A study of the determinants of geographical concentration of inventors within firms ‘’The

Master Thesis MSc: BA – Strategic Innovation Management

A study of the determinants of geographical

concentration of inventors within firms

‘’The effect of innovative knowledge on the geographic distance between inventors within an organization’’

During the Covid-19 pandemic, working from home has become common due to government restrictions and nationwide lockdowns. This has forced firms to rethink their business models and the organization of their innovation activities. An important challenge for innovation companies is that remote work can be detrimental given the well documented benefits of geographical proximity for innovation. In this study I characterize the typology of firm that can relocate its workforce of inventors in scattered locations respecting social distancing, without suffering potential costs from decreased interaction. As well as provide guidance for legislators when deciding on measures in response to exogenous shocks and other crisis, to help keep the economy as stable as possible during difficult times. Whereas there is considerable research on the clustering of firms, there is fewer research on the exact location of inventors working for a given firm. My paper contributes directly to the latter topic. I propose that firms that rely on internal knowledge spillover and experience a high burden of knowledge achieve better results when their inventors are geographically close to each other. To test that assumption, I use a dataset including several million patent grants by the USPTO ranging from 1976 to 2006. These grants are collapsed at the firm level to give the research a different dimension. The results show a positive relationship between high internal knowledge spillover and inventor geographical concentration, and a negative relationship between a high burden of knowledge and inventor geographical concentration with highly innovative firms within my sample giving opposite results

Keywords: Innovation, The Burden of Knowledge, Knowledge Spillover, Patents, Inventor geographical concentration, Exogenous Shocks, Inventor

Word Count: 12,306

Jeroen Heerspink: S2952513, J.H.Heerspink@student.rug.nl 2020-2021 University of Groningen

Faculty of Economics and Business

2

CONTENTS

1. INTRODUCTION ... 3

2. LITERATURE REVIEW ... 5

3. THEORY ... 8

3.1.1 The Burden of Knowledge ... 8

3.1.2 Unintended spillovers ... 9

3.1.3 Reliability on physical infrastructure. ... 10

3.1.4 Tacitness of knowledge/communication costs. ... 11

3.2 Canonical typologies of firms ... 13

3.2.1 Geographically close. ... 13 3.2.2 Geographically scattered ... 14 4. DATA COLLECTION ... 15 4.1 Sources ... 15 4.2 Sample ... 15 4.3 Measurements... 15 4.3.1. Dependent variable ... 16 4.3.2 Explanatory variables ... 16 4.3.3 Controls ... 17 4.4. Results ... 18 4.4.1. Main results ... 18 4.4.2. Robustness Check ... 19 4.4.3. Descriptive Statistics... 20 5. DISCUSSION ... 25 5.1. Theoretical implications ... 25 5.2. Managerial implications ... 26

5.3. Limitations and future research ... 27

6. CONCLUSION ... 28

REFERENCES ... 29

APPENDIX ... 35

Appendix A. NBER category labels ... 35

Appendix B. Variable Labels ... 36

Appendix C. Distribution Control Variable ‘Firm Industry Type’... 36

3

1. INTRODUCTION

With the recent Covid-19 pandemic haunting the world and the consequences it brings to normal every day life, there is also a big impact on the way business is conducted around the globe. Look for instance at a phenomenon that has been long proven and studied in academic research, knowledge clusters (Moretti, 2019; Peri, 2005). These geographic clusters, such as US based Silicon Valley and the German based Baden-Wuerttemberg automotive cluster, benefit from knowledge spillovers that are caused by their geographical closeness and have been forced to change overnight. Due to the Covid-19 pandemic governments had to make the decision to stimulate every organization in their jurisdiction to work from home when possible. Whereas this obviously would not work for the factory employees at say BMW, their office personnel including the innovation department are stimulated and forced to work from their homes. This in theory would mean that they lose the benefit from being close together geographically. Furthermore, Enkel & Heil (2004) found that Cross-industry innovation leads to market disruption and more exploratory innovation. Therefore, knowledge clusters have become more and more differentiated in order to benefit not only from other players within the own industry but to benefit from valuable knowledge from other industries as well. Finding suitable knowledge and partners has thus been made much harder due to the fact that geographical knowledge clusters are losing their geographical closeness due to the Covid-19 pandemic. This begs the question what is needed for firms to not be hindered or even benefit from their inventors being geographically scattered.

4

to deal with exogenous shocks. And consequently, do inventors still need to be geographically close when communication tools such as Microsoft Teams, Skype and Zoom are innovating at a blistering pace and had to substitute face-to-face communication almost overnight?

Additionally, research by Griffith, Lee & van Rheenen on home-bias adds to this story. Home-bias is a phenomenon that has been widely accepted but not overly studied. The authors challenge the accepted way of thinking by using a patent database. Within their research the most predominant finding is that they find evidence that ‘’home bias has declined over time, being much stronger in the pre-1990 period than the post-1990 period’’ (Griffith, Lee & Rheenen, 2011, p238) This finding suggests that within the period of 1990 until 1999 information was more easily flowing over state and national borders. There is however an anomaly in this finding, both the High-Tech ICT sector and the Pharma sector are believed to still score high in this home-bias. Suggesting that these sectors would be harmed by having their inventors geographically scattered. Of course, in these aforementioned high-tech- and pharma sectors innovation goes quick, and whereas the internet was still relatively new and underdeveloped in the 90’s, the 21st _{century has brought huge innovation within these sectors and the} communication between firms in general. Besides, due to the current crisis situation many firms have been forced to participate in these giant leaps due to the new normal and work from home restrictions.

Another perspective provided by Agrawal, Cockburn & McHale (2006). Here the conclusion is that employees and thus inventors can be socially close without being geographically close. However, for this to work the employee/inventor should first develop a good relationship with the firm. This finding is particularly relevant when we consider the current Covid-19 pandemic and the need to work from home. Especially when taking into account the current example of Twitter where a statement was released that their employees are allowed to work from home indefinitely (BBC, 13th may 2020). Showing that being geographically close is not as important for High-tech firms as was previously understood by scholars such as Agrawal et al. who mention ‘’Yet the tendency for high technology industries (where knowledge-intensive inputs are essentially weightless) to be geographically clustered suggests that proximity to sources of knowledge flows as inputs to R&D is critically important’’ (Agrawal, Cockburn & McHale, 2006, p.589). Is it thus possible to create this social relationship when being geographically distant? And does this negatively affect innovation performance?

5

data for over 40 years (Marco et al. 2015). The purpose of this research is therefore to deepen and broaden the base of technological change and innovation research through the use of patent datasets by looking at the USA as whole, and deepening on the firm level. Consequently, this research aims to help closing the gap that exists in literature regarding inventor geographical clustering on the firm level.

To add to the patent research and to be able to test the hypothesis in my research using a quantitative research method I constructed a patent-based dataset of 3259 observations with data provided by the NBER, USPTO and Patentsview over the timespan of 1976-2006. Additionally, within this dataset I run the same tests on a shrunken down version including only firms that have more than 50 granted patents within the dataset. The analysis of both resulted in interesting and opposite findings for the complete dataset and the dataset with the highly innovative firms. With the consequent finding being that not all dimensions operationalized have the hypothesized effect on inventor geographical concentration.

To contribute to the research field, provide guidance to local and national governments and to fill the research gap that exists for firm-level inventor geographical concentration. I aim to answer the following research question within this paper and provide guidance in current uncertain economic times during the worldwide Covid-19 pandemic and help deal both firms and legislators with their difficult tasks of stimulating innovation and the economy as a whole:

How can a firm organize their inventive activities with regard to their inventors being geographically close/scattered and what could be the impact of an exogenous shock such as a

pandemic on these firms?

The theory within research will discuss conditions that should hold for firms to organize their inventors in close geographical proximity. Additionally, I describe the typology of firm that will be forced to keep their inventors in close geographical proximity and firms that do not rely on this geographical closeness. I focus on four dimensions that are key at explaining firm dependence on face-to-face interaction and thus inventor geographical concentration.

2. LITERATURE REVIEW

6

these flows due to the fact that this process does not leave a paper trial (MacPherson & Krugman, 1992) However, the culmination of these knowledge flows related to innovation are visible and a well understood part of research, in the form of patents (Jaffe, Trajtenberg & Henderson, 1993; Moretti, 2019). Within patents a trial of knowledge flows starts to appear when researching the citations to other patents that are attached in the application. After all, ‘’Innovations are often based on knowledge created within previous innovations’’ (Aghion & Jaravel, 2015, p. 553). Continuous innovation like mentioned by Aghion & Jaravel is an innovative cornerstone for intra-firm and inter-firm innovations alike.

The way this knowledge travels from inventor to inventor is possible in several different ways and with multiple intentions. Knowledge can be taught or demonstrated, like what happens in our education system for example, or it can unintentionally spillover from host to host during meetings, informal chats near the coffee machine, social interaction or imitation. The image of communication tools, and the internet replacing our social contacts both in our spare time and at work is one shown often in movies and stories especially regarding high-tech firms. However, like in other industries face-to-face (F2F interaction is just as important in high-tech sectors, especially because knowledge is often tacit and hard to codify (Griffith, Lee & van Rheenen, 2011). The recent Covid-19 pandemic has understated that F2F interaction is, in most industries, invaluable not only for the transfer of knowledge and the subsequent innovation but also for the mental health of innovators and colleagues alike.

With the knowledge that F2F interaction is important to transfer knowledge between two or more hosts, it begs the question what kind of influence an exogenous shock such as the Covid-19 Pandemic has on the results of organizations and industries that would normally rely on these F2F encounters. Consequently, how they deal with a disruption of their routines. Is it possible to substitute F2F contact, with all the additional informal interactions, with online communication tools such as Microsoft Teams, Skype and Google Meet? Which organizations and industries are up to that task and which do not have a choice? Distance innovation could, when looking at it from a different perspective, add to the innovation strength of a firm. After all, innovation with geographically distant innovators opens up the access to more specialized and diverse knowledge available worldwide (Tabata & Johnsrud, 2008).

7

Innovation can be seen as a form of interactive learning (Lundvall, 1992) within learning co-operation is important, this can be strengthened by social interaction and existing social capital. A high level of social interaction and thus social capital improves trust and enhances mutual knowledge. This has the effect that transaction costs of knowledge transfer decrease (Asheim, Coenen & Vang, 2007). This concept adds a sociological approach to standard economic mechanisms. Asheim, Coenen & Vang (2007), summarize the importance of this sociological approach for inventive activities with the following elements; ‘’Importance of interaction, networking, co-operation, social capital and spatial proximity’’ (p. 5). All these elements stimulate collective learning and the subsequent collective innovation. F2F interaction is a special process that in contrast to digital communication gives both parties the opportunity to simultaneously send and receive messages where feedback is instantaneous (Nohria & Eccles, 1992; Storper & Venables, 2004). This is important for the transferring of tacit knowledge so often needed for innovation. A concept adding to F2F is Buzz,

Bathetlt, Malmberg & Maskell (2004) describe Buzz as the following: Buzz refers to the information and communication ecology created by face-to-face contact, co-presence and co-location of people and firms within the same industry and place or region. This buzz consists of specific information and continuous updates of this information, intended and unanticipated learning processes in organized and accidental meetings, the application of the same interpretive schemes and mutual understanding of new knowledge and technologies, as well as shared cultural traditions and habits within a particular technology field, which stimulate the establishment of conventions and other institutional arrangements. Actors continuously contributing to and benefiting from the diffusion of information, gossip and new by just ‘being there’. (p. 38)

This process, as is the case with F2F communication, stimulates the transfer of non-deliberate knowledge also known as knowledge spillover. In summary F2F interaction allows for multi-dimensional communication and is thus the preferred way to transfer tacit knowledge between two or more actors. The literature teaches that Buzz can be transmitted both electronically and F2F (Asheim et al., 2007). However, in practice it is often difficult to create Buzz, especially when the bonding process with colleagues and the firm gets damaged by a prolonged period of geographical distance.

8

exogenous shock such as an economic crisis, conflict or a global pandemic and the possibility of actors in the inventive process being stuck at a relatively distant location compared to his colleagues increases, making is even more necessary to understand the impact of geographic closeness or scatteredness between inventors within a single firm.

With my research I aim to answer my research question and help both firms and governmental bodies with the choice of organizing inventive activities from a distance or to fully focus on bringing their inventors in close geographical proximity in the difficult times of the Covid-19 pandemic, other exogenous shocks and the future in general. Is it possible for all organizations to change their inventive activities to deal with such shocks and to potentially profit from knowledge outside of their conventional knowledge pools?

3. THEORY

Recently the world has changed in a drastic way, what started as a pandemic outbreak in China spread to the whole world in less than a year. Legislators had to intervene to keep the economy from collapsing and impose strict lockdowns in order for healthcare to keep functioning. The Covid-19 virus comes in waves and has an enormous impact on our day to day lives and the way firms organize their activities. This exogenous shock has forced employees from all types of industries to work remotely from their home. Additionally, it has had a massive impact on a firm’s activities which can be especially problematic for a firm’s inventive activities where F2F interaction is presumably relevant for the success of these activities. In this section I discuss conditions that should hold for firms to organize their inventors in close geographical proximity. I describe the typology of firm that will be forced to keep their inventors in close geographical proximity and firms that do not rely on this geographical closeness. I focus on four dimensions that are key at explaining firm dependence on F2F interaction and thus inventor geographical concentration.

3.1.1 The Burden of Knowledge.

9

and the ‘Death of the Renaissance Man’ is innovation getting harder?’’ With Leonarda Da Vinci as the stereotypical image of the Renaissance man.

The Burden of Knowledge is not a generic score that applies to every industry and research field. Instead it is a concept that can be used to explain the changing way of working within both research and industries alike. One of the most well-known examples is the invention of the micro processer. Malone (1995) describes that the invention of the microprocessor was the idea of one person but the invention of several people. Experts in their field were necessary to get from the first concept stage to actually realizing the design. In more and more industries these lone inventors do not possess all the knowledge or expertise to realize their inventions. This has resulted in a growth of average team size with regard to innovation in the U.S. with on average 17% per decade, and an increase of specialization of about 6% in the same decade (Jones, 2005). The Burden of Knowledge is real and we need bigger and better teams to deal with this concept in certain domains and industries.

Research with regard to teamwork has predominantly focused on F2F interaction and the results presented are thus related to this physical interaction between team-members (Montoya, Massey & Song, 2001). F2F teamwork and the created cohesion that comes with it has proven to improve an organizations performance including their inventive activities (Montez, Moreno & Morales, 2005) If this same level of performance or cohesion can be achieved in virtual teams has yet to be seen. Innovation in teams is therefore obviously more dependent on F2F interaction whereas the remaining lone inventors or Renaissance men can afford to be geographically and socially isolated from their organization and colleagues. Therefore, if an industry or organization relies heavily on teamwork to overcome the present burden of knowledge, F2F interaction is necessary to be efficient and effective in the inventive process and the inventors of the organization need to be in close proximity of the organization and each other.

3.1.2 Unintended spillovers.

Knowledge can flow in multiple different ways, both intended knowledge transfers but also through unintended spillovers. These knowledge flows are difficult to see and follow sporadically even called invisible, it leaves no paper trail (Macpherson & Krugman, 1992). However, is this really the case? Aghion & Jaravel (2015) mention that inventions are often based on previous inventions and knowledge. This thus shows a clear trail of a knowledge flow and could be the result of a knowledge spillover. A knowledge spillover is thus a knowledge flow that occurs between two or more parties where often the sharing party does not have the intention to share this knowledge. It has to be mentioned that there lies a negative stigma on the term spillover. However, in intra-firm cases this spillover is often positive of nature.

10

own is already valuable but becomes a multitude more valuable and rarer when recombination of this occurs. This recombination and cross-pollination are often difficult to achieve as employees frequently convey to group thinking within their own departments and expertise. It is therefore especially important for employees to socially interact with each other as informal talks work as a way to get to know the beliefs and interests of a colleague. This is a situation where unintended spillovers of knowledge occur. Consequently, achieving these informal talks, coffee breaks and F2F interaction are easier when these employees are all present at the same geographical location and can ‘bump’ into each other. ‘’after all, intellectual breakthroughs must cross hallways and streets more easily than oceans and continents’’ (Glaeser, Kallal, Scheinkman & Shleifer, 1992, p. 1127).

Even with the advancements of communication technology F2F interaction is still the preferable way of communication for the majority of the employees and is more valuable for the bonding experience. The more employees, within the same department and cross departments, interact with each other the more chance unintended spillovers of knowledge occur that gives the new owner of the knowledge more possibilities to recombine knowledge and stimulate their innovative efforts for the firm and the efficiency they can work at. This intra-firm mechanism can be mirrored to the inter-firm perspective that is well known and researched and has created knowledge clusters such as Silicon-Valley.

3.1.3 Reliability on physical infrastructure.

Infrastructure defined as ‘’The basic physical and organizational structures and facilities (e.g. buildings, roads, power supplies etc.) needed for the operation of society or enterprise’’ (Infrastructure, 2020) In my research I will focus more on firm specific infrastructure such as labs, production facilities and other specific research facilities. Some organizations need their inventors to be at a fixed location to use the tools available at these facilities.

An example of this reliability on physical infrastructure is the pharmaceutical firm Pfizer and their partner BioNTech, which in November 2020 communicated to the world that they have developed a covid-19 virus vaccine with a reliability of 90% (Pfizer, 2020). To develop this vaccine Pfizer and BioNTech relied on their high-tech laboratories with equipment, that is not easily transferred and sometimes even one-of-a-kind, to produce the vaccine and test the basics. To further test the vaccine, Pfizer and BioNTech had research centers start the testing on human subjects which in turn relies on their test subjects physically appearing at the research center. Both are difficult if not impossible with their innovators and researchers in scattered geographical proximity.

11

world’s largest and most complex scientific instruments’’ (About CERN, n.d.) The infrastructure they are most reliant on and what could be seen as one of the most impressive pieces of infrastructure in the world is the Large Hadron Collider (LHC), a 27-kilometre ring of superconducting magnets with the ability to boost particles at super speeds through the tube. Here they collide at four different locations to create new particles. As expected, this immense piece of infrastructure is dependent on a specific location and in turn the researchers working with the LHC are bound to that specific geographical location to conduct their research and to be able to innovate.

What these examples show is that physical infrastructure is essential in some industries and that these are linked to specific geographical locations. ‘’Geography, in other words, serves as a vessel in which entrepreneurs, venture capitalists, and other agents of innovation, organize an infrastructure that brings together the crucial resources and inputs for the innovation process’’ (Feldman & Florida, 1994). As some organizations are able to switch to a digital infrastructure, others will not or simply cannot afford to do so due to their reliance on physical infrastructure. For those last organizations the Covid-19 pandemic has been even more challenging or even meant that these organizations have not been able to innovate and operate, resulting in losses and potential bankruptcy. Therefore, it can be important for governing bodies and legislators to make exceptions on the rules and regulations that impede movement in response to exogenous shocks as the Covid-19 pandemic. If strict rules are put in place that forbids employees to go to their place of work it will mean that some industries and organizations will fail that rely on their physical infrastructure to be efficient and innovative. This can have far-reaching effect on the national- and global economy which could result to be even more impactful for the governing bodies and their territory than just the outfall of the virus.

3.1.4 Tacitness of knowledge/communication costs.

12

The transfer of tacit knowledge is more difficult than it is to transfer codified/explicit knowledge. E.g. a recipe is easier to share than 20 years of sales experience. Experience and thus tacit knowledge needs time, frequent contact and demonstrations for a transfer to be successful. Often, when graduates start with their first job, they get assigned a mentor who has experience in the field and can help teach this graduate what is important and how to deal with certain situations. The more often there is contact between the mentor and the mentee the more the mentee learns and the more tacit knowledge gets transferred. This is in line with research by Tsang (2002), who describes that intimate human contact (F2F) is necessary between a set of employees to be able to bundle the information they both possess. F2F contact is therefore important for employees and inventors to bundle and transfer their possessed tacit knowledge. Furthermore, F2F interaction is needed to be able to demonstrate the tacit knowledge that an inventor possess as this has been proven to be an easier way to transfer tacit knowledge (Teece, 1986). The aforementioned demonstrating is easier, especially in industries that work with prototypes, when inventors are in close geographical proximity.

The acquisition of the Absorptive Capacity that is necessary for tacit knowledge to be transferred between inventors will be accelerated when there is a bonding period between inventors and the firm itself (Subramanian & Soh, 2017). This period of F2F interaction helps inventors understand each-other and the goals of the firm. This will help them transfer knowledge that the firm and their co-inventors deem valuable. However, the more relationships the inventor has within the firm the more difficult it will be to create strong and cohesive ties that are necessary to transfer tacit knowledge (Wuyts & Dutta, 2014). Important is that these relationships are effectively managed and stimulated by mangers and other senior personnel (Li et al., 2012), only this way the maximization of tacit knowledge transfer is possible.

13

Considering all the relevant literature I bundle and hypothesize the following:

H1: A firm will cluster its inventors in close geographical proximity when at least one of the following conditions apply: I) To overcome the burden of knowledge teamwork is required.; II) Co-inventors spillovers are a relevant input in the innovative process; III) Physical infrastructure is required for innovation; IV) Knowledge is predominantly tacit of nature

Due to the complexity to operationalize both the explanatory variables considering the requirement of physical infrastructure for innovation and the tacitness of knowledge I will not be able to test these two mechanisms in my study. I will therefore focus on the first two mechanisms mentioned in the theory and H1. Accordingly, in the theory I have discussed all the possible mechanisms to offer a complete characterization of the typology of firm that needs to cluster inventors versus firms that do not. I would stimulate future research to add these variables to give an even more descriptive picture of the effects of the proposed mechanisms in my study.

As argued by van Aken, Berends and Van der Bij (2012) the use of the theory testing approach is based on already established literature streams where there is still a knowledge gap present. This is the first step in the theory testing approach. With this starting point in mind I will follow the other 4 steps given by the authors: (2) Find measurements for the variables used in the research; (3) Data collection and analysis of the data; (4) Interpretation of the analysis in step 3 in regard of the significance of the acquired analytic results; (5) Theorize the relevance of the findings for the existing theory and propose a follow-up (van Aken et al., 2012) Step 1 and 2 have already been followed in the introduction, literature review and theory sections above. From here on out I will focus on the data and the analysis of the data itself. Afterwards I will interpret the results from my analysis and discuss implications for several fields and further research. Thereafter, I will give my conclusion.

3.2 Canonical typologies of firms

To give some more context I will give a few examples of organizations/industries that do require their inventors to be geographically close and some that do not have that requirement. These conclusions will be substantiated through the use of the mechanisms explained in the theory section. Adding substance to the theory section and the hypotheses and give a starting point for the expected results.

3.2.1 Geographically close.

14

of sharing this tacit knowledge is through research centers such as the one at Porsche, where inventors can share their knowledge by demonstrating. The sharing of this tacit information is often unintended and happens during a normal conversation or a coffee break for example. These so-called unintended spillovers happen more frequently when inventors have the possibility to speak F2F and in an informal manner. This stimulates the cross-pollination of ideas, something that is of utmost importance in an industry that relies on the recombination of many different knowledge sources and technologies. Furthermore, to realize the innovative ideas, production facilities where prototypes can be created and tested in for instance wind tunnels are necessary making these firms heavily reliant on physical infrastructure that is accessible on only certain geographical locations. Lastly, teamwork is getting more important in this industry every day due to the Burden of Knowledge. More and more recombination of knowledge is necessary to create actual ‘new’ ideas in an industry that is relatively mature.

Bayer, another example of an industry that relies on their inventors being geographically close, with Bayer as the face of the pharmaceuticals industry. Just like Porsche, Bayer is situated and headquartered in Germany and has a long history. This history and experience can be a reason why these firms generally react well to an exogenous shock, although their size plays a massive role to. Within Bayer experience plays a big role, the more tests with regard to certain drugs are done the more understanding is created on the reaction of elements present. Often this is difficult to codify precisely and is thus tacit. This knowledge is more easily shared when innovators/chemists work together in the same lab with the corresponding informal chats. Here unintended knowledge spillover is realized. Moreover, these innovators/chemists are bound to their labs with expensive and sometimes one-off-a- kind machines. Tying these innovators to a certain location. Additionally, due to the complexity of the pharma industry the Burden of Knowledge might be even greater than it is for the automotive industry, the pharma industry goes back thousands of years and discovering new working combinations will get even more difficult in the future. Furthermore, the speed required for the Covid-19 Vaccine will make teamwork essential. An important sidenote for Bayer is that this current Covid-19 exogenous shock is quite profitable due to their use of labs and testing material.

3.2.2 Geographically scattered

15

gives a clear angle regarding my research. This could have several reasons. First, because of the relatively young age of the company it has been focused on the online domain and has more experience working via the internet plus the writing of code can be seen as codified knowledge that can be shared easily between developers. secondly, Twitter does not rely on physical infrastructure for their inventive activities however, it does rely on a physical location for their servers and a good digital infrastructure. Lastly, teamwork within twitter is less predominant and has already been focused on working via the internet/communication tools. Additionally, due to the relatively young age of this industry the burden of knowledge experienced in other industries is less predominant and innovation is thus ‘easier’ in this young industry.

4. DATA COLLECTION

Within this research, I will make use of several already existing databases. These databases are the NBER patent data file (Hall, Jaffe & Trajtenberg, 2001), a file that consists of all patens that were granted and applied for between the years of 1976 and 2006 at the United States Patents and Trademarks office. The COMPUSTAT database, this is a database that has been previously linked to the NBER patent data file and includes information about the firms behind the patent applications. Additionally, I will use a patent citation dataset provided by the NBER. To complete the data, I will use the data on inventors provided by Patentsview on their website.

4.2 Sample

All the datasets will be pooled in a single dataset. This merging will be done via the firm identifier observation GVKEY. The pooled data will consist of data describing values in the timeframe between 1976 – 2006. When the data is all present in the master file it will be collapsed at the GVKEY level and thus the firm level. Even though the data is collapsed in the dataset I use many more observations to produce each variable, citation data on patents will for instance be used to produce a measurement of internal knowledge spillover regarding patents present in a given firm. The average amount of inventors per patent per firm will be produced to be able to measure the present burden of knowledge in a given firm and so on For the analysis I will use the cross-walk of GVKEY-patent numbers provided by the NBER to match all the variables and create a final firm-level dataset on which I will base my full analysis. This results in a final dataset with a cross-section of 3,259 unique firms.

4.3 Measurements

16 4.3.1. Dependent variable

Inventor geographical concentration. For this variable I will make use of a common measure of statistical dispersion, the Gini Coefficient, also called the Gini index/ratio (Lambert & Aronson, 1993). This measurement is most often used to measure income inequality but can be used to measure inequality among values of frequency distribution. Including in my case the inequality of distance between inventors within a specific firm. To create this measurement data on the geographical location of the inventor will first be linked to a firm indicator (GVKEY). For the location of the inventor I will use the county associated with the inventor instead of the current state as this give me a more in-depth variable. I create two different indexes on geographical concentration per GVKEY based on the share of inventors of a given GVKEY across US counties.

Cindex: Sum of squared shares. This variable will be used in the models I will do a regression analysis on because this specific concentration index adjusts for the fact that a firm might have inventors in different locations however, most will be in only one location. This measurement is preferred over my second measurement bcindex due to the adjustment for the fact that a firm might still have inventors in many locations, but have most of the inventors concentrated in just one location. The bcindex measurement explained below is simply based on the number of locations in which the firm has inventors, it does however not exploit the distribution of inventor shares across locations. That is why cindex will be used in my main regressions whereas bcindex will be used in my robustness checks.

Bcindex: 1 – (Total number of locations the firm employs at least one inventor) / (Highest number of locations in which a firm employs inventors). This variable will be used in robustness checks but not in the main regressions of my models.

4.3.2 Explanatory variables

Internal Knowledge spillover. For the internal knowledge spillover variable, I will use the measurement presented by Hall et al. (2001) where citations to patens of ‘others’ are close to the specific definition of knowledge spillovers. ‘’If patent B cites patent A, it implies that patent A represents a piece of previously existing knowledge upon which patent B builds’’ (Hall et al. 2001, p. 14) Within this measurement I focus specifically on self-citations that are made within the firm, these resemble an intra-firm knowledge spillover. To measure this, I will use the total of self-citations in patents within the intra-firm over the total number of citations in firm patents. To construct this variable, I will use Citation data related to patents presented by the NBER. With this dataset I will calculate the percentage of self-citations from the total self-citations on the GVKEY (firm) level.

The Burden of Knowledge. As teamwork is quite an abstract construct this measurement will be

17

measurement on the GVKEY level will be created to be used in the analysis. The more co-authors and inventors the patents receive the more teamwork was necessary to come up with the invention and thus the application for the patent. Inventors that have had a big enough share in the development of the patent will want to be named in the application because this helps their reputation as an inventor. Therefore, I believe this measurement will give a good indication in the teamwork necessary within a firm’s inventive activities and the consequent Burden of Knowledge associated with the firm.

I have not been able to construct variables for the measurements described below due to the complexity in operationalizing them. As I find the measurements important to the theory of my research, I have chosen to include them in this study. In future research I would attempt to operationalize them as follows:

Tacitness of knowledge / Communication costs. Tacitness is difficult to measure because the very

definition of tacit knowledge implies that it cannot be easily measured or codified. Therefore, I have chosen one of the few measurements used in academic research before. Within the patent dataset I will use the number of academic inventors that are used as co-authors / inventors of the patent application (Lee, 2012). Within Lee’s research academic inventors are inventors that work for Universities and other academic institution and thus not exclusively for the firm itself. Due to the lack of time and available dataset this measurement has not been operationalized in my research. I would urge future research to try and include this variable to add even more meaning to the results of this research.

Dependence on physical infrastructure. This variable is so reliant on the industry type that there

are as far as I know no academic measurements for this variable. However, due to my control variables including industry type this variable will be included with less depth. There is no clear-cut way to tell if a firm relies on certain physical infrastructure as this is too firm specific. Therefore, just as the tacitness of knowledge variable, this variable will not be able to be operationalized within my research.

4.3.3 Controls

18

total number of patents granted per firm in my dataset. With this control I will be able to make a distinction between firms that are perceived as highly innovative and firms that are not. To achieve this, I will also test my models with a restriction of a minimum of 50 granted patents per firm. Finally, I have tried to operationalize a variable on ‘R&D expenditures’ however, due to limited data in COMPUSTAT the addition of this variable would have rendered my existing dataset useless due to the amount of missing values.

4.4. Results

Finally, I will test my hypotheses. To test this hypothesis, I will perform a multiple regression analysis. Due to the nature of the data used, hierarchical regression will be most suitable to test H1. Therefore, the first step will be to only include my control variables ‘Firm Size’, ‘Firm Industry type’ and ‘Patent Amount’. Next, the independent variables ‘The Burden of Knowledge’ and ‘Internal knowledge spillover’ will be added. additionally, a Robustness check will be done. Within this regression a positive and statistically significant coefficient on ‘internal knowledge spillover’ accepts the hypothesis that internal spillovers are important for the clustering of inventors and vice versa. The same holds for a positive and statistically significant coefficient on ‘The Burden of Knowledge’, that accepts the hypothesis that a higher perceived Burden of Knowledge is important for the clustering of inventors and vice versa. If both ‘internal knowledge spillover’ and ‘The Burden of Knowledge´ have a positive and statistically significant coefficient H1 as a whole (with the exception of the non-operationalized variables) will be accepted. To be even more thorough in my testing I have created a second dataset that includes al firms that have 50 or more granted patents in the master data. This way I will be able to distinct between the full dataset and firms that are considered highly innovative in the 1976-2006 period. 4.4.1. Main results

Due to aforementioned data issues I will be unable to operationalize all variables and thus also I will not be able to test my complete H1. Therefore, I will be forced to focus on testing the two explanatory variables I was able to operationalize.

In Model 1 I only included my dependent variable ‘Inventor geographical concentration’ and all my control variables ‘Firm Size’, ‘Firm Industry Type’ and ‘Patent amount’. This can thus be seen as a baseline model. All control variables show neither a positive nor negative result in relationship with my dependent variable. Neither control variable shows any sign of significance in this model. To summarize model 1 in table 3 both of my control variables on their own have little to no influence on my dependent variable ‘Inventor geographical concentration’.

19

concentration’ and ‘Internal knowledge spillover’ (β = 0.19, p > 0.01) and a negative relationship between ‘Inventor geographical concentration’ and ‘The Burden of Knowledge’ (β = -0.04, p > 0.05)

Model 4 in Table 3 includes my dependent variable and all my control variables, it also introduces both my independent variables. Here a difference is visible between the relationships of my independent variables ‘Internal Knowledge spillover’ and ‘The Burden of Knowledge’ with my dependent variable ‘Inventor geographical concentration’. Whereas the former shows a positive significant relationship (β = 0.17, p > 0.01) the latter shows a relatively small negative significant relationship (β = -0.03, p > 0.05). The effect of both independent variables on my dependent variable has dampened slightly from both my 2nd and 3rd model. The mixed process model does therefore not provide support for Hypothesis 1 completely. Assumption I of Hypothesis 1 is rejected whereas assumption II of Hypothesis 1 is supported.

To summarize Model 4 in Table 2, there is little to no influence of my variables on the geographical concentration of inventors with in a specific firm as only 0.70% of the variance in my dependent variable ‘Inventor geographical concentration’ can be explained by my control- and independent variables. In Table 2 I present the main effect for the determinant ‘Inventor geographical concentration’. Here I observe that the Log-likelihood of Model 4 in relation to Model 1 increases from LL = -2220 to LL = -2218. Although this is a small increase it does confirm the goodness of fit from Model 1 to Model 4.

To be able to give more distinction to the findings I corrected my dataset to only include highly innovative firms in my sample (Patent amount>50) and ran my analysis again. This resulted in a dataset of 1233 firms and some interesting findings. Whereas my control variables have not given significant results in the complete dataset, in my highly innovative dataset ‘Firm industry type’ has a negative significant relationship with my dependent variable (β = -0.04, p > 0.05) in model 5 and (β = -0.04, p > 0.05) in model 7. Even more interesting though are the results when adding my independent variables to the models. Both effects of the variables seem to have turned the other way around. ‘Internal knowledge spillover’ shows a negative non-significant relationship with my dependent variable in model 8 (β = -0.12,) whereas ‘The Burden of Knowledge’ shows a positive significant relationship with my dependent variable (β = 0.14, p > 0.01). As a result, Assumption I of Hypothesis 1 is neither rejected nor supported. Assumption II of Hypothesis 1 is supported in my dataset consisting of the top inventive firms in my dataset.

4.4.2. Robustness Check

20

spillover’ and my dependent variable ‘Inventor geographical concentration’ is no longer positive but still significant (β = -0.28, p > 0.001) against (β = 0.17, p > 0.01) in Model 4. The relationship between ‘The Burden of Knowledge’ and ‘Inventor geographical concentration’ however stays negative but loses its significance (β = -0.02) against (β = -0.03, p > 0.01) in Model 4. For the highly innovative firms the following changes were reported: The relationship between ‘Internal Knowledge spillover’ and my dependent variable ‘Inventor geographical concentration’ is negative significant (β = -0.54, p > 0.001) against (β = -0.12) in Model 8. The relationship between ‘The Burden of Knowledge’ and ‘Inventor geographical concentration’ however stays positive but loses its significance (β = 0.03) against (β = 0.14, p > 0.01) in Model 8.

TABLE 1

Pearson Correlation Matrix

1 2 3 4 5 Dependent variable (1) Inventor geographical concentration 1.00 Independent variables (2) Internal knowledge spillover .0608** 1.00 (3) The Burden of Knowledge -.0456** -.2080** 1.00 Control variables

(4) Firm industry type -.0066 .2092** -.3081** 1.00

(5) Firm Size -.0554** -.0146 .0221 .0348* 1.00

(6) Patent amount -.0160 .0690** .0364* -.0477** .4248** 1.00 **. Correlation is significant at the 0.01 level (2-tailed)

*. Correlation is significant at the 0.05 level (2-tailed) 4.4.3. Descriptive Statistics

Before starting my analysis, I tested the validity of my variables. During this test I checked for extremes, outliers, independence of the observation and if the sample size used in the analysis will be sufficient to draw conclusions from. Without this analysis, results could become skewed in either direction due to an over citated patent or ‘impossible’ value for Inventor geographical concentration.

21

between 1976-2006 and includes variables on ‘Firm size, ‘Firm industry type’, ‘Patent amount’

‘Internal Knowledge spillover’, ‘The Burden of Knowledge’ and ‘Inventor geographical concentration’. All in all, it consists of 3259 different observations that are collapsed on the firm level

in table 4, and 1233 unique observation ins table 5 where I account for the number of patents that are granted for a specific firm. A sidenote has to be made for the variable describing ‘Firm industry type’, as this is a categorical variable the descriptives in table 4 and 5 underneath are not of the upmost importance for this variable. The distribution of the variable can be seen in Appendix C. The lion’s share of my data describes firms within the Drugs & Medical, Computers & Communications and Electrical & Electronics industry types.

To determine the linear correlation within my dataset I conduct a Pearson Correlation Matrix including my Dependent Variable, and Independent- and Control variables. Table 1 above shows the results of this test. The closer the correlation figure of the two variables to 1. The more linear the relationship between both variables can be interpreted. Some interesting findings in my dataset are the negative correlations between ‘The Burden of Knowledge’ and both ‘Internal knowledge spillover’ and ‘Inventor geographical concentration’.

Some interesting findings for both Table 1 and Table 4 are: with 2.26 inventors per patent on average a small number of inventors is involved in the average patent. Furthermore, 1 out of 5 patents granted include a citation to a previous patent granted to the firm. Showing that building on previous granted patents is of upmost importance for most firms included in my dataset. Finally, when looking at Table 1. It is interesting to find two different effects of my independent variables on ‘Inventor

geographical concentration’. Whereas, inventors within firms that experience a relatively higher degree

of ‘Internal Knowledge Spillover’ are on average more geographically concentrated (r = .0608, p<0.01). Firms that experience a higher degree of ‘The Burden of Knowledge’ are on average more geographically scattered (r = -.0456, p<0.01). As expected, when accounting for the highly innovative firms, table 5 shows that these firms are bigger in size then their less innovative counterparts. Making for a sizable difference in the control variable ‘Firms size’ and ‘Patent amount’. Other variables remain relatively similar as the statistics for the whole dataset.

TABLE 2

Determinants of Inventor Geographical concentration

Dependent variable: Inventor geographical concentration Estimation method: Ordinary Least Squares Method

Model 1 Model 2 Model 3 Model 4 Robustness

Independent variables

Internal Knowledge spillover 0.19***

(0.06)

0.17** (0.06)

-0.28*** (0.04)

The Burden of Knowledge -0.04*

(0.02) -0.03* (0.02) -0.02 (0.01) Control Variables Firm Size -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) -0.00*** (0.00)

Firm Industry Type -0.00

(0.01) -0.01 (0.01) -0.01 (0.01) -0.02 (0.01) 0.02** (0.01) Patent amount 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) -0.00*** (0,00) AIC 4446 4448 4447 4448 452.6 BIC 4473 4479 4478 4484 489.1 Log-likelihood -2220 -2219 -2219 -2218 -220.3 Observations 3259 3259 3259 3259 3259 R-squared .0032 .0055 .0070 .0085 .0552 Adj R-squared .0022 .0043 .0058 .0070

***p<0.001, ** p<0.01, * p<0.05, + p<0.1

23

TABLE 3

Determinants of Inventor Geographical concentration (Patent amount > 50)

Dependent variable: Inventor geographical concentration Estimation method: Ordinary Least Squares Method

Model 5 Model 6 Model 7 Model 8 Robustness

Independent variables

Internal Knowledge spillover -0.18+

(0.11)

-0.12 (0.11)

-0.54*** (0.07)

The Burden of Knowledge 0.14***

(0.03) 0.14*** (0.03) 0.03 (0.02) Control Variables Firm Size -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) -0.00*** (0.00)

Firm Industry Type -0.04*

(0.01) -0.01 (0.02) -0.03* (0.02) -0.01 (0.02) 0.05*** (0.01) Patent amount 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) -0.00*** (0.00) AIC 1594 1590 1595 1591 322.6 BIC 1614 1615 1621 1622 353.3 Log-likelihood -792.9 -789.8 -792.6 -789.6 -155.3 Observations 1233 1233 1233 1233 1233 R-squared .0100 .0317 .0122 .0326 .0481 Adj R-squared .0076 .0285 .0089 .0286

***p<0.001, ** p<0.01, * p<0.05, + p<0.1

N Min. Max. Mean. S.D. Dependent variables cindex 1233 .027 1 .345 .234 bcindex 1233 .561 .999 .971 .038 Independent variables P_self_cites 1233 .002 .842 .218 .164 Avg_Inventor 1233 1.109 5.167 2.304 .661 Control variables Emp 1233 0 1049.713 22.482 46.961 Category_id 1233 1 6 3.510 1.133 Patent_amount 1233 50 46052 936.328 3161.05 Label Gvkey 1233 - - - - TABLE 4 Descriptive Statistics

N Min. Max. Mean. S.D.

Dependent variables cindex 3259 .027 1 .424 .250 bcindex 3259 .561 .999 .986 .027 Independent variables P_self_cites 3259 .002 1 .207 .210 Avg_Inventor 3259 1 5.167 2.260 .815 Control variables Emp* 3259 0 1049.713 11.252 39.675 Category_id 3259 1 6 3.502 1.245 Patent_amount 3259 2 46052 365.570 1994.226 Label Gvkey 3259 - - - -

*17 Missing values of Emp have been replaced using the single imputation technique by replacing the missing values with the mean of the variable associated.

TABLE 5

5. DISCUSSION

Within my research I have tried to theoretically and empirically contribute to innovation literature and research on patent data. This study focuses on firm level data whereas previous research often concentrated on industry level data (Agrawal et al., 2012; Bathelt et al., 2004). Another distinct factor of my research is that I build on studies done on patent data by Griffith, Lee and Reenen (2011). With the world digitalizing more from day to day the way economies and employees work change rapidly. Moreover, exogenous shocks such as the current Covid-19 pandemic make these changes exponentially more important. This thus begs the question, if geographical concentration of inventors is necessary to be able to be a successful and innovative firm. For Example, with current communications technologies does distance play as big of a role as in the past or is geographic distance compensated by technology? To ground my study, I use secondary data constructed by the NBER project (Hall et al., 2001) and provided by the USPTO. Together with the merge of datasets provided by Patentsview I was able to create a dataset with 3259 observations at the firm level (1233 for highly innovative firms). Building on previous research conducted on knowledge economies (Powell & Snellman, 2004), and patents and innovation (Acs, Anselin & Varga, 2002).

To my knowledge this is the first study that uses patent data that studies the determinants of geographical concentration of inventors within a given firm. As many find it easy to assume that inventors being geographically close is always preferable over them being scattered to achieve the optimal innovation results. I beg to ask the question if this assumption is true, or remanent of grooved thinking in this ever-expanding global economy. Therefore, I examine whether firms that experience a relatively high degree of internal knowledge spillover and a relatively high burden of knowledge cluster their inventors in geographical closeness to achieve their results.

26

firms that rely on previous inter-firm knowledge to innovate. I do wonder if this assumption still holds with the disappearance of relative distance due to modern communication technology.

Secondly, I study the effect of The Burden of Knowledge (Jones, 2005) on the geographical concentration of inventors within a firm. Here my findings were different than anticipated and hypothesized. The result implies that the higher the burden of knowledge and thus the more difficult it is to come up with an innovation due to for instance maturity of the industry, the less geographically concentrated the inventors within a firm are. This could be due to the fact that innovators need to fish in a more diverse pool of knowledge to be able to find valuable knowledge that innovates when the difficulty of innovation becomes higher. I expect that this effect has only increased in the years after my data, partly due to the ease of remote working but also due to an increase of The Burden of Knowledge for a lot of firms and industries. However, for highly innovative firms the benefits of inventor geographical closeness is present when The Burden of Knowledge is perceived as high, an opposite finding with regard to the analysis of my complete dataset. This might be due to the fact that highly innovative firms in my dataset are significantly larger in size in relation to my complete dataset giving these firms access to a more diverse knowledge pool within their own employees. All in all, I do not reject nor support Hypothesis 1 fully. Furthermore, my results might differ slightly for firms within different industries.

5.2. Managerial implications

As my research is useful and applicable both in theory and in practice, I will describe a few actions and consequences that can help not only firms structure their inventive process but also help legislators make decisions that stimulate their economy both in times of prosperity and in times of crisis.

For firms and their inventive structure, it can be of importance to concentrate their inventors geographically close when the bulk of the inventive activities focus on continuous innovation. As the inventors rely more and more on previous patents granted to the firm it is of importance for these personnel to be close to the personnel that worked on the cited patent. F2F interaction will ensure a less costly and lengthy knowledge flow process and help with the implementation of both tacit and explicit knowledge from cited to citing patent.

27

For legislators, the results in my research can be implicated as follows: Especially in difficult times such as the still ongoing Covid-19 pandemic legislators could possibly stimulate innovation by providing funding and the possibility of inventors working together in the same geographical area when continuous innovations provide the majority of the income for firms. For these firms being able to keep their inventive activities in the traditional way of being geographically close increases the likelihood of their income being stable and with that contribute to a healthy economy. This could also be the case for firms that operate in relatively infant industries where routines and best practices are still to be developed. Forcing firms in these industries to start innovating with their inventive personnel being geographically scattered could hinder these firms from battling with the established economy and hamper overall progress of the economy. For established and highly innovative firms it is important to keep innovating in a rapid pace with their inventors geographically concentrated. This will keep important parts of a countries economic engine running and might even improve one’s economic position when other legislative areas decide to not ease restrictions for certain firms/industries.

5.3. Limitations and future research

With the aforementioned implications that this research contributes there also exist several limitations and different angles for future research. First of all, due to time and data constraints two of my explanatory variables lack a suitable measurement and have not been taken into account in the empirical section of this paper. The variables ‘Tacitness of Knowledge’ and ‘Dependence on physical infrastructure’ are therefore not included in the models and results. Due to their potential importance on the firm level and to improve on this research, future studies could include these variables to give more meaning to the produced results and help both firms and legislators in their decision-making regarding innovation during exogenous shocks or out of the ordinary situations that occur either due to fault of their own or when no one is to blame.

28

Finally, the data used in this paper is data provided by the United States Patent and Trademark Office, USPTO in short, and has a focus on the world’s largest economy the USA. The generalizability of this research can therefore be questioned. A study with the focus on another economic giant, Europe, could lead to different results even though the characteristics differ only marginally from the US economy. More interestingly perhaps would be a study on inventive activities within an emerging economy such as China or India. This would give a look on the other side of the mountain and possibly explain the reason why such economies are closing in on the western economies and their decennia long hegemony and could possibly overtake these economies as the leading economy in the world. Results from such research could teach underdeveloped economies to stimulate their inventive activities and in relation to that their economy in general. Western economies could in their turn anticipate on results and implement the best practice in these emerging economies for their own gain.

6. CONCLUSION

Within this paper I aim to answer the question both theoretically and empirically, what influences the possible scatteredness of inventors on the firm level. This has become exponentially more important due to the current Covid-19 Pandemic and the consequent lockdowns proposed and executed by most of the world. I do this by examining multiple firm level variables and their granted patents between 1976 and 2006. Within this paper I aimed to answer the question how firms can organize their inventive activities (inventors geographically close or scattered) and what could be the potential impact on these firms during an exogenous shock such as the current Covid-19 Pandemic.

Additionally, the results of this study show a positive association of inventor geographical concentration when internal knowledge spillover in the firm is high and a negative association of inventor geographical concentration when The Burden of Knowledge and thus the difficulty to innovate in high. Moreover, these results seem to be mirrored for firms that are seen as highly innovative (Number of granted patens >50) It is therefore important for managers to fit their inventive structure to the kind of innovation they are seeking, taking in regard the industry the firm is present in, the maturity of their inventions/industry and the innovativeness of their own firm. For legislators the results will help them to consider measures that will have the smallest impact on a healthy economy or to keep the economy healthy for as long as possible. With this research being on the firm level I provided an interesting angle for future research on innovation and the importance of geographical distance even with regard of the limitations this research is subject to.

29

firms that are prepared for innovating while being more geographically distant will experience less hinderance from governments restrictions such as lockdowns etc. Firms that do not have this experience will play catch-up and are forced to make quick decisions to be able to handle the pressure.

REFERENCES

About CERN. (n.d.). CERN. Retrieved November 9, 2020, from https://home.cern/about

Acs, Z. J., Braunerhjelm, P., Audretsch, D. B., & Carlsson, B. (2009). The knowledge spillover theory of entrepreneurship. Small Business Economics. https://doi.org/10.1007/s11187-008-9157-3

Acs, Z. J., Anselin, L., & Varga, A. (2002). Patents and innovation counts as measures of regional production of new knowledge. Research Policy. https://doi.org/10.1016/S0048-7333(01)00184-6

Aghion, P., & Jaravel, X. (2015). Knowledge spillovers, innovation and growth. Economic Journal. https://doi.org/10.1111/ecoj.12199

Agrawal, A., Cockburn, I., Galasso, A., & Oettl, A. (2012). Why are Some Regions More Innovative than Others? The Role of Firm Size Diversity. National Bureau of Economic Research. https://doi.org/10.3386/w17793

Agrawal, A., Cockburn, I., & Rosell, C. (2010). Not Invented Here? Innovation in company towns. Journal of Urban Economics. https://doi.org/10.1016/j.jue.2009.10.004

Agrawal, A., Cockburn, I., & McHale, J. (2006). Gone but not forgotten: Knowledge flows, labor mobility, and enduring social relationships. Journal of Economic Geography. https://doi.org/10.1093/jeg/lbl016

Agrawal, A., Galasso, A., & Oettl, A. (2017). Roads and innovation. Review of Economics and Statistics. https://doi.org/10.1162/REST_a_00619

Asheim, B., Coenen, L., & Vang, J. (2007). Face-to-face, buzz, and knowledge bases: Sociospatial implications for learning, innovation, and innovation policy. Environment and Planning C: Government and Policy. https://doi.org/10.1068/c0648

Azoulay, P., Zivin, J. S. G., & Wang, J. (2010). Superstar extinction. Quarterly Journal of Economics. https://doi.org/10.1162/qjec.2010.125.2.549

30

Bathelt, H., Malmberg, A., & Maskell, P. (2004). Clusters and knowledge: Local buzz, global pipelines and the process of knowledge creation. Progress in Human Geography. https://doi.org/10.1191/0309132504ph469oa

Bharadwaj, A. S. (2000). A resource-based perspective on information technology capability and firm performance: An empirical investigation. MIS Quarterly: Management Information Systems. https://doi.org/10.2307/3250983

Bloom, N., Jones, C., Reenen, J., & Webb, M. (2017). Are Ideas Getting Harder to Find. LSE Research Online Documents on Economics. https://doi.org/10.3386/w23782

Carlino, G. A., Hunt, R. M., Carr, J. K., & Smith, T. E. (2012). The Agglomeration of R&D Labs. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2149008

Christie, J. (2020, May 12). Keeping our employees and partners safe during #coronavirus. Blog.Twitter. https://blog.twitter.com/en_us/topics/company/2020/keeping-our-employees-and-partners-safe-during-coronavirus.html

CoLaborator. (n.d.). Innovate Bayer. https://innovate.bayer.com/programs/colaborator

Duranton, G., Martin, P., Mayer, T., & Mayneris, F. (2011). The Economics of Clusters: Lessons from the French Experience. In the Economics of Clusters: Lessons from the French Experience. https://doi.org/10.1093/acprof:oso/9780199592203.001.0001

Factory of the future. (n.d.). Porsche.

https://www.porsche.com/international/aboutporsche/innovation/innovation-factory/

Feldman, M. P., & Florida, R. (1994). The Geographic Sources of Innovation: Technological Infrastructure and Product Innovation in the United States. Annals of the Association of American Geographers. https://doi.org/10.1111/j.1467-8306.1994.tb01735.x

Foss, K., & Foss, N. J. (2005). Resources and transaction costs: How property rights economics furthers the resource-based view. Strategic Management Journal. https://doi.org/10.1002/smj.465

Glaeser, E. L., Kallal, H. D., Scheinkman, J. A., & Shleifer, A. (1992). Growth in Cities. Journal of Political Economy. https://doi.org/10.1086/261856