONALISATION DESCRIPTION
TECHNOLOGY MARKET
Continuous Size of the technology market in class i and year t
TEAM_SIZE Continuous Mean team size in class i
and year t
DISTANCE Continuous Mean distance between
inventors per patent in class i and year t
YEAR Categorical Ranging from 1980 to 2010
CLASS Categorical Patent class
ACLASS Categorical Sub-category code
AACLASS Categorical Main technology category
APP Continuous Sum of applications in class
i and year t
TRANSACTIONS Continuous Sum of transactions in class
i and application year t
CITATIONS Continuous Sum of backward citations
in class i and year t
40 Appendix C
Model 3 Model 4 Model 7 Model 8
Chi-sq 82.93 82.93 13.73 13.73
P value 0.0000 0.0000 0.0033 0.0033
Table 8: Results Hausman test Appendix D
Model 3 Model 4 Model 7 Model 8
Chi-sq 1.1^8 2.0^7 34452.93 35654.34
P value 0.000 0.000 0.000 0.000
Table 9: Results Wald test for heteroskedasticity
Appendix E
Model 1 Model 2 Model 5 Model 6
Chi-sq 57383.28 57923.19 3.39 2.43
P value 0.000 0.000 0.065 0.1187
Table 10: Results Breusch-Pagan test for heteroskedasticity
Appendix F
Model 3 Model 4 Model 7 Model 8
F value 12.372 14.857 0.689 0.621
P value 0.0006 0.002 0.4083 0.4323
