A Study of Inventors: Incentives, Productivity, and Mobility (Innovation und Entrepreneurship) (German Edition) [2007 ed.] 3835006509, 9783835006508

Citation preview

Innovation und Entrepreneurship Hrsg.: Nikolaus Franke und Dietmar Harhoff

Karin Hoisl

A Study of Inventors Incentives, Productivity, and Mobility

GABLER EDITION WISSENSCHAFT

Karin Hoisl A Study of Inventors

GABLER EDITION WISSENSCHAFT Innovation und Entrepreneurship Herausgegeben von Professor Dr. Nikolaus Franke, Wirtschaftsuniversität Wien, und Professor Dietmar Harhoff, Ph.D., Universität München

Innovative Konzepte und unternehmerische Leistungen sind für Wohlstand und Fortschritt von entscheidender Bedeutung. Diese Schriftenreihe vereint wissenschaftliche Arbeiten zu diesem Themenbereich. Sie beschreiben substanzielle Erkenntnisse auf hohem methodischen Niveau.

Karin Hoisl

A Study of Inventors Incentives, Productivity, and Mobility

With a foreword by Prof. Dietmar Harhoff, Ph. D.

Deutscher Universitäts-Verlag

Bibliografische Information Der Deutschen Nationalbibliothek Die Deutsche Nationbalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über abrufbar.

Dissertation Universität München, 2006

1. Auflage Januar 2007 Alle Rechte vorbehalten © Deutscher Universitäts-Verlag | GWV Fachverlage GmbH, Wiesbaden 2007 Lektorat: Brigitte Siegel / Stefanie Brich Der Deutsche Universitäts-Verlag ist ein Unternehmen von Springer Science+Business Media. www.duv.de Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlags unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Die Wiedergabe von Gebrauchsnamen, Handelsnamen, Warenbezeichnungen usw. in diesem Werk berechtigt auch ohne besondere Kennzeichnung nicht zu der Annahme, dass solche Namen im Sinne der Warenzeichen- und Markenschutz-Gesetzgebung als frei zu betrachten wären und daher von jedermann benutzt werden dürften. Umschlaggestaltung: Regine Zimmer, Dipl.-Designerin, Frankfurt/Main Gedruckt auf säurefreiem und chlorfrei gebleichtem Papier Printed in Germany ISBN 978-3-8350-0650-8

Foreword

Inventions are the raw material of innovative success at the corporate and the national level. This statement is uncontroversial, yet it poses major problems for managers and for decisionmakers in public policy alike. How are we to manage, motivate and remunerate those individuals whose creativity and ingenuity gives rise to new inventions? Inventions are strongly driven by human capital. Most of the inventive output is accounted for by a relatively small number of highly talented individuals. The latter aspect has been widely neglected in both theoretical and empirical innovation literature since the late 1970s. In contrast to most of the literature which focuses on invention output at the firm and team level, Karin Hoisl addresses in her analysis the contributions made by individual inventors. In her dissertation, Hoisl undertakes a study of the characteristics of individual inventors and of the inventive process itself. Hoisl applies econometric methods to a large data set that was collected in the course of a project funded by the European Commission. In a questionnaire survey, more than 3,000 German inventors responded to questions on their personal attributes, the process of invention and the institutional settings for inventive activities in firms. The thesis consists of three chapters. The first two chapters focus on inventor productivity and on inventor mobility as well as on the complex causal relationship between these two phenomena. The third chapter addresses and discusses efficient motivation and incentive systems for inventors and studies in particular the German Employees’ Inventions Act. The thesis presented by Karin Hoisl delivers major new research results which promote our understanding of inventive processes and inventor motivation. It is a welcome complement to the existing management literature. Europe will not be able to maintain or extend current levels of wealth and technical progress unless we learn how to manage inventive processes better. This thesis is an encouraging step forward. Prof. Dietmar Harhoff, Ph.D.

V

Acknowledgements

This dissertation thesis was carried out during the past three and a half years. Even though the work on this thesis has often been highly demanding and laborious, the greater part has been exciting, instructive, and enjoyable. At the outset, I want to thank those persons who have supported me most and thus made this thesis possible: First and foremost, I want to thank Prof. Dietmar Harhoff, my doctoral advisor, for his exceptional and motivating support and his endless patience. I would like to thank Prof. Harhoff for accepting me as a doctoral student and for providing me with the opportunity to work on the PatVal project. During the collaboration with Prof. Harhoff I learned much about how to conduct scientific research and how to write scientific papers. I am very much indebted to him. Furthermore, I am grateful to Prof. Anton Meyer, my second advisor, for his excellent support and his advice, commencing during the MBR program and continuing later during my dissertation. The data used within the several parts of this thesis originate from a coordinated survey carried out in six European countries: France, Germany, Great Britain, Italy, Spain, and the Netherlands. Financial support from the European Commission (Contract No. HPV2-CT2001-00013) for this project is gratefully acknowledged. I am also grateful to all my colleagues at the Institute for Innovation Research, Technology Management and Entrepreneurship for their support during the last three years. In particular, I owe my thanks to Marc Gruber, who has become a friend during the last years. He provided me with fruitful input for my dissertation, and numerous times I benefited from his research experience. Furthermore, we share a fondness for gummy bears. I wish him all the best for the professorship he accepted recently. I want to thank Celine Schulz, who spent a couple of weekends with my papers to correct my English. Additionally, her valuable comments on methodology and content during several fruitful discussions always hit the mark. Further on, I thank Uli Lossen, with whom I do not only share my birthday, but we also shared one room for a long time. His constructive criticism during the course of this thesis was very helpful. Additionally he saved countless hours of my time due to his EXCEL solutions. Many thanks VII

also go to Christian Tausend and Felix Treptow who always were there when I needed encouragement and who had a major part in the great time we had during the last years. I am grateful to the “patent-group”: Joachim Henkel, Stefan Wagner, and Georg von Graevenitz for their support and valuable comments. Special thanks go further on to Rosemarie Wilcox for her support especially during the last months of my dissertation and for the final language check of my papers. For providing me with the essential know-how on intellectual property and for ongoing encouragement I wish to thank Dr. Jürgen Lachnit, partner at the law firm of Weickmann & Weickmann. Thanks also go to Ben Neuburger, who commented my papers from the point of view of a future patent attorney. I also want to thank my friend Una Schulze for her support. Additional thanks go to my longtime student assistant Thomas von Eggelkraut-Gottanka (who has become a colleague in the meantime) for his help. I never had to check his work. Further thanks are given to Silvia Knittl, who helped in building the dataset for a major part of my dissertation. Thanks also go to Alina Resnik and to Jes Villa for their aid. Last but not least, a number of hard-working student assistants of the PatVal project should be mentioned in deep gratitude for supporting the project: Lydia Bojerivanova, Mehmed Celik, Paolo Colonna, Nina Kreuzpointner, Miriam Lorber, Elke Müller, Stephanie Pörschmann, Yu Quian, Dominik Rottenkolber, and Christian Schmid-Eickhoff. My deepest and most heartfelt gratitude goes to my family, especially to my parents, my sister, and to my boyfriend Bernd, who are of capital importance to me. My parents support me in every way they can, they always believe in me and provide me with their love. I owe them so much. I want to thank my boyfriend Bernd for his love, endless patience, and for his continued interest in what I am doing. Karin Hoisl

VIII

Table of Contents

1

Introduction ...................................................................................................................... 1 1.1

2

Motivation .................................................................................................................. 1

1.2

The Complexity of Invention and Review of the Literature ...................................... 3

1.3

Summary of the Three Empirical Studies .................................................................. 6

A Closer Look at Inventor Productivity – What Makes the Difference? ................ 11 2.1

Introduction .............................................................................................................. 11

2.2

Theoretical Background and Empirical Evidence.................................................... 14

2.2.1

Literature Review................................................................................................. 14

2.2.2

Scientists vs. Engineers ........................................................................................ 17

2.3

Hypotheses ............................................................................................................... 19

2.4

Data Source and Description of the Variables ......................................................... 22

2.4.1

Description of the Data ........................................................................................ 22

2.4.2

Description of the Variables................................................................................. 24

2.4.2.1

Motivation for the Different Measures of Inventive Output........................ 24

2.4.2.2

Dependent Variables .................................................................................... 28

2.4.2.3

Explanatory Variables .................................................................................. 30

2.5

Descriptive Statistics und Multivariate Results ....................................................... 31

2.5.1

Descriptive Statistics............................................................................................ 31

2.5.2

Multivariate Specification.................................................................................... 42

2.6

Panel Regression ...................................................................................................... 50

2.6.1

Description of the Methodology .......................................................................... 50

2.6.2

Multivariate Results ............................................................................................. 53

2.7

Conclusion................................................................................................................ 61

IX

3

Tracing Mobile Inventors - The Causality between Inventor Mobility and Inventor Productivity..................................................................................................... 65 3.1

Introduction .............................................................................................................. 65

3.2

Hypotheses ............................................................................................................... 68

3.2.1

Hypotheses on Inventor Productivity................................................................... 68

3.2.2

Hypotheses on Inventor Mobility......................................................................... 68

3.2.3 Hypotheses on the Causality between Productivity and Mobility ....................... 71 3.3

Data Source and Sample .......................................................................................... 75

3.3.1

Description of the Data ........................................................................................ 75

3.3.2

Variables............................................................................................................... 77

3.3.2.1

Dependent Variables .................................................................................... 77

3.3.2.2

Explanatory Variables .................................................................................. 80

3.4

Descriptive Statistics and Multivariate Results ....................................................... 82

3.4.1

Multivariate Specification.................................................................................... 89

3.4.3

Discussion of the Results ..................................................................................... 91

3.5

Difference-in-Differences Estimation .................................................................... 101

3.5.1

Data Source, Matching, and Descriptive Results............................................... 102

3.5.2

Description of the Results .................................................................................. 106

3.6

4

Descriptive Results............................................................................................... 82

3.4.2

Conclusion.............................................................................................................. 116

Institutionalized Incentives for Ingenuity – Patent Value and the German Employees’ Inventions Act .......................................................................................... 121 4.1

Introduction ............................................................................................................ 121

4.2

Invention Processes and the German Employees’ Inventions Act......................... 123

4.2.1

Salient Features of Invention Processes............................................................. 123

4.2.2 Institutionalized Compensation Schemes in Various Countries ........................ 125

X

4.2.3

Historical Aspects of the German Employees’ Inventions Act ......................... 128

4.2.4

Regulations of the German Employees’ Inventions Act as of 1957 .................. 129

4.2.5

The Impact of the German Employees’ Inventions Act..................................... 132

4.3

Research Questions and Hypotheses...................................................................... 135

4.4

Data Source and Sample ........................................................................................ 136

4.4.1

Data Source – the German Inventor Survey....................................................... 136

4.4.2

Variables............................................................................................................. 138

5

4.5

Survey Evidence - Descriptive Statistics ............................................................... 140

4.6

Multivariate Analysis ............................................................................................. 150

4.7

Conclusions ............................................................................................................ 154

Summary of the Results and Outlook ........................................................................ 155

Bibliography ......................................................................................................................... 161 Annex 1.................................................................................................................................. 175 Annex 2.................................................................................................................................. 1767 Annex 3.................................................................................................................................. 181 Erratum................................................................................................................................. 200

XI

List of Figures

Figure 2.1: Figure 2.2:

Absorptive capacity (according to Cohen/Levinthal 1990) ........................... 21 Yearly means of the number of claims specified in the patent applications and yearly means of inventor team size ......................................................... 25

Figure 2.3:

Average inventor team size by firm size........................................................ 26

Figure 2.4:

Patent activity index per technological area................................................... 27

Figure 2.5:

Distribution of applications per inventor (whole counts) .............................. 38

Figure 2.6:

Distribution of applications per inventor (fractional counts)......................... 38

Figure 2.7:

Distribution of the number of citations per inventor...................................... 39

Figure 2.8:

Relationship between number of patent applications and number of

Figure 2.9:

Productivity differences by age groups.......................................................... 57

Figure 2.10:

Productivity differences by age groups (additional control for the priority

citations (scatter plot)..................................................................................... 40

years of the patents) ....................................................................................... 59

Figure 3.1:

Composition of the sample............................................................................. 76

Figure 3.2:

Distribution of the number of applications per inventor................................ 87

Figure 3.3:

Distribution of inventor productivity ............................................................. 88

Figure 3.4:

Distribution of the number of moves per inventor......................................... 88

Figure 3.5:

Distribution of inventor mobility per 10 years inventive activity.................. 89

Figure 3.6:

Untreated control group design with pre- and posttest (Cook/Campbell 1979)................................................................................. 102

Figures 3.7a/b: Difference-in-differences estimator calculated for the mean share of granted patents (4 year period and 3 year period)........................................ 108

Figure 4.1:

Share of salary received as inventor compensation for this patent .............. 141

Figure 4.2:

Share of salary received as inventor compensation for all patents,

Figure 4.3:

Effect of the German Employees’ Inventions Act on incentives for

multiple inventors are included once ........................................................... 141 innovation..................................................................................................... 147

XIII

Figure 4.4:

Incentives emerging from the German Employees’ Inventive Act (positive effect to motivation) ...................................................................... 148

Figure 4.5:

Disincentives emerging from the German Employees’ Inventive Act (negative effect to motivation) ..................................................................... 149

XIV

List of Tables

Table 2.1:

25 top-productive inventors (rank 1 to rank 12) ............................................ 33

Table 2.2:

Descriptive statistics....................................................................................... 36

Table 2.3:

Pearson correlation coefficients ..................................................................... 37

Table 2.4:

Number of applications (whole counts) by number of technical area ........... 41

Table 2.5:

Number of applications (whole counts) by level of education ...................... 41

Table 2.6:

Number of applications (whole counts) by firm size ..................................... 42

Table 2.7:

Cook-Weisberg test for heteroskedasticity .................................................... 43

Table 2.8:

OLS regression with heteroskedasticity-robust standard errors .................... 44

Table 2.9:

Robust fixed effects panel estimation (Model 1) ........................................... 55

Table 2.10:

Robust fixed effects panel estimation (Model 2) ........................................... 58

Table 2.11:

Robust fixed effects panel estimation (Model 3) ........................................... 61

Table 3.1:

Example 1 (applicant sequence of inventor 1) ............................................... 78

Table 3.2:

Example 2 (applicant sequence of inventor 2) ............................................... 79

Table 3.3:

Example 3 (applicant sequence of inventor 3) ............................................... 79

Table 3.4 :

Descriptive statistics....................................................................................... 83

Table 3.5:

Pearson correlation coefficients ..................................................................... 85

Table 3.6:

Differences between mobile and non-mobile inventors according to different incentives......................................................................................... 85

Table 3.7:

Share of mobile inventors by level of education............................................ 86

Table 3.8:

Share of mobile inventors by regional characteristics ................................... 86

Table 3.9:

2SLS regression with heteroskedasticity-robust standard errors ................... 93

Table 3.10:

Descriptive statistics..................................................................................... 105

Table 3.11:

T-Test of difference-in-differences estimations for 4 and 3 year periods

Table 3.12:

Difference-in-differences regression estimation (OLS regression) ............. 112

before and after the event of a move ............................................................ 107

XV

Table 4.1:

Descriptive statistics .................................................................................... 142

Table 4.2:

Inventor compensation by age and education .............................................. 144

Table 4.3:

Inventor compensation by firm size and number of inventors..................... 145

Table 4.4:

Inventor compensation by monetary patent value ....................................... 145

Table 4.5:

Inventor compensation by strategic patent value ......................................... 146

Table 4.6:

Means of compensation for this patent ........................................................ 147

XVI

Chapter 1 1 Introduction 1.1 Motivation Prior to the industrial revolution in the late 18th and the early 19th century, firms carried out virtually no systematic research and development (R&D) processes. During that time, research processes were simply conducted as “trial and error”. Inventive activity did not depend in any major way on science. Most firms were relatively small, and large firms were rather uncommon. Inventions were conducted almost exclusively by individual inventors (Bruland/ Mowery 2004). During the early 19th century industrial revolution in Europe and the U.S. there was a shift from a labor-intensive to a more capital-intensive economy. For instance, it led to the mechanization of the textile industry and to the development of industries like iron, steel, and mechanical engineering. In the middle of the 19th century, the second industrial revolution took place which led to an evolution of new industrial sectors, such as the chemical industry and electricity. During the late 19th century a number of extraordinary scientific discoveries were achieved. Darwin’s theory of natural selection published in “The Origin of Species” in 1859 and Kekulé’s discovery of the structure of the benzene molecule in 1865 are just two of many examples.1 This change of the technical focus was accompanied by changes in firm structure. The German chemical firms were pioneers during this phase of organizational change. Large vertically integrated enterprises emerged that operated their own R&D laboratories. Inventions became a result of systematic R&D processes which were carried out by teams of researchers. Additionally, the chemical industry in Germany profited from remaining in close touch with university research. Universities did not only provide a valuable source of knowledge, but also offered a source of highly qualified research staff (Freeman 1982).

1

See http://scienceworld.wolfram.com/biography/DarwinCharles.html and http://scienceworld.wolfram.com/ biography/Kekule.html (access on November 10, 2005).

1

The shift from unsystematic to planned R&D processes brought about major changes for the generation and protection of knowledge. In particular, inventors who are the source of inventions were treated with more attention. Additionally, the need for protection of the knowledge created within the scope of R&D projects emerged. An explanation for these changes is provided by the strategic management literature. In particular, the literature points out that potential sources of sustainable competitive advantage of firms are resources which are (a) valuable, (b) rare, (c) imperfectly imitable, and (d) which do not have strategically equivalent substitutes. Additionally, the assumption is made that firm resources may be heterogeneous and immobile (Barney 1986, 1991). Human capital is one example for a source of sustainable competitive advantage. Although human capital is certainly heterogeneously distributed across firms, it is not immobile. Employees can change their employer taking with them part of the knowledge stock of the firm. Dosi (1988) proposed that labor mobility facilitates the transfer of tacit knowledge that is otherwise immobile. Assigned to R&D, the mobility of key R&D personnel jeopardizes the innovative ability of a firm and consequently its competitiveness. The patent system actually facilitates identifying key inventors.2 This might pose severe problems to firms. Assume a firm applied for a patent. As from 18 months after the application date, the patent document will be published and every competitor who searches the databases of the patent offices will find information on the respective inventors. Consequently, R&D management would rather prefer to keep inventor names and addresses secret to decrease the probability of loosing an important inventor who received a job offer from a competitor. Since the non-disclosure of inventor related information is not possible, firms have to undertake special efforts to keep valuable inventors from leaving. In particular, providing incentives to key inventors increases their commitment to the firm. Yet knowledge is not only jeopardized by inter-firm mobility (moves between firms) but also by intra-firm mobility. Allen and Katz (1985) found that engineers and scientists are often attracted by higher wages to undertake administrative roles. Thus, key inventors are often promoted into management positions to keep them from leaving the firm. This “job-switch” usually terminates their participation in inventive activity in the firm.

2

2

According to § 37 (1) German Patent Law the applicant shall name the inventor(s) in the patent application and affirm that to his knowledge no other person has contributed to the invention. In case the applicant does not fulfill this obligation without cause, the patent office can refuse to grant the patent. See http://www.patentgesetz.de/ (access on November 10, 2005).

The protection of knowledge by intellectual property rights forms a second example of a sustainable competitive advantage. The German Patent Law, which came into force in 18773, aims at creating incentives for technical development. In exchange for property rights, applicants are made to disclose details of their new technical invention to the public. For providing access to the underlying knowledge, the applicant is rewarded a limited period of exclusivity for the usage of his invention (Harhoff 2005, DPMA 2002). Since patent protection impedes the imitation of technical achievements or at least forces competitors to a costly invent-around it helps firms to protect their competitive advantage over competitors. The above mentioned issues on inventors and on the protection of knowledge pose major challenges to the management of firms to maintain or increase the firms’ competitiveness and also to policy makers since innovation has a large share in economic growth and technological leadership of countries. Nevertheless, little empirical research has been done on one of the most important sources of innovation - the inventors. Especially the following three questions have so far remained unanswered: Who are these inventors? How do they invent? and Why do they invent? This dissertation thesis seeks to contribute to a better understanding of innovation processes and the involved inventors in general and specifically on (1) the determinants of inventor productivity, (2) the relationship between inventor mobility and inventor productivity and (3) the establishment of efficient incentive systems to commit key inventors to the firm. To do so, a novel and unique dataset is employed, including information on more than 3,000 German inventors from a recent survey as well as information on almost 40,000 European patents.

1.2 The Complexity of Invention and Review of the Literature Although R&D processes have become more systematic, inventions are in many cases still not predictable. Inventions can result from problem solving processes of private individuals, who do not have any technical or scientific education. Inventions may also arise from failed research projects or as by-products of other research projects. The following three inventions which were made during the 20th century, demonstrate examples that fall into these categories.

3

The first patent law was passed in 1474 by the city of Venice, followed by the ”Statute of Monopolies” which was enacted in England in 1624. In several Germany states, the concept of protecting intellectual property through patents dated back to the 18th century. In 1877, the first federal German Patent Law was established to create legal provisions and standards for patent protection, replacing the rather vague privileges used so far.

3

Melitta Bentz, a German housewife, was looking for a way to brew a perfectly filtered cup of coffee without the unpleasant taste of coffee grains at the bottom of her cup. After experiments with different materials, Melitta Bentz found her son’s blotting paper the most effective. She cut out a round piece of blotting paper, put it in a metal cup, and poured boiling water over the coffee. Melitta Bentz realized the value of her invention and applied for a patent at the Patent Office in Berlin. On July 8, 1908, Melitta Bentz received a patent for her "Filter Top Device lined with Filter Paper." In December 1908 the Melitta Bentz Company was founded. The Melitta coffee making method still enjoys great popularity, now being sold worldwide in over 100 countries.4 Cellophane was invented by Jacques E. Brandenberger, a Swiss textile engineer. In 1900 Brandenberger was searching for an idea to make textiles waterproof. He experimented with liquid viscose, but this material made the cloth too stiff. Although his experiments failed, he discovered that the coating peeled off in a transparent film, usable as clear, protective packaging layer. Even though the invention did not provide a solution for the original problem, a new and better use was found for cellophane.5 Finally, the microwave oven was invented by Percy L. Spencer. It was a by-product of another technology discovered during a radar-related research project in 1946. Spencer was an ingenious inventor working for the Raytheon Company. One day, he stood in front of a magnetron which is the power tube that drives a radar set, and noticed that the chocolate bar in his pocket had begun to melt. For a second experiment he used unpopped popcorn. Starting from these experiments, Spencer developed the microwave oven at Raytheon.6 These examples provide first indication of the complexity and the unpredictability of invention processes. To understand and after all to manage these processes one has to start with the analysis of the source of invention, i.e., the inventor. While the literature on innovation in the early fifties and sixties contributed a large number of insights regarding invention processes and the role of inventors, much of the subsequent work in this field veered away from the individual inventor as a unit of observation. Instead the literature started to focus on research and development (R&D) at the firm level (Bound et al. 1984, Griliches et al. 1987, Lanjouw/Schankerman 2004). In fact, inventors have still

4

See http://inventors.about.com/library/inventors/blcoffee.htm and http://www.melitta.info/index.htm (access on November 10, 2005)

5

See http://inventors.about.com/library/inventors/blcellophane.htm (access on November 10, 2005).

6

See http://inventors.about.com/gi/dynamic/offsite.htm?site=http://web.mit.edu/invent/iow/spencer.html, (access on November 10, 2005).

4

remained a black box. In order to provide a starting point for the discussion in this thesis, some of the literature that deal with individual inventors will be summarized here briefly. The first two papers describe the modification of inventive activity over time and provide some insights into the socio-demographic characteristics of inventors. In 1957 Schmookler conducted a survey on U.S. inventors to analyze the changing character of invention over half a century. He questioned 122 inventors of 74 different U.S. patents. Additionally, he used data from U.S. census reports for the years 1900, 1920, 1930, 1940, and 1950. These data were merged with patent statistics for the respective years. Results showed that invention changed from an activity dominated by individual inventors to an activity dominated by firms. Whereas inventions in the past stemmed from individuals without a technical background, inventions in later years were frequently made by technical professionals. What has remained relatively stable is that inventions are primarily a part-time job. Only less than half of all inventions are made by full-time inventors. Finally, Schmookler found that about half of the inventions are still being created by inventors without any university degree (Schmookler 1957). Amesse et al. (1991) described the results of an inventor survey, conducted in 1986. The units of observation were 374 individual Canadian inventors who were responsible for 352 patents. The analysis aimed at discovering socio-demographic characteristics of inventors, to explore the inventive process, and to analyze the commercialization of the inventions. The authors found that individual Canadian inventors were highly educated men with an income level that was higher than the average Canadian income. The inventors were mostly engineers and were likely to be self-employed. 50% of the inventions were made within a short time span and at low cost. 43% of the inventions were commercialized, either in a company already owned by the inventor or at a newly founded company (Amesse et al. 1991). The following two papers deal with the question of what motivates inventors to invent. In 1931, Rossman questioned 710 inventors about their motives and incentives behind their inventions. The most important motives of inventing turned out to be “love of inventing”, “desire to improve”, and “financial gain”. Less important were “altruistic reasons” and “laziness” (Rossman 1931). Macdonald (1984) used data from individual Canadian inventors, who applied for a patent in 1978. The author compared the motivation of inventors to patent with the motivation to invent. Results showed that inventors applied for a patent especially for financial purposes. 37% stated that the main reason for patenting was “to make money out of the invention”. The second most important reason was “to prevent others from making money out of the invention”. Financial purposes were less important for making the invention. 31% of the

5

inventors invented “to solve specific problems”, 20% wanted “to be useful for the society”. Of course, making money was also of importance (25%). The last paper goes a step further and analyzes whether individual inventors are responsible for more or less valuable patents compared to inventors employed with firms. In particular, Dahlin et al. (2004) tried to answer the question of whether independent inventors could be considered “heroes” or “hobbyists”, that is, if individual inventors produced more or less valuable inventions. The authors examined 225 U.S. tennis racket patents, granted between 1981 and 1991. Results showed that independent inventors were a heterogeneous group who were responsible for both, very important and also very unimportant patents. Inventors who turned out to be independent inventors in the area of tennis rackets were often listed in patents owned by firms in other technical areas. This meant that many independent inventors could in fact be professional engineers or scientists in other technical areas.

1.3 Summary of the Three Empirical Studies This thesis consists of three empirical analyses which are self-contained as chapters. The three chapters deal with inventor-related issues, in particular, inventor productivity, inventor mobility, and inventor compensation. The empirical analyses of the Chapters 2 to 4 are based on a dataset which was collected in the course of a project, sponsored by the European Commission. The project called PatVal (“The Value of European Patents: Empirical Models and Policy Implications Based on a Survey of European Inventors”, project number HPV2CT-2001-00013) was carried out by research groups from six European universities and aimed at creating a database of patent characteristics by surveying European inventors and by combining the survey responses with bibliographical and procedural data derived from European patent documents.7 The purpose of the first analysis (Chapter 2) is to create an appropriate productivity measure for inventors based on patent data. Conventional methods of measuring inventor output include pure and quality adjusted patent counts (Narin/Breitzman 1995, Ernst et al. 2000). The application of such measures is limited, since patent counts do not equal productivity. For instance, an increasing patent propensity (firms apply for more patents per unit of R&D expenditure due to strategic reasons) leads to biased results. To deal with strategic patenting behavior, citation counts are used as an alternative output measure. Since patent counts

7

6

See Annex 3 for The PatVal questionnaire.

represent a quality measure whereas citation counts represent a quantity measure, citations are assumed to be less affected by an increasing patenting activity. Furthermore, inventors working with large firms are different from inventors working with small or medium-sized firms. Inventors in large firms are often highly specialized and work as members of large inventor teams on many different projects in parallel. These organizational differences could also lead to biased results. To overcome biases caused by organizational differences, fractional patent and citation counts are employed. To identify determinants of inventor productivity data on more than 3,000 German inventors (who hold at least one granted European patent) are used. To trace the patent counts of each inventor over time, the EPOLINE database of the European Patent Office was used to search for all patents belonging to the 3,000 inventors with priority dates between 1977 and 1999, resulting in almost 30,000 EP patents. Whereas former research focused either on inventor related determinants or on environment related determinants of productivity, this study is the first to integrate both types of determinants. Results show that environmental and inventor related factors are important with respect to output quality (citation counts). In particular, firm size, the inventor’s level of education and the use of external sources of knowledge turn out to be important determinants of quality adjusted productivity. Output quantity (patent counts) rather seems to be determined by decisions made by R&D management, for instance, whether to file a patent application for an invention or not. To shed more light onto the relationship between productivity and the age of the inventor an additional panel regression estimation is conducted in Chapter 2. Results reveal that the age of an inventor considerably influences his or her productivity. To accommodate inventors dropping out of R&D over time, e.g., due to advancement in management positions, the sample is sub-divided into three groups: inventors who kept on inventing for their whole professional life, inventors who spent at least a major part of their professional life on inventing, and finally, inventors who stopped inventing only after a short period of time. Overall, results show that those inventors who are highly productive at the beginning of their career leave R&D. Assuming that these inventors are promoted into management positions, the question arises whether firms rather weaken their own competitive position by “keeping” their key inventors from inventing. Chapter 3 deals with the causality between inventor productivity and inventor mobility. The literature reveals that hiring a productive inventor from another firm can lead to knowledge transfers. Firms characterized by a technological gap can hire productive inventors in order to catch up with their rivals (Gilfillan 1935). Following this point of view, the productivity of inventors should increase the probability of observing a move. However, not only the new employer profits from the knowledge which is transferred by the mobile inventor. Inventors who move are exposed to a new environment that also affects the inventors' activity and 7

possibly their productivity. This effect is well-known from the empirical labor economics literature. Liu (1986) as well as Topel and Ward (1992) propose that mobility can lead to an increase in the match quality between employer and employee. A better match quality should lead to an increase in the inventor’s own productivity. This study builds on the labor economic and the knowledge transfer literature and allows for a simultaneous relationship between productivity and mobility. Again, this paper makes use of the combination of PatVal data and information from the EPOLINE database. For this analysis, the dataset was extended. Refinement of the search procedure as well as the extension of the priority years from 1999 to 2002 resulted in a total of almost 40,000 EP patents belonging to the 3,000 German inventors. An instrumental variables technique is employed to estimate the causal relationship between productivity and mobility. Results show that mobility increases productivity. A possible interpretation for this finding is that a move improves the employer-employee match. Consequently, a better match quality leads to a higher productivity. Results further reveal that, in contrast, increasing productivity reduces mobility. Possibly, efficient incentive systems may help to keep productive inventors from leaving the firm once management has recognized the importance of their contribution. Additionally, the level of education has no significant influence on inventor productivity. Making use of external sources of knowledge, on the contrary, increases productivity. Finally, firm size has a positive impact on productivity. Firm size also influences inventor mobility, although negatively. Furthermore, the temporal concentration of inventive activity, and the inventive environment are major determinants of mobility. Whereas the probability of a move decreases with the duration of inventive activity, it is higher in large cities compared to rural areas. To provide a better understanding of the impact of a certain move on inventive performance an additional quasi-experimental design is employed, comparing the performance of an inventor before and after a given move. Difference-in-differences estimation reveals that the value of patent applications increases once the move has occurred. Moreover, the share of patents granted as well as the share of patents opposed by a third party increase as a result of the move. Additionally, patents receive more references and also more citations in the postmove period. Again these findings support the importance of match quality. The results pose a challenge to R&D management. Firms should try to improve the match quality between employee and employer, since it has a positive impact on productivity. Possible steps towards a better match quality are individually designed motivation and

8

incentive systems for inventors or the establishment of a so-called “dual ladder” career systems, as proposed by Allen and Katz (1985), providing more career chances for engineers within the firm. In the third part of the dissertation (Chapter 4), motivation and incentive systems of inventors are analyzed. Germany is one of the few countries in which the monetary compensation for inventors is not only determined in negotiations between employer and employee inventor, but also by relatively precise legal provisions. Aside from the discussion of the characteristics of the German Employees’ Invention Act (GEIA), this chapter analyzes which incentives or disincentives it creates. The data reveal that the GEIA creates substantial monetary rewards for productive inventors. However, the qualitative responses from the survey also point to a number of dysfunctional effects. Some of the inventors consider compensation too low compared to their inventive performance or complain about the lack of transparency concerning the determination of an appropriate compensation. The phenomenon of co-inventorship of superiors (superiors are mentioned as co-inventors, not due to their inventive performance or participation in the inventive process, but due to their position within the firm) turns out to be considerably less important than expected. This result is surprising, since it has been given considerable attention in the literature. The multivariate analysis proceeds in two steps. First, the variables used are related to the (presumably) most important determinant of inventor compensation – the patent’s value. The value of the patents is obtained from the PatVal questionnaire. Since the patent value is an ordinal variable, an ordered probit is used for the analysis. The second step – the analysis of the compensation share variable – also treats the data as ordinal. The multivariate analysis reveals that the patent’s value, the number of inventors, and the associated variables with the inventor’s position in the company have the expected impact, while surprisingly industry dummy variables are insignificant. This means that, after controlling for the mentioned influencing factors, the GEIA is applied fairly consistently across different industries and technical fields. This means that the GEIA is in principle appropriate to compensate inventors for their merits and to work as an incentive system. But results also show that inventors are to some extent dissatisfied due to a perceived lack of transparency as to the amount of the compensation or a delay of payment. Hence, there is still room for improvement on the part of the firms to use the full capacity of the GEIA to motivate inventors.

9

Chapter 2 2 A Closer Look at Inventor Productivity – What Makes the Difference? 2.1 Introduction In December 2004, the German producer SIEMENS AG honored its top inventors for the tenth year in a row. The 13 winners were responsible for about 600 inventions made in 2004, accounting for approximately 7.5% of all reported inventions in the same year. The SIEMENS Inventor’s Award is aimed at enhancing inventive and innovative activities of the roughly 50,000 employees in the field of R&D. In 2004, SIEMENS spent 5.1 billion Euros in R&D (accounting for 6.8% of sales), resulting in 23.2 million Euros per working day.8 Another example for an award to motivate inventors is the WIPO Award scheme launched in 1979. This scheme aims at improving the reputation of inventors through recognition of their merits and also at promoting inventive activities. Between 1979 and 2000, 561 prizes were awarded to inventors from 77 countries. The WIPO does not interfere in the selection of the nominees. Independent national or international organizations nominate potential candidates to be selected by an inventor’s award committee, composed of representatives of government authorities, academia, associations of inventors, chambers of commerce, and industry.

I would like to thank the conference audience at the 5th EURAM (European Academy of Management) Conference in May 2005, the conference audience at the 2nd ZEW Conference on Economics of Innovation and Patenting in September 2005, as well as the conference audience at the AEA (INNOVATIONS and INTELLECTUAL PROPERTY VALUES - Econometric Studies) Conference in October 2005 for helpful comments. Special thanks go to Francesco Lissoni for his valuable comments on the differences between scientists and engineers. The survey responses used in this analysis originate from a coordinated survey effort in Italy, France, Spain, the Netherlands, the United Kingdom and Germany. The author thanks the European Commission, Contract N. HPV2-CT-2001-00013, for supporting the creation of the joint dataset. This paper makes use of the German survey responses which contain information relating to inventor compensation paid under the German Employees’ Inventor Act.

8

See http://www.siemens.com/index.jsp?sdc_p=cz3s5uo1233409pnfl0mi1033645&sdc_sid=33336152884&sd c_bcpath=1026937.s_5%2C&, (access on November 11, 2005).

11

“Outstanding research activities and numerous patented inventions” are given as criteria for the decision to grant a prize.9 Previous research on the productivity of scientists and engineers has revealed that productive inventors are relatively rare. Treating all inventors equally would probably lead to a decrease of the motivation of the most productive inventors. To design efficient incentive systems, such as the two examples described above, firms have to identify their key inventors. Conventional methods of measuring inventor output include pure and quality adjusted patent counts (Narin/Breitzman 1995, Ernst et al. 2000). These measures are to a certain extent a good proxy for productivity, however the application of such measures is limited since patent counts do not equal productivity. To illustrate, Hall (2004) proposes that there has been a strategic shift in the patenting behavior of U.S. firms in certain industries which resulted in a “patent explosion”. In particular, the number of patent applications per unit R&D has increased over time. This increasing patent propensity will lead to biased results when using uncorrected output measures, since younger inventors today tend to patent more inventions than older inventors did in the past when they were the same age (Hall et al. 2005). Apart from strategic patenting behavior, output measures based on patent counts are also at risk of being biased in favor of large firms since large firms have more resources at their disposal to hire and retain high quality researchers. Additionally, failing to control for organizational differences, e.g., the organization of the inventive process, could lead to biased results. Kim et al. (2004) propose that R&D is organized differently in large firms than in smaller firms. Inventors play a smaller role in any single R&D project but are involved in more projects at the same time. These organizational differences could lead to an overestimation of the number of patents per inventor unless fractional counts are applied. This paper aims to determine an appropriate measure of inventive productivity based on patent data. Four different measures of inventive activity are employed: whole patent counts, fractional patent counts10, citation counts, and fractional citation counts. This paper suggests using citation counts as an alternative output measure to overcome biases caused by strategic patenting behavior. Since the number of citations a patent receives is a measure for its quality (Harhoff et al. 1999), citation counts seem to be relatively independent of the increasing patenting activity. Additional this paper proposes the use of fractional counts of patents and citations as a means to deal with the organizational differences described by Kim et al. (2004). Since inventor team size increases with firm size, assigning a fraction of a patent to an inventor with respect to the size of the inventor team seems to be appropriate to control for

9

See http://www.wipo.int/innovation/en/wipo_awards/invention.htm, (access on November 11, 2005).

10

Fractional patent count means that a fraction of a patent is assigned to an inventor with respect to the number of co-inventors.

12

possible biases coming from firm size. Productivity11 is measured by relating these four different output measures to the age of the inventor. The different output measures are used to identify determinants of inventor productivity. Whereas former research only looked at determinants related either to the inventor or the inventive environment, this study is the first to integrate both types of determinants. This paper uses survey data on 3,049 German inventors, who hold at least one granted European patent. The inventors were requested to provide demographic information as well as information on the R&D process underlying their patented invention. To trace the patent counts of each inventor over time, the EPOLINE database of the European Patent Office (EPO) was used to search for all patent applications belonging to the 3,049 inventors with priority dates between 197712 and 1999. This search resulted in a total of 29,971 EP patent applications. Empirical results show that the characteristics of the inventor, for instance, the level of education or the external sources of knowledge only have a significant effect on productivity when output quality (citation counts) is considered. Output quantity (patent counts) rather seems to be determined by decisions made by the firms. R&D management, e.g., decides whether to file a patent application for an invention or not. To shed more light onto the relationship between productivity and age, additional fixed effects panel regression estimation will be conducted. To do so, the inventors’ patent applications were sorted into groups according to the age of the inventor at the time of the application of the patent. In particular, nine five-year age groups were constructed which represent the time structure of the panel. Then the remaining variables were categorized according to this time structure (i.e., to the nine age groups). Results reveal that the age of the inventor considerably influences his productivity. To accommodate different patenting activities of inventors over time, the sample is sub-divided into three groups: inventors who kept on inventing for their whole professional life, inventors who spent at least a major part of their professional life in inventive activity, and finally, inventors who stopped inventing after a short period of time. Results reveal that the longer

11

In the following, the terms “inventive output” and “inventive activity” will be used synonymously and refer to the cumulative number of patent applications or the cumulative number of citations per inventor. The term “productivity” will only be applied in case the cumulative output is related to the age of the inventor.

12

Although the first European patent application was filed on June 1, 1978 with the European Patent Office, priority dates from 1977 are found in the dataset. A priority from 1977 occurs when an application claims priority under the Paris Convention from a counterpart application filed less than one year earlier in another country; such a subsequent application has a priority date equal to that of the earlier application (Art. 4 Paris Convention). See http://www.wipo.int/treaties/en/ip/paris/trtdocs_wo020.html (access on November 10, 2005).

13

inventors remain in R&D, the higher their productivity. A possible interpretation may be that inventors who remain in R&D get more experienced and consequently increase their productivity. However, it seems more reasonably to assume that the causality runs the other way around, inventors who are more productive stay in R&D, whereas less productive inventors leave R&D for another job, e.g., in sales. The remainder of this paper is divided into six sections. The following section provides an overview of the theoretical and empirical literature on inventor productivity. The third section proposes hypotheses derived from existing literature. Section 4 contains the description of the data used in the empirical part of this paper as well as the description of the dependent and the explanatory variables. In section 5 descriptive statistics will be provided. Furthermore, a log linear regression will be estimated to identify determinants of productivity differences. Section 6 provides a fixed effects panel regression estimation to analyze the age-performance relationship of inventors more detailed. Finally, section 7 discusses the results and provides implications for further research.

2.2 Theoretical Background and Empirical Evidence 2.2.1 Literature Review Previous research on the distribution of productivity of scientists and engineers reveals that productivity is highly concentrated within a small number of researchers. In 1926, Lotka examined the number of name entries appearing in the decennial index of Chemical Abstracts between 1907 and 1916 as well as the names appearing in the index of Auerbach’s “Geschichtstafeln der Physik” (Lotka 1926: 317). Plotting the logarithmic number of persons responsible for one, two, or more contributions against the logarithmic number of contributions resulted in an almost linear relationship with a gradient of approximately two. Based on these results, Lotka formulated the “inverse square law of scientific productivity” (Lotka 126: 320). According to Lotka’s Law, the number of researchers producing n scientific contributions is proportional to 1/n2. Assigned to the example of Chemical Abstracts, this means that when 100 scientists appear exactly once in the index of Chemical Abstracts, it would be a share of 1/22 ( ˆ 25) who would appear twice within the same time period, and so on.

14

Price (1965) also showed that productivity is highly concentrated within a small number of key scientists. He formulated the “square root law of elitism” (Ernst et al. 2000: 186) suggesting that a scientific community in a particular research field contains an elite group of scientists, almost identical to the square root of all community members. This elite group is responsible for about 50 percent of the entire scientific output within this research field. Further interesting insights into productivity distributions were provided by Spilerman (1970). He proposed the mathematical formulation as well as the empirical test of a model, providing an explanation of the “distribution of disorders”. In his paper, Spilerman examined the causes of radical disorders in the U.S. during the 1960s. Results revealed that disorders can be “adequately described by a negative binomial distribution” (Spilerman 1970: 635), i.e., the outbreak of violence increases the probability of subsequent disorder. In 1974, Allison and Stewart applied the findings of Spilerman (1970) to productivity differences among scientists. The authors found that scientists, who produce little, produce even less later on, and vice versa. Such reinforcement models implicate two things: (1) highly skewed productivity distributions, and (2) increasing dispersion or inequality over time (Allison/Stewart 1974). Narin and Breitzman (1995) were the first to apply Lotka’s Law to the distribution of patents. They studied the output of inventors in R&D departments of four semiconductor firms. Their findings suggest that a relatively small number of key inventors is responsible for ten or more patents whereas a large number of inventors are responsible for only one patent. The top 1% of inventors are five to ten times as productive (in number of patents) as the average inventor. The top decile of the inventors is at least three to four times as productive as the average inventor (Narin/Breitzmann 1995). A second study, confirming the results of Narin and Breitzman, was conducted by Ernst et al. (2000). The authors used a sample of inventors of 43 German companies, active in the chemical, electrical, and mechanical engineering industry. Ernst et al. employed quality-weighted patent counts as a measure for inventor productivity. The following indicators for patent quality were employed: the grant rate (number of patents granted divided by the total number of applications per inventor), the share of valid patents (share of patents for which the renewal fees had still been paid), the citation ratio (number of citations received divided by the total number of patents), and the share of US patents in the inventors’ patent portfolios. Results show that “key inventors are characterized by a large number of patent applications which are of high quality” (Ernst et al. 2000: 184). Another line of research tried to analyze the determinants of productivity differences between researchers. Two hypotheses are proposed by Allison and Steward (1974): the Sacred Spark Hypothesis and the Accumulative Advantage Hypothesis. According to the Sacred Spark Hypothesis, differences in productive capacity arise due to substantial, predetermined 15

dissimilarities among scientists. Cole and Cole (1973), for instance, state that scientists differ distinctly in their ability and motivation for doing research. The Accumulative Advantage Hypothesis, on the other hand, provides a generalization of the Matthew Effect which can be described as follows: “the accruing of greater increments of recognition for particular scientific contributions to scientists of considerable reputation and the withholding of such recognition from scientists who have not yet made their mark” (Merton 1968: 58). Allison and Stewart (1974) as well as David (1994) and Stephan (1996) propose that the allocation of recognition and resources make highly productive scientists even more productive or lead at least to the maintenance of output productivity. The Accumulative Advantage Hypothesis is also supported by Price (1976: 292) who suggests that a possible reason for the concentration of productivity is that “success seems to breed success”. Cross-sectional studies of Cole and Cole (1967) and Hagstrom (1968) also confirm this hypothesis. Cole and Cole (1967), for example, examined the relationship between quantity and quality of scientific output of 120 physicists. As a quantity measure, the total number of papers listed in Science Abstracts was employed. The quality was measured by using citation counts from the Science Citation Index (SCI). Results show that physicists whose papers were frequently cited continued to be highly productive, whereas scientists responsible for less frequently cited papers did not show high productive capacity. The authors, therefore, conclude that scientific productivity is strongly associated with recognition. Another interesting finding is that output quality is even more strongly related to recognition, for instance, by receiving awards, than output quantity (Cole/Cole 1967). Another empirical study conducted by Arora, David, and Gambardella (1998) raised the question about the effect of reputation on the volume of academic output. Past research publication performance is found to have an important effect on the expected amount of research funding and therefore, on future publication productivity (Arora et al. 1998). In a number of studies the relationship between age and productivity among technical personnel or scientists has been analyzed13. Early findings show a maximum of productivity at the age of about 40 and a decline afterwards. This decline was explained by a decrease in motivation and risk-taking as well as by difficulties in keeping up with technological change (Dalton/Thompson 1971; Lehman 1966; Oberg, 1960). Later studies detected a curve with two modes, one before the age of 40, the second approximately at the age of 50 (Pelz/Andrews 1966; Vincent/Mirakhor 1972). These findings were criticized by Zuckerman and Merton (1972). Studying Nobel Prize winners, the authors showed that these scientists remained highly productive over time. A decline in productivity due to seniority was explained by differences between two groups: a small group of key scientists who increase or

13

16

See Goldberg/Shenhav (1984) for a summary of the relevant literature on the relationship between age and productivity.

at least maintain their productivity level, and another, larger group showing a decrease in productivity over time. These findings are concordant with the Matthew Effect described above (Goldberg/Shenhav 1984: 111; Zuckerman/Merton 1972: 497). Jones (2005) also uses data on Nobel Prize winners. Additionally, 20th century great inventors are included in his analysis which shows an upward trend in the age at which scientists and engineers begin their careers. A reason for this delayed start is an increase of the age at the time of the highest educational degree. Thus, scientists and engineers spend more time on education. This time shift is not compensated by a shift in the productivity of innovators beyond middle age. The combined effects lead to a decline of the overall innovative output of younger innovators. In particular, Jones observes a 30% decline in life-cycle output over the 20th century. Furthermore, the author finds that “the mean age of great achievement for both Nobel Prize winners and great technological inventors rose about 6 years over the course of the 20th century” (Jones 2005: 2). Overall, economic studies agree on the fact that the productivity distribution among scientists is highly skew. The literature also provides initial evidence that quantitative productivity is not negatively correlated with the quality of output. Researchers, however, do not agree on the reasons for the disparities in research productivity. Allison and Stewart (1974) argue that the disparity is due to differences in the assignment of resources and recognition. The relationship between age and productivity has been analyzed in numerous studies. Whereas Dalton and Thompson (1971), for instance, find a maximum productivity at the age of 40, Zuckerman and Merton (1972) propose that key scientists increase or maintain their productivity level. A second group of less productive inventors, on the contrary, show a decrease in productivity over time. To shed more light onto the controversial discussion of the determinants of productivity, this paper aims at analyzing productivity differences more closely using improved productivity measures. Most notably, this paper combines the above mentioned determinants in one analysis, allowing for interactions between different determinants.

2.2.2 Scientists vs. Engineers Seminal literature that examines the distribution of research productivity uses scientific output for their analysis. Lotka (1926) and Price (1965) for instance employ the number of scientific articles published by a researcher. Allison and Stewart (1974) also use scientists as their unit of observation to analyze the determinants of research productivity. Hence, before hypotheses from the existing literature are derived, this section will discuss the feasibility of transferring these research results from scientists to engineers in R&D departments. In particular, the results of Ritti (1971) and Allen (1977) are considered which compare the 17

research process and incentives of scientists and engineers. Additionally, the process of publishing a scientific paper is compared to the examination process at the European Patent Office. In 1977, Allen compared nineteen R&D projects to analyze characteristics that distinguish scientists from engineers. Two of them were scientific projects. The remaining 17 were technological projects. Results showed that scientists received their ideas from the scientific literature, whereas engineers hardly use scientific literature but rather employed customers or suppliers as their external sources of knowledge (Allen 1977). Scientists and engineers also differ in their motivation to conduct research. Whereas the publication of research results, reputation outside the company, and professional autonomy are important work goals for scientists, engineers are motivated by the increase in the profits of their firms, the gain in knowledge of company management policies, and the opportunity to explore new ideas (Ritti 1971). Additionally, scientists and engineers differ in the extent of the area of discretion to select their research projects. As Moon (2005) points out, senior researchers at universities have the freedom to select their own research projects. In corporations, the R&D managers rather than the engineers hold the authority to decide on the main research direction (Moon 2005). An additional difference regards the decision behavior of the scientists or engineers with respect to their research results. Scientists have the freedom to decide whether to publish their scientific results or not. Engineers, on the contrary, have to report their results to their employer. Decision makers, for instance the R&D manager or the head of the intellectual property department, decide if a patent application should be filed. In case, patent protection is desired, the patent attorney has to formulate the application document. It is furthermore possible that one invention is protected by two or more patents, for instance due to strategic reasons. This would lead to an increase in the number of patents per inventor without an increase in the number of inventions. Hence, engineers have only little control as to their patent output. Scientists, on the contrary, determine their publication output. What scientists and engineers have in common is that their research output has a positive effect on their reputation. Scientists are eager to build up a reputation outside of their own company, for instance by presenting their research results at conferences. Engineers also try to achieve a good reputation (inside the company) in order to get promoted (Allen 1977). Additionally, engineers could use their patent portfolio to be more visible on the labor market, for instance, to get a job offer from a competitor.

18

Finally, the publication process can be seen as an analogue to the process of patent grants at the patent office. A paper submitted to a journal is reviewed by an expert in the particular research area. Before an article is published, the paper has to be revised und resubmitted to the editor. Papers which are of minor quality or novelty are rejected by the reviewer. Almost the same applies to the examination process at the European Patent Office. Patent applications have to pass through an examination procedure. The examination process at the European Patent Office consists of an initial search and subsequent examination phase. During the examination process, the patent examiner either informs the applicant that the patent will be granted or requires the applicant to agree to changes in the application. Once an agreement has been found between the applicant and the examiner, the patent is granted (Harhoff/Wagner 2005). Patents that do not meet the requirements for patentability (novelty, inventive step and commercial applicability) are rejected by the examiner. These comparisons of characteristics of engineers and scientists and of the publication and patenting processes reveal that there are differences between these two groups of researchers. Nevertheless, there are a lot of analogies as well. In both cases the output has to withstand an examination process. Additionally, both groups need to build up a reputation to receive appropriate returns for their merits. It is therefore assumed that the analysis of inventor productivity can profit from considering the results of the literature that deals with scientific productivity. Possible differences which arise during the analysis will be discussed in the conclusion of this paper.

2.3 Hypotheses Shockley (1957) proposed that productivity is affected by many “mental factors”, such as the ability to detect important problems, technical skills and persistence. Since then, a large number of authors have considered the relationship between education and ability, especially the appropriateness of education as a proxy for ability.14 Guilford (1968) suggested to consider the intellectual ability of a person “as a somewhat generalized skill that has developed through the circumstances of experience, within a certain culture, and that can be further developed by means of the right kind of experience” (Guilford 1968: 619). Griliches (1970) proposed that “ability is the product of ‘learning’, even if it is not all a product of ‘schooling’” (Griliches 1970: 93). Moreover, ability can be considered as some measure of “achievement” determined by the following equation:

14

See Becker (1964) and Denison (1964) for a survey of the relevant literature.

19

A D 0  D 1 S  D 2 QS  D 3 LH  D 4 G  v where S is the level of schooling, QS is the quality of schooling, LH are the learning inputs at home, and G is the original genetic endowment. Griliches proposes to “confess ignorance” with respect to the potential determinants of ability and to define ability as gross output of the schooling system (Griliches 1970). This paper, according to the existing literature, controls for intellectual ability using the level of education of the inventors. The following relationship is expected: H.1:

Inventors with a high level of education tend to show higher productivity15 than inventors with a low level of education.

Beyond education, external sources of knowledge can positively influence inventor productivity. Possible sources of knowledge are, among others, the patent literature, customers, competitors, and scientific institutions. The literature provides evidence that knowledge transfer from different sources spurs innovation. Patent documents allow inventors not only to catch up on the state-of-the-art but also to collect relevant research information. Los and Verspagen (2003) characterize patent documents as a “potential source of ‘ideacreating’ knowledge spillovers” (Los/Verspagen 2003: 3). Allen (1977: 45), von Hippel (1988: 97) and Freeman (1991: 499) highlight the importance of customers and competitors regarding the innovativeness of firms. These external sources of knowledge play an important role in information sharing (Afuah 2000). The literature described above analyzes the influence of knowledge transfer on innovative output at the firm level. However, the results should also apply to the inventor level. Using different sources of knowledge should enable inventors to increase their inventive output, hence: H.2:

Inventors making use of patent literature, customers’ knowledge or competitors’ knowledge are more productive than inventors who do not use these external sources of knowledge.

15

20

Productivity is measured as the cumulative output of an inventor divided by the age of the inventor minus 25. The denominator (age-25) represents the number of productive years of an inventor. Hence, the productivity measure represents the mean annual output of an inventor.

Additional external sources of knowledge are university research and scientific literature. Jaffe (1989) analyzes the effects of knowledge transfer between university research and research labs. Using data on 29 U.S. states for the years 1972-1977, 1979, and 1984, he finds a positive relationship between corporate lab patenting and university research in three technical fields: drugs, chemicals and electronics. Allen (1977) compares nineteen parallel R&D projects (two scientific projects and 17 technological projects) to analyze characteristics, distinguishing engineers from scientists. Results show that scientists receive ideas from the literature, whereas engineers hardly use scientific literature and rather employ customers or suppliers as external sources of knowledge (Allen 1977).

Technical Knowledge (Invention)

Education

Absorptive Capacity External Sources of Knowledge (University Research, Scientific Literature)

Figure 2.1: Absorptive capacity (according to Cohen/Levinthal 1990)

A possible explanation for these differences provides the concept of “absorptive capacity” (Cohen/Levinthal 1989, 1990). Absorptive capacity - the ability of a firm to recognize the value of external information, to assimilate and to apply it to commercial R&D - is required to profit from spillovers. Applying the concept of Cohen and Levinthal to the inventor level, it can be assumed that inventors generate knowledge (inventions) by using external sources of knowledge (e.g. scientific literature). The inventors’ absorptive capacity determines the extent to which the external knowledge can be assimilated and employed. Absorptive capacity in turn depends on the extent to which the inventor is used to using scientific sources of knowledge. It is, therefore, assumed that inventors who did doctoral or postdoctoral studies are more able to benefit from scientific research. Figure 2.1 illustrates the proposed dependencies.

21

The following relationship is proposed: H.3:

Inventors who have conducted scientific research (doctoral studies) increase their productivity more by using university research or scientific literature than inventors who do not conduct scientific research.

Mansfield (1986) uses data from a random sample of 100 U.S. manufacturing firms representing twelve different industries. He finds that most industries exhibit a positive correlation between firm size (measured by sales) and patented inventions. Bound et al. (1984) use panel data on U.S. firms from 1972 to 1978. Results show that R&D expenditures have a highly significant influence on the quality of patents. In terms of absolute patent counts large firms stay ahead of small firms, although small firms tend to patent more per R&D dollar (Bound et al. 1984). Idson and Oi (1999) propose a positive relationship between labor productivity and firm size because large firms are generally early adopters of new technology. Additionally, they have more resources at their disposal to hire and retain high quality researchers. Kim et al. (2004) use longitudinal worker-firm matched data in the semiconductor and pharmaceutical industries. In both industries the authors find that inventor productivity increases with firm size. Research expenditures, sales and number of employees were used as alternative size measures confirming the result of a positive correlation in each case. Since it is plausible that inventors who have more resources at their disposal show a higher inventive output, the following hypothesis is proposed: H.4:

Inventors who are employed with a large firm show a higher productivity than inventors working at small firms.

2.4 Data Source and Description of the Variables 2.4.1 Description of the Data The data used in this chapter were collected in the course of a project sponsored by the European Commission. The project called PatVal (“The Value of European Patents: Empirical Models and Policy Implications Based on a Survey of European Inventors”) started in January 2002. The project aims at creating a database of patent characteristics based on a survey of European inventors named in European (EP) patents and from information drawn from the patent documents.

22

To achieve this goal, research groups from six European universities participated in this project. In each of the six countries (France, Germany, Great Britain, Italy, Spain, and the Netherlands) domestic inventors were surveyed simultaneously regarding their granted EP patents. The following analysis relies only on the German dataset. Therefore, units of observation are inventors who lived in Germany at the time of application of the respective patent. 10,500 EP patents listing inventors living in Germany were chosen by stratified random sampling based on a list of all granted EP patents with priority dates between 1993 and 1997 (15,595 EP patents). A stratified random sample was used in order to oversample potentially important patents. The sample of 10,500 patents hence includes all patents an opposition16 was filed against by a third party (1,048) and patents which were not opposed but received at least one citation (5,333), and a random sample of 4,119 patents drawn from the remaining 9,212 patents. The information was obtained using a questionnaire which was divided into six sections: section A asked for personal information about the inventors; section B for information about their educational backgrounds. Section C contained questions on employment and mobility of the inventors. Section D asked for details characterizing the invention process, i.e., the existence and extent of collaborations and important sources of knowledge. Section E included questions on the inventors’ rewards as well as the effectiveness of the German Employees’ Invention Act to provide a basis of an incentive system. Finally, section F dealt with the economic and strategic value of the patents.17 As the addressee of the survey, the first inventor listed on the patent document was chosen. To control for address changes we tried to confirm or correct the address of the first inventor using the web version of the white pages as well as the EPOLINE database of the European Patent Office (EPO). When the address could not be confirmed or changed, we searched for the second inventor. The search was continued until either the addresses of the first inventor or of one of the co-inventors could be confirmed, corrected, or changed. In case we did not manage to find any of the inventors in step one or two, the questionnaire was addressed to the first inventor using the address given on the underlying patent document. The dispatch of the questionnaires took place between May and October 2003. A cover letter with an attached questionnaire was sent to each inventor. The letter contained a link to a web questionnaire in order to give the inventors the possibility to choose between the paper-based and the web-based questionnaire. To date, 3,346 responses were received (mail: 3,184, online:

16

According to Harhoff et al. (1999) an opposition and the number of citations a patent received from subsequent patents are positively correlated with the value of a patent.

17

See Annex 3 for a copy of the questionnaire.

23

162), resulting in a response rate of 32%.18 The sample contains 2,761 inventors who answered one questionnaire, 282 inventors with two questionnaires, 4 with three questionnaires, 1 inventor who answered four questionnaires, and 1 inventor who filled out 5 questionnaires.19 Hence, the sample used in this paper is made-up of 3,049 different inventors (responsible for 3,346 EP patents). The data from the questionnaire were merged with bibliographic and procedural information on the respective patents obtained from the online EPOLINE database. The database contains information on all published EP patent applications as well as all published PCT applications since the founding of the EPO in 1978. The dataset corresponds to the EPOLINE data as of March 1st, 2003 and covers over 1,260,000 patent files with application dates ranging from June 1st, 1978 to July 25th, 2002. For this study, inventor address data were available up to 1999. To trace the productivity of each inventor over time, the EPOLINE database was further used to search for all patent applications belonging to the 3,049 inventors with priority dates between 1977 and 1999. The search procedure20 resulted in a total of 29,971 EP patent applications.

2.4.2 Description of the Variables

2.4.2.1

Motivation for the Different Measures of Inventive Output

Former empirical studies on inventor productivity used pure and quality adjusted patent counts to measure inventive output (Narin/Breitzman 1995, Ernst et al. 2000). These measures are at risk of being biased due to: ƒ

differences in the organization of the inventive process across firms and due to

ƒ

a strategic shift in the patenting behavior of firms over time.

Both problems as well as their handling in the following regression model will be discussed below.

18

Correcting for neutral non-responses, for instance inventors were not reached (due to a change in their residential address), or they passed away, resulted in a response rate of 33%.

19

Inventors who were responsible for more than one patent in the underlying time period and who were chosen more than once by stratified random sample were provided with up to five questionnaires.

20

The search procedure is described at length in Annex 1 and Annex 2.

24

Organizational Differences Whereas Figure 2.2 shows that the average size of inventor teams has remained almost stable over time, Figure 2.3 reveals that the average team size varies considerably with firm size. Whereas an inventor team consists of an average of two inventors in small firms, in large firms an average of four inventors are jointly responsible for an invention. This finding is not surprising. For instance, Kim et al. (2004) showed that inventors in large firms contribute less in any single R&D project but are involved in more projects at the same time. Using whole patent counts which assign a whole patent to each co-inventor, such organizational differences lead to biased results. In particular, the number of patents per inventor would be overestimated. Fractional patent counts, on the other hand, assign a fraction of a patent to the inventor with respect to the number of co-inventors. Using fractional patent counts should reduce these organizational effects, although fractional counts raise the additional problem of quantifying the contribution of each co-inventor. Since information regarding the contribution of each co-inventor is not available, the patents will be attributed equally across the members of the inventor team. For instance, if there were two inventors of a patent, each inventor is assigned half a patent.

16 14 12 10 8 6 4 2

19 77 19 78 19 79 19 80 19 81 19 82 19 83 19 84 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99

0

application year mean (no. of claims)

mean (no. of inventors)

Figure 2.2: Yearly means of the number of claims specified in the patent applications and yearly means of inventor team size (N = 28,542)

25

average number of inventors . per application

5 4 3 2 1 0 50,000

firm size groups (number of employees)

Figure 2.3: Average inventor team size by firm size (N = 28,542)

Patent Propensity Over the last years, the annual number of patent applications increased both in the US (Hall 2004) and in Europe (Harhoff 2005, Harhoff 2006). One possible reason is that patenting has been extended to new technological areas such as genes, software, or business methods which were previously not patentable. Additionally, firms apply for more patents per unit of R&D expenditure due to strategic reasons. Hall (2004) uses U.S. patent data of about 1,400 U.S. manufacturing firms between 1980 and 1989 to explore the sources of patent growth in the U.S. since 1984. Results reveal that the increase of patent applications has taken place especially in the electrical, electronics, computing and scientific instruments industry. This “patent explosion” is assumed to be a result of a strategic shift in patenting behavior of U.S. firms in these industries (Hall 2004). As well as in the U.S., the annual number of patent applications at the European Patent Office has grown rapidly in recent years. Figure 2.4 displays all European patents with application dates between 1977 and 1999 assigned to 30 technological areas, based on their IPC classes.

26

6 Biotechnology Telecom

5

patent activity index

IT 4 Pharma

3

MedTech

2

1

1985

1995

19 77 19 78 19 79 19 80 19 81 19 82 19 83 19 84 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99

0

application years

Electr./Energy Audiovisual Telecom IT Semiconductors Optical Analysis/Measurement MedicalTechn. NuclearTechn. OrganicChem. Polymers Pharma/Cosmetics Biotechnology Agric&Foods Petrol/MaterialsChem. SurfaceTechn. Materials ChemEngineering Matprocessing/Textile Handl/Printing Agric&FoodProcess Environment MachineTools Motors ThermProcesses MechElements Transportation SpaceTech/Weapons ConsGoods ConstrTechn.

Figure 2.4: Patent activity index per technological area (N = 1,186,321 EP applications)

For each of the technical areas i a patent activity index was calculated, defined as the number of patent applications in technical area i in year t divided by the number of applications in the same area in 1985. 1985 is used as a base year21 which means that the patenting activity index is 1 for all technical areas in 1985. Results show that patenting activity increased mainly in the following technical areas: telecom, IT, biotechnology, pharmaceuticals, and medical technology. This finding is in accordance with the above described results of Hall (2004). Furthermore, Figure 2.4 shows that especially as of 1995, an increase becomes apparent in every technical area. An increasing patenting activity over time leads to biased results when using uncorrected output measures, since younger inventors today (keeping all other variables constant) tend to patent more inventions than older inventors did in the past when they were the same age (Hall et al. 2005). To avoid these biases, an alternative output measure is employed. Following Ernst et al. (2000), who found that inventive quantity does not rule out inventive quality, the

21

1985 was used as the base year, since the years between 1977 and 1984 are characterized by a strong increase in the number of European applications, which is not caused by an increasing patent activity but by the diffusion of the European patent.

27

quality of the patent applications is used as a second dependent variable. According to Harhoff et al. (1999) the number of citations a patent application received from subsequent patent applications within a certain period of time is an appropriate proxy for the quality of the application. Not only the number of patent applications but also the number of citations has increased over time (Harhoff/Wagner 2005). However, this increase has probably occurred as a result of the increasing number of claims per patent and not due to an increasing patent propensity of firms (see Figure 2.2).22 Therefore, in the following multivariate analysis, the number of claims will be included as a control variable.23 Again fractional citation counts will be employed to control for the above described firm size and team effects.

2.4.2.2

Dependent Variables

productivity – The variable is defined as the cumulative output of an inventor, divided by age of the inventor in 1999 minus 25. A way of justifying this measure would be to assume that inventors become active at the age of 25 and continue to work with constant productivity.

PROD

22

23

cumulative output age1999  25

An application that seeks patent protection at the European Patent Office has to pass an examination process. During this examination process, a search report is prepared by the patent examiner. The search report contains patent and non-patent documents constituting the relevant prior art to be taken into account in determining whether the underlying invention is new and involves an inventive step. According to the Guidelines for Examination in the European Patent Office (see http://www.european-patent-office.org/legal/ gui_lines/pdf_2003/gui_03_full_e.pdf, access on November 18, 2005) the patent examiner should direct his search to the most important characteristics of the invention. Therefore, the search is conducted on the basis of the claims that describe the scope of protection for which patent protection is designated. An increasing number of claims per patent, hence, leads to an increase in the number of references included in the search report. Since references in the search report form the basis for calculating the number of citations a patent received by a subsequent patent, an increasing number of claims indirectly increases the number of citations per patent. An alternative approach for future work would be correcting for an increase in the number of citations per patent over time: propensity to be cited

§ claimst  4 years ,i  claimsti · ¸¸ number citations within 5 years ˜ ¨¨1  claimsti © ¹

The “propensity to be cited” measure controls for an increase in the number of claims across different technical areas i over time (in years t).

28

To measure a coefficient for age instead of assuming a proportional relationship between age and the number of applications, age is also included in the regression. Due to the skewness of the productivity distribution (Lotka 1926) the logarithm of the dependent variable is employed. As discussed in the section before, four different output measures will be used: number of applications (whole counts) - This variable includes the total number of patent

applications per inventor. Each inventor is assigned a whole patent irrespective of the number of co-inventors. number of applications (fractional counts) - The number of fractional applications refers to

the number of applications with respect to the size of the inventor team. The variable is defined as the sum of all fractional patents per inventor. number of citations (whole counts) - This variable includes the number of citations24 a

patent application received within 5 years following the publication of the search report added up for the total number of patent applications per inventor. In accordance with Price (1976), who counts the publication of a paper as its first citation “success”, the application of an EP patent is supposed to be its first patent citation. In the following, the number of citations is adjusted by adding one in order to be able to calculate the logarithm of this variable. number of citations (fractional counts) - The number of citations with respect to the

number of co-inventors is used as an alternative output measure. The variable is defined as the number of fractional citations a patent application received added up for the total number of applications per inventor. Also this variable is adjusted by adding one to enable a logarithmic transformation of this variable. The following transformation of the fractional counts variables makes it possible to transfer the fractional counts term to the right hand side of the estimation equation. First of all, this transformation permits to compare the models using whole patent (citation) counts with the models using fractional patent (citation) counts since the same dependent variable (log(whole counts)) is used. Secondly, it becomes possible to estimate an effect of the fractional counts term, since

24

In 1955 Eugene Garfield proposed the Science Citation Index providing bibliographic information, author abstracts, and cited references of a large number of scientific and technical journals (Narin 2000). Using the Science Citation Index, it became possible not only to count the number of articles published but also to include the number of citations a published article receives from subsequent articles. Since then numerous researchers used citation data in order to account for quality adjusted productivity (Turner/Mairesse 2002, Allison/Stewart 1974, Bayer/Folger 1966, Sher/Garfield 1966). This method is transferred to patent counts or inventive productivity.

29

log( fractional counts )

§ whole counts · ¸ log¨¨ fractional counts ˜ whole counts ¸¹ ©

§ fractional counts · ¸¸ . log( whole counts )  log¨¨ © whole counts ¹

The term

fractional counts will in the following be referred to as “correction term for whole counts

fractional counts”.

2.4.2.3

Explanatory Variables

level of education - The questionnaire included a question asking the respondents for their

terminal degree. In order to simplify the analysis, the education variable was aggregated to three groups: (1) secondary school, high school diploma, or vocational training (reference group), (2) vocational academy (Berufsakademie) or university studies, and (3) doctoral or postdoctoral studies. sources of knowledge – university research, scientific literature, patent literature, customers, and competitors. The questionnaire included a question relating to the

importance of different sources of knowledge for the development of an invention.25 Answers were collected on a scale from one (absolutely not important) to five (very important). A dummy variable was created for each source of knowledge, combining categories 1 (absolutely not important) to 3 (partly important) as well as categories 4 (important) and 5 (very important). The latter implies a use of the respective knowledge source. firm size - number of employees. The firm size was also obtained from the questionnaire. A

set of eight dummy variables was generated in order to account for variation across different firm sizes. The intervals range from “less than 50 employees” to “more than 50,000 employees”. Except for the first group (“less than 50 employees” = reference group), the dummies were included in the analysis. oppositions - The variable contains the share of granted patents per inventor that were

opposed by a third party within the opposition term of nine months after grant.

25

30

Although the answers to the questionnaire were related to specific patents, the answers seem to be transferable to all patent applications of an inventor. It is assumed that inventors basically tend to use special sources of knowledge, for example, due to positive experiences in the past. This assumption proves true, when comparing the answers of inventors who filled out more than one (five at the most) questionnaires. The answers of those inventors match completely or are at least highly correlated. The spearman correlation coefficients for the five different sources of knowledge range between 0.84 and 0.73.

status - This variable provides information on the status of the patent applications. Two share

variables were included accounting for the share of applications that were either refused by the examiner or withdrawn by the applicant, for instance, due to the results of the search report. The status variables as well as the opposition variable are included to control for the value of the patent applications. claims - This variable contains the number of claims added up for the total number of

applications per inventor. The claims define the scope of an invention for which patent protection is requested. Within the multivariate analysis, this variable is used to control for an increase in the number of references in the search report caused by an increase in the number of claims per application over time.

2.5 Descriptive Statistics und Multivariate Results 2.5.1 Descriptive Statistics

The empirical analysis is based on the responses of 2,63026 inventors who are responsible for a total of 26,601 EP patent applications. Table 2.1 presents data on the 25 most productive inventors (1% of all inventors in the sample) ranked by their total number of patents. These top 25 inventors account for 3,103 out of 26,601 patents, i.e., for 11.7%. The share of citations received by the patents of the top 25 inventors is even more concentrated. In particular, the inventors listed in Table 2.1 are responsible for patents that received 15.1% of all citations received by the 26,601 EP patent applications (35,926 citations). The top inventor is responsible for 258 patent applications which received a total of 721 citations (2% out of 35,926 citations received by all 2,630 patents). The second best inventor accounts for 201 EP applications and 374 corresponding forward citations. The results change when fractional counts are considered. The inventor with the highest number of fractional counts (80.43) is ranked 6th when whole counts are considered. For the data underlying this paper, the most productive inventors are almost solely employed with chemical firms. 9 out of the 25 leading inventors are employed with BAYER AG, another 8 with BASF AG. The top 25 inventors are further characterized by a high educational level. 20 out of 25 inventors have earned a doctoral degree, another four have

26

2,630 of the 3,349 questionnaires were filled out completely with regard to the above described variables.

31

some other university degree. Only one inventor out of the 25 had not received a high school diploma. 17 inventors had studied chemistry. The remaining 6 inventors have a degree in engineering, veterinary medicine or agronomy. These results suggest that industry effects have an influence on the patent output. In particular, the chemical industry has higher patent rates than other industries. Industry dummies will be included in the multivariate analysis to control for these industry differences

32

33

55

59

59

52

64

64

50

53

63

50

68

60

1

2

3

4

5

6

7

8

9

10

11

12

27

BASF AG

FESTO AG & Co

BASF AG

Clariant GmbH

BASF AG

BASF AG

HOECHST AG

BAYER AG

BAYER AG

BASF AG

BASF AG

BASF AG

employer

111

116

116

118

125

151

161

168

185

198

201

258

22.53

67.29

25.1

59.6

38.02

19.42

80.43

47.64

64.42

64.12

42.61

38.89

240

112

184

131

194

159

240

472

305

259

374

721

number of number of number of applications applications citations (fractional) received (whole)

27

doctoral studies

university studies

doctoral studies

doctoral studies

doctoral studies

doctoral studies

doctoral studies

doctoral studies

university studies

doctoral studies

doctoral studies

doctoral studies

highest level of education

chemistry

mechanical eng.

chemistry

inorganic chemistry

chemistry

agronomy

chemistry

chemistry

chemistry

chemistry

chemistry

chemistry

university degree (discipline)

2

6 to 8

3 to 5

3 to 5

6 to 8

2

6 to 8

6 to 8

3 to 5

3 to 5

3 to 5

3 to 5

number of diff. techn. areas

organic chemistry (74%)

mechanics (72%)

organic chemistry (81%)

organic chemistry (96%)

organic chemistry (74%)

1977

1983

1978

1978

1980

1989

1977

organic chemistry (95%)

1977

organic chemistry (66%) polymers (23%)

1977

1977

1977

1979

first priority

polymers (71%)

organic chemistry (94%)

organic chemistry (89%)

organic chemistry (97%)

organic chemistry (73%)

primary technical area

1998

1998

1997

1997

1998

1997

1997

1998

1998

1998

1998

1998

last priority

33

Clariant acquired a 50.1% stake of Hoechst’s specialty chemicals company Colour-Chem Ltd. in 1997. Today Clariant is headquartered in Muttenz near Basel (CH). See Clariant Annual report 2004, http://www.clariant.com/C12568C5004FDBD7/vwLookupDownloads/2004_Annual_Report_E.pdf/$FILE/2004_Annual_Report_E.pdf (access on November 11, 2005).

Table 2.1: 25 top-productive inventors (rank 1 to rank 12)

age in 1999

rank

34

49

54

47

67

53

41

49

61

57

40

66

48

61

13

14

15

16

17

18

19

20

21

22

23

24

25

Rolic AG

Basell Polyolefine

BAYER AG

BAYER AG

Heidelberger Druck.

BAYER AG

Degussa AG

BAYER AG

Agfa-Gevaert AG

BASF AG

BAYER AG

BAYER AG

BAYER AG

employer

77

79

81

82

82

90

92

94

98

100

100

110

110

27.32

24.74

55.33

13.31

41.57

33.02

28.05

26.68

29.4

21.34

17.05

28.09

31.71

280

143

122

122

136

64

63

65

178

296

179

250

153

number of number of number of applications applications citations (fractional) received (whole)

university studies

doctoral studies

doctoral studies

doctoral studies

university studies

doctoral studies

doctoral studies

doctoral studies

doctoral studies

vocational training

doctoral studies

doctoral studies

doctoral studies

highest level of education

34

Table 2.1 continued: 25 top-productive inventors (rank 13 to rank 25)

age in 1999

rank

physics

mechanical eng.

chemistry

veterinary medicine

precision eng.

organic chemistry

chemistry

chemistry

chemistry

chemistry

organic chemistry

organic chemistry

university degree (discipline)

1978

3 to 5

3 to 5

2

3 to 5

6 to 8

2

6 to 8

3 to 5

1984 1979

organic chemistry (42%), optics (33%)

1979

1990

1981

1977

1980

1987

surface (72%)

petrol chemistry (91%)

organic chemistry (88%)

print (78%)

organic chemistry (99%)

organic chemistry (72%)

polymers (81%)

1978

1984

organic chemistry (90%)

1977

1983

first priority

polymers (58%) petrol chemistry (20%)

polymers (90%)

polymers (97%)

primary technical area

polymers (30%) organic more than chemistry (25%) pharma 8 (17%)

6 to 8

2

6 to 8

3 to 5

number of diff. techn. areas

1997

1998

1995

1998

1997

1997

1998

1997

1998

1996

1997

1996

1997

last priority

The inventors hold patents in at least 2 different technical areas and “more than 8” at the most. Chemicals and polymers are the most frequent technical areas. As to the temporal concentration of inventive activity, the inventors look quite different. Whereas the inventors ranked 2 to 5 invented for at least 22 years, there are also inventors who exhibit 8 years of inventive activity (rank 7 and 22). Part of this variation arises due to age differences. The inventors, listed in Table 2.1, are aged between 40 and 66. I will address the relationship between inventive performance and duration of inventive activity in more detail in section 6 using panel estimation techniques. Table 2.2 presents selected descriptive statistics of the variables described in the previous section. Each inventor is on average responsible for about 10 EP patent applications.28 The total number of applications per inventor ranges between 1 and 258. The standard deviation amounts to 16.3. Fractional application counts range between 0.1 and 80.4 and have their mean at 4.4. The patent applications received an average of 13.7 citations, ranging from 0 to 721. Additionally, the cumulative number of applications per inventor contain on average 103.3 claims. The cumulative number of claims per inventor ranges between 1 and 2,613. Furthermore, Table 2.2 provides information on the legal status of the patents. On average 7% of the inventors’ granted patents were opposed by a third party, on average 11% of the applications had been withdrawn, and 2% had been refused by the EPO. Statistics of the EPO reveal that on average 29.7% of EP patent applications between 1980 and 1990 had been withdrawn by the applicant and 5.2% had been refused by the EPO (Harhoff/Wagner 2005). A possible reason for the low rates of withdrawal and refusal within this data is the fact that the dataset includes only patents of German inventors. The difference may especially arise due to a different behavior of German applicants in drafting patent applications. In particular, German applicants perform extensive search of prior art before filing a patent application. This should result in lower rates of withdrawal and refusal. The responding inventors were between 28 and 80 years old in 1999 with a mean at 49.8 years. Furthermore, the respondents were characterized by a very high level of education. Asked about their highest educational degree, 13% of the inventors indicated “high school diploma” or “vocational training”. More than 50% have a university degree and 34% a doctoral or post-doctoral degree.

28

In the PatVal questionnaire the inventor was asked to provide the number of EP patents he is responsible for. The correlation between the number of patents deriving from the EPOLINE database and the number of EP patents from the questionnaire reveals a correlation coefficient of 0.74 (0.76 when the logarithm of the variables is used). The correlation coefficient points at a high match between the information from the two different data sources.

35

Variable number of applications (whole) number of applications (fractional) number of citations received (whole) number of citations received (fractional) total number of claims age of the inventor in 1999 share of granted patents opposed share of applications refused share of applications withdrawn level of education (terminal degree) secondary school/vocational training / high school diploma university studies doctoral/post-doctoral studies external sources of knowledge university research scientific literature other patents customers competitors firm size (no. of employees)

Mean 10.14 4.36 13.66 5.17 103.30 49.78 0.07 0.02 0.11

S.D. 16.29 6.04 32.80 9.93 159.24 9.74 0.17 0.05 0.16

Min. 1 0.09 0 0 1 28 0 0 0

Max. 258 80.43 721 135.01 2,613 80 1 0.50 0.75

0.13 0.53 0.34

0 0 0

1 1 1

0.23 0.62 0.65 0.73 0.56 47,776

0 0 0 0 0 1

1 1 1 1 1 550,000

93,649

Table 2.2: Descriptive statistics (N = 2,630)

Users of the patented inventions as well as other patent documents turned out to be the most popular sources of knowledge. 73% of the inventors believe customers or users to be an important source of knowledge and 65% make use of other patent documents. Only 23% of the respondents believe university research to be important for making inventions. On average the patent assignees’ firms have 47,776 employees. The number of employees ranges between 1 and 550,000. The high standard deviation, amounting to 93,649, can be interpreted as a sign of a strongly heterogeneous distribution of this variable. In order to account for productivity variation across different firm sizes, for the following multivariate analysis firm size groups are used.29

29

36

However, it turns out that results using log(number of employees) as a measure of firm size are quite similar to the ones described in the next section. Especially, log(number of employees) turns out to have a positive and highly significant effect on productivity.

number of appl.

1.000 0.889* 0.935* 0.131* -0.058* 0.081* 0.188* 0.066* * significant at 5% or lower number of applications number of citations number of claims age in 1999 share_opposition share_refused share_withdrawn firm size

number of citations

number of claims

age in 1999

share_ opposition

share_ refused

share_ withdrawn

1.000 0.830* 0.136* -0.016 0.059* 0.134* 0.048*

1.000 0.113* -0.051* 0.080* 0.182* 0.037

1.000 0.064* 0.059* 0.078* -0.061*

1.000 -0.032 -0.097* -0.059*

1.000 0.061* 0.049*

1.000 0.001

firm size

1.000

Table 2.3: Pearson correlation coefficients (N = 2,630)

Table 2.3 lists the Pearson correlation coefficients for interval scaled variables. The dependent variable “number of applications” is positively correlated with age, the variables concerning the legal status of the EP patents, and firm size. The share of granted patents opposed by third parties possesses a negative correlation. Regarding “number of citations”, Table 2.3 presents quite similar results. The number of citations is also negatively correlated with the share of granted patents opposed, although not significantly. Overall, the correlation coefficients of the independent variables are quite small. Figure 2.5 and Figure 2.6 report the distribution of the number of patent applications per inventor. Whereas Figure 2.5 plots a histogram of whole application counts, Figure 2.6 shows the distribution of fractional counts. Due to the requirements of the stratified random sample, we used for the PatVal project, each inventor is (jointly) responsible for at least one granted patent with priority date between 1993 and 1997. Assigning the whole patent application to each inventor of an inventor team, 15% of the inventors are responsible for 1 patent application. The bigger part (58%) holds 2 to 10 applications. Only 3% are responsible for more than 50 applications. Fractional counts assign 1 application to 25% of the respondents and 2 to 10 patent applications to 45%. Only 0.3% of the inventors have more than 50 patent applications.

37

1200 966

1000

frequency

800 547

600

412

397 400

241 200 53

7

5

2

101 to 150

151 to 200

> 200

0 1

2 to 5

6 to 10

11 to 20

21 to 50 51 to 100

cumulative number of applications per inventor

Figure 2.5: Distribution of applications per inventor (whole counts) (N = 2,630)

700

660

600 531

frequency

500 394

400

419

359

300 200 143

116

100 8 0 1 to 2

> 2 to 3

> 3 to 5

> 5 to 10

> 10 to 15

> 15 to 50

> 50

cumulative number of applications per inventor (fractional)

Figure 2.6: Distribution of applications per inventor (fractional counts) (N = 2,630)

Figure 2.7 displays a histogram of the distribution of the citations the inventors received for their patent applications. The tail of the distribution (more than 60 forward citations) is displayed separately in the right hand upper corner. Nearly 20% of the inventors generated patent applications that received no citations at all. 2.1% of the inventors are responsible for applications that received more than 100 cumulative citations.

38

25 0.5

Fraction [%][%] fraction

Fraction[%] [%] fraction

20

15

10

0.4 0.3 0.2 0.1

721

296

250

194

164

153

142

130

123

112

108

93

101

87

82

77

71

65

61

0

5

Number of citations received (>60 citations) cumulative number of citations received (> 60 citations)

60

56

52

48

44

40

36

32

28

24

20

16

12

8

4

0

0

cumulative number of citations received

Figure 2.7: Distribution of the number of citations per inventor (N = 2,630)

The scatter plot displayed in Figure 2.8 reveals an approximate linear relationship between the cumulative number of patent applications per inventor and the cumulative number of citations the inventors’ patents received. More importantly, it shows a non-constant variation in the number of citations over the values of patent applications, referred to as heteroskedasticity, i.e., the variation in citations differs depending on the number of patent applications. Inventors responsible for a low number of patents receive only a small number of citations, and the variations in citation counts across such inventors are small. For highly productive inventors, the mean number of citations will be higher and there will also be greater variability among such inventors, resulting in heteroskedasticity. These results support the finding of Ernst et al. (2000) that highly productive inventors are also responsible for important patents.

39

total number of citations

total number of patent application

Figure 2.8: Relationship between number of patent applications and number of citations (scatter plot) (N = 2,630)

Table 2.4 to Table 2.6 which are described in the following, give first insights into relationships between dependent and independent variables. The relationship between the number of applications per inventor and the number of different technical areas an inventor’s patent applications are assigned to is reported in Table 2.4. Results show that the number of different technical areas is highly correlated with the number of applications. Especially, the number of applications per inventor increases with the number of technical areas. Inventors patenting in one area are on average responsible for four patents. Two technical areas lead to a mean of more than twice as many applications (9.3 applications). Inventors active in more than eight technical classes (more than 25% of all existing technical classes) on average are responsible for 53 EP applications. Overall, the number of applications is monotonically increasing with an increasing number of technical areas. An analysis of variance (ANOVA) confirms the relationship (F = 191.10, p = 0.000).

40

Number of different technical areas (groups)

1 technical area 2 technical areas 3 to 5 technical areas 6 to 8 technical areas More than 8 technical areas Total

Number of applications per inventor (whole counts) Number of observations 1,159 727 654 86 4 2,630

Mean 4.02 9.30 18.03 37.73 52.50 10.14

Note: In an ANOVA, the effect of the number of different technical areas turned out to be highly significant (F = 191.10, p = 0.000).

Table 2.4: Number of applications (whole counts) by number of technical area (N = 2,630)

Table 2.5 displays the relationship between the level of education of the inventors and whole application counts. Whereas inventors who earned a high school diploma or served a vocational training on average are responsible for 6 patent applications, inventors with a university degree are slightly more productive (mean = 7 applications). Inventors with doctoral or postdoctoral studies are more than twice as productive compared to the two other groups, holding an average of 16 applications (ANOVA: F = 100.17, p = 0.000).

Level of education (groups)

Number of applications per inventor (whole counts) Number of observations

secondary school/vocational training / high school diploma university studies doctoral/post-doctoral studies Total

333 1,396 901 2,630

Mean 6.09 7.24 16.12 10.14

Note: In an ANOVA, the effect of the level of education turned out to be highly significant (F = 100.17, p = 0.000).

Table 2.5: Number of applications (whole counts) by level of education (N = 2,630)

Finally, Table 2.6 presents the relationship between inventive output (number of applications) and firm size of the inventors’ employer (number of employees).

41

Firm Size in number of employees

Number of applications per inventor (whole counts) Number of observations 152 231 196 361 422 231 488 549 2,630

Less than 50 employees 51 – 250 employees 251 – 500 employees 501 – 1,500 employees 1,501 – 5,000 employees 5,001 – 10,000 employees 10,001 – 50,000 employees More than 50,000 employees Total

Mean 4.95 4.96 6.58 6.65 8.11 11.07 14.74 14.39 10.14

Note: In an ANOVA, the effect of the firm size turned out to be highly significant (F = 22.42, p = 0.000).

Table 2.6: Number of applications (whole counts) by firm size (N = 2,630)

Table 2.3 already contained a first reference to a positive correlation between the dependent variable and firm size. Results displayed in Table 2.6 reveal that inventive output increases almost monotonically with an increasing firm size (ANOVA: F= 22.42, p = 0.000). This result can be considered as a first indicator in favor of hypothesis 5.

2.5.2 Multivariate Specification

Due to the skewness of the productivity distribution known from the literature (Lotka 1926), a logarithmic transformation of the dependent variable is used. The hypotheses derived in section 2.3 as well as the transformation of the dependent variables described in section 2.4.2.2 lead to the following specification:

log(cumulative output k )

E 0  ¦ x kj E j  H k j

referring to the cumulative output (number of applications or number of citations) of an inventor k. The above described log-specification is estimated using an OLS regression. A Cook-Weisberg test (Breusch/Pagan 1979; Cook/Weisberg 1983), testing the assumption that the residuals are homoskedastic, was conducted. Results show that H0 (constant

42

variance) has to be rejected for all models displayed in Table 2.8. Due to these results, an OLS estimate using robust standard errors is employed.30 Table 2.7 provides the results of the Cook-Weisberg test.

dependent variable chi2 (1) Prop > chi2

(a) log (no. of applications) 78.39 0.000

Model 1 (industry dummies not included) (b) (c) log (no. of log (no. of applications) citations) 78.40 83.99 0.000 0.000

(d) log (no. of citations) 89.89 0.000

dependent variable chi2 (1) Prop > chi2

(a) log (no. of applications) 92.79 0.000

Model 2 (industry dummies included) (b) (c) log (no. of log (no. of applications) citations) 94.01 93.00 0.000 0.000

(d) log (no. of citations) 95.24 0.000

Table 2.7: Cook-Weisberg test for heteroskedasticity corresponding to the OLS regressions displayed in Table 2.8, H0: constant variance (N = 2,630)

Table 2.8 presents the results of different OLS regressions. Model (1) contains the control variables and the independent variables required for testing the hypotheses. Model (2) additionally includes variables controlling for variation between technological areas.

30

The Stata regress command includes a robust option for estimating the standard errors using the HuberWhite sandwich estimators; see http://www.ats.ucla.edu/stat/stata/webbooks/reg/chapter4/statareg4.htm (access on November 12, 2005).

43

44

44

Table 2.8: OLS regression with heteroskedasticity-robust standard errors (N = 2,630)

Robust standard errors in brackets / * significant at 10%; ** significant at 5%; *** significant at 1%

Model 1 (industry dummies not included) (a) (b) (c) (d) dependent variable log (no. of log (no. of log (no. of log (no. of applications) applications) citations) citations) log(no. fractional applications/no. applications) 0.005 [0.019] log(no. fractional citations/no. citations) -0.084*** [0.026] log(years of inventive activity) 0.144*** 0.143*** 0.255*** 0.265*** [0.021] [0.021] [0.030] [0.030] log(total number of claims) 0.790*** 0.790*** 0.912*** 0.908*** [0.008] [0.008] [0.011] [0.011] share of granted patents opposed -0.120** -0.120** 0.242*** 0.240*** [0.049] [0.049] [0.083] [0.082] share of applications refused 0.463*** 0.461*** 0.661*** 0.694*** [0.164] [0.164] [0.227] [0.221] share of applications withdrawn 0.354*** 0.354*** 0.253*** 0.250*** [0.051] [0.051] [0.071] [0.071] Observations 2630 2630 2630 2630 R-squared 0.883 0.883 0.836 0.837 F-test (df) 679.69(21) 650.02(22) 553.05(21) 535.50(22) 0.154*** [0.021] 0.786*** [0.008] -0.105** [0.049] 0.524*** [0.166] 0.332*** [0.051] 2630 0.888 323.23(50)

(a) log (no. of applications)

0.150*** [0.021] 0.787*** [0.008] -0.105** [0.049] 0.518*** [0.167] 0.330*** [0.051] 2630 0.888 316.86(51)

0.033 [0.020]

0.276*** [0.030] 0.904*** [0.011] 0.265*** [0.081] 0.719*** [0.218] 0.209*** [0.071] 2630 0.845 263.34(50)

Model 2 (industry dummies included) (b) (c) log (no. of log (no. of applications) citations)

-0.050* [0.027] 0.282*** [0.030] 0.902*** [0.011] 0.264*** [0.081] 0.732*** [0.215] 0.210*** [0.071] 2630 0.845 260.04(51)

(d) log (no. of citations)

45

0.026 [0.023] 0.058 [0.037] -0.030 [0.027] -0.019 [0.038] -0.029 [0.021] 0.013 [0.036] 0.030* [0.018] -0.009 [0.018] 0.003 [0.016] 2630 0.888 316.86(51)

(a) log (no. of applications) 0.026 [0.023] 0.056 [0.037] -0.030 [0.027] -0.019 [0.038] -0.031 [0.020] 0.012 [0.036] 0.029 [0.018] -0.011 [0.018] 0.003 [0.016] 2630 0.888 323.23(50)

0.02 [0.033] 0.142*** [0.050] -0.043 [0.038] -0.018 [0.055] -0.015 [0.028] -0.027 [0.051] 0.053** [0.026] -0.019 [0.026] -0.003 [0.023] 2630 0.845 263.34(50)

Model 2 (industry dummies included) (b) (c) log (no. of log (no. of applications) citations)

Table 2.8 continued: OLS regression with heteroskedasticity-robust standard errors (N = 2,630)

Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Model 1 (industry dummies not included) (a) (b) (c) (d) dependent variable log (no. of log (no. of log (no. of log (no. of applications) applications) citations) citations) level of education - highest degree (reference group: high school diploma or less) university studies 0.027 0.027 0.018 0.018 [0.023] [0.023] [0.033] [0.033] doctoral/postdoctoral studies 0.084** 0.085** 0.188*** 0.176*** [0.036] [0.037] [0.050] [0.050] source of knowledge - universities -0.031 -0.031 -0.049 -0.049 [0.027] [0.027] [0.038] [0.038] doctoral studies * knowledge_university -0.040 -0.040 -0.050 -0.049 [0.038] [0.038] [0.056] [0.056] source of knowledge - literature -0.019 -0.019 0.009 0.002 [0.020] [0.020] [0.029] [0.029] doctoral studies * knowledge_literature 0.035 0.035 0.006 0.000 [0.037] [0.037] [0.052] [0.051] source of knowledge - other patents 0.036** 0.036** 0.065** 0.061** [0.018] [0.018] [0.026] [0.026] source of knowledge - user -0.030* -0.030* -0.041 -0.043* [0.018] [0.018] [0.026] [0.026] source of knowledge - competitors 0.001 0.001 -0.005 -0.004 [0.017] [0.017] [0.024] [0.024] Observations 2630 2630 2630 2630 R-squared 0.883 0.883 0.836 0.837 F-test (df) 679.69(21) 650.02(22) 553.05(21) 535.50(22)

45

0.019 [0.033] 0.139*** [0.050] -0.043 [0.038] -0.019 [0.055] -0.018 [0.028] -0.029 [0.050] 0.052** [0.026] -0.021 [0.026] -0.003 [0.023] 2630 0.845 260.04(51)

(d) log (no. of citations)

46

46

-2.181*** [0.083] 2630 0.883 679.69(21)

-2.178*** [0.083] 2630 0.883 650.02(22)

-2.515*** [0.120] 2630 0.836 553.05(21)

-2.559*** [0.120] 2630 0.837 535.50(22)

0.052 [0.041] 0.143*** [0.043] 0.106*** [0.040] 0.155*** [0.039] 0.204*** [0.042] 0.251*** [0.041] 0.277*** [0.041] included Chi2(29) =3.77, p=0.00 -2.138*** [0.102] 2630 0.888 316.86(51)

(a) log (no. of applications) 0.049 [0.041] 0.138*** [0.043] 0.095** [0.039] 0.145*** [0.038] 0.193*** [0.042] 0.238*** [0.039] 0.264*** [0.040] included Chi2(29) =3.73, p=0.00 -2.156*** [0.102] 2630 0.888 323.23(50)

-0.016 [0.059] 0.083 [0.061] 0.074 [0.056] 0.176*** [0.055] 0.226*** [0.061] 0.273*** [0.056] 0.318*** [0.056] included Chi2 (29) =5.21, p=0.00 -2.611*** [0.133] 2630 0.845 263.34(50)

Model 2 (industry dummies included) (b) (c) log (no. of log (no. of applications) citations)

Table 2.8 continued: OLS regression with heteroskedasticity-robust standard errors (N = 2,630)

Robust standard errors in brackets / * significant at 10%; ** significant at 5%; *** significant at 1%

Observations R-squared F-test (df)

Constant

Wald test

Model 1 (industry dummies not included) (a) (b) (c) (d) dependent variable log (no. of log (no. of log (no. of log (no. of applications) applications) citations) citations) firm size in number of employees (reference group: less than 51 employees) 51 - 250 employees 0.055 0.055 -0.017 -0.027 [0.041] [0.042] [0.060] [0.060] 251 - 500 employees 0.152*** 0.153*** 0.108* 0.092 [0.044] [0.044] [0.063] [0.063] 501 - 1,500 employees 0.115*** 0.116*** 0.098* 0.070 [0.040] [0.041] [0.057] [0.058] 1,501 - 5,000 employees 0.167*** 0.169*** 0.204*** 0.175*** [0.039] [0.040] [0.056] [0.057] 5,001 - 10,000 employees 0.234*** 0.236*** 0.276*** 0.247*** [0.042] [0.043] [0.062] [0.063] 10,001 - 50,000 employees 0.302*** 0.305*** 0.365*** 0.324*** [0.039] [0.041] [0.057] [0.059] more than 50,000 employees 0.321*** 0.323*** 0.395*** 0.359*** [0.039] [0.040] [0.056] [0.058] technical area dummies not included not included not included not included

-0.021 [0.059] 0.075 [0.062] 0.057 [0.057] 0.160*** [0.056] 0.210*** [0.062] 0.253*** [0.058] 0.299*** [0.057] included Chi2(29) =4.86, p=0.00 -2.635*** [0.134] 2630 0.845 260.04(51)

(d) log (no. of citations)

First of all, the results concerning the different productivity measures are reported. The Models (a) and (b) (columns 1, 2, 5, and 6) use the logarithm of the total number of applications per inventor as the dependent variable. Models (c) and (d) employ the logarithmic number of citations, received by the total number of applications per inventor (= cumulative citations). Models (a) and (c) (columns 1, 3, 5, and 7) use whole application or citation counts as dependent variables. The remaining models ((b) and (d)) include the correction term for fractional counts (described in section 2.4.2.2) on the right hand side of the equation. Whereas the coefficient of the correction term for fractional application counts is not significant at the 10% level (columns 2 and 6), the correction term for fractional citation counts is significant in both models (Model 1(d) and Model 2(d), columns 4 and 8). Also the number of claims which was included in the regression to control for an increasing number of citations per patent application31 has a significant impact on the dependent variable (Models (c) and (d)). The number of claims further contributes to the explanation of the inventors’ output using the number of applications as an output measure. Additionally, Table 2.8 provides evidence that the cumulative output (applications or citations) rises less than proportionally with the age of the inventor. Age was calculated as the age of the inventor in 1999 minus 25, assuming that an inventor becomes active at the age of 25. The age variable can, therefore, be interpreted as the “years of inventive activity”. In Models 2(a) and 2(b), using patent applications as a dependent variable, the inventive activity coefficient amounts to 0.15. Thus, when the years of inventive activity increase by 1%, the number of applications rises by 0.15%. This effect is significant at the 1% level. In case the number of citations is employed (Models 2(c) and 2(d)), the coefficient of the years of inventive activity amounts to 0.28. Thus, when inventive activity increases by 1%, the number of citations rises by 0.28%. Interpreting the years of inventive activity as a proxy for the experience of an inventor, results reveal that experience is more important to make better inventions rather than more inventions. As already mentioned, the relationship between age and productivity will be analyzed more closely in the following section using a panel regression estimation. The control variables: the share of granted patents opposed by a third party, the share of applications refused, and the share of applications withdrawn, contribute to the explanation of inventive output. The share of patents either refused by the EPO or withdrawn by the applicant seems to be associated with the cumulative inventive output (applications and citations). This result may arise in the event that firms try to patent any invention. Patent

31

As described in section 2.4.2.1, an increasing number of claims per patent leads to an increasing number of references in the search report. Since the patent references are used to calculate the number of citations, the number of claims also increases the number of citations a patent receives.

47

applications that do not meet the requirements for patentability (novelty, inventive step and commercial applicability) are than rejected by the EPO during examination procedure. A withdrawal by the applicant takes place in case the firm is either no longer interested in obtaining a patent for the underlying invention or forestalls a rejection by the EPO. Additional control variables, such as the share of patent applications the inventors applied for in each technical area were also included. These variables control for heterogeneity that would otherwise lead to biased results of the explanatory variables. Table 2.8 illustrates that inclusion of this information also has some additional explanatory power. In the following, the results of Model 2, which controls for the distribution of applications across technical areas, are reported. Table 2.8 reveals that the correction term for fractional application counts (section 2.4.2.2) does not have additional explanatory power. Hence the results of Model 2(a) are discussed in more detail in the following. As the correction term for fractional citation counts (section 2.4.2.2), on the other hand, does have a significant effect, the results of Model 2(d) are also reported. The main differences between the results of Model 1(a) and Model 2(a) arise with respect to the level of education and with respect to the external sources of knowledge. Whereas doctoral studies are significantly associated with inventive output in Model 1(a), the effect is no longer significant after including the industry control variables. A possible explanation may be that the proportion of inventors who earned a doctoral or post-doctoral degree is the highest in the following disciplines: chemistry, pharmaceuticals and medicine. At the same time, the chemical and pharmaceutical industry relies heavily upon patents to protect inventions against imitation (Cohen et al. 2000). Furthermore, these industries produce the most important patents. It is therefore assumed that controlling for industry differences absorbs the effect of the educational degree on inventive output. Additionally, adding age to the regression model in part controls for the skills of an employee, since inventors who earned a higher educational degree are generally older compared to inventors with a lower level of education (Kim/Marschke 2005). Overall, the data support the conclusion that inventors with a high level of education tend to show a higher productivity. Furthermore, the significant effect of external sources of knowledge disappears after controlling for industry differences. Model 1(a) shows that inventors who make use of patent documents are 4% (Table 2.8, column 1) more productive than inventors who do not use this source of knowledge. Inventors attaching great importance to users as an external source of knowledge are 3% (Table 2.8, column 1) less productive compared to inventors who do not use information derived from users. Initially, the result concerning users as sources of knowledge was not expected since the involvement of users in the R&D process leads to knowledge spillovers. According to Foray (2004), user innovations are often innovations 48

without R&D and, therefore, almost never patented or even published or cited. Hence patent counts and citation counts are a very limited measure to account for knowledge spillovers from users. The coefficients of including users and patent literature as external sources of knowledge are no longer significant after having controlled for industry effects (Table 2.8, column 5). Again industry differences seem to explain the use of different sources of knowledge. Therefore, the hypotheses 2 and 3 are not supported in case the number of applications is used as a dependent variable. A set of dummies for firm size, measured by the number of employees (51 – 250 employees, 251 – 500 employees, 501 – 1,500 employees, 1,501 – 5,000 employees, 5,001 – 10,000 employees, 10,001 – 50,000 employees, more than 50,000 employees), show that an increasing firm size almost monotonically rises inventive output, compared to the reference group (less than 50 employees). The coefficients (except for 51 – 250 employees) are significant at the 1% level. As already discussed, this increase with firm size is a result of large firms having more resources to hire and retain high quality researchers. Additionally, the availability of more resources during the R&D process should also result in more patent applications per inventor. Overall, hypothesis 4 is supported by the data. In the following, the results of Model 2(d) are compared to the results of Model 2(a). In Model 2(d) the inventive output is measured using the cumulative number of citations per inventor. According to the literature, citation counts are a measure for the quality of output (Harhoff et al. 1999). As already reported before, a correction term for fractional citation counts was included in the regression which turned out to be significant. A coefficient of 0.05 for the correction term for fractional citation counts implicates the following optimal output measure regarding citation counts: § fractional citation counts · ¸¸ log( whole citation counts )  0.05 ˜ log¨¨ © whole citation counts ¹ 0.05 § § fractional citation counts · ·¸ ¸¸ log¨ whole citation counts ˜ ¨¨ ¨ © whole citation counts ¹ ¸¹ ©

^

log whole citation counts

0.95

˜  fractional citation counts

0 , 05

`

University studies alone again do not significantly influence qualitative output. Although the effect of doctoral studies decreases slightly after including industry control variables (column 4 vs. column 8), the effect remains significant. In particular, the patents of inventors who have a doctoral degree receive 14% (column 8) more citations compared to the control group (inventors who have a high school diploma or less). Overall, the coefficients are significant at the 1% level which at least in part supports hypothesis 1.

49

Again in contrast to the results employing quantitative output as a dependent variable, using other patents as an external source of knowledge increases the cumulative number of citations. Inventors who make use of patent documents receive 5% (column 8) more citations than inventors who do not use this source of knowledge. This result supports hypothesis 2 with respect to qualitative output. Hypothesis 3 which states that inventors who have a university or a doctoral degree are able to increase their output by using scientific sources of knowledge is not at all supported by the data. Compared to Model 2(a), firm size coefficients in part become insignificant when citation counts are used (column 5 vs. columns 7 and 8). For large firms, size still has an important effect on the cumulative output of an inventor. For instance, inventors employed with firms that have 5,001 to 10,000 employees hold patents that receive 21% (column 8) more citations as compared to inventors in the control group (firms with less than 51 employees). Inventors employed with firms that have more than 50,000 employees receive even 30% more citations (column 8). Firms that have less than 1,500 employees do not significantly differ from the control group (less than 51 employees). A possible explanation for the fact that small and medium sized firms are not associated with qualitative output, whereas large firms do significantly influence output quality, is that only large firms can provide enough resources to increase output quality. Finally, hypothesis 5 is supported.

2.6 Panel Regression 2.6.1 Description of the Methodology

As described in the last section, the age of the inventor has a significant effect on his productivity. One drawback of the previous multivariate analysis was that due to the aggregation of the data, it was not possible to trace the inventors’ productivity over time. To shed more light onto the relationship between productivity and age an additional panel data analysis will be conducted. To do so, the inventors’ patent applications were sorted into groups according to the age of the inventor at the time of the application of the patent. In particular, nine five-year age groups were constructed: 25-29 years, 30-34 years, 35-39 years, 40-44 years, 45-49 years, 50-54 years, 55-59 years, 60-64 years, and >64 years.

50

Before providing the results, the methodology of panel regression estimations will be described in general. The basic model can be written as y it

E 1 xit  c i  u it

where i indexes the different individuals and t the different time periods. ci denotes an unobserved individual effect, representing all factors affecting y that do not change over time, e.g., the educational degree of an inventor or his gender. uit is called the idiosyncratic error term (Wooldridge 1999). Two different methods exist that could be used for estimating the described unobserved effects panel data model: (1) the fixed effects estimator which uses the variation in explanatory variables over time to estimate regression coefficients. Inventor specific characteristics which are time invariant are automatically dropped from the equation procedure. Regression analysis is employed to provide unbiased, consistent estimators. (2) The random effects estimator which makes assumptions about the unobserved individual effect ci uses a GLS estimation. An advantage of the random effects model is that the coefficients of time invariant explanatory variables are estimated (Ruud 2000). The fixed effects approach uses a fixed effects transformation to eliminate the unobserved individual effect ci. Assume again the following model with a single explanatory variable for each unit i and the time periods t: y it

E 1 xit  c i  u it , t = 1, ... , T

(1)

As a first step, for each i, equation (1) is averaged over time, leading to the following equation: yi

E 1 xi  ci  u i , t = 1, ... , T

(2)

T

¦y where y i

t 1

T

it

, x1 and u i are calculated accordingly. Since ci is time-invariant it remains

unchanged in both equations. As a second step, equation (2) is subtracted from equation (1) for each t, leading to the following result: y it  y i

E1 ( xit  xi )  u it  u i

t = 1, ... , T

The parameters of the linear model with fixed individual effects can be computed by least squares regression (OLS) of yit* = ( y it  y i ) on the same transformation of xit where the 51

averages are group specific means. An OLS estimator which is based on the time-demeaned variables is called the within estimator (Greene 2001, Wooldridge 1999). For the random effects approach again the unobserved effects model is used: y it

E 0  E1 xit  ci  u it , t = 1, ... , T

(3)

Equation (3) represents a model with a single explanatory variable for each unit i and the time periods t. The intercept E0 is included to make the assumption that the unobserved individual effect ci has zero mean. An additional assumption of the random effects model - to obtain consistent coefficients - is the orthogonality between ci and xit. In other words, ci and xit have to be uncorrelated (Wooldridge 1999). Compared to the fixed effects model that uses a fixed effects transformation to eliminate the unobserved individual effect ci, the random effects model puts the unobserved individual effect ci into the error term. In case the composite error term is defined as Q it

ci  u it then

equation (3) can be written as y it

E 0  E1 xit  Q it , t = 1, ... , T

(4)

Because ci is the composite error in each time period, the Q it are serially correlated over time under the random effects assumptions: Corr (Q it ,Q is )

V c2 , tzs V c2  V u2

where V c2 = Var(ci) and V u2 = Var(uit). To solve the serial correlation problem, GLS estimation is employed. The GLS transformation is derived as follows: define

O 1  >V u2 /(V u2  TV c2 )@

1/ 2

which is between zero and one, than the transformation turns out to be y it  Oy i

E 0 1  O  E1  xit  Oxi  Q it  OQ i

where the overbar again denotes the time averages. Whereas the fixed effects estimator subtracts the entire time averages from the corresponding variable, the random effects estimator subtracts only a fraction of the time averages. Hence, this transformation allows estimating the coefficients of time invariant explanatory variables. The size of the fraction depends on V c2 , V u2 , and the number of time periods T. For instance, the smaller T, the 52

smaller O and consequently also the fraction of the time averages that is subtracted. In case

O equals one, the results of the random effects estimation are identical with the fixed effects estimation results (Wooldridge 1999, 2001).

2.6.2 Multivariate Results

The key consideration in choosing between a random effects and a fixed effects approach is whether ci and xit are uncorrelated which is – as described above - an assumption of the random effects model. To test this assumption Hausman (1978) proposed a specification test based on the differences between the random effects and the fixed effects estimates. In particular, the null hypothesis tests if the coefficients estimated by the efficient random effects estimator are the same as the ones estimated by the consistent fixed effects estimator (Wooldridge 2001). Results of the Hausman test show that H0 has to be rejected (Chi2 = 223.96, p = 0.000) which means the random effects estimators are not consistent. Therefore, in the following a fixed effects approach will be employed. The following fixed effects regression model will be estimated: log(citation countsit  1)

E 0  G 1 * (d _ age25  29) t  ...  G 8 * (d _ age ! 64) t

 E1 * ( priority82  87) it  E 2 * ( priority88  93) it  E 3 * ( priority ! 93) it  E 4 * (no _ claims) it  E 5 * (teamsize) it  E 6 j * ( status ) itj  E 7 k * (tech _ area) itk  u it

(t

1,...,9)

where i denotes the different inventors and t indexes the time period. Age groups of the inventors represent the nine time periods: 25-29 years, 30-34 years, 35-39 years, 40-44 years, 45-49 years, 50-54 years, 55-59 years, 60-64 years, and >64 years. The time periods do not change across i, which is why they have no i subscript. j indexes the status variables (application granted, withdrawn, or refused, patent opposed) and k denotes different technical areas.32 To measure the productivity of the inventors again the number of citations a patent application received within 5 years following the publication of the search report added up for the total number of patent applications per inventor and time period was used. As the number of citations is characterized by a highly skewed distribution, the logarithm of this variable was included in the regression. To accommodate zero values, one was added to the total number of citations before calculating the logarithm.

32

As proposed by Schmoch (OECD 1994) six main technical areas were used as control variables: electrical engineering, instruments, chemistry/pharmaceuticals, process engineering/equipment, mechanical engineering/machinery, and consumption.

53

As explanatory variables, age dummies were included in the panel regression to account for changes of productivity caused by the age of the inventor. Additionally, control variables for the priority years were factored into the regression model: priority year 1977-1981 (reference group), priority year 1982-1987, priority year 1988-1993, priority year >1993. The priority variables were essential to control for an increasing number of claims over time.33 Finally, further determinants of inventor productivity, e.g., the status of the patent applications or technical areas, were employed as control variables. To accommodate different patenting activities of inventors over time, the sample was subdivided into three groups. The first group includes inventors who were observable for at least five periods (25 years) within the panel (hereinafter referred to as long-term inventors). The second group comprises inventors observable for three to four periods (15 to 20 years) (hereinafter referred to as medium-term inventors). Inventors who were only observable during two periods (10 years) were sorted into the last group (hereinafter referred to as others). Whereas the long-term inventors (5 or 6 periods) represent inventors who kept on inventing for their whole professional life, the medium-term inventors (3 to 4 periods) include inventors who spent at least a major part of their professional life on inventing. Finally, others (2 periods) comprise two types of inventors: first inventors who stopped inventing and left R&D for another job, e.g., in sales or marketing. Second, inventors who were still at the beginning of their career in 1999 (inventors who were about 40 years old in 1999 or younger) and who could due to truncation of the data only be observed for two periods. Table 2.9 displays the results of the fixed effects panel estimation. Model 1 includes dummy variables for the age of the inventors. Model 2 (Table 2.10) additionally controls for an increasing number of citations over time by including control variables for the priority years of the patent applications. Model 3 (Table 2.) finally includes further determinants of productivity: the number of claims, the size of the inventor team, the status of the patent, and the technical area based on the IPC classes listed on the patent documents. Model (a) is estimated for the full sample of inventors. Models (b) - (d) refer to the three sub-samples described before.

33

54

Earlier in this paper it was shown (Figure 2.2) that the number of claims increased over time. Claims form the basis of the references included in the search report by the patent examiners at the EPO. The references in turn provide the basis of the calculation of the citations per patent.

Model 1 (a)

(b)

(c)

(d)

log(no. of citations + 1) sub-samples full sample age: 25 - 29 years 0.307*** [0.109] age: 30 - 34 years 0.621*** [0.073] age: 35 - 39 years 0.466*** [0.062] age: 40 - 44 years 0.228*** [0.053] age: 45 - 49 years 0.128*** [0.048] age: 50 - 54 years (reference group)

log(no. of citations + 1) 5 to 6 -0.420 [0.327] -0.089 [0.175] 0.145 [0.118] 0.293*** [0.098] 0.303*** [0.104]

log(no. of citations + 1) 3 to 4 0.412*** [0.129] 0.737*** [0.085] 0.636*** [0.073] 0.245*** [0.063] 0.120** [0.057]

log(no. of citations + 1) 2 0.969*** [0.208] 1.095*** [0.160] 0.741*** [0.144] 0.384*** [0.120] 0.045 [0.095]

age: 55 - 59 years

-0.344*** [0.114] -1.011*** [0.160] -1.484*** [0.292] 2.209*** [0.074] 929 184 14.81 (8,737) 0.144

-0.272*** [0.063] -0.621*** [0.088] -0.849*** [0.158] 1.433*** [0.046] 3538 1056 25.84 (8,2474) 0.078

-0.263*** [0.090] -0.386** [0.159] -0.288 [0.356] 0.830*** [0.099] 1990 995 9.46 (8,987) 0.074

age: 60 - 64 years age: > 64 years Constant Observations Number of inventors F-test (n1, n2)

-0.302*** [0.053] -0.704*** [0.080] -0.904*** [0.145] 1.378*** [0.039] 7236 3046 28.39 (8,4182) 0.069

R-squared Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Table 2.9: Robust fixed effects panel estimation (Model 1) (N = 7,236)

First of all, the outcomes of Model 1(a) will be discussed using results based on the full sample of inventors (Table 2.9, column 1). Results suggest that inventors aged between 24 and 29 receive 31% more citations compared to the reference group (50-54 years). Inventors aged between 30 and 34 receive 62% more citations. The early literature in this field proposes a maximum of productivity at the age of about 40 and a decline afterwards (Dalton/Thompson 1971; Lehman 1966; Oberg 1960). Model 1(a) does not confirm the findings of this earlier research. The number of citations rather reaches its maximum at the age of 30 to 34. As from the age of 34 the number of citations received decreases monotonically. Additional information is provided when dividing the sample into the three sub-samples according to the number of periods the inventors were observable in the panel dataset. Models 55

1(b) to 1(d) (Table 2.9, columns 2 to 4) provide the regression results for the three subsamples. Figure 2.9 displays the differences in the productivity-age relationship between the three sub-samples. The three curves present the logarithm of the medium number of citations the inventors received for patents applied for at the age of, e.g., 25 to 29 or 30 to 34. The upper curve represents long-term inventors who were observable for five or six periods. As proposed by the literature, the relationship between productivity and age is inverted ushaped and has its maximum at an age of about 40 years. The medium curve represents medium-term inventors who were observable for three to four years. These inventors still spent a considerable share of their professional career in R&D (15 to 20 years) but are supposed to have stopped inventing at a certain point in time. Figure 2.9 shows that mediumterm inventors are at the age of 25 to 35 even more productive than the long-term inventors (5 to 6 periods). After the age of about 35 the productivity of the medium-term inventors decreases steadily. This could mean that these inventors are characterized by a very high level of productivity and are consequently promoted. These inventors may then stop inventing or at least spend only part of their time on inventive activities leading to a lower observable productivity. Finally, the lower curve represents others who were only observable for two periods (about 10 years). Others receive, almost as from the beginning of their career, less citations compared to the other two groups. These inventors could first of all drop out of the sample since they are unsuccessful inventors and change to an administrative position or another position within the firm (or leave the firm completely). Second, these inventors may be still very young in 1999. Therefore, truncation of the data impedes observing these inventors any longer. Young inventors may be mistakenly sorted into sub-sample three (others). In the event these inventors are indeed on average more productive than the unsuccessful inventors, the first two or three age groups of the lower curve (including these young inventors) should suffer from an overestimation of productivity. Overall, it becomes clear that the patent applications of inventors remaining in R&D for a longer time receive more citations. Additionally, the findings of the literature are supported but only for long-term inventors.

56

3,5 3,0

log(productivity) ..

2,5 2,0 1,5 1,0 0,5 0,0 -0,5

25_29

30_34

35_39

40_44

45_49

50_54

55_59

60_64

>64

-1,0 -1,5

age groups long-term inventors (5 to 6 p eriods)

medium-term inventors (3 to 4 p eriods)

others (2 p eriods)

Figure 2.9: Productivity differences by age groups; subdivided into three groups by number of periods observed (N5_6 = 929, N3_4 = 3,538, N2 = 1,990), graph of: log(no _ cit it  1) E 0  G 1 * d _ age25  29 t  ...  G 8 * d _ age ! 64 t  u it where the coefficient G1 is the percentage change in productivity between the reference group and the first age group. G2 to G8 have the same interpretation with respect to the remaining age groups. E0 is the intercept for the reference group and E0 + G1 is the intercept for the first age group.

The second model (Table 2.10) includes control variables for the mean share of priorities within the different age groups. The share of priorities between 1977 and 1981 forms the reference group. These variables were included in the regression to control for a time trend, in particular, for an increasing number of references over time due to an increasing number of claims per patent. Results of Model 2 (Table 2.10) confirm this finding for the whole sample (Model (a)) as well as for the three sub-samples (Models (d)-(d)). In particular, the number of citations increases over time. The coefficients are highly significant at the 1% level.

57

Model 2 (a)

(b)

(c)

(d)

log(no. of citations + 1) full sample

log(no. of citations + 1) 5 to 6

log(no. of citations + 1) 3 to 4

log(no. of citations + 1) 2

sub-samples reference group: age: 50 - 54 years age: 25 - 29 years 1.852*** 2.249*** [0.200] [0.662] age: 30 - 34 years 1.943*** 2.160*** [0.155] [0.524] age: 35 - 39 years 1.553*** 1.777*** [0.122] [0.379] age: 40 - 44 years 1.005*** 1.330*** [0.089] [0.270] age: 45 - 49 years 0.506*** 0.753*** [0.061] [0.171] age: 55 - 59 years -0.581*** -0.695*** [0.063] [0.157] age: 60 - 64 years -1.235*** -1.724*** [0.104] [0.274] age: > 64 years -1.763*** -2.697*** [0.174] [0.458] reference group:(mean) priority year: 1977 - 1981 (mean) priority year: 1982 - 1987 0.805*** 1.122*** [0.085] [0.184] (mean) priority year: 1988 - 1993 1.480*** 1.780*** [0.111] [0.320] (mean) priority year: > 1993 1.620*** 2.101*** [0.151] [0.482] Constant -0.530*** 0.352 [0.171] [0.381] Observations 7236 929 Number of inventors 3046 184 42.37 16.11 F-test (n1, n2) (11,4179) (11,734) R-squared 0.126 0.207 Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

1.923*** [0.255] 2.007*** [0.198] 1.646*** [0.154] 0.984*** [0.109] 0.500*** [0.073] -0.558*** [0.078] -1.149*** [0.122] -1.691*** [0.208]

1.855*** [0.298] 1.869*** [0.226] 1.388*** [0.188] 0.874*** [0.149] 0.311*** [0.106] -0.522*** [0.105] -0.902*** [0.189] -1.164*** [0.327]

0.590*** [0.102] 1.342*** [0.134] 1.478*** [0.192] -0.211 [0.200] 3538 1056 33.45 (11,2471) 0.134

0.950*** [0.303] 1.469*** [0.308] 1.649*** [0.327] -1.121*** [0.375] 1990 995 10.86 (11,984) 0.102

Table 2.10: Robust fixed effects panel estimation (Model 2) (N = 7,236)

Model 2(a) provides the results of the full sample. Similar to Model 1(a), the turning point of productivity is reached at the age of 30 to 34 years. Afterwards, productivity again decreases but more rapidly compared to Model 1. Figure 2.10 in turn provides the productivity-age relationship for the three sub-samples. It becomes apparent that long-term inventors (5 to 6 periods) are most productive and others (2 periods) again turn out to be least productive. Again the productivity of the first two to three 58

age groups of others seem to be overestimated due to truncation of the data. However, the shape of the three curves has changed considerably. Especially the upper curve (long-term inventors) no longer shows a turning point at the age of 45 but at the age of 34. Additionally the upper curve has its maximum at a productivity level of about 2.75 (Figure 2.9) compared to a maximum productivity level of 1.25 provided in Figure 2.10. This change arises due to the correction of the time trend (earlier citations are weighted more strongly). Overall, comparing Figure 2.9 and Figure 2.10 reveals that including controls for a time trend leads to a downward correction of the inventors’ productivity.

1,5 1,0

log(productivity) ..

0,5 0,0 -0,5

25_29

30_34

35_39

40_44

45_49

50_54

55_59

60_64

>64

-1,0 -1,5 -2,0 -2,5 -3,0

age groups long-t erm inventors (5 to 6 periods)

medium-t erm invent ors (3 to 4 periods)

ot hers (2 periods)

Figure 2.10: Productivity differences by age groups (additional control for the priority years of the patents); subdivided into three groups by number of periods observed (N5_6 = 929, N3_4 = 3,538, N2 = 1,990), graph of: log(no _ cit it  1) E 0  G 1 * d _ age25  29 t  ...  G 8 * d _ age ! 64 t  E1 * priority82  87 it  E 2 * priority88  93it  E 3 * priority ! 93it  u it

Finally, Model 3 which is displayed in Table 2. includes determinants of productivity as further control variables. I will refrain from describing the results of Table 2., since the findings are very similar to those of Model 2 (Table 2.10) as regards the signs of the effects and the significance. Simply the absolute magnitude of the coefficients exhibits a decrease due to the explanatory power of the additional control variables.

59

Model 3 (a) (b) (c) (d) log(no. of log(no. of log(no. of log(no. of citations+1) citations+1) citations+1) citations+1) sub-samples full sample 5 to 6 3 to 4 2 reference group: age: 50 – 54 years age: 25 – 29 years 0.910*** 1.295* 0.976*** 1.046*** [0.220] [0.727] [0.275] [0.329] age: 30 – 34 years 1.168*** 1.183** 1.222*** 1.233*** [0.172] [0.580] [0.215] [0.253] age: 35 – 39 years 0.973*** 1.110*** 1.047*** 0.916*** [0.134] [0.422] [0.167] [0.203] age: 40 – 44 years 0.625*** 0.836*** 0.601*** 0.570*** [0.096] [0.299] [0.117] [0.154] age: 45 – 49 years 0.320*** 0.508*** 0.313*** 0.187* [0.062] [0.179] [0.075] [0.104] age: 55 – 59 years -0.399*** -0.492*** -0.363*** -0.425*** [0.065] [0.166] [0.080] [0.107] age: 60 - 64 years -0.858*** -1.249*** -0.740*** -0.715*** [0.109] [0.297] [0.130] [0.188] age: > 64 years -1.140*** -1.872*** -1.071*** -0.727** [0.184] [0.500] [0.220] [0.334] reference group:(mean) priority year: 1977 - 1981 (mean) priority year: 1982 - 1987 0.582*** 0.878*** 0.373*** 0.755** [0.087] [0.199] [0.104] [0.318] (mean) priority year: 1988 - 1993 0.995*** 1.245*** 0.855*** 1.068*** [0.120] [0.356] [0.143] [0.332] (mean) priority year: > 1993 1.086*** 1.532*** 0.965*** 1.192*** [0.159] [0.504] [0.198] [0.349] Observations 7236 929 3538 1990 Number of inventors 3046 184 1056 995 31.41 11.15 24.94 8.10 F-test (n1, n2) (22,4168) (22,723) (22,2460) (22,973) R-squared 0.166 0.259 0.177 0.147 Robust standard errors in brackets / * significant at 10%; ** significant at 5%; *** significant at 1%

Table 2.11: Robust fixed effects panel estimation (Model 3) (N = 7,236)

60

(a) log(no. of citations+1) sub-samples full sample (mean) no. of claims 0.010*** [0.004] (mean) no. of inventors 0.069*** [0.015] (mean) share withdrawn 0.385*** [0.087] (mean) share refused 0.436** [0.199] (mean) share granted 0.671*** [0.067] (mean) share opposed 0.305*** [0.082] n.s. (mean) share technical areas Wald test

Model 3 (b) (c) log(no. of log(no. of citations+1) citations+1) 5 to 6 3 to 4 0.012 0.008* [0.010] [0.004] 0.037 0.071*** [0.037] [0.017] 0.283 0.479*** [0.230] [0.105] 0.515 0.495** [0.497] [0.230] 0.840*** 0.762*** [0.217] [0.083] 0.510** 0.279*** [0.244] [0.095] n.s. Chi2(5)=2.14

(d) log(no. of citations+1) 2 0.015*** [0.006] 0.079*** [0.025] 0.228 [0.140] 0.293 [0.362] 0.439*** [0.098] 0.332** [0.134] n.s.

p=0.059 0.355** 1.689*** 0.522*** -0.059 [0.159] [0.531] [0.201] [0.402] Observations 7236 929 3538 1990 Number of inventors 3046 184 1056 995 31.41 11.15 24.94 8.10 F-test (n1, n2) (22,4168) (22,723) (22,2460) (22,973) R-squared 0.166 0.259 0.177 0.147 Robust standard errors in brackets / * significant at 10%; ** significant at 5%; *** significant at 1% Constant

Table 2.11 continued: Robust fixed effects panel estimation (Model 3) (N = 7,236)

2.7 Conclusion The purpose of this paper was to develop a productivity measure that controls for output quality, firm size effects as well as an increasing patenting activity over time. Over the last two decades firms have begun to apply for more patents per unit of R&D expenditure due to strategic reasons. To deal with strategic patenting, citation counts are used as an alternative output measure. Since patent counts represent a quantity measure whereas citation counts represent a quality corrected output measure, citations are assumed to be less affected by an increasing patenting activity. Additionally, organizational differences of large firms compared to smaller firms have been taken into account. Inventors working with large firms are highly specialized and work as members of large inventor teams on many different projects in parallel. Failing to control for these organizational differences would lead to biased productivity measures, since the number of patents per inventor working in large firms would be overestimated compared the cumulative number of patents of inventors employed with small firms. To deal with these differences, correction terms for fractional counts (section 61

2.4.2.2) were included in the regression. Results reveal that whereas the correction term for fractional application counts is not significant at the 10% level, the correction term for fractional citation counts has a significant effect. Results show that output quantity is almost not dependent on the characteristics of the inventor (with exception of age). A reason for these results may be that the personal attributes or abilities of one single inventor do not matter since more than 90% of the inventions are made by inventor teams. It is likely that inventors who themselves have no university or doctoral degree but work in an inventor team with university graduates profit from their colleagues. Another explanation may be that - as already explained in section 2.2.2 inventors have less control over their quantitative output than over their qualitative output. R&D management, for instance, determines whether to file a patent application for an invention or how much to spend on R&D. Therefore, counting the patents of an inventor seems to be a rather inadequate measure for inventor productivity. Concerning the output quality, results indeed reveal that both, the allocation of resources (e.g., represented by firm size) and the inventive ability (e.g., represented by the educational level of an inventor) are important determinants of inventive output. Apparently, personal characteristics or skills of an inventor and the resources he has at his disposal are complements rather than substitutes in enhancing qualitative output. These findings suggest that the number of citations an inventor’s patents received within a certain period of time is a more appropriate productivity measure than pure patent counts. A second purpose of this paper was to analyze the age-performance relationship of inventors more closely, in particular, to trace inventor productivity over time. To do so, an additional panel regression model was estimated. Overall, results of the panel estimation provide clear evidence that the age of an inventor considerably influences his productivity. Time series data tracing the inventor’s professional life were needed to estimate this influence correctly. The results further suggest that one has to distinguish between long-term inventors and inventors who dropped out of R&D for certain reasons (medium-term inventors, others) to avoid biased results. Whereas long-term inventors remain visible in terms of patents over the whole period under consideration, medium-term inventors are no longer visible after they left R&D. Comparing the mean productivity of both groups over time leads to an underestimation of productivity of the inventors who stopped inventing earlier. Furthermore, there is considerable evidence that failing to control for an increasing number of citations over time (caused by an increasing number of claims per patent application) would also lead to biased results. Additionally, future research is needed to shed more light onto the inventors’ life cycle, for instance, onto reasons for leaving R&D. It is also necessary, to analyze career systems for R&D personnel offerd by firms more closely. In case, firms do not provide a dual ladder 62

career system for management and R&D, a move into a management position may be the only way for a productive inventor to get promoted. It will be interesting to analyze whether transferring productive inventors into management positions causes damage to the innovative potential of the firm or whether able inventors who agree to move to a management position do an even better job as a manager. Personnel interviews with inventors will help to answer these questions.

63

Chapter 3 3 Tracing Mobile Inventors - The Causality between Inventor Mobility and Inventor Productivity 3.1 Introduction Kai-Fu Lee, an expert on speech recognition and search technologies, graduated in computer science at Columbia University in 1983, and earned a Ph.D. in computer science from Carnegie Mellon University. In 1998, Lee moved to Microsoft to found the Chinese Microsoft research division in Beijing. In 2000 he became vice president of interactive services at Microsoft. In July 2005, Lee left Microsoft to work for Google. While working for Microsoft, Lee had signed a non-compete agreement, which bared him from working in research areas competing with Microsoft within one year after leaving the company. On July 19, 2005, Microsoft sued Google and Lee claiming that Lee was violating his non-compete agreement since working for Google would unavoidably lead to the disclosure of technical know-how to Google. On July 28, the Washington State Superior Court enacted a preliminary injunction, which prevented Lee from working on Google projects that competed with Microsoft. After a second hearing in September 2005, Lee was still not allowed to work in his main research area: speech recognition and search technology. On December 22, 2005 Google and Microsoft announced that they had entered into a private agreement, which put an end to the dispute between the two companies. The Google-Microsoft story gives first insights into possible consequences of a key employee leaving a firm. Kai-Fu Lee is an expert in the field of speech recognition and search technology. A move from Microsoft to Google not only weakened the position of Microsoft in this research field but also strengthened the position of the competitor. For Microsoft a legitimate reason to take court action. Given this story, it is interesting to analyze whether productive R&D personnel is especially jeopardized to get job offers from competitors and,

The survey responses used in our analysis originate from a coordinated survey effort in Italy, France, Spain, the Netherlands, the United Kingdom and Germany. The author thanks the European Commission, Contract N. HPV2-CT-2001-00013, for supporting the creation of the joint dataset. This paper makes use of the German survey responses.

65

consequently, leaves more often or whether the Google-Microsoft case is only a special case that received particular public attention. On the one hand, moving R&D personnel is exposed to a new environment that affects their activity. For instance, Topel and Ward (1992) propose that mobility can lead to an increase of the match quality between employer and employee. A better match quality should lead to an increase in the inventor’s own productivity. Furthermore, the inventor may profit from the knowledge of his new colleagues. This could also increase the productivity of an inventor. One might, therefore, expect that mobility increases productivity34. On the other hand, the literature reveals that hiring a key inventor from another firm can lead to knowledge transfer. Firms characterized by a lower technology level can use this knowledge to catch up and thus are motivated to attract productive inventors (Gilfillan 1935). In particular, the transfer of tacit knowledge, that is otherwise immobile, is facilitated by inventor mobility (Dosi 1988). One could, therefore, assume that the causality runs from productivity to mobility: the more productive an inventor is, the higher the probability to observe a move. Nevertheless, one has to bear in mind that inventors who are very valuable to their employers may be treated with particular attention. In particular, employers try to keep these inventors from leaving the firm by providing certain incentives. Whereas existing research on inventors implicitly assumes causality to point in one direction (from mobility to productivity or from productivity to mobility), this study ex-ante allows for a simultaneous relationship. One reason for the lack of literature which deals with this causality is the absence of appropriate data. First of all, a matching problem exists with respect to name and address information derived from the patent documents.35 Furthermore, bibliographic and procedural data hardly suffice to represent the most important determinants of productivity or mobility. Additional information is needed on the inventor himself, for instance, on the inventor’s age or educational background. This paper makes use of data collected in a large-scale survey of 3,049 German inventors who hold at least one granted European patent. The inventors were requested to provide demographic information as well as information on the R&D process underlying their patented invention. To trace the mobility and the productivity of each inventor over time, the EPOLINE database of the European Patent Office was used to search for all patent applications belonging to the 3,049 inventors with priority dates between 1977 and 2002, resulting in a total of 39,417 EP patent applications.

34

The productivity of inventors is measured by relating the number of applications per inventor to the age of the inventor.

35

See for instance Hall (2004): The Patent Name-Matching Project, http://emlab.berkeley.edu/users/bhhall/ pat/namematch/namematch.html (access on November 28, 2005).

66

Since, as described above, a simultaneous relationship between productivity and mobility is proposed, instrumental variables techniques will be employed. The results concerning inventor productivity are consistent with the findings in Chapter 3. In particular, the level of education has no influence on inventor productivity. Making use of external sources of knowledge, on the contrary, has a significant effect on productivity. In particular exploiting the knowledge from scientific literature decreases inventive output. Finally, firm size has a positive impact on productivity. Firm size also influences inventor mobility, although negatively. Furthermore, the temporal concentration of inventive activity and the inventive environment are major determinants of mobility. Whereas the number of moves decreases with the duration of inventive activity, it is higher in large cities compared to rural areas. Results reveal a simultaneous relationship between inventor productivity and inventor mobility. Whereas mobility increases productivity, an increase in productivity decreases the number of moves. One drawback of the regression analysis described before is that it disregards the time structure of the data. Time series data can help to provide a better understanding of the impact of a certain move on inventive performance. To examine the effect of a particular move an additional quasi-experimental design will be employed comparing the performance of an inventor before and after a selected move. Difference-in-differences estimation reveals that patent applications in the post-move period are more important than those in the pre-move period. In particular, the share of patents granted as well as the share of patents opposed by a third party is higher in the post-move period. Additionally, patents receive more references and also more citations after the move has occurred. The remainder of this paper is organized as follows. Section 2 contains the derivation of the hypotheses from the literature. A description of the dataset as well as the operationalization of the variables used in the empirical part of the paper are provided in section 3. Section 4 provides descriptive statistics, followed by a two stage least squares regression model to analyze the causality between inventor productivity and inventor mobility. Section 5 contains a difference-in-differences estimation to provide a closer look at the impact of one particular move on inventive performance. Finally, section 6 discusses the estimation results and provides implications for further research.

67

3.2 Hypotheses The following section provides the hypotheses to be tested in the multivariate analysis. Section 3.2.1 presents the hypotheses concerning the productivity of inventors. Section 3.2.2 contains the hypotheses regarding inventor mobility. Finally section 3.2.3 provides two hypotheses regarding the causality between productivity and mobility.

3.2.1 Hypotheses on Inventor Productivity

Since the determinants of productivity36, either related to the inventor himself or to his inventive environment, were already summarized in Chapter 3 of this thesis37 I will refrain from deriving the hypotheses of inventor productivity again. The following hypotheses will be tested in the multivariate analysis: P.1:

Inventors with a high level of education tend to show higher productivity than inventors with a low level of education.

P.2:

Inventors making use of patent literature, users’ knowledge or competitors’ knowledge are more productive than inventors who do not use these external sources of knowledge.

P.3:

Inventors who conducted scientific research increased their productivity more by using university research or scientific literature than inventors who did not conduct scientific research.

P.4:

Inventors who are employed with a large firm tend to show a higher productivity than inventors working at small firms.

3.2.2 Hypotheses on Inventor Mobility

In the following section, hypotheses with respect to the determinants of mobility are derived from the literature. Again, the determinants include personal characteristics of the inventor and his work environment.

36

This paper builds on prior research, providing an improved productivity measure as well as determinants of inventor productivity. For a detailed overview of the theoretical and empirical literature on inventor productivity, see Chapter 3.

37

For the derivation of the hypotheses from the existing literature, see section 2.3 of this thesis.

68

Spence (1973) suggests that hiring an employee constitutes an investment under uncertainty since the employer is not sure of the capabilities of an employee at the time he hires him. But certain characteristics of the individual are observable and hence can be used to decrease this uncertainty. Tuma (1976) uses a sample of 695 Chicago family heads to analyze the relationship between education and social mobility. The data include a continuous job history, whereby jobs are treated as a proxy for the social position. The results show that an increase in education (years of completed schooling) indeed increases the rate of mobility. The level of education of the inventor can be used as a signal for qualification. Therefore, inventors with a higher level of education may get more job offers and consequently may move more often. It is, hence, assumed that M.1:

Inventors with a higher level of education change employers more often than inventors with a lower level of education.

Additionally, monetary incentives can determine the decision of an inventor to change employers. Doeringer and Piore (1971) propose that mobility within firms occurs more often compared to mobility between firms. Mobility within firms or “osmotic mobility” means that the tasks of an employee change gradually. Osmotic mobility quite often leads to an advancement and can be seen as a compromise between promotion by seniority and by ability (Doeringer/Piore 1971). These findings suggest that inventors who attach importance to monetary rewards are less likely to move. Otherwise, Allen and Katz (1985) find that career systems of engineers and scientists are completely different. Engineers and scientists are often forced into administrative roles to receive higher salary levels. In general, career prospects are less promising for technical professionals. In cases, where progress in terms of salary or advancement is impossible within the current employment, a change of employer could help sustain their motivations. Therefore, the following relationship is expected: M.2:

Inventors who classify “increase in salary” and “advancement” as important motives for inventive activity change employers more often than inventors who do not classify these motives as important.

Furthermore, improvement of working conditions can be a motive to change the employer. Freeman (1978), for instance, finds that job satisfaction is a major determinant of labor market mobility. This finding is confirmed by Clark et al. (1998) who use data of the German Socio-Economic Panel to examine the effects of job satisfaction on employees’ future termination behavior. Results show that workers who are dissatisfied with their jobs are more

69

likely to quit compared to highly satisfied workers. Results of the PatVal survey, for instance, reveals that inventors are frequently dissatisfied with the amount of inventor compensation they receive according to the German Employees’ Invention Act. Since inventors do not want to jeopardize their careers by enforcing their claim to remuneration by legal action, continuous dissatisfaction may lead to a change of the employer. Hence the following hypothesis is proposed: M.3:

Inventors who classify “improvement of working conditions” as important as a motive for inventive activity change employers more often than inventors who do not think that “improvement of working conditions” is important for inventive activity.

Topel/Ward (1992) use longitudinal employee-employer data containing records for over one million individuals between 1957 and 1972. The authors find that jobs are more stable in large firms. This means that the turnover rate in the smallest class is double that of the largest class (1-9 vs. 1000-2499 employees). The authors provide the following explanation for their results: large organizations encompass transitions that would otherwise occur between smaller ones. Especially, large firms provide internal job markets.38 Careers, therefore, can develop within the firm and the employees need not move out. Also the probability that so-called “dual ladder” career systems, as proposed by Allen and Katz (1985), are established, providing more career chances for engineers, increases with firm size. Therefore, the following relationship is expected: M.4:

Inventors employed by large firms change employers less often than inventors employed with small firms.

Finally, Marshall (1964) recognized that workers may be economically more valuable to one firm than to all other firms. The author stated that firm-specific human capital may be a reason for this phenomenon. Becker (1962) was the first to systematically formulate a theory of investment in human capital. He proposed that workers decide how much firm specific human capital to accumulate. Quit rates fall with tenure because workers with more tenure tend to have more specific human capital. Parsons (1972) also finds that large investments in firm-specific human capital, either by the firm or the worker, are likely to lead to reduced labor mobility, since the economic cost of worker-job separations is increased. An example

38

70

See, for instance, Althauser (1989) for a review of theoretical and empirical studies on internal labor markets.

for firm specific human capital is the technical specialization of an inventor. A highly firm specific technical concentration of inventive activity can lead to a lower value of an inventor in the labor market. Thus, the following hypothesis is proposed: M.5:

Inventors whose patent applications are concentrated in a small number of technical areas change employers less often than inventors whose patenting activity is diversified.

3.2.3 Hypotheses on the Causality between Productivity and Mobility

A review of additional relevant literature concerning inventor productivity and inventor mobility reveals two different streams: on the one hand, papers use labor economics models to analyze the relationship between mobility and match quality. This literature emphasizes that mobility can lead to a better match quality, resulting in a higher productivity of an employee. The second stream of literature analyzes mobility as a mechanism of knowledge transfer. In particular, knowledge is transferred by productive employees. Therefore, productivity should also increase the probability of a move. First, important results derived from the labor economics literature are presented. Jovanovic (1979) provides a model of permanent job separations39. His results show that the probability of a resignation increases with job tenure. The author explains the finding with the fact that the worker accumulates information about the match quality40 over the years on the job. Therefore, uncertainty about match quality declines with tenure and poor matches are more likely to be terminated. Liu (1986) uses data derived from a labor force survey conducted among employees in the manufacturing sector of Singapore in 1974. He finds that an increase in productivity over time is attributable to inter-firm and intra-firm mobility of workers. A reason for this relationship is that inter-firm and intra-firm mobility lead to a better match quality between “unobservable worker productivity and the productivity required by the firm for the job” (Liu 1986: 1145). The literature supports the perception that mobility can positively affect labor productivity. It is assumed that match quality between employer and employee is also relevant for inventors

39

Permanent job separation implies that the worker definitely changes the employer, temporary separations, on the contrary, are, for instance, temporary stays abroad or temporary layoffs, presuming a return of the worker (Jovanovic 1979).

40

As a measure for match quality, differences in wages, not explained by observable factors like education or labor market experience, may be employed (Widerstedt 1998).

71

to increase inventive output. A move can hence be interpreted as a search process to improve match quality. The second stream of literature deals with mobility as a means to transfer knowledge between firms. In 1935, Gilfillan proposed that labor mobility enables firms characterized by a lower technology level to catch up with technological leaders. Arrow (1962: 615) finds that “mobility of personnel among firms provides a way of spreading information”. Dosi (1988) goes even one step further and suggests that it is not only information or codified knowledge that is transferred by mobile inventors. Hiring employees from another firm is a way of transferring tacit knowledge that is otherwise immobile. Song et al. (2003) use data on engineers who moved from U.S. to non-U.S. firms. Results reveal that inter-firm knowledge transfer is more likely when the technological expertise of the hired engineer is distant from that of the hiring firm. This finding is not surprising since non-competition clauses in employment contracts or so-called employee non-compete agreements41 can impede knowledge transfer, in cases when the core technological area of the new employer is similar with that of the former employer (Fleming/Marx 2005, Sauter 2000). However, knowledge is not only transferred between firms but also from science into commercial practice, e.g., from university to an R&D department of a firm. Zucker et al. (2002) analyze star scientists at U.S. universities, characterized by “high quality intellectual human capital” (Zucker et al. 2002: 561). Quality of human capital is measured in terms of number of citations to genetic-sequence-discovery articles. In cases where the scientists exhibit knowledge which is relevant to firms commercializing biotechnology, results show that at least some of them move from university to firms early in the process (after a shorter duration in the university). The authors argue that star scientists are recruited by firms since scientific knowledge which is of tacit nature, may be best transferred to industry through the mobility of these scientists. This relationship should also apply to star inventors, moving between different employers. Gersbach and Schmutzler (2003) develop a game-theoretical model to endogenize technological spillovers. The authors assume that if engineers change firms, spillovers take place. Spillovers are therefore the result of a strategic decision of a firm: “firms have to take costly actions to acquire spillovers, and they may be able to prevent spillovers at some costs” (Gersbach/Schmutzler 2003: 180). In particular, firms can keep their employees from leaving the firm by offering sufficiently high wages. Assuming that the firms are able to observe the

41

72

Employee non-compete agreements “restrict a departing employee from accepting new employment with a competitor, for a specified time and geographical jurisdiction” (Fleming/Marx 2005: 4).

quality of an R&D employee, valuable employees will be treated favorably in order to prevent labor mobility and thus to prevent knowledge spillovers to competitors. One would thus expect that valuable employees get offers from competitors, but in equilibrium mobility does not actually occur. Kim and Marschke (2005) use firm-level panel data and show that the risk of a scientist’s move reduces the firm’s R&D expenditures but raises the firm’s propensity to patent. The authors explain their findings by the fact that firms patent research results to protect themselves from their employees. Scientists develop technical knowledge while working in the employer’s laboratory. Sometimes employees quit to join a rival or to start their own business. At their new job they can exploit the technical knowledge gathered at their former employer. Thus, firms rely on patents instead of secrecy to protect their intellectual property. To measure labor mobility the authors use the turnover experience of all scientists and engineers derived from the Current Population Survey (CPS), conducted by the U.S. Census Bureau. In particular, the shares of scientists and engineers in each industry and year who changed their employer at least once during the previous year are employed. Kim and Marschke (2005) point out that the empirical analysis faces a problem of reverse causality which means that the causality could also point in the direction from patents to mobility and not only from mobility to patents. They argue that an increasing patent rate may lead to a signaling effect and may, therefore, induce some scientists to move. To deal with this problem, the authors control for possible endogeneity of the mobility variable by using generalized method of moments (GMM) to estimate the relationship between mobility and patents. Results reveal a positive relationship between the firms’ patenting propensity and mobility rates of scientists and engineers. In particular, results suggest that a reduction of industry-specific mobility by 50% lowers the firms’ patent-R&D ratio by 2% (Kim/Marschke 2005). With the exception of Trajtenberg (2005) no other research has been carried out on the simultaneous relationship between productivity and mobility of inventors. Trajtenberg (2005) addresses the causality between mobility and productivity of 1,565,780 inventors listed on U.S. patent documents. Overall, 216,581 (33%) of the inventors are movers which means that these inventors changed their employer at least once. The data further reveal that 13% of the inventors moved across U.S. states and 1.9% moved across countries, i.e., from the U.S. to another country. Just as the labor economics literature, Trajtenberg (2005) finds evidence that mobility has a positive impact on inventive output. To analyze the effect of mobility on output value the author uses patent citation data. Results show that a recent change of the employer has a positive impact on the value of a patent applied for at the new employer. Furthermore, Trajtenberg (2005) shows that a move to a new employer has a stronger influence on the 73

number of citations received by the subsequent patent than moving geographically. Once an inventor decides to change the employer, moving to a company results in better patents compared to moving to a government agency or getting self-employed (Trajtenberg 2005). These findings suggest that an inventor who moves is able to increase his productivity. The author provides his own explanation why productivity should also increase the probability of a move. First of all, the assumption is made that inventors ex ante have a more accurate estimation of the expected value of their patents than their employer. This assumption is plausible since the inventors have more information about their own invention and know more about the state of the art in the respective technical area. Additionally, the value of a patent is hard to observe ex ante. From the literature it is known that, for instance the number of claims or the number of citations a patent receives from subsequent patents are a proxy for the value of a patent (Harhoff et al. 1999). Whereas the number of claims is observable at the time of filing, the number of citations is only observable ex post. Trajtenberg (2005) uses different proxies for patent value as explanatory variables for the probability to observe a move. He finds that inventors who hold patents which include more claims are less likely to move, whereas inventors who hold patents that received more citations are more likely to move. These results may be explained by the asymmetric distribution of information between employee and employer. Claims, as described above, are a proxy for the value of a patent which is observable ex ante. Thus, employers successfully keep their key inventors (who hold patents with many claims) from leaving the firm. Citations, on the other hand represent a proxy for patent value that is not observable ex ante. In this case the inventor has a head start of information over his employer. Consequently, Trajtenberg (2005) proposes that in case the inventor has better information on the expected value of his patents than his employer and he expects the value to be high, this inventor is more likely to move. Due to the lack of information on the part of the employer, the employer does not try to keep the inventor from leaving. Thus, the results show that the causality could also point in the reverse direction from productivity to mobility. On the one hand it is plausible that an improvement of match quality may lead to an increase in inventive productivity. On the other hand, asymmetric information makes it difficult to impede mobility of high quality inventors. Overall, a simultaneous relationship between productivity and mobility of inventor is expected. Consequently, the following two hypotheses are proposed:

74

C.1:

Inventors who moved more often between firms tend to show a higher productivity.

C.2:

Productive inventors move more often than less productive inventors.

This summary contains the key findings of theoretical and empirical literature that deals with labor mobility and productivity. It becomes clear that existing research with the exception of Trajtenberg (2005) and Kim and Marschke (2005) implicitly assumes causality to point in one direction. The following study improves on the current literature by (1) ex-ante allowing for a simultaneous relationship of productivity and mobility and (2) by including inventor characteristics as explanatory variables.

3.3 Data Source and Sample 3.3.1 Description of the Data

Data were collected in the course of the European project (called PatVal) sponsored by the European Commission. Units of observation are inventors who lived in Germany at the time of application of the respective patents. 10,500 EP patents attributed to inventors living in Germany were chosen as a stratified random sample based on a list of all granted EP patents with priority dates between 1993 and 1997 (15,595 EP patents). A stratified random sample was used in order to oversample potentially important patents. The first inventor listed on the patent document was chosen as addressee. Each inventor was provided with a cover letter together with a questionnaire. 3,346 responses were received, resulting in a response rate of 32%. The sample contains 2,761 inventors who answered one questionnaire and 288 inventors who filled out two to five questionnaires.42 Hence, the sample used in this paper contains 3,049 different inventors (representing 3,346 EP patents). The data from the questionnaire was merged with bibliographic and procedural information on the respective patents obtained from the online EPOLINE database. The dataset is a counterpart of the EPOLINE data as of March 1st, 2003 and covers approximately 1,200,000 patent files with application dates ranging from June 1st, 1978 to July 25th, 2002. To trace the productivity and the mobility of each inventor over time, the EPOLINE database was used to search for all patents belonging to the 3,049 inventors with priority dates between 1977 and 2002. The search procedure resulted in a total of 39,417 EP patents. For the instrumental

42

Inventors who were responsible for more than one patent in the underlying time period and who were chosen more than once by stratified random sample, were provided with up to five questionnaires.

75

variables technique that is employed to analyze the causality between productivity and mobility, the data are aggregated at the inventor level creating cross-sectional data.

Nges = 3,346 patents (total responses) data source: PatVal data 2,761 inventors (1 questionnaire)

288 inventors (2 or more questionnaires)

3,049 different inventors

352 inventors with one EP patent

39,417 EP patent applications

data source: EPOLINE data

2,697 inventors with multiple EP patents

1,673 non-movers (62%)

1,024 movers (38%)

Figure 3.1: Composition of the sample

For inventors holding only one patent (352 inventors) it is not possible to observe a move. Therefore, these inventors were excluded from the sample. The final sample contains 2,697 inventors who are responsible for at least two patents during the time period under consideration. Figure 3.1 provides an overview of the composition of the sample. Prior to the description of the variables, some limitations of using patent data for tracing mobility und productivity should be mentioned. First of all, a matching problem exists due to a lack of standardization of the spelling of inventors’ names. This lack of standardization complicates the identification of inventors, especially of inventors with common last names. This may lead to an underestimation of patents per inventor and, consequently, to an underestimation of the number of moves. Second, identical names may refer to different inventors. Even if additional information, such as the name of the patent applicant, is applied, this could lead to an overestimation of the number of patents per inventor. Third, incomplete address data and female inventors who changed their name due to a marriage may also lead to wrong matches.43

43

76

Annex 1 and Annex 2 provide more details on the matching procedure.

If the matching procedure works well, it is possible to identify a move, but only if the inventor applied for another patent after he changed the employer. If an inventor moved but did not apply for any patents after this move the data will not reveal the change of the employer. This could result in an underestimation of moves. Furthermore, this may lead to a selection bias since the probability to observe a move increases with the number of patents per inventor, i.e. the probability to observe a move is higher for productive inventors. Information from the PatVal questionnaires on the mobility of less productive inventors was used to reduce this bias. Let us further assume that the patent documents of two successive patents contain different applicants. The fact that different applicants are listed does not automatically mean that the inventor changed jobs. A possible explanation for two different applicants is, for instance, a strategic alliance between two companies or a merger after which patent applications are filed under the applicant name of the new company. These effects may lead to an overestimation of mobility. The classification of “move” and “no move” will be described in more detail in the following section. The results from the PatVal questionnaires, including questions related to the mobility of the inventors, in particular to the employment before, during, and after the invention was made, were utilized to confirm the matching and mobility outcomes. However, the mentioned limitations have to be taken into account when deriving implications from the results.

3.3.2 Variables

3.3.2.1

Dependent Variables

productivity – The variable is defined as the number of applications44 per inventor, divided

by the age of the inventor in 2002 minus 25. A way of justifying this measure would be the assumption that inventors become active at the age of 25 and continue to invent with constant productivity. PRODUCTIVITY

44

number of applications age2002  25

The results of chapter 3 show that citation counts are a more appropriate measure for inventor productivity. In particular, citations which are a proxy for output quality are more dependent on the inventor himself. Patent counts in contrast are largely determined by the firm that is the R&D management decides whether to file a patent application or how much to spend on R&D. Using citation data, however, requires a five year period after publication of the search report in order to compare citation counts between patents. Whereas chapter 3 applied patent applications between 1978 and 1999, this paper employs applications up to the year 2002. The years between 1999 and 2002 contain important information on mobility, which would otherwise be disregarded due to missing citation data. In the following, the better mobility information is preferred to the improved productivity measure; therefore, the number of patent applications is used as an output measure.

77

mobility – Based on the full sample of 39,065 patents45, a variable was created indicating the

number of moves per inventor. A move is defined as a change of the employer. Since almost two-thirds of the inventors have not moved at all one is added to the number of moves to calculate the logarithm of this variable. MOBILITY

lognumber of moves  1

The classification of “move” (the inventor changed the employer) and “no move” (the inventor did not change the employer) was corrected manually on the basis of the applicants listed in the EP documents. I made the assumption that the applicant listed on the patent document is also the employer of the respective inventor. To test this assumption, the responses from the PatVal-questionnaire were employed. The questionnaire included a question which asked the inventor whether the applicant listed on the patent document is also their employer. The results revealed that 92% of the questioned inventors are employed with the applicant of the patent. Since the firm applying for the patent is almost surely the employer of the inventors, it is assumed that this assumption should not lead to large biases, assigning it to all patent applications in the sample. The following three examples of chronological applicant sequences for particular inventors give some insight into the problem of distinguishing between move and no move: The first example displayed in Table 3.1 shows a sequence of 7 patents, applied for by two different applicants. Whereas the first change of the applicant is classified as a move, the second change is interpreted as an invention that was made during the employment with SIEMENS, which applied for a subsequent patent. This case was found quite frequently in the data. 26.4% of the mobile inventors have at least one patent application that belongs to this category.

PRIOYEAR 1988 1989 2000 2001 2001 2002 2002

APPLICANT SIEMENS SIEMENS SIEMENS Phillips SIEMENS Phillips Phillips

move no move

Table 3.1: Example 1 (applicant sequence of inventor 1)

45

78

39,417 – 352 = 39,065 (the patents of 352 single inventors were excluded from the dataset, see Figure 1)

Table 3.2 shows a second example: in this case, the inventor is the applicant of one of the patents, and additionally, the applicants before and after this patent match completely (here: SIEMENS), it is assumed that this invention is a free invention which means that the applicant did not claim the right to this invention according to the German Employee Invention Act.46 Therefore, it is taken for granted that the inventor has not changed his employer. The data reveal that 3.7% of the mobile inventors have applied for at least one patent in their own name during employment with another firm.

PRIOYEAR 1988 1989 2000 2001 2001 2002

APPLICANT SIEMENS SIEMENS “inventor” SIEMENS SIEMENS SIEMENS

no move

Table 3.2: Example 2 (applicant sequence of inventor 2)

The last example (Table 3.3) contains two patents from different applicants (SIEMENS and BASF) which were applied for on the same day. This case is also not treated as a move, since it is assumed that these two patents derive from research cooperation between these two firms. The data reveal that about 17.2% of the mobile inventors hold at least one pair of patent applications that belongs to the last category.

PRIO DATE

01/05/2000 01/05/2000

APPLICANT SIEMENS SIEMENS BASF SIEMENS SIEMENS SIEMENS

no move

Table 3.3: Example 3 (applicant sequence of inventor 3)

46

A more detailed description of the German Employee Invention Act is presented in Harhoff and Hoisl (2005).

79

3.3.2.2

Explanatory Variables

age - The age of the inventor was obtained from the questionnaire and represents the age at

the time of the survey. Age is included in the productivity regression to estimate a coefficient for age instead of assuming the coefficient to be 1, i.e., to take a proportional relationship between patent counts and age for granted. The findings of Trajtenberg (2005) give first evidence that younger inventors are more likely to move. Kim and Marschke (2005) also propose that inter-firm mobility is higher among young inventors since they have fewer skills. Therefore, the age of the inventor is also a control variable in the mobility model. level of education - The questionnaire contained a question asking the respondents for their

highest attained degree. In order to simplify the analysis, the education variable was aggregated into three groups: (1) secondary school, high school diploma, or vocational training (reference group), (2) vocational academy or university studies, and (3) doctoral or postdoctoral studies. external sources of knowledge - university research, scientific literature, patent literature, users, and competitors. The questionnaire included a question relating to the

importance of different sources of knowledge for the development of an invention.47 Answers were again collected on a scale from one (absolutely not important) to five (very important). A dummy variable was created for each source of knowledge, combining categories 1 (absolutely not important) to 3 (partly important) as well as categories 4 (important) and 5 (very important). The latter implies a use of the respective knowledge source. incentives - increase in salary, advancement, improvement of working conditions. The

inventors were asked on the importance of different incentives for inventive activity. Answers were collected on a scale from one (absolutely not important) to five (very important). A dummy variable was created for each incentive, combining categories 1 (absolutely not important) to 3 (partly important) as well as categories 4 (important) and 5 (very important). For the latter group the dummy becomes 1; 0 otherwise.

47

80

Although the answers to the questionnaire were related to specific patents, the answers seem to be transferable to all patents of an inventor. It is assumed that inventors basically tend to use special sources of knowledge, for example, due to positive experiences in the past. This assumption proves true, when comparing the answers of inventors who filled out more than one (five at the most) questionnaires. The different sources of knowledge are found to be equally important for all surveyed patents per inventor. Those answers that do not show a perfect match are at least highly correlated. The spearman correlation coefficients for the five different sources of knowledge range between 0.84 and 0.73.

technical area - Based on their International Patent Classification (IPC) codes, the patent

applications were classified into 30 technical areas. This classification was proposed by Schmoch (OECD 1994). technical concentration - share of patent applications in the same technical area. Using

the 30 technical areas, a Herfindahl index was calculated. For each inventor, the number of applications in the technical area i divided by the total number of applications was calculated, in the following denoted by p. The Herfindahl index (HI), consequently, corresponds to the sum of squared shares of applications: HI

¦p

2 i

i

If all applications belong to one technical area, technical concentration is at its maximum and the Herfindahl index is equal to 1. firm size - number of employees. The firm size was also obtained from the questionnaire. A

set of eight dummy variables was generated in order to account for variation across different firm sizes. The intervals range from “less than 50 employees” to “more than 50,000 employees”. Except for the group “less than 50 employees” (reference group), the dummies were included in the analysis. oppositions - The variable contains the share of granted patents per inventor that were

opposed by a third party within the opposition term of nine months after grant. status - These variables provide information on the status of the patent applications. Three

variables were included representing the shares of applications that were either granted, refused by the examiner or withdrawn by the applicant, for instance, due to the results of the search report. The status variables as well as the opposition variable are included to control for the value of the applications. claims - This variable contains the number of claims added up for the total number of patents

per inventor. The claims define the scope of an invention for which patent protection is requested. As proposed by Trajtenberg (2005), the number of claims is included as a control variable for an observable characteristic of the inventions at the time of filing. temporal concentration - This variable controls for temporal effects, i.e. this measure

reveals, whether an inventor kept on inventing constantly during his career or whether he carried out his inventions within a short period of time. The index was calculated as follows:

81

TEMPCON

number of applicationst (max) number of applications

where t(max) is the application year, in which the inventor holds the maximum number of applications. In the event the inventor’s applications are all applied for in the same priority year, the index is at its maximum, and equals 1. regional characteristics - This set of dummies indicates whether the inventions were made in

a city with more than 1 million inhabitants or in a city with between 500,000 and 1 million inhabitants. The reference group relates to inventions made in rural areas or cities with fewer than 500,000 inhabitants.48

3.4 Descriptive Statistics and Multivariate Results 3.4.1 Descriptive Results

Table 3.4 presents descriptive statistics. The final sample consists of 2,409 different inventors49, of which 37% changed their employer at least once. In the following, these inventors are classified as mobile. The data reveal a slightly higher share of mobile inventors compared to Trajtenberg (2005) who found that 33% of the inventors moved at least once. This difference is somewhat surprising since we sampled German inventors listed on EP patent applications whereas Trajtenberg used US patent data. US employees generally change their employer more often than German employees. Moreover, residential mobility in Germany is rather uncommon compared to the US (Harhoff/Kane 1997, Evers-Koelman et al.

48

Although the answers to the questionnaire were related to specific patents, the answers concerning the environment of the invention seem to be transferable to all patents of an inventor. To test this assumption, 30 mobile inventors were chosen by random to analyze whether the address of these inventors changed over time. Mobile inventors were used since these inventors are rather at risk of changing the home address than inventors who have not changed their employer. Results reveal that only three out of 30 mobile inventors changed their address. Whereas one inventor moved abroad (from a large city in Germany to a small town in Great Britain), the second one moved within Germany (both cities had a comparable size and have been sorted in the same city size group). The third one moved within the same city. The last two moves are thus of no relevance since they were sorted in the correct group. Overall, 1 out of 30 inventors are characterized by a address change relevant for the “inventive environment” variable. This share of inventors (3%) should not lead to large biases when transferring the answers related to one specific patent to all patents of the inventors.

49

The sample used within this analysis only includes inventors employed with firms. Academic inventors were excluded from the sample. Finally, 2,409 questionnaires were filled out completely with regard to the above mentioned variables. The firm size variable contains the largest number of missing values. In particular, 313 inventors did not indicate the number of employees of their firm.

82

1988, Schneider et al. 1985). Secondly, a lower share of mobile inventors in our data should have resulted from the different calculation of the mobility variable. Whereas this paper employs a mobility variable that was corrected manually, Trajtenberg counted each change of the employer as a move. Due to the large number of inventors (1,565,780 inventors) used in Trajtenbergs analysis, a differentiation between a change of the employer not resulting in a move (described in section 3.3.2.1) had probably not been possible. An explanation for the larger share of mobile inventors may be the oversampling of potentially important patents. Possibly, inventors responsible for these patents are more productive and move more often. To avoid biases caused by the sampling method the following multivariate analysis will control for the value of the patent applications.

Variable Mean mobility (dummy variable) 0.37 number of moves 0.64 number of patents 14.69 number of claims 157.02 share of granted patents opposed 0.06 share of applications refused 0.02 share of applications withdrawn 0.12 age of the inventor in 2002 54.04 level of education (terminal degree) secondary school/vocational 0.12 training / high school diploma university studies 0.52 doctoral/post-doctoral studies 0.36 external sources of knowledge universities 0.22 literature 0.63 other patents 0.66 users 0.73 competitors 0.57 incentives increase in salary 0.67 advancement 0.59 improvement of working conditions 0.64 technical concentration 0.68 temporal concentration 0.36 firm size (no. of employees) 48,880 regional characteristics more than 1 million inhabitants 0.10 500,000 to 1 million inhabitants 0.13 less than 500,000 inhabitants 0.77

S.D.

1.10 20.02 211.91 0.11 0.05 0.15 9.76

0.26 0.19 93,488

Min. 0 0 2 5 0 0 0 28

Max. 1 12 308 3,027 1 1 1 84

0

1

0 0

1 1

0 0 0 0 0

1 1 1 1 1

0 0 0 0.14 0.08 1

1 1 1 1 1 550,000

0 0 0

1 1 1

Table 3.4 : Descriptive statistics (N = 2,409)

83

Each inventor is on average responsible for 14.7 EP patents, the number of patents per inventor ranges between a minimum of 2 patents and a maximum of 308 patents. On average 6% of the inventors’ granted patents were opposed by a third party, on average 12% of the applications had been withdrawn by the applicant, and 2% had been refused by the patent examiner. Respondents were aged between 28 and 84 with a mean at 54 at the time they answered the questionnaire. Furthermore, the responding inventors are characterized by a high level of education. 12% have a high school diploma or went through a vocational training, 52% have a university degree, and 36% have a doctoral or post doctoral degree. Users as well as other patent documents turned out to be the most popular sources of knowledge utilized during the invention process: 73% of the inventors believe users to be an important source of knowledge and 66% make use of other patent documents, whereas only 22% of the respondents believe university research to be important for making inventions. Furthermore, the inventors were asked about the importance of different incentives for their inventive activity. An increase in salary is classified as an important incentive by 67%. Compared to the other incentives, advancement seems to be less critical, as only 59% of all inventors rank advancement to be important for inventive activity. Technical concentration has its mean at 0.68, ranging between 0.14 and 1. This means that the inventors make on average more than two-thirds of their inventions in one technical area. The temporal concentration of the inventive activity has its mean at 0.36, ranging between 0.08 and 1. A mean of 0.36 implies that inventors on average applied for 36% of their patents in one year which means that inventive activity is not too concentrated within a short time. On average, the patent assignees’ firms have 48,880 employees. The number of employees ranges between 1 and 550,000. In the multivariate analysis firm size groups are used. Finally, the inventors were asked about the environment of the invention that is whether the inventions were made in large cities or in rural areas. 10% of the respondents stated that the inventions were made in a city with more than 1 million inhabitants, while 13% reported that the invention was made in a city with 500,000 to 1 million inhabitants. Finally, 77% of the inventors made their inventions in rural areas or cities with fewer than 500,000 inhabitants. Table 3.5 lists the Pearson correlation coefficients for interval scaled variables with respect to the number of applications and the number of moves per inventor. The variable “number of applications” is positively correlated with inventor age, the variables concerning the legal status of the EP patent applications, and firm size. The share of applications opposed by third parties is negatively correlated with the number of applications per inventor. The variable

84

“number of moves” is positively correlated with the number of applications for which an inventor is responsible and with age. Moreover the number of moves is negatively correlated with temporal and technical concentration as well as with firm size. Overall, the correlation coefficients of the independent variables are quite small.

number number number of of appof moves lications claims

age

share_ opposition

share_ refused

share_ withdrawn

temporal concentration

techn. firm concensize tration

1.00 0.27* -0.02

1.00 -0.04*

number of moves

1.00 0.14* 1.00 number of claims 0.18* 0.92* 1.00 age 0.05* 0.09* 0.05* 1.00 share_opposition -0.01 -0.06* -0.06* 0.11* 1.00 share_refused 0.02 0.04 0.03 0.06* -0.003 1.00 share_withdrawn 0.12* 0.12* 0.09* 0.09* -0.07* 0.05* temporal concentration -0.20* -0.42* -0.40* -0.16* -0.001 -0.03 technical concentration -0.19* -0.14* -0.15* -0.04* 0.05* -0.04 firm size -0.06* 0.06* 0.03 -0.06* -0.06* 0.07* * significant at 5% or lower number of applications

1.00 -0.08* -0.11* -0.02

1.00

Table 3.5: Pearson correlation coefficients (N = 2,409)

Variable

incentives increase in salary advancement * improvement of working conditions * i

(1)

(2)

All inventors (N = 2,409)

Mobile inventors (N = 892)

Meani

Meani

(3) Non-mobile inventors (N = 1,517) Meani

0.68 0.59

0.69 0.62

0.66 0.56

chi2 = 2.67, p = 0.103 chi2 = 7.08, p = 0.008

0.64

0.65

0.63

chi2 = 0.62, p = 0.430

In a Chi2-Test, the difference between movers and non-movers with respect to the different variables listed in column 1 turned out to be significant at the 5%-level or lower. Since dummy variables are nominally scaled, the values within the column are rather shares than means. In particular, the results present the share of inventors who attach importance to different incentives.

Table 3.6: Differences between mobile and non-mobile inventors according to different incentives (N = 2,409)

Table 3.6 describes the relationship between a move and different incentives for inventive activity. In particular, the table reports means for all inventors in the sample and also separately for movers and non-movers. Results show that movers turn out to be different from non-movers especially with regard the importance of advancement. Whereas 62% of the movers attach importance to advancement as an incentive for inventive activity, only 56% of 85

the non-movers regard advancement as important. The difference is significant at the 1% level. Table 3.7 and Table 3.8 display the relationship between a move and the level of education as well as regional characteristics.

Level of education (terminal degree in groups)

Inventors Number of observations

secondary school/vocational training/high school diploma university studies doctoral/post-doctoral studies Total

Share of mobile inventors

298

0.29

1,254

0.36

857

0.41

2,409

0.37

Note: In an ANOVA, the effect of the level of education turned out to be highly significant (F = 17.41, p = 0.000).

Table 3.7: Share of mobile inventors by level of education (N = 2,409)

Table 3.7 shows that inventors who have a higher level of education are more likely to move. The share of mobile inventors increases monotonically with an increasing level of education.

Regional characteristics (groups)

Inventors Number of observations

Share of mobile inventors

more than 1 million inhabitants

236

0.42

500,000 to 1 million inhabitants

303

0.51

less than 500,000 inhabitants

1,870

0.34

Total

2,409

0.37

Note: In an ANOVA, the effect of the level of education turned out to be highly significant (F = 20.86, p = 0.000).

Table 3.8: Share of mobile inventors by regional characteristics (N = 2,409)

Table 3.8 comprises regional characteristics, i.e., whether the inventions were made in large cities, small cities, or in rather rural areas. Results show that the share of movers is highest in cities with 500,000 to one million inhabitants. The share of movers is lowest in rural areas with less than 500,000 inhabitants. An ANOVA reveals that the differences are highly significant. 86

The following four figures aim at a more detailed description of the dependent variables.

700

563

600

574 496

frequency .

500

392 400 300

276

200

89

100

11

4

4

101 to 150

151 to 200

> 200

0 2

3 to 5

6 to 10

11 to 20

21 to 50 51 to 100

cumulative number of applications per inventor

Figure 3.2: Distribution of the number of applications per inventor (N = 2,409)

Figure 3.2 presents the distribution of the cumulative number of applications per inventor. Due to the requirements of the stratified random sample, we used for the PatVal project, each inventor is (jointly) responsible for at least one granted patent with priority date between 1993 and 1997. Furthermore, to distinguish between movers and non-movers, inventors listed on only one patent application were excluded from the sample. Therefore, the inventors under consideration are at least responsible for two EP patent applications. Overall, 275 (11%) of the inventors are responsible for 2 applications. The bigger part (23% or 24%) is listed on 3 to 10 or 11 to 20 applications. Only 4% are responsible for more than 50 patent applications. Figure 3.3 provides the distribution of inventor productivity. Productivity was calculated as the cumulative number of applications per inventor divided by the age of the inventor in 2002 minus 25. The histogram displayed in Figure 3.3 supports the findings of Lotka (1926) that the distribution of productivity among researchers is highly skew. Due to the skewness of the productivity distribution a logarithmic transformation of the productivity variable is used in the following multivariate analysis.

87

800 700

frequency

600 500 400 300 200 100 0 < = 0.1

> 0.1 to 0.25

> 0.25 to > 0.5 to 1 0.5

> 1 to 2

> 2 to 5

>5

productivity (number of applications/age-25)

Figure 3.3: Distribution of inventor productivity (N = 2,409)

Figure 3.4 reports the distribution of the number of moves per inventor. 1,526 (63%) inventors have not moved at all. 516 (21%) changed their employer once, 217 (9%) inventors changed their employer twice and only 27 (1%) inventors moved more than five times.

1,800 1,517

frequency

1,500 1,200 900 525

600

217

300

86

37

27

4

5 or more

0 0

1

2

3

number of moves

Figure 3.4: Distribution of the number of moves per inventor (N = 2,409)

Finally, Figure 3.5 displays the average number of moves per 10 years of inventive activity. The move – age relationship is calculated as follows:

88

number of moves . (age2002  25) / 10

1,800 1,600

1,517

frequency

1,400 1,200 1,000 800 600

466

400

190 88

200

148

0 0

> 0 to 0.5

> 0.5 to 0.75

> 0.75 - 1

>1

mobility per ten years inventive activity

Figure 3.5: Distribution of inventor mobility per 10 years inventive activity (N = 2,409)

Results reveal that 466 (19%) inventors changed their employer on average between > 0 and 0.5 times per ten years inventive activity. Only 148 (6%) inventors moved on average more than once per ten years. Since Figure 3.3 and Figure 3.4 also point at a distribution of mobility that is highly skewed, the logarithm of the mobility variable is used in the following regression estimation.

3.4.2 Multivariate Specification

In this paper, an endogenous relationship between productivity and mobility of inventors is expected. Assume a linear regression model y

EX  u relating y to X with an error term u,

representing factors other than X that affect y. Ignoring the endogeneity problem would lead to a violation of the zero conditional mean assumption E(u|X) = E(u) = 0 which means that the error term u must not depend on the value of the independent variables. A violation of this assumption would lead to biased estimators of ȕ. To avoid this problem, a simultaneous relationship between productivity and mobility is estimated using Two-Stage Least Squares (2SLS) estimators (Wooldridge 1999). The method of instrumental variables (IV) was first proposed by Reiersol (1945). Extensions of applicability of the IV estimation were suggested by Sargan (1958) who defined the theoretical treatment of the IV method as it is employed today.

89

IV estimation is applicable for simultaneous or causal relationships if it is reasonable to maintain that some regressors are determinants of one dependent variable (e.g., PRODUCTIVITY) but not of the other variable (e.g., MOBILITY). These variables constitute instruments for PRODUCTIVITY in the MOBILITY regression estimation. This strategy permits a consistent estimation of the mobility equation. The productivity equation can be estimated in the same way, using a second IV regression estimation (Mullahy/Sindelar 1996). Considering inventor productivity (PRODUCTIVITY) and inventor mobility (MOBILITY), the following equation systems are presumed:

MOBILITY

f ( PRODUCTIVITY , X 1 ,..., X n , incentives, tech _ con, temp _ con, reg _ char , N )

PRODUCTIVITY

g ( MOBILITY , X 1 ,..., X n , source _ know, H )

MOBILITY is a function of: PRODUCTIVITY X1 – Xn

incentives, technical concentration, temporal concentration, and regional characteristics,

the endogenous variable, a number of exogenous variables, which are also assumed to determine PRODUCTIVITY, and which represent additional exogenous variables that only affect MOBILITY. These additional exogenous variables will instrument for MOBILITY in the PRODUCTIVITY regression.

The regional characteristics of the invention (whether the invention was made in a large city or rather in a rural area), for instance, are assumed to serve as instrumental variables. Inventions made in larger cities should have a larger signaling effect leading to a higher probability of getting a job offer by a competitor. The productivity of an inventor, on the contrary, should remain unaffected by environmental differences.

90

PRODUCTIVITY is a function of: MOBILITY X1 – Xn external sources of knowledge

the endogenous variable, a number of exogenous variables, which are also assumed to determine MOBILITY, and which represent additional exogenous variables that only affect PRODUCTIVITY. These additional exogenous variables will instrument for PRODUCTIVITY in the MOBILITY regression.

External sources of knowledge can positively influence inventor productivity. Patent documents, for instance, allow inventors to collect relevant research information about the state of the art or about inventions made by competitors. Additionally, scientific literature is assumed to have a positive impact on inventor productivity. Inventors can use this source of knowledge to catch up on the actual state of basic research. Furthermore, basic research could form a source of idea creating for applied research. The use of patent and scientific literature should not have a significant influence on the mobility of inventors, since reading patents or scientific articles does not lead to a personal contact between the inventor and the applicant or the author of the article. Thus, there is no reason to believe that the inventors would receive information from job vacancies in a company. Granovetter’s theory of “the strength of weak ties” also confirms that personal contact is needed to establish weak ties (Granovetter 1974, 1983). Montgomery (1991) confirms the applicability of Granovetter’s results to the labor market. In particular, the author describes the importance of personal contacts as a source of employment information.

3.4.3 Discussion of the Results

Table 3.9 provides the results of the 2SLS regression estimation. Model (1) contains control variables and explanatory variables required to test the hypotheses. Model (2) additionally includes variables controlling for variation between technical areas. In both models 2(a) and 2(b) technical areas which control for heterogeneity that would otherwise lead to biased results with respect to the estimated coefficients, have additional explanatory power. Therefore, in the following the results of Model (2) are described in more detail. I first discuss the results with respect to the productivity equation (Table 3.9, column 4).

log(age-25) was included as an independent variable to account for a relationship between age and productivity which may be not proportional. A coefficient of -0.74 implies that the number of applications rises less than proportional with age (slope: 1-0.74 = 0.26). Thus, 91

when age increases by 1%, productivity rises by 0.26%. The effect is significant at the 1% level. It is reasonably to believe that this increase of patent applications over time is an effect of experience. Inventors growing older gain experience which may lead to a higher productivity. An increasing number of patent applications with age may also arise due to the hierarchical position of the inventor. R&D managers increasingly act as advisors in different R&D projects. Consequently, an R&D manager is involved in more projects in parallel, resulting in more patents per inventor. Part of this increase may also occur due to a changing patenting behavior over time. In particular, inventors today tend to patent more than inventors did in the past. Inventors who were in their 40s or 50s or even older at the time of the survey profit from this increasing patent propensity.

92

93

Table 3.9: 2SLS regression with heteroskedasticity-robust standard errors

Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Model (1) (industry dummies not included) (a) (b) dependent variable log(mobility) log(productivity) log(productivity) -0.836* [0.432] log(mobility) 0.504*** [0.091] log(years of inventive activity) -0.640* -0.757*** [0.354] [0.048] log(total number of claims) 0.697** 0.763*** [0.325] [0.014] share of granted patents opposed -0.133 -0.136* [0.118] [0.071] share of applications refused 0.188 0.094 [0.249] [0.172] share of applications withdrawn 0.429*** 0.174*** [0.143] [0.060] level of education, terminal degree (reference group: high school diploma or less) university studies 0.071** -0.019 [0.033] [0.027] doctoral/postdoctoral studies 0.156*** -0.056 [0.045] [0.045] Observations 2409 2409 F-test (df) 10.99(22) 497.81(22) 0.072** [0.033] 0.117*** [0.043] 2409 5.86(51)

-0.022 [0.029] -0.076 [0.048] 2409 206.88(51)

Model (2) (industry dummies included) (a) (b) log(mobility) log(productivity) -0.864* [0.482] 0.593*** [0.101] -0.640* -0.743*** [0.385] [0.051] 0.717** 0.745*** [0.359] [0.015] -0.081 -0.154** [0.118] [0.076] 0.166 0.17 [0.257] [0.188] 0.429*** 0.123* [0.147] [0.063]

93

94

94

dependent variable

Table 3.9 continued: 2SLS regression with heteroskedasticity-robust standard errors

Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Observations F-test (df)

source of knowledge - competitors

source of knowledge - user

source of knowledge - other patents

doctoral studies * knowledge_literature

source of knowledge - literature

doctoral studies * knowledge_university

source of knowledge - universities

incentive - improvement of working conditions

incentive - advancement

incentive - increase in salary

Model (1) (industry dummies not included) (a) (b) log(mobility) log(productivity) 0.011 [0.030] 0.073** [0.030] -0.037 [0.025] 0.012 [0.028] -0.096** [0.045] -0.088*** [0.023] 0.124*** [0.044] 0.040** [0.020] -0.023 [0.020] 0.007 [0.019] 2409 2409 10.99(22) 497.81(22) Model (2) (industry dummies included) (a) (b) log(mobility) log(productivity) 0.010 [0.030] 0.070** [0.030] -0.035 [0.026] 0.006 [0.030] -0.064 [0.047] -0.091*** [0.024] 0.098** [0.046] 0.031 [0.022] -0.007 [0.021] 0.007 [0.020] 2409 2409 5.86(51) 206.88(51)

95

Table 3.9 continued: 2SLS regression with heteroskedasticity-robust standard errors

Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Model (1) (industry dummies not included) (a) (b) dependent variable log(mobility) log(productivity) firm size in number of employees (reference group: less than 51 employees) 51 - 250 employees -0.080 0.054 [0.071] [0.050] 251 - 500 employees -0.034 0.218*** [0.090] [0.052] 501 - 1,500 employees -0.078 0.202*** [0.083] [0.049] 1,501 - 5,000 employees -0.127 0.284*** [0.091] [0.051] 5,001 - 10,000 employees -0.031 0.309*** [0.113] [0.052] 10,001 - 50,000 employees -0.083 0.423*** [0.129] [0.052] more than 50,000 employees -0.094 0.463*** [0.140] [0.053] technical concentration -0.196*** [0.047] temporal concentration -0.791*** [0.301] Observations 2409 2409 F-test (df) 10.99(22) 497.81(22) -0.074 [0.070] -0.061 [0.090] -0.087 [0.085] -0.164* [0.090] -0.076 [0.107] -0.135 [0.118] -0.155 [0.131] -0.210*** [0.048] -0.823** [0.326] 2409 5.86(51)

2409 206.88(51)

0.057 [0.054] 0.230*** [0.057] 0.212*** [0.053] 0.298*** [0.056] 0.299*** [0.057] 0.401*** [0.058] 0.451*** [0.060]

Model (2) (industry dummies included) (a) (b) log(mobility) log(productivity)

95

96

96

-2.661*** [0.201] 2409 497.81(22)

Table 3.9 continued: 2SLS regression with heteroskedasticity-robust standard errors

2409 10.99(22)

-1.442 [0.900]

Robust standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Observations R-squared1 F-test (df)

Constant

Wald test

Model (1) (industry dummies not included) (a) (b) dependent variable log(mobility) log(productivity) regional characteristics (reference group: less than 500,000 inhabitants) city with more than 1 mio. inhabitants 0.088** [0.036] city with 500.000 to 1 mio. inhabitants 0.222*** [0.039] distribution of patents across technical areas not included not included 0.072* [0.039] 0.211*** [0.038] included Chi2(29)=1.91 p=0.002 -1.615 [0.999] 2409 5.86(51)

included Chi2(29)=3.86 p=0.000 -2.637*** [0.224] 2409 206.88(51)

Model (2) (industry dummies included) (a) (b) log(mobility) log(productivity)

Table 3.9 (columns 2 and 4) shows that the level of education is not associated with inventive output. Inventors who have a university or a doctoral degree do not show a higher productivity compared to the reference group (inventors who earned a high school diploma or less). This finding is surprising since many studies have pointed to a positive relationship between the educational degree and inventive output (e.g., Shockley 1957). In case a positive relationship between education and productivity does actually exist, the question poses why the data are not revealing this relationship. Three possible explanations will be given in the following. First, inventive output of productive inventors may be subject to a more severe selection process. Assume that patents of key inventors are often applied for in many different countries leading to higher application costs, this should lead to a more accurate estimation of the probability of a grant or of the market potential of an invention. This could result in a lower number of patent applications of key (highly educated) inventors. Secondly, the most productive inventors are probably involved in larger and more capital-intensive projects. Generally, these projects are long-term projects leading to a smaller number of patents within a certain period of time. Finally, it is possible that the importance of personal attributes or abilities of one single inventor is overstated considerably in the literature since more than 90% of the inventions are made by inventor teams. Possibly inventors, who themselves have no university or doctoral degree but work in an inventor team with university graduates, profit from their colleagues. Therefore, the composition of the inventor team rather than the characteristics of an individual inventor is of importance for inventive output. This assumption has to be analyzed in more detail in future research focusing on the inventor team as unit of observation. Nevertheless, hypothesis P1 is not supported by the data. Model 1(b) reveals that exploiting the knowledge from other patent documents increases productivity. Inventors who make use of patent documents are 4% (Model 1(b), column 2) more productive than inventors who do not use this source of knowledge (significant at the 5% level). The effect of using patent literature is no longer significant after including the control variables for industry effects (Model 2(b), column 4). Industry effects seem to explain the use of patent documents as external source of knowledge, leaving no explanatory power for this variable. Therefore, hypothesis P2 is also neglected by the data. Making use of scientific literature hence reduces productivity by 9% (Model 2(b), column 3). The coefficient of the variable “use of scientific literature” is significant at the 1% level. A reason for this negative effect may be that inventors who attach importance to scientific literature conduct basic research rather than applied research. Since basic research compared to applied research results in longer and more extensive R&D processes, basic research should result in a lower application rate per years of inventive activity. Furtheron, Model 2(b) supports the proposition that applying scientific knowledge requires absorptive capacity. Doctoral or post-doctoral studies increase the influence of scientific literature on productivity 97

by 1%50 (Model 2(b), column 4). The interaction between doctoral studies and spillovers from university research is not significant. These results, at least in part, support hypothesis P3. Firm size is positively associated with productivity. The coefficients (except for 51 – 250 employees) are significant at the 1% level. Productivity rises almost monotonically with firm size. A productivity increase with firm size can arise due to large firms adopting new technologies earlier. Additionally, they have more resources to hire and retain high quality researchers and to provide incentives for inventive activity. A second reason for this relationship may be that R&D is organized differently in large firms. Possibly, scientists in large R&D departments are highly specialized and play a smaller role in a particular R&D project but are involved in different projects at the same time (Kim et al. 2004). Overall, hypothesis P4 is supported by the data. The control variables: the share of patents opposed and the share of applications withdrawn, contribute to the explanation of inventive productivity. The cumulative number of claims also explains inventor productivity. The share of patents opposed is negatively associated with inventor productivity, the number of claims positively. The share of patents withdrawn by the applicant is also positively associated with the cumulative inventive output. This positive relationship may arise due to the high patenting activity of several firms. A withdrawal by the applicant takes place in case the firm is either no longer interested in obtaining a patent for the underlying invention or forestalls a rejection by the EPO. In the following, the results of the mobility equation will be discussed. Model 2(a) (Table 3.9) relates mobility to a number of explanatory variables, characterizing the inventor as well as the work environment. The set of dummies controlling for the level of education of the inventor shows that an increasing level of education raises the number of moves. A university degree raises the number of moves per inventor by about 7%, a doctoral or post-doctoral degree by about 8% (compared to the reference group: high school or less). These findings support hypothesis M1 that inventors with a higher level of education move more often. This finding complies with the existing literature; in particular, the level of education which is observable is a factor in reducing uncertainty in job negotiations (Spence 1973). Furthermore, a number of dummy variables were included in the regression estimation to control for the effect of different incentives. An improvement of working conditions does not significantly influence inventor mobility. Advancement, as expected, has a significant effect

50

98

The overall effect is calculated by adding up the effect of source of knowledge_literature (-0.091) and the effect of the interaction term doctoral studies * knowledge_litarature (0.098).

on mobility. Classification of advancement as important for inventive activity increases the number of moves by 7% (Model 2(a), column 3). Possibly, inventors who regard advancement as an important incentive for inventive activity are more receptive to job offers from competitors. This finding also supports the proposition of Allen and Katz (1985) that career opportunities for technical professionals are often unsatisfactory, resulting in a quit. Whereas hypothesis M3 is not supported, hypothesis M2 is supported by the data. As expected, an increase in firm size has a negative effect on mobility (Model 2(a), column 3), although the effect of firm size is not significant at the 10% level. Firms with 1,501 to 10,000 employees form an exception: being employed with these firms decreases the number of moves by 16% (Model 2(a), column 3) compared to the reference group (less than 51 employees). These findings in part support hypothesis M4 that inventors employed with large firms are less likely to move. First of all, jobs with large firms are more stable. Secondly, R&D departments of large firms dispose of more resources which are of great interest to the inventors. It is somewhat surprising that firm size does not significantly affect mobility. Possibly, the productivity variable absorbs part of the firm size effect. Due to strategic reasons, in particular large firms tend to patent more per unit R&D (Hall 2004). Using citation counts rather than patent counts to measure productivity will also enable to analyze the firm size-mobility relationship more precisely. Since the number of citations a patent received is a measure for its quality (Harhoff et al. 1999), citation counts are assumed to be less affected by an increasing patenting activity of firms. Finally, hypothesis M5 is also supported by the data. Inventors whose inventions are concentrated in a smaller number of technical areas are less likely to move. In particular, an increase in technical concentration by 1% decreases mobility by 0.21 (Model 2(a), column 3). This result is in accordance with the findings in the literature. Technical specialization leads to an increase in firm-specific human capital, resulting in a lower value of the inventor to the job market. A set of control variables was further factored into the regression. First of all, the age of the inventor was included. Results show that mobility decreases with age. This finding complies with the results of Trajtenberg (2005) who shows that inventors tend to move earlier in their patenting career. Kim and Marschke (2005) suggest that inter-firm mobility is observed more often among younger inventors since they have fewer skills and are less productive. Another explanation for the negative relationship may be that at the beginning of their career inventors undergo a search and sorting process to find the right employee-employer match, resulting in more moves. In addition, temporal concentration of inventive activity is used to show whether an inventor kept on inventing constantly during his inventive life or whether he developed his inventions 99

within a short period of time. Results reveal that a higher temporal concentration decreases the number of moves. An explanation for this finding could be that inventors, who keep on inventing continuously, are more visible and are of more interest to other firms. A set of dummy variables was included to control for the environment of the invention. The dummies indicate whether the invention was made in a city with more than 1 million inhabitants or in a city with 500,000 to 1 million inhabitants. The reference group relates to inventions made in rural areas or cities with fewer than 500,000 inhabitants. Both coefficients are highly significant and possess a positive sign which means that inventors who are active in larger cities move more often. Again, this is not surprising, since large cities provide more job opportunities. In rural areas, inter-firm mobility often forces employees to an interregional move leading to an increase in mobility costs for the inventor.

Finally, the findings concerning the causality between inventor productivity and inventor mobility will be provided. Results confirm that there is a simultaneous relationship between inventor productivity and inventor mobility. Model 2(b) (Table 3.9) shows that in case the number of times an inventor changes his job doubles his productivity will increase by 59.3%. Assuming an inventor who has changed his job once, he will increase his productivity by about 60% if he changes once again. One additional move of an inventor, who has moved twice before, will increase his productivity by 30%. The coefficient is significant at the 1% level. A doubleling of productivity decreases the number of inter-firm moves of an inventor by 86%. Assume, e.g., an inventor who holds 5 patents. An increase by one patent decreases the number moves by about 17%. The effect is significant at the 10% level. Consequently, the data support hypothesis C1 that inventors who move increase their productivity. Hypothesis C2 that inventor productivity affects mobility is also supported. But, the latter relationship shows a different sign as proposed in hypothesis C2. This result may be explained by the fact that productive inventors have found good matches and may not want to move. It is also possible that productive inventors receive job offers from competitors but they do not change because incentive systems within their firm encourage them to stay.

100

3.5 Difference-in-Differences Estimation Results of the 2SLS regression analysis showed that there is a bi-directional causal relationship between inventor mobility and inventor productivity. One of the major drawbacks of the regression analysis described before is that it disregards the time structure of the data. Time series data can help provide a better understanding of the impact of a certain move on inventive performance. The 2SLS results are based on the whole professional life of an inventor and do not reflect the impact of one particular move on the quantity or quality of inventive output in the aftermath of this move. To compare the performance of an inventor before and after a selected move an additional quasi-experimental design will be employed. To distinguish between differences attributable to the move and differences caused by other variables, a comparison group will be constructed containing inventors who have not moved during the time period under consideration but who otherwise are similar to the mobile inventors, for instance, with respect to the age or the educational background. The use of quasi-experiments to analyze treatment effects in the absence of truly experimental data has gained wide acceptance in empirical research (see e.g., Solon 1985, Krueger 1990, Card/Krueger 1994). Quasi-experiments are characterized by the lack of one of the decisive particularities of a (randomized) experiment: a randomized assignment of the units of observation to the treatment and to the comparison group. The units are instead sorted into the two groups by self-selection (Meyer 1995). One of the most often used quasi-experimental designs is the difference-in-differences estimation approach51 which aims at analyzing the impact of some treatment on a certain group of subjects under consideration. To do so, the performance of the treatment group is compared relative to the performance of a control group52 for the periods before and after the treatment. The difference-in-differences estimator is based on the strong assumption that in absence of the treatment, the average outcomes for the treatment and the control group would have followed parallel paths over time (Abadie 2003). Therefore, the control group shows what would have happened to the treatment group in the absence of any treatment. Difference-in-differences estimation requires knowledge of the specific point in time when the treatment occurred. Moreover, a control group is needed which contains different, but similar units of observations who have not received any treatment but which experienced as many of the other influences as possible that affected the treatment group (Meyer 1995,

51

In psychology this approach is also called the “Before and After Design with an Untreated Comparison Group” (Meyer 1995: 154).

52

In the literature the terms control group and comparison group are used synonymously. Hereinafter the term control group is used.

101

Cook/Campbell 1979: 207f). The described design is diagrammed in Figure 3.6, where X stands for the treatment, O for an observation, and subscripts 1 through n refer to a sequential order of observations (O1 … On). The subscript A is used for observations of the treatment group, the subscript B is used for the control group. A dashed line between the treatment and the control group means that the groups were not randomly formed.

OA1

OA2

OA3

OB1

OB2

OB3

X

OA4

OA5

OA6

OB4

OB5

OB6

Figure 3.6: Untreated control group design with pre- and posttest (Cook/Campbell 1979)

The following analysis makes use of this quasi-experimental design to gain a better understanding of the impact of a particular move on the quality and on the quantity of the patent applications. In the following, the data and the matching of the treatment group and the control group are described. Subsequently, the results of the difference-in-difference estimation are presented. In addition to the results of a t-test, OLS regression estimations with a further control variable for inventors who moved repeatedly during their inventive life are presented.

3.5.1 Data Source, Matching, and Descriptive Results

The data used in this analysis include patent applications with priority dates between 1985 and 1999. The lower limit was chosen since the years between 1977 and 1984 are characterized by a strong increase in the number of European patent applications, which was caused by the diffusion of the European patent after the founding of the European Patent Office in 1978. Hence, I assume that as of mid 1980, European patent data are a sufficiently reliable source of data to use for a quasi-experimental design. The upper limit was chosen due to the limitations in the availability of citation data. To count the number of citations a patent received from subsequent patents and to compare citations between patents applied for in different years, the number of citations received within a five year time lag from the publication of the search report is employed.53

53

Since the search report is published about one year after the applicant applied for the EP patent, patent data as at 2005 are needed to calculate five years citation lags for patents with priority year 1999.

102

Inventors assigned to the treatment group had to be at least 25 years old in 1986, since it is assumed that inventors do not become active before the age of 25. Additionally, treatment inventors had to have changed their employer at least once between 1990 and 1995. The move-period was chosen in order to analyze three- and four-year windows before and after the move. Since difference-in-differences estimation requires knowledge of the specific point in time when the move took place, information on the applicants of the patents and the priority dates were used to calculate a proxy for the exact date of the move. To identify whether a move actually occurred, applicant names listed on the EP patent documents were employed. In the event two successive patent documents belonging to the same inventor contained two different applicants, it was assumed that the inventor changed his employer in the time period between these two patent applications. Since the exact time of move was not available, the move date was estimated by taking the midpoint between the two application dates (the last patent before and the first patent after the move). In case inventors changed their employer more than once between 1990 and 1995 one of these moves was selected at random. Finally, a sample of 553 mobile inventors was chosen. To construct the control group for these 553 mobile inventors, a matching approach was used to match each treated unit with a non-treated control unit. In particular, matched treatmentcontrol pairs are necessary to avoid spurious effects of inventor characteristics or industry effects between the treatment and the control group that are not attributable to the treatment itself. In case it is not possible to identify eligible characteristics for the control group, one could pick inventors randomly from the group of non-treated units. The characteristics within this analysis were chosen based on the findings of the 2SLS regression estimation. According to these results relevant matching characteristics that have an effect on inventive output quantity and quality are the age of the inventor, his educational background as well as the main technical area in which the inventor is active. Consequently, the control group was constructed with the help of these three criteria. Additionally, control inventors were chosen who were also responsible for at least one patent before and after the move of the “twin” inventor. This assumption resulted in congruent three and four year windows of the mobile and the matched control inventor. Different time periods could have resulted in biases due to a different patent behavior at different points in time (Hall 2004). In case two or more inventors were potential candidates for matched pairs, one of these inventors was again selected at random. Overall, matched pairs could be found for 352 mobile inventors, resulting in a dataset of 704 inventors who have been responsible for a total of 11,273 patent applications between 1985 and 1999. Table 3.10, provides some descriptive statistics of the variables that will be tested in the following difference-in-differences estimation. The variables were selected to compare the inventive performance of the underlying inventors in the pre and post period of the move. 103

Whereas columns 1 to 4 present descriptive statistics for the full sample, columns 5 to 12 provide summary statistics for the treatment and the control group separately. In the following, the results based on the full sample will be provided. The two subsamples will only be addressed in case the means of the two subsamples differ considerably. Each of the 704 inventors is on average responsible for about 16 EP patent applications. The total number of applications per inventor ranges between 2 and 308. The mean number of patents of the mobile inventors amounts to 14.5 whereas the mean number of patents of the control inventors amounts to 17.6 patents. The difference arises due to the fact that one of the control inventors holds 315 patents whereas the mobile inventor with the maximum number of patents is responsible for only 119 patents.

104

105

21.10 0.10 0.22 0.05 0.17 81.02 1.09 50.30 0.98 218.54 3.92 1.37

0.05

0.69

0.02

0.16

65.04

4.23

28.34

1.51

167.10

10.64

3.16

1

2.33

5

0

0

1

2

0

0

0.11

0

2

10.08

32.50

2,986

7.25

819

9.60

1,188

0.72

0.25

1

0.67

308

Full sample (N = 704) S.D. Min. Max.

16.01

Mean

3.11

11.22

165.23

1.52

26.60

4.29

59.91

0.17

0.02

0.68

0.05

14.47

1.34

4.05

190.80

1.05

40.48

1.17

61.94

0.17

0.05

0.22

0.09

15.35

1

2.5

5

0

0

1.11

3

0

0

0.11

0

2

10.08

26.67

1,310

7.25

373

8.50

378

0.70

0.25

1

0.50

119

Mobile inventors (N = 352) Mean S.D. Min. Max.

3.21

10.05

168.96

1.49

30.08

4.18

70.17

0.16

0.02

0.69

0.05

17.55

1.40

3.71

243.40

0.91

58.51

1.01

96.22

0.17

0.05

0.21

0.10

25.53

1

2.33

7

0

0

1

2

0

0

0.17

0

2

8

105

32.50

2,986

5.25

819

9.60

1,188

0.72

0.25

1

0.67

315

Control inventors (N = 352) Mean S.D. Min. Max.

Table 3.10: Descriptive statistics of the full sample (N = 704) as well as of the treatment group (N = 352) and the control group (N = 352)

Variable number of patent applications share of granted patents opposed share of applications granted share of applications refused share of applications withdrawn cumulative number of references number of references per patent application cumulative number of citations number of citations per patent application cumulative number of claims number of claims per patent application size of the inventor team (per patent)

On average, 5% of the inventors’ granted patents were opposed by a third party, on average 16% of the applications had been withdrawn by the applicant, and 2% had been refused by the patent examiner. 70% of the applications were finally granted. The cumulative number of applications per inventor received an average of 65.04 references made by the patent examiners at the EPO. Each patent application received on average 4.23 references. Furthermore the cumulative number of patent applications per inventor received an average of 28.3 citations from subsequent patents (each application received on average 1.51 citations). Whereas Table 3.10 shows that the control inventors’ patents altogether received more references and also more citations, the number of references and the number of citations per patent application are almost identical for both groups. The inventors’ applications altogether contained an average of 10.64 claims. The number of claims per patent had a mean of 10.64. Finally, the average inventor team size varied between 1 and 10.08 with a mean of 3.16.

3.5.2 Description of the Results

To calculate the difference-in-differences estimator Gˆ1 one has to take the mean value of each

group’s outcome (treatment and control group) before and after the treatment and then calculate the “differences-in-differences” of the means. Therefore, the following equation is constructed:

Gˆ1

(mobile post  control post )  (mobile pre  control pre )

' post  ' pre

where the average value of the treatment group is denoted by mobile and the average value of the control group by control. Pre and post stand for the pre-treatment and the post-treatment period and a bar indicates an average over the inventors. To test whether Gˆ1 is statistically different from zero one could either conduct a t-test, testing

the Ho hypothesis that ' post

' pre . The results of the t-tests for the variables under

consideration are presented below. The same results can be achieved by using an OLS regression framework which will be described later.

106

(mobilepost – controlpost) (mobilepre – controlpre) diff.-in-diff. |t-value|

(mobilepost – controlpost) (mobilepre – controlpre) diff.-in-diff. |t-value|

(mobilepost – controlpost) (mobilepre – controlpre) diff.-in-diff. |t-value|

share of applications granted54 4 years 3 years 0.018 0.043

share of applications refused 4 years 3 years -0.005 -0.004

share of applications withdrawn 4 years 3 years -0.021 -0.026

-0.034

-0.031

-0.001

0.0002

0.038

0.034

0.052* 1.84

0.074** 2.51

-0.004 0.413

-0.005 0.43

-0.058** 2.31

-0.060** 2.24

share of applications opposed 4 years 3 years 0.007 0.014

number of references per patent application 4 years 3 years 0.275 0.320

number of citations per patent application 4 years 3 years 0.134 0.190

-0.016

-0.022

0.035

0.003

-0.066

-0.099

0.021 1.20

0.035* 1.83

0.240 1.55

0.318** 2.06

0.200 1.34

0.289* 1.85

number of claims per application 4 years 3 years 1.224 1.357

number of applications 4 years 3 years -1.293 -1.063

number of inventors per patent 4 years 3 years -0.271 -0.307

1.188

1.144

-1.011

-0.707

-0.045

-0.073

0.036 0.09

0.214 0.48

-0.281 0.57

-0.355 0.99

-0.226** 2.02

-0.233** 1.97

Table 3.11: T-Test of difference-in-differences estimations for 4 and 3 year periods before and after the event of a move (* significant at 10%; ** significant at 5%; *** significant at 1%) (N = 352)

First of all, the results concerning the status variables are provided. Figures 3.7a and b chart the calculation of the difference-in-differences estimators Gˆ1,t for the mean share of applications granted.

54

The share of patents which have not been granted include patents which were either refused by the patent examiner or withdrawn by the patent applicant as well as the number of patent applications which are still pending. Analysis of the share of patents pending revealed that in both groups (the mobile inventors and the control inventors) 13% of the patents are pending.

107

mean granted m eanshare shareof of applications applications granted

0,82

0,78

4 year period

0,803

0,8 'pre

0,768 0,76 0,74 0,720

0,72 0,7

G1, 4years = 'post - 'pre = (0.720-0.702) - (0.768 - 0.803) = 0.018 – (-0.035) = 0.053

0,702

'post

0,68 pre

post

mean granted m eanshare share of of applications applications granted

time

0,82

0,78

3 year period

0,801

0,8 'pre

0,770 0,766

0,76

'post

0,74 0,723

0,72 0,7 0,68

G1, 3 years = 'post - 'pre = (0.766 - 0.723) - (0.770-0.801) = 0.074

pre

post time mobile

control

Figures 3.7a/b: Difference-in-differences estimator calculated for the mean share of granted patents (4 year period and 3 year period)

A positive difference-in-differences estimator – as illustrated above – implies that the incident of a move has a positive impact on the mean share of applications granted. In particular, the mean share of applications granted increases by 5% when a window of 4 years before and after the move is considered and by 7% with respect to a 3 year window. Hence the effect of a move on the grant rate seems to decrease over time. Additionally, Figures 3.7a and 3.7b show that the trend lines cross over. The important point here is the pattern of switching mean differences. This means that the ex ante low scoring treatment group, including mobile inventors, has overtaken the higher scoring control group. A possible explanation for this cross over is again the matching approach. An increasing match quality between employer and employee after the move could lead to a better performance of the inventor such as, e.g., a 108

higher grant rate. The overall decrease of the mean share of applications granted (between 4% and 10%) - observable for the treatment as well as for the control group – occurs due to truncation of the data. In particular, the average time lag between the application date and the date the application is granted at the EPO is 4.2 years (Harhoff/Wagner 2005). Whereas more than 72% of the patent applications with priority year 1995 were granted and less than 10% are pending, only 41% of the patents were granted and 38% are pending if the patents were filed in 1997. Finally, when 1999 is the priority year, only 12% of the applications were granted and 77% are pending. Interestingly, a move seems to have no impact on the share of patents refused by the patent examiner. The difference-in-differences estimator is not significant at the 10% level - neither for the 4 year window nor for the 3 year window. Hence, the Ho hypothesis that ' post

' pre

cannot be rejected. In case the share of patent applications withdrawn by the applicant is considered, a move has a negative impact. Whereas the share of withdrawals prior to the move is larger for the treatment group than for the control group, it becomes smaller afterwards. Overall, the share of applications withdrawn decreases by about 6% (3 and 4 year period). Again this result suggests an increase in the value of the applications or at least an increase of importance of the applications for the new employer. Withdrawals on the part of the applicants take place in case the applicant fears that the invention does not meet the requirements for patentability (novelty and inventive step) or the applicant is no longer interested in receiving patent protection for the invention. One could, therefore, also assume that the inventions of the movers are more in line with the patent portfolio of their new employer compared to the old one. The share of oppositions received within the opposition term of nine months after the patent was granted is lower in the treatment group before the move took place and higher afterwards. The difference-in-differences estimator reveals a 4% increase of the opposition rate due to the move. According to Harhoff and Hall (2003) the number of oppositions a patent received is a proxy for the value of the patent. Opposition results hence confirm that patent applications of the mobile inventors become more important after the move. Although the signs point in the same direction, the difference-in-differences estimator is no longer significant in case the 4 year window is considered. The results suggest that this opposition effect diminishes over time.55

55

Future research will also include information on the firms that filed the oppositions. In particular, the question will be addressed whether any of the oppositions were filed by former employers of the mobile inventor.

109

In addition to the results described above, reference and citation counts underline the proposition that a move has a positive effect on the value of after-move patent applications. However references and citation counts only exhibit a significant difference in case the 3 year window is employed. Table 3.11 shows that the treatment group and the control group receive the same average number of references during the three years prior to the move. In the postmove period, in contrast, there is a difference between the two groups. Overall, a move leads to an increase in the number of references by 0.32 references per application. As to citation counts Table 3.11 gives evidence that whereas the control group’s patents received slightly more citations per patent in the pre-move period, the mobile inventors are ahead in the postmove period. The difference-in-differences estimator indicates that the number of citations increases by 0.29 citations per patent as a result of the move. A possible explanation for this outcome could first of all be an increase in the value of the patent applications in the postmove period. Secondly the outcome may be explained by the fact that the inventor who moved works within the same technical area at his new job. Two R&D teams working in the same area produce patent applications which form potential state of the art to be referenced by the patent examiner of the EPO during the search process. The references from the search reports are used to calculate the number of citations from subsequent patents, therefore, more references also lead to more citations. To confirm this assumption, the citing patents have to be analyzed more closely. In particular one would have to find out whether part of the citations received by the after-move patents derive from the former employer of the mobile inventor. The former employer could be interpreted as kind of a second source of selfcitations.56 The number of claims per patent and the number of applications themselves do not change due to the move - neither in the 3 nor 4 year window before and after the move. Finally, the inventor team size is affected by a move. In general inventors who move work in smaller inventor teams afterwards. This result is surprising since former research confirmed that inventor teams are larger in big firms (especially in large pharmaceutical firms) and inventors working in large firms are less likely to move. Additionally inventors who move typically change from small to larger firms (Topel/Ward 1992, Kim et al. 2004). The difference-indifferences estimator reveals that team size decreases by 0.23 inventors (3 and 4 years) as a result of the move. Although the difference-in-differences estimator is significant at the 5% level, the effect seems to be rather irrelevant since a coefficient of 0.23 corresponds to less than a fourth person. The regression results provided in the next section, however, reveal that the team size effect is no longer significant.

56

Self-citation corrected citation counts will be employed in future research.

110

Overall the results of Table 3.11 show that a move seems to have a larger impact during the 3 year period as compared to the 4 year period. A possible explanation is that inventors who changed their employer do a better job during the first years after the move. Possibly, the inventors make special efforts during their first years to impress their new employer or to gain respect from their colleagues. Over time the differences caused by the move decrease and finally disappear. It is also possible that inventors are able to profit from knowledge they have “transferred” from their former job. Over time this advantage disappears since the inventor faces new tasks and his present knowledge does not suffice to solve these problems. A third explanation may be that inventors who move to a new company bring new know-how, ideas and skills to the company. A combination with existing know-how and skills with new ideas could lead to an increase in innovative activity. Over time, know-how and skills become more and more similar leading to a reduction of inventive activity. Another explanation may be that the units of observation are still inventors during the first years after the move but leave R&D for a job in sales or marketing after about three years. Intra-firm mobility (e.g., from R&D to sales) leads to invisibility of the inventor in terms of patents and consequently, the mean inventive performance seems to decrease over time. The results provided here hence confirm the causal relationship between inventor mobility and patent quality obtained from the 2SLS regression.57 One of the advantages of the difference-in-differences estimation, which is at the same time a drawback, is that it considers the impact of one specific move on patent quality. As described before, the selected moves took place between 1990 and 1995. In case the inventors moved repeatedly between 1990 and 1995, one of these moves was selected at random. But there are also moves that have taken place before 1990 and/or after 1995. These moves have not been considered at all. To overcome this drawback, an additional dummy regression analysis will be employed. Although an OLS regression framework would lead to the same results as that of the t-tests, it has the advantage that additional control regressors such as multiple moves can be included in the analysis. The following equation including a control dummy for inventors who moved more than once during their inventive life will be estimated: y

E 0  G 0 * mobile  G 1 * post  G 2 * (mobile * post )  G 3 * mult _ mobil  u

where mobile is a dummy variable, taking the value one in case the inventor moved and zero otherwise. post is a time dummy variable taking the value one in case the time period after the

57

When only the four year period before and after the move is used to analyze differences in inventive performance, the sample contains 401 matched pairs. To find out whether an increase in the sample by 15% does change the results, the t-tests and the dummy regressions were also conducted using the larger sample. Results show that the absolute values of the coefficients changed only slightly (third position after the decimal point), the signs and the significance remained completely stable. Therefore, in the following the smaller sample is used for the reason of comparison between the two time periods.

111

treatment is considered and zero otherwise. (mobile * post) is defined as interaction between mobile and post and, therefore, is one if the two dummies mobile and post both take the value

one. mult_mobile, finally, is defined as a dummy variable taking the value 1 in case the mobile inventors moved more than once during the time under consideration (1985 – 1999). In the following, I will refer to these inventors as multiple movers.

share of applications granted

d_mobile d_post d_mobile * d_post

Model 1a 4 years 3 years -0.034 -0.031 [0.022] [0.023] -0.101*** -0.078*** [0.022] [0.023]

Model 2a 4 years 3 years -0.035 -0.037 [0.025] [0.026] -0.101*** -0.078*** [0.022] [0.023]

Model 1b 4 years 3 years -0.001 0.0002 [0.007] [0.007] -0.004 -0.003 [0.007] [0.007]

Model 2b 4 years 3 years -0.0002 0.001 [0.008] [0.008] -0.004 -0.003 [0.007] [0.007]

0.052* [0.032]

0.074** [0.033]

0.052* [0.032] 0.001 [0.022]

0.074** [0.033] 0.011 [0.023]

-0.004 [0.010]

-0.005 [0.010]

-0.004 [0.010] -0.001 [0.007]

-0.005 [0.010] -0.002 [0.007]

0.803*** [0.016] 1408 0.018 8.51(3)

0.801*** [0.016] 1408 0.008 3.86(3)

0.803*** [0.016] 1408 0.018 6.38(4)

0.801*** [0.016] 1408 0.019 2.95(4)

0.028** * [0.005] 1408 0.002 0.78(3)

0.027** * [0.005] 1408 0.001 0.54(3)

0.028*** [0.005] 1408 0.002 0.59(4)

0.027*** [0.005] 1408 0.001 0.42(4)

d_mult_mobile

Constant Observations R-squared F-test (df)

share of applications withdrawn

d_mobile d_post d_mobile * d_post

Observations R-squared F-test (df)

share of applications opposed

Model 1c 4 years 3 years 0.038** 0.034* [0.019] [0.020] -0.001 -0.002 [0.019] [0.020]

Model 2c 4 years 3 years 0.039* 0.036 [0.021] [0.022] -0.001 -0.002 [0.019] [0.020]

Model 1d 4 years 3 years -0.015 -0.022 [0.013] [0.014] -0.023* -0.027** [0.013] [0.014]

-0.058** [0.027]

-0.060** [0.028]

-0.058** [0.027] -0.003 [0.019]

-0.060** [0.028] -0.004 [0.020]

0.021 [0.018]

0.155*** [0.013] 1408 0.007 3.42(3)

0.157*** [0.014] 1408 0.007 3.32(3)

0.155*** [0.013] 1408 0.007 2.57(4)

0.157*** [0.014] 1408 0.007 2.50(4)

d_mult_mobile

Constant

share of applications refused

0.072** * [0.009] 1408 0.002 1.12(3)

0.035* [0.020]

Model 2d 4 years 3 years -0.019 -0.028* [0.014] [0.016] -0.023* -0.027** [0.013] [0.014]

0.021 [0.018] 0.009 [0.013] 0.072** 0.078*** * [0.010] [0.009] 1408 1408 0.003 0.003 1.46(3) 0.96(4)

0.035* [0.020] 0.011 [0.014] 0.078*** [0.010] 1408 0.004 1.26(4)

Table 3.12: Difference-in-differences regression estimation (OLS regression) (N = 1,408)

112

d_mobile

d_post d_ mobile * d_post

number of references per patent application Model 1e Model 2e 4 years 3 years 4 years 3 years 0.035 0.003 0.074 0.040 [0.124] [0.128] [0.141] [0.146] 0.106 [0.124]

0.064 [0.128]

0.106 [0.124]

0.064 [0.128]

0.240 [0.175]

0.318* [0.181]

0.240 [0.175] -0.071 [0.124]

0.318* [0.181] -0.069 [0.128]

d_mult_mobile

number of citations per patent application Model 1f 4 years 3 years -0.066 -0.099 [0.121] [0.128] 0.690*** 0.615*** [0.121] [0.128] 0.200 [0.172]

Constant

4.202*** 4.228*** 4.202*** 4.228*** 1.997*** [0.088] [0.090] [0.088] [0.090] [0.086] Observations 1408 1408 1408 1408 1408 R-squared 0.008 0.009 0.009 0.014 0.034 F-test (df) 3.89(3) 4.11(3) 2.99(4) 3.15(4) 16.23(3) Standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

number of claims per application

d_mobile d_post d_ mobile * d_post

Observations R-squared F-test (df)

1.976** * [0.090] 1408 0.021 9.95(3)

-0.690*** -0.615*** [0.121] [0.127] 0.200 [0.171] 0.520*** [0.121]

0.289 [0.180] 0.501*** [0.128]

1.997*** [0.085] 1408 0.046 16.92(4)

1.976*** [0.090] 1408 0.031 11.38(4)

number of applications

Model 1g 4 years 3 years 1.188*** 1.144*** [0.366] [0.388] 0.731** 0.630 [0.366] [0.388]

Model 2g 4 years 3 years 0.818* 0.816* [0.417] [0.442] 0.731** 0.630 [0.366] [0.388]

Model 1h 4 years 3 years -1.011* -0.707 [0.579] [0.441] 1.597*** 1.071** [0.579] [0.441]

Model 2h 4 years 3 years -1.819*** -1.293** [0.658] [0.502] 1.597*** 1.071** [0.577] [0.440]

0.036 [0.518]

0.214 [0.549]

0.036 [0.518] 0.679* [0.368]

0.214 [0.549] 0.600 [0.390]

-0.281 [0.818]

-0.355 [0.624]

-0.281 [0.817] 1.481** [0.580]

-0.355 [0.623] 1.074** [0.442]

9.576*** [0.259] 1408 0.021 10.01(3)

9.607*** [0.274] 1408 0.02 9.38(3)

9.576*** [0.259] 1408 0.023 8.37(4)

9.607*** [0.274] 1408 0.021 7.63(4)

5.446*** [0.409] 1408 0.015 6.91(3)

4.460** * [0.312] 1408 0.012 5.53(3)

5.446*** [0.408] 1408 0.019 6.83(4)

4.460*** [0.311] 1408 0.016 5.63(4)

d_mult_mobile

Constant

0.289 [0.181]

Model 2f 4 years 3 years -0.350** -0.372** [0.138] [0.145]

Table 3.12: (continued): Difference-in-differences regression estimation (OLS regression) (N = 1,408)

113

number of inventors per patent

d_mobile d_post d_ mobile * d_post d_mult_mobile Constant

Model 1i 4 years 3 years -0.045 -0.073 [0.121] [0.123] 0.087 0.056 [0.121] [0.123]

Model 2i 4 years 3 years -0.256* -0.250* [0.138] [0.140] 0.087 0.056 [0.121] [0.123]

-0.226 [0.171]

-0.226 [0.171] 0.387*** [0.121] 3.202*** [0.085] 1408 0.011 3.86(4)

-0.233 [0.174]

-0.233 [0.174] 0.323*** [0.123] 3.215*** [0.087] 1408 0.010 3.48(4)

3.202*** 3.215*** [0.086] [0.087] Observations 1408 1408 R-squared 0.004 0.005 F-test (df) 1.74(3) 2.34(3) Standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%

Table 3.12: (continued): Difference-in-differences regression estimation (OLS regression) (N = 1,408)

E 0 is the intercept which captures the average value of the dependent variable for inventors in the control group prior to the treatment. Gˆ2 is the true causal effect of the treatment (= move) on the outcome for the treatment group. Table 3.12 provides the results of the OLS regression analysis. Whereas Model 1 refers to the results without any further control variables, Model 2 includes a dummy variable for multiple movers. The first column of each model provides the results of the 4 year window, the second column the results of the 3 year window. Model 1 provides the same results as the t-tests with the exception that the number of citations received as well as the team size are no longer significant. Apparently, the other dummy variables leave no explanatory power to the interaction term. However, it has to be mentioned that the p-value of the interaction term in the citation regression (Model 2f) is only slightly above the 10% level (p = 0.111). In the following, only the results of Model 2 are presented. In particular, the impact of multiple movements on inventive output is analyzed. Results show that the control dummy does not have a significant effect in the regressions with status variables as dependent variables (Models 2a-2c). As explained above, the specific move under consideration has an impact on the share of patents granted and on the share of patents withdrawn. Whether the inventors are single movers or multiple movers does not make any difference. The share of

114

patents opposed by a third party is also not affected by multiple moves (Model 2d), the same applies to the number of references per patent (Model 2e). In case the number of citations is considered (Model 2f), the coefficient of multiple movements has a significantly positive impact on the dependent variable. In particular, multiple movers receive on average 0.5 more citations (3 year and 4 year window). Table 3.10 presents that the mean number of citations per patent amounts to 1.51. Therefore, an increase by 0.5 means that the number of citations increases by one third which is quite a large effect. Consequently the patent applications of multiple movers are considerably more valuable compared to single movers or non-movers. The multiple movers dummy also affects the number of claims per patent (Model 2g) – although only when the 4 year period is considered. Table 3.10 displays that the number of claims per patent increases by 0.7. Comparing this result with the descriptive statistics provided in Table 3.10 (the mean number of claims per patent amounts to 10.6), 0.7 equals an increase by about 6%. Model 2h reveals that multiple movers are responsible for more patent applications. The effect is significant at the 5% level, both for the 3 and the 4 year window. The interesting result here is that an increasing output quantity of multiple inventors seems to be a long running process. In particular, inventors who moved repeatedly, have on average applied for 1.1 (3 year window) and 1.5 (4 year window) more patents as compared to inventors who moved once or less. This result is not surprising, since inventors have to accustom themselves to R&D processes at their new employer and should therefore be able to increase their output over time. Finally, the multiple mover dummy has also a significantly positive effect on inventor team size (Model 2i). Results show that inventor teams of multiple movers include 0.4 more inventors (0.3 in case the 3 year period is considered). Again comparison with the mean team size provided in Table 3.10 (the mean team size per patent amounts to 3.16 inventors), reveals that the number of co-inventors increased by 13% (10%). Overall this result implies that inventors who move repeatedly apparently move from smaller firms to larger firms (which are characterized by larger inventor teams) which is again in line with the existing literature (Kim et al. 2004). Overall, it is shown that the effect of the multiple move dummy is quite large. What has so far not been answered is the question which effects this dummy variable actually captures. First of all, the multiple mover dummy can not be interpreted causally with respect to the move under consideration. It rather refers to the whole time period under consideration (4 years or 3 years before and after the move). Three interpretations of the multiple mover dummy seem 115

possible: first of all, the dummy could represent experience effects. Possibly multiple movers may due to their “move-experience” be able to settle in or to adjust to a new environment faster. In addition, experienced movers may be capable to make better use the knowledge from their new colleagues. Again these results point at the importance of absorptive capacity at the individual inventor level (Cohen/Levinthal 1990). Second, inventors who move repeatedly may be different from single movers or non-movers as to their personal characteristics. For instance, multiple movers may be more flexible or cosmopolitan compared to the reference group (single and non-movers). These characteristics can again help multiple movers to settle in faster and consequently to increase inventive performance. Third, it is possible that the multiple mover-effect reveals that there is another move which has a stronger impact on output quality than the move selected for the difference-indifferences analysis. To shed more light into this discussion further research should analyze multiple movers more closely.

3.6 Conclusion In this paper, first the causality between inventor productivity and inventor mobility was analyzed using a 2SLS regression estimation to deal with the endogeneity problem between productivity and mobility. Secondly, to get a better understanding of the impact of one particular move on inventive performance, a quasi-experimental design was employed. In particular, a difference-in-differences estimation was used to compare output quality of a group of mobile inventors and a non-mobile control group in the pre and post move period. Results show that the level of education has no significant influence on inventor productivity. Using external sources of knowledge during the invention process significantly affects productivity. Exploiting knowledge from scientific literature decreases productivity. The concept of absorptive capacity has some explanatory power. Inventors who have a doctoral degree are able to increase productivity by using scientific literature. Finally, firm size has a positive impact on productivity. Firm size also influences inventor mobility, although negatively. Furthermore, the level of education of the inventor, the technical concentration of inventive activity as well as the environment of the invention, in particular the fact whether the invention was made in large cities or rather rural areas, increase the number of moves. One of the key finding of this paper is that there exists a simultaneous relationship between inventor mobility and inventor productivity: An increase in the number of moves per inventor increases productivity. This outcome confirms the findings of the literature that mobility can lead to a better match between employer and employee, resulting in a higher productivity of 116

the employee. As Liu (1996: 1145) suggests inter-firm mobility rather leads to a better match quality between “worker productivity and the productivity required by the firm for the job”. This result could also mean that a move increases the technical skills or the experience of an inventor - for instance, due to knowledge spillovers from colleagues - resulting in a higher productivity. In contrast, increasing productivity decreases mobility. The literature proposes that labor mobility can help to transfer tacit knowledge which is otherwise immobile (Dosi 1988). The very fact that the mobility of an inventor facilitates inter-firm knowledge flows should suffice to give a reason for a positive relationship. Results of the 2SLS regression model nevertheless do not confirm this positive relationship. At first view this could mean that productive inventors are not at risk of being poached. But it could also mean that productive inventors do not want to move. Not because they do not receive any job offers from a competitor but because incentive systems within their employing firm work quite well. An example for such a motivation and incentive system is the IBM Fellows program. The program aims at promoting creativity among the company’s most exceptional inventors. In particular “IBM Fellows are given broad latitude to identify and pursue projects in their area of expertise to advance IBM’s technological leadership”.58 Indeed, the IBM Fellows program does not only aim at enhancing creativity but is also an efficient mean to keep key inventors from leaving the firm. IBM Fellows are provided with a research budget that is not tied to any research objective. Thus, the IBM Fellow status deters even Nobel Prize winners from leaving IBM.59 Another possible explanation for the negative causality between productivity and mobility can be special contracts or agreements, for instance, a non-compete agreement between the inventor and his employer. It is common practice that inventors, leaving their employer, are not allowed to work on the same area or project as before one (or more) year(s) after mobility took place. Non-compete agreements restrict employment options of the inventors outside the firm and, therefore, limit the inventors’ bargaining power over their employer (Fleming/Marx 2005). This could either keep inventors from leaving at all or at least make the inventors less attractive for the job market. This question could be addressed in further research by using the number of serious job offers or intra-firm mobility rather than de facto changes of the employer as a dependent variable.

58

See http://www2.hursley.ibm.com/IBM_Fellows.html and http://www.research.ibm.com/resources/awards_ fellows.shtml (access on November 10, 2005).

59

An example for an IBM Fellow who was awarded with a Nobel Prize in 1973 is Leo Esaki, http://nobelprize.org/physics/laureates/1973/esaki-bio.html (access on November 10, 2005).

117

The difference-in-differences estimation confirmed a positive effect of a particular move on output quality. Inventors are able to increase their grant rate in the aftermath of a move. Granted patents are opposed more frequently and patent applications also receive more references and more citations. Results also suggest that a move has a larger impact during a three year window before and after the move compared to a four year window. A possible explanation is that the inventors make special efforts during their first years to impress their employer or to gain respect from their colleagues. Over time the differences caused by the move decrease and finally disappear. Another explanation may be that inventors leave R&D for another job, e.g., in sales. Intra-firm mobility makes the inventor invisible in terms of patents and leads to an increase in his mean output over time. A dummy regression controlling for multiple movers finally revealed that the effect size of the multiple move dummy is quite large. Multiple movers receive on average more citations per patent. In particular, the number of citations increases by one third. Due to multiple movements the mean number of claims per patent increases by 6% and the number of coinventors increase by almost 15%. Finally multiple movers are responsible for more patent applications. Inventors who moved repeatedly, have on average applied for 1.5 more patents in case the 4 year window is employed. Three interpretations of this dummy seem reasonable: (1) multiple movers are due to their “move-experience” able to settle in faster, (2) multiple movers are different as to their personal characteristics, e.g., they are more flexible, also leading to a faster adjustment to a new working environment, and (3) the multiple mover-effect reveals that there is another move which has a stronger impact on output quality than the move selected for the differencein-differences analysis. To shed more light into this discussion further research should analyze multiple movers more closely. The results of this paper have certain implications for the management of R&D personnel. First, the matching between employee and employer seems to be of particular importance. A good match does not only explain about general differences in the productivity between inventors but also about increases in inventive performance after a certain move. For R&D management as well as for inventors these results imply that both parties should try to maximize match quality. Since match quality is hardly to observe ex ante, R&D management could try to offer different contracts to inventors, resulting in a self-selection of heterogeneous individuals to these contracts. Second, the characteristics of a single individual seem to matter less when considering inventive output. This result suggests that the composition of the inventor team could form a major determinant of inventive output. Therefore, further research should look more closely at inventor teams, especially on the effects of team composition on productivity. Possible 118

determinants of team productivity may be a heterogeneous distribution of the characteristics and skills of the team members as well as team size. Finally, results revealed the importance of move-experience for inventive performance. In particular, multiple movers turned out to hold more important patents. To shed more light into the impact of multiple moves on inventive performance, future research should analyze the life cycle of inventors. Possibly, not only the number of moves per inventor has an impact on his performance. Experience gained from previous projects the inventor was involved in or the colleagues the inventor worked together with in the past could also be of importance. Therefore, in upcoming research I will focus on the relevance of the composition of inventor teams for output quality and on the development of inventor teams over time.

119

Chapter 4 4 Institutionalized Incentives for Ingenuity – Patent Value and Ƈ

the German Employees’ Inventions Act 4.1 Introduction

In 2001, the EU15 countries invested a total of €175.5 billion in research and development activities. Approximately two thirds of this amount were R&D expenditures made by business enterprises, the remainder shared by publicly financed research institutions and academia. Almost 70 percent of these investments are expenditures for R&D personnel. Yet, it is surprising how little attention is given to the transformational process behind these numbers. Clearly, the development of ideas and new concepts is financed by the funds just summarized. But the actual work is done by inventors, either in corporate labs, in publicly financed research institutions or in academic research, and their motivation and incentives should matter greatly for Europe’s chances of becoming more successful in global technology markets. While the innovation literature in the early 50s and 60s contributed a large number of insights into invention processes60, much of the subsequent work in this field turned away from the individual inventors to consider the overall design of processes and organization. The underlying paper does not follow this trend, but seeks to return to a topic that is probably a highly neglected one in contemporary innovation studies – the motivation and performance of individual inventors. Using a novel dataset with extensive information on the context of

Ƈ

This paper is a joint paper with Dietmar Harhoff.

We would like to thank the conference audience at the 3rd EPIP (European Policy for Intellectual Property) Conference in Pisa in April 2004 as well as the conference audience at the 5th GEABA (Annual Meeting of the German Economic Association of Business Administration) Conference October 2004 for helpful comments. The survey responses used in our analysis originate from a coordinated survey effort in Italy, France, Spain, the Netherlands, the United Kingdom and Germany. The authors thank the European Commission, Contract N. HPV2-CT-2001-00013, for supporting the creation of the joint dataset. This paper makes use of the German survey responses which contain information relating to inventor compensation paid under the German Employees’ Inventor Act.

60

Consider for example the work in Ritti (1971) and Allen (1977).

121

invention processes, the value of inventions, and the biography and mobility of inventors, we describe a rather unique institutional setting in which the compensation of inventors is determined by law. The backdrop of our study is the German Employees’ Inventions Act (GEIA), once hailed as progress towards giving inventors a fair share of the benefits that they produce in creative work, and nowadays criticized as heavy-handed government intervention in processes public administrators should better not meddle in. There are few processes that go unregulated in Germany. The compensation for employeeinventors is one of the processes which many countries have left to be negotiated between the parties involved, the inventors and the corporations. Only very few countries have embedded rules and provisions in the civil code or in legal provisions. There has been a long-standing debate in Germany (but also in other countries) whether such regulation really creates proper incentives for ingenuity and inventions. Some authors have pointed out that the regulation may actually be counterproductive in that it induces strategic behavior among inventors and firms which may be harmful to innovation incentives. Since much of the potential behavior cannot be observed easily, we use inventor responses from a large-scale survey to explore some of these issues. We find that inventor compensation in Germany allocates some returns to inventors – in a few cases exceeding gross salaries by a factor of two and more. But in most cases, the compensation appears quite moderate from the inventor’s perspective. In our survey data, we measure the inventor’s compensation as the compensation received for a particular patent (or all patents to which the inventor contributed) divided by the gross salary without inventor compensation. The extent to which inventors profit from their inventions can be shown to depend most strongly on the invention’s value and quality. Second, the number of inventors has a very plausible impact which appears to be largely in accordance with the legal provisions. Variables associated with the inventor’s rank in the organization also impact significantly on the compensation share variable which is again consistent with the legal provisions. While inventor productivity (the average annual number of inventions made over the inventor’s active career) contributes positively to patent value, it has a negative coefficient in our compensation regressions. Similarly, educational attainment is negatively associated with the compensation share variable. These results presumably reflect the legal rule that individuals higher up in the organization will profit less from a service invention – simply since contributing to such inventions is part of their normal job for which they are compensated by relatively high salaries. Taken these and other results together, it appears that the mechanism is basically meritocratic in nature, and that it follows the legal provisions quite closely. Our survey responses also yield qualitative insights into the functioning of the compensation schemes. Most inventors (59.5 percent) view the legal regulation positively, largely because of its financial effect in their favor. Others emphasize the fact that the risk and costs of 122

patenting are born by the employer. Among the 28.3 percent who view the compensation rules largely negative, two opinions are dominant – that the compensation is not large enough, and that the compensation scheme lacks transparency. A criticism frequently encountered in the literature concerns the tendency that superiors appear among the inventors although they have not contributed to the invention. Inventors may accommodate such requests (or even suggest an inclusion of other decision-makers) in order to maximize the chance of having an invention being protected by patents. We do not find much evidence of such strategic behavior – only 1.8 percent of the respondents mention it. These results support a cautious positive assessment of the German Employees’ Inventions Act, but we cannot compare the compensation rules to a setting in which bonuses would be negotiated bilaterally between employer and employee-inventor. The remainder of the paper proceeds as follows. In section 2, we describe the history and provisions of the German Employees’ Inventions Act. We also summarize the results from previous studies that have analyzed this nexus. In section 3, we discuss our research questions in more detail and specify the hypotheses to be tested later. Section 4 presents the data and discusses the key variables that we have devised for the multivariate analysis. In section 5, we present some descriptive evidence bearing on our main question, while the multivariate results are presented in section 6. The final section provides a brief discussion of our results and concludes.

4.2 Invention Processes and the German Employees’ Inventions Act 4.2.1 Salient Features of Invention Processes

To assess the efficacy of institutionalized incentives for inventors, it is helpful to consider the salient features of invention processes first. We do not attempt to draw a real-life picture of such processes here, but a basic understanding of their features is important to understand the basic incentive problems. We make three observations on inventions and patents: first, productivity among inventors appears to be very heterogeneous in the sense that few inventors produce the lion’s share of inventions within an R&D department; second, the value of individual patent rights follows a highly skew distribution which can be approximated quite well using a log-normal distribution function; third, inventions are often made by or within teams. Thus, the dynamics of R&D teams is highly relevant for our study. We discuss these points in turn.

123

The productivity of inventors is quite heterogeneously distributed. An early study of this issue was undertaken by Lotka (1926). Lotka analyzed the publication output of scientists. His research resulted in the following equation, often referred to as Lotka’s Law - the number of scientists producing exactly n papers is proportional to 1/n2. Hence, when 100 scientists publish exactly one paper per time period, it would be a share of 1/22, i.e. 25 who would produce two papers in that time period, about 11 (1/32) with three publications, and so on. One may argue that such differences in productivity may merely be a function of experience (and age). Allison/Steward (1974) undertook a study to distinguish between time-invariant effects associated with the individual inventor, and experience effects due to the cumulative advantage of scientists. Cumulative advantage means that scientists become more productive during their professional life. As a measure of productivity, the authors used the number of citations received by scientific articles of US scientists in university departments. Results showed that both, pre-existing differences and cumulative advantage affect differences in scientists’ productivity. Note that in this study, output is measured using the quantity of output, not its quality. But as it turns out, the relatively good performance in terms of publication numbers is not achieved by trading quantity off against quality. Ernst et al. (2000) conducted a survey on inventors of 43 German firms and show that highly productive inventors are frequently responsible for the most valuable patents. Narin/Breitzman (1995) extended Lotka’s findings on patented inventions in the semiconductor industry. As a result, they find that output is even more concentrated than Lotka’s Law suggests. A multitude of attempts have been made to measure patent value using value indicators drawn from patent documents available in patent databases of national or regional patent offices. Indicators used are e.g., the number of citations received (Trajtenberg 1990, Harhoff et al. 1999, Lanjouw/Schankerman 1999), the number of claims (Lanjouw/Schankerman 1999 and 2001), the incidence of an opposition or litigation (Lanjouw/Schankerman 1999, Harhoff/Reitzig 2001, Harhoff/Hall 2003), or patent renewal data (Pakes 1986, Schankerman/Pakes 1986, Lanjouw et al. 1998). Results of most of these approaches have one element in common: they provide evidence for a highly skewed distribution of patent value (i.e., Harhoff et al. 1999, Scherer et al. 2000). Scherer et al. (2000) show that the top decile of German patents in 1977 accounted for 88 percent of the total value. Giummo (2003) analyzes the time profile of returns to patented inventions, as well as its cumulative value. In his analysis, information on patent value was gained from compensation records of German inventors of six major German companies (the variable we are analyzing later on). Calculation of the amount of compensation requires valuation of the surveyed invention. Therefore, compensation records are often considered a good proxy for patent value (up to a factor of proportionality). Building on this assumption, Giummo confirms previous findings concerning the skewness of the distribution of patent value. 124

The third characteristic of modern invention processes is team production. In our data, only 24 percent of the inventions covered by our survey were made by individual employeeinventors. Inventor teams consist on average of 2.5 inventors. Incentive systems need to take this characteristic into account by coming up with a sharing rule for inventor compensation. While the negotiations between co-inventors can be acrimonious, even more complex cases arise in the context of sequential inventions. Suppose that team A has made a basic invention upon which the inventions of several other inventor teams are based. If subsequent inventions replace the original one in the marketplace, the earlier inventors may see their compensation erode because too little emphasis is given to the pioneering nature of their contribution. In our survey responses, some of these conflicts will turn out to be important.

4.2.2 Institutionalized Compensation Schemes in Various Countries

In Germany, the rights and liabilities within an employer–employee-inventor relationship are governed by a specific legal institution. Comparable legal regulations only exist in Denmark, Finland, Norway and Sweden. We briefly consider a few features of these systems before turning to the German institutions. The Swedish Employees' Invention Act is of dispositive nature, i.e., the legal provisions may be amended by the employer or the employee as long as the employee’s basic right to compensation is not affected. Basically, the Swedish Employees' Invention Act distinguishes between two types of employee inventions: the work-related invention and the invention arising outside the context of employment. The rights on work-related inventions are fully transferred to the employer. For the second type, the rights to the invention remain with the employee. The employee may apply for a patent before reporting the invention to the employer, however, she must offer the employer the right to use the invention (Rebel 1993). The Danish Employees' Invention Act is similar to the Swedish law. The right to the invention remains with the employee-inventor. The inventor is obliged to report all inventions to the employer. For inventions which were made in the course of the employee's normal work the employer can claim the right to the invention. The claiming of the right has to be declared no later than four months from receipt of the invention report. Disagreements are brought before a board of arbitration. The inventor’s claim to a reasonable compensation is deemed to be satisfied with his regular salary (Rebel 1993). In the United Kingdom, France, Italy, Austria, the Netherlands and Japan, regulations concerning employee-invention are part of the respective national patent laws. According to

125

Section 39 (1) of the English Patents Act61 inventions “made in the course of the normal duties of the employee” or made in the course of “duties falling outside his normal duties, but specifically assigned to him” belong to the employer. The remaining inventions belong to the inventor himself. Compensation is to be paid only if the invention is of outstanding benefit to the employer (Section 40 (1) English Patents Act). Disputes concerning the compensation are submitted to court or are decided by the comptroller within the firm (Section 41 English Patents Act).62 The French Patent Law also assigns inventions which are made in fulfillment of an employment contract to the employer. The employee-inventor may come into possession of additional compensation (compensation beyond normal salary) if a claim to compensation is regulated by plant agreement (i.e., between employer and works council) or by contractual agreements between employer and inventor. Disputes concerning compensation have to be resolved by an arbitration commission or in court (Reitzle et al. 2000). The Italian legal regulations differ from the French ones with respect to inventor compensation. If no special arrangements have been made, the inventor is entitled to a reasonable compensation, depending on the economic value of the invention. The amount payable to the inventor decreases with the degree of involvement of the employer in the creation of the invention (Rebel 1993). In Austria, employees explicitly referred to as “inventors” are excluded from receiving compensation or only receive a limited payment. Inventors are considered to be sufficiently compensated for inventive efforts with their regular monthly salary. However, special agreements leading to some compensation for the inventors are allowed. The amount of remuneration depends on the economic value of the invention (Rebel 1993). Almost the same regulations are applied in the Netherlands. An inventor obtains an additional compensation if and only if she has not already been sufficiently remunerated by her regular salary. The amount of compensation is again derived from the economic value of the invention which is determined by the employer (Rebel 1993). The Japanese Patent Act basically assigns the right to the invention to the employee (§ 35 Japanese Patent Act). The employer receives the right to a non-exclusive license and is not obligated to pay compensation. Assignment of an employee-invention to the employer may be

61

See http://www.patent.gov.uk/patent/legal/consolidation.pdf for the English Patents Act of 1977 (access on August 18, 2005).

62

See Littler/Pearson (1979) and Orkin (1984) for a more detailed description of compensation of employeeinventors in the UK.

126

regulated in advance by contract or employment negotiations. In this case, the right to the invention is passed to the employer or the employer receives an exclusive license. The employee is then entitled to receive a reasonable remuneration on the part of the employer (Yano 1992).63 In Switzerland and Liechtenstein employee-inventions are subject to civil law. According to Art 332 Obligation Law (Obligationenrecht), patentable and not patentable inventions made during the employee's normal work duties are to be reported to the employer. Rights to the invention are assigned to the employer. Payment of compensation is compulsory. The amount of payment depends on the economic value of the service invention, the duties and position of the employee in the firm, the contribution made by the employer and by third parties, and finally on the extent internal equipment has been used for making the invention. Disputes have to be solved by labor court (Rebel 1993). In the United States64 and in Canada, there exist no special legal provisions pertaining to inventor compensation. Basically, the inventor principle is applied which means that the invention belongs to the inventor. Therefore, patents are always applied for in the inventor's name, and are then assigned to the employer. Conditions concerning an assignment of the invention to the employer are to be regulated by contractual agreements. Typical employment contracts in the US therefore specify the following obligations for employee-inventors: first, the employee-inventor has to notify the employer of each invention made. Second, the employee-inventor has to keep secret any invention or company related information, and finally, the inventor has to confer all rights to the invention to the employer during the employer-employee relationship. The employee-inventor in return has no legal claim to compensation. Compensation may also be determined in the employment contract. In cases where no contractual agreements exist between the employer and the employee-inventors and where the employer was instrumental in making the invention (e.g., by providing the inventor with the necessary tools, materials, or financial resources), the employer receives a “shop right”. Due to this shop right, the employer obtains a non-exclusive license. In exchange, the employer pays a license-fee of 1 US$ representing a symbolic inventor compensation (Leptien 1996: 98; Rebel 1993).

63

In a recent case, the inventor of blue laser diode technology, Shuji Nakamura, was awarded inventor compensation at the amount of US$ 188.7 million on January 30th 2004. See http://www.compoundsemi.com/documents/articles/news/3693.html (access on July 29, 2005).

64

See Merges (1991) for the description of the U.S. regulations concerning service inventions. Kline (1992) and Savitsky (1991) also provide insights into the compensation practices in the US.

127

4.2.3 Historical Aspects of the German Employees’ Inventions Act

While regional patent systems existed in German states already in the 18th century, the first federal German Patent Law came into force in 1877. Given the limited number of employeeinventors at the end of the 19th century, there was little need for specific legal regulation concerning employee-inventors. But the increase in the number of employee-inventors at the beginning of the 20th century led to an increasing demand for legislation (Kurz 1997, 1). A first cooperative agreement between employers and employees in the chemical industry was negotiated during the First World War. The Chemists’ Collective Bargaining Agreement65 (Chemikertarifvertrag) for academically trained employees was passed in 1920. As of 1934, considerable efforts were made to replace collective labor agreements (Tarifverträge) by a law regulating the cooperation between employer and employee-inventor. Finally, in 1939 the parties involved came to a compromise. But the draft law was rejected by the government due to its complexity. In 1942, as the Second World War turned out to become an economic war, the minister of armament single-handedly enforced a legal regulation, the Provision on the Handling of Inventions of Subordinates (Verordnung über die Behandlung von Erfindungen von Gefolgschaftsmitgliedern)66. The regulation already contained a number of provisions that were converted into today’s Act on Employees’ Inventions without substantial modifications. The Provision on the Handling of Inventions of Subordinates (PHIS), for instance, included the obligation of the subordinate to report the invention (§3 PHIS), the obligation to keep the invention secret (§6 (3) PHIS), the claiming right of the employer within 6 months after receiving the report of an invention (§4 PHIS), and the claim of the subordinate to reasonable remuneration. Mode and amount of the remuneration had to be negotiated between employer and subordinate (§5 PHIS) (Kurz 1997: 91ff). On March 20, 1943, the Remuneration Guidelines for Subordinates’ Inventions (Richtlinien für die Vergütung von Gefolgschaftserfindungen) were added to the provisions. The guidelines provided instructions to estimate the degree of the inventive effort, necessary to calculate the amount of remuneration. The degree of inventive effort depended on (1) the conceptual formulation of the problem, i.e., the degree of the subordinate’s own initiative, (2) the solution of the problem, and (3) the position of the subordinate within the firm (Kurz, 1977: 103). In the aftermath of the war, employers and unions were keen to reestablish the German patent system as soon as possible. The draft law, becoming the first government bill (Regierungsentwurf) for an Act on Employees’ Inventions in 1952, was prepared by the

65

Reichstarifvertrag für die akademisch gebildeten Angestellten der chemischen Industrie as of April 27, 1920.

66

See Reichsgesetzblatt I, 1943, p. 466.

128

Federal Ministry of Justice (Bundesjustizministerium) and also by the German Association for Intellectual Property Rights and Copyright Protection (Deutsche Vereinigung für gewerblichen Rechtsschutz und Urheberrecht). The government bill proposed the establishment of a board of arbitration at the German Patent Office. Due to a lengthy discussion about amendments, the German Bundestag could not pass the draft law during the legislative period between 1949 and 1953. Thus the Federal Ministry of Justice worked on a new proposal. In 1955, the second government bill was brought forward. The second proposal excluded technical improvement proposals67, release under reserve of a use (Freigabe unter Benutzungs-vorbehalt) was replaced by restricted claiming of inventions, and the provisions on inventions of university professors and scientific assistants68 were added. By means of this proposal passing of the law became possible. In 1957 the German Employees’ Inventions Act became effective. The need for this legislation arose because of a conflict between the German Employment Law and the Patent Law. According to the German Employment Law, the results of the work of an employee belong to the employer, whereas the Patent Law assigns the property of an invention to the inventor himself. The Employees’ Inventions Act produced a balance between employer and employee. Whenever the rights to the patent are transferred to the employer, the employer must in return pay the employee-inventor a reasonable compensation. Moreover, the law was supposed to strengthen incentives for inventors in corporations (Leptien 1996: 83).

4.2.4 Regulations of the German Employees’ Inventions Act as of 1957

In its current form, the German Employees’ Inventions Act applies to all patentable inventions (patented or not) or inventions which are eligible for a utility model as well as to any other technical improvement proposals made by employees (§§2, 3 ArbNErfG69). The Act applies to inventions made by inventors in organizations which are governed under German

67

The first government bill assigned the monopoly principle (Monopolprinzip) to inventions and the supplementary benefit principle (Sonderleistungsprinzip) to technical improvement proposals. According to the monopoly principle, the inventor was granted a compensation for providing his employer with a monopoly (the right from the patented invention). According to the supplementary benefit principle, the right to a compensation arises from an effort not bounded by contract. Since the new proposal aimed at aligning the law with the monopoly principle, technical improvement proposals were excluded from the second government bill (Kurz 1997: 232).

68

According to §42 ArNErfG, containing the handling of university inventions, inventions of professors, lecturers and scientific assistants were free inventions. Meanwhile, §42 ArbNErfG, and in particular the socalled professorial privilege, has been revoked. On February 7, 2002, the modified §42 ArbNErfG became effective, which treats university inventions as inventor-employee inventions and therefore, as service inventions.

69

Arbeitnehmererfindungsgesetz (ArbNErfG)

129

law or in German subsidiaries of international organizations. It provides a set of rules concerning rights and liabilities of both the employer and the employee. The Act distinguishes between service inventions and free inventions. Service inventions are inventions which either result from the obligatory activity of the employee in the company or “(…) are substantially based on experience or activities of the company” (§4 ArbNErfG). Other inventions, for instance, inventions made by employees during their leisure time or by self-employed inventors, are free inventions. According to §5 ArbNErfG the employee is obligated to report a service invention to the employer immediately.70 Within the period of four months from the receipt of the report of the invention, the employer can claim the invention on a restricted or unrestricted basis (§6 ArbNErfG). If the employer does not claim the invention the legal title to the invention is released to the inventor. In case of an unrestricted claim to the invention, all rights to the invention are transferred to the employer, and the employer is obliged to file a national patent application for the invention. A restricted claim provides the employer with a non-exclusive right to use the invention which implies that the employer is not allowed to grant licenses on the patented invention (Reitzle et al. 2000). Restricted claims turn out to be quite infrequent – in our data, only 2.6 percent of all patents are claimed by the employer on a restricted basis. In the case of a restricted claim, the employer has no obligation to file a German patent application. An inventor, who wants his invention to be protected by a patent, has to file the application in his own name. Once the invention is claimed, either in restricted or unrestricted form, the employer has the obligation to reasonably compensate the inventor. The inventor’s right to remuneration arises as soon as the employer has claimed the right to the service invention (unrestricted claiming of right) or as soon as the employer has claimed the right to the invention and uses it (restricted claiming of right). Guidelines for the Remuneration of Employees’ Inventions in Private Employment71 were first issued by the Federal Minister for Labor and Social Affairs (Bundesminister für Arbeit und Sozialordnung) in 1959. These guidelines are based upon the Remuneration Guidelines for Subordinates’ Inventions from March 20, 1943. They regulate in some detail how the compensation is determined. The compensation is supposed to be proportional to the value of the invention. According to Section 1 of the guidelines, three different methods exist for calculating the value of the invention:

70

Free inventions also have to be reported without delay. In case the employer does not contradict that the invention is free, it is at the employee’s disposal (§§ 18, 19 ArbNErfG).

71

See Bundesanzeiger No. 156 of 18.08.1959, Annex. The Guidelines were amended by Sept. 1, 1983, see Bundesanzeiger 1983, p. 9994. The guidelines are not legally binding provisions. They only provide an informative basis for calculating the inventors’ compensation. However, the Board of Arbitration at the German Patent and Trademark Office as well as the courts check the appropriateness of compensation by means of these guidelines (Leptien 1996, 86).

130

•

by using a licensing analogy, i.e., by determining the license fee that would have to be paid for the use of a comparable invention owned by a third party,

•

by calculating the benefits from the invention accruing to the employer, i.e., the difference between costs and revenues resulting from the use of the invention, or

•

by estimation of the value of the invention, i.e., by determining the price which would have had to be paid by the company to buy the invention from a free inventor.

The estimation of the value of the invention provides the basis for the calculation of the compensation payable to the inventor. In a second step, the share of value accruing to the inventor(s) is determined. According to §9 (2) ArbNErfG, the proportion attributable to the inventor(s) depends: •

on the economic exploitability of the service invention, i.e., the value of the invention which is determined according to the three above described methods,

•

on the duties and position of the employee in the company, i.e., the share of the inventor in the creation of the service invention decreases the more it is expected of him by reason of his position and by the amount of salary paid to him at the time of the report of the invention, and also

•

on the degree of involvement of the company in the creation of the service invention, i.e., the share of the inventor in the creation of the service invention increases the greater his own initiative in recognizing the problem, and the smaller the company’s support with technical assistance.

If more than one employee-inventor is responsible for a service invention, the relative contributions of the inventors have to be specified. §12 (2) ArbNErfG constitutes that the compensation must be determined for each inventor separately. Each inventor has to be informed about the total amount of remuneration and the share received by the other coinventors. Disputes arising between employees and employer regarding the inventors’ compensation can be brought before the Board of Arbitration at the German Patent and Trademark Office in Munich or Berlin (§§28-36 ArbNErfG). The Arbitration Board issues a proposal for a settlement. This proposal is binding for both parties unless a written opposition is filed within one month. Should an appeal be filed against the proposal, the proceedings before the Arbitration Board are deemed to have been unsuccessful and the filing of an action with the court having jurisdiction (the respective district court) is possible. On average, fewer than 100 disputes per year are negotiated before the Arbitration Board (GPTO 2003). Compared to the 131

annual number of patent applications to which the German Employees’ Inventions Act applies72, this number is quite small.

4.2.5 The Impact of the German Employees’ Inventions Act

Since its inception in 1957, the German Employees’ Inventions Act has been subject to many controversial discussions. Within the last 20 years, a number of economic and legal studies have analyzed the advantages and disadvantages of the law and of the associated institutions. The Act aimed at creating a social balance between employer and employee, as well as providing incentives for inventive activities. Several theoretical and empirical analyses have examined to what extent this original objective of the Employees’ Invention Act has been attained. In the following section, we summarize some results from this literature. We first address literature, providing potential advantages of the German law. According to Merges (1999), the Employees’ Invention Act enhances the degree of legal certainty for employee-inventors. Due to the law, inventors are entitled to receive compensation in exchange for the assignment of the rights to the invention. A transfer of rights to the employer is economically plausible, since the employer may have made specific investments in complementary assets to exploit the employee’s invention. To ensure employment, firms have to balance risks by holding a patent portfolio. Successful inventions can compensate for losses (Merges 1999). Apart from spreading the risk, employee-inventors would not be able to afford costs associated with patent applications. The following two empirical analyses highlight the importance of remuneration for inventive activity: already in 1931, Rossman asked 710 inventors about their motives and incentives which cause them to invent. The most important motives of inventing turned out to be “love of inventing”, “desire to improve”, and “financial gain” (Rossman 1931: 523f). The relevance of monetary incentives, in particular inventor compensation, was also confirmed by Staudt et al. (1990) who conducted a survey of 522 employee-inventors. Respondents were drawn by random sampling from a list of all German patents published in 1987 and held by German applicants. Results show that more than 70 percent of the inventors rank inventor compensation as important. Less important are advancement, trainings, or flexible working hours (Staudt et al. 1990).

72

In 2002 e.g., the German Patent and Trademark Office (GPTO) received 51,513 patent applications from enterprises which are governed by German law and thus by the inventor compensation scheme (GPTO 2003).

132

Results from the above described analyses indicate that inventors basically regard the legal regulations as motivating for inventive activity. But the literature also points to some dysfunctional effects that the Employees’ Invention Act may have. For example, delayed payments are generally thought to provide relatively weak incentives for inventive effort. A study of 10 major inventions conducted by Globe at el. (1973) shows that inventions are brought to market several years after the date of patent application. The results of this study show that the shortest time lag from invention to commercialization and therefore to payment of the remuneration amounts to 6 years. Brockhoff (1997) argues that large firms frequently are organized in terms of profit centers and therefore are keen to delay payment. Additionally, exorbitant administration efforts impede annual payment. As a rule, compensation is paid in two- or three-year intervals (Brockhoff 1997). Another frequently discussed issue concerns the actual calculation of the remuneration. First of all, the German Employees’ Inventions Act does not provide detailed provisions concerning the calculation of the compensation. Inventors, therefore, complain about the strong influence that employers may have on how remuneration is determined. Furthermore, in many cases there is a choice among potential reference quantities (e.g., sales or production quantity) which can influence the amount of compensation considerably (Gaul 1988). In this context, Kersten (1996) criticizes that compensation is calculated as proportion of overall turnovers achieved with a product. Therefore, radical innovations (which initially generate very low levels of revenues, but a large relative increase in revenues) are disadvantaged in comparison to incremental innovations or modifications of existing technology. To counteract such effects, Kersten proposes to limit compensation to the actual increase in sales due to the invention. Additionally, the allocation of remuneration between co-inventors may lead to a reduction of employee inventors’ motivation. Each inventor is entitled to be informed on the total amount of remuneration and the shares that each co-inventor receives. However, the attribution of performance is difficult and often leads to controversies. Furthermore, an increase in the number of co-inventors reduces compensation for every single inventor. The motivation within a research team may suffer from such disputes (Staudt et al. 1992, Heimbach 1992). Moreover, the profitability of new products is also impacted by efforts made by employees other than inventors. Manly (1978) criticizes that legal regulations “ (...) single out one cog in the innovative wheel – the inventor”. The author especially argues that today's R&D processes are characterized by cooperation within interdisciplinary teams of specialists from different functions within the firm. The German Employees' Invention Act in contrast is only applicable for employee-inventors. Moreover, Staudt et al. (1992) find that 27.9 percent of the questioned inventors complain about superiors being mentioned in invention reports because of their hierarchical position, 133

but not due to their contribution to the invention. The phenomenon of executives being included as co-inventors without having made a contribution to the invention is also reported by Brockhoff (1997) and Schmeisser (1986). Delay of payment, intransparent calculation of remuneration and unfair allocation of remuneration between co-inventors are only three examples of causes for disputes between employer and employee-inventor. According to §28 ArbNErfG, the Board of Arbitration may be called upon in case of a dispute. However, both Giummo (2003) and Manly (1978) find that the number of conflicts brought before the Arbitration Board at the German Patent and Trademark Office is relatively small when compared to the overall number of patents for which such a conflict could in principle arise.73 But they differ in their interpretation of this indicator. Manly (1976) interprets the limited amount of disputes as a sign of an effective operation of legal regulations in Germany. Conversely, Giummo (2003) argues that inventor employees are unlikely to jeopardize their careers by initiating a legal conflict with their employer. In his interpretation, the low number of conflicts is not informative about the actual effectiveness of the legal provisions governing inventor compensation. Another possible interpretation may be that inventors are not sufficiently informed about the legal provisions of the German Employees’ Inventions Act. Leptien (1996) surveyed 116 inventors of German firms active in the electrical engineering, mechanical engineering and chemical industries. One of the major findings is that 13 percent of the inventors are inadequately informed about the regulations of the Employees’ Invention Act. Staudt et al. (1992) confirm that employee-inventors have only partial knowledge of their rights. Given the controversy surrounding inventor compensation in general and the German institutions in particular, it is not surprising that some observers have called for the abolition of the law. For example, Brockhoff (1997) proposes to replace the collective legal regulation with an individual incentive system for employee-inventors, where compensation is a result of negotiations between inventor and employee.

73

In 2002 the patent stock of the GPTO amounted to 376,744 patents. This number includes patents granted by the EPO with effect in Germany. According to the GPTO annual report, more than 80% of the applications filed with the GPTO are attributable to German firms. Therefore, patents coming under the German Employees' Invention Act at least amount to 300,000. In comparison, the Arbitration Board at the GPTO received 95 requests in 2000, 81 requests in 2001, and 87 requests in 2002 (GPTO 2003).

134

4.3 Research Questions and Hypotheses While a number of studies have looked at particular features of the German Employees’ Inventions Act, little representative large-sample evidence has been produced. The most reliable information on the amounts paid out to inventors comes from Giummo’s (2003) study. But his figures are not representative, since they reflect inventor compensation in a few large corporations. The questions that we address in this study are the following: •

Does the German Employees’ Inventions Act create substantial rewards?

•

Which type of inventor profits the most from the German Employees’ Inventions Act?

•

How does compensation differ across industries, technical fields, etc.?

•

Do inventors consider the law important in providing suitable incentives?

•

Do inventors point to significant disincentives created by the law?

To answer these questions, we collected data on inventor compensation (measured as the share of gross salaries before bonus payments) associated with a particular patent, characteristics of the patent and of the associated invention process as well as information on inventor biographies. Given that the German Employees’ Inventions Act regulates compensation relatively precise, our hypotheses for the empirical tests are easily derived: H1. Inventor compensation increases with the value of the patent right. H2. Inventor compensation decreases with the number of co-inventors. H3. Inventor compensation decreases with the inventor’s rank in the organization. H4. Measured as a share of gross salaries, inventor compensation does not vary across industries. H5. Measured as a share of gross salaries, inventor compensation does not vary with firm size. The first three hypotheses reflect largely the regulatory components of the German Employees’ Inventions Act. The latter two state that inventor compensation follows typical patterns of industry and firm size. A large number of studies have documented wage differentials between industries and between firms of different size. We measure inventor

135

compensation as the share of gross salaries without inventor-specific remuneration; in our regressions industry and firm size variables should therefore have no statistically discernable effect if inventor compensation is proportional to gross wages.74

4.4 Data Source and Sample 4.4.1 Data Source – the German Inventor Survey

Data underlying this survey was collected within the scope of a European project sponsored by the European Commission. The project named PatVal (The Value of European Patents: Empirical Models and Policy Implications Based on a Survey of European Inventors) started in January 2002. The main objective of the PatVal project is to create a database of characteristics of the invention process. The data were obtained from a survey of European inventors which were named in EPO patent grants. The survey responses were combined with information drawn from the patent documents and an extended patent database. Research groups from six European universities collaborated on this project. In each of the six countries (France, Germany, Great Britain, Italy, Spain, and the Netherlands) domestic inventors were asked simultaneously about their granted EP patents as well as the invention process leading to the specific patent. A detailed description of the research design and of descriptive statistics is presented in Giuri et al. (2005). This survey only relies on the German dataset of the PatVal survey. Therefore, units of observation are inventors who lived in Germany at the time of application of the respective patent. 10,500 EP patents containing inventors living in Germany were chosen by a stratified random sample from a list of all granted EP patents with priority date between 1993 and 1997 (15,595 EP patents). A stratified random sample was used in order to oversample potentially important patents. To do this, the sample contains all patents an opposition had been filed against by a third party (1,048) as well as patents which were not opposed but received at least one citation (5,333). Out of the remaining patents (9,212) a random sample of 4,119 patents was drawn. Within this sample, 118 inventors moved to another country in the meantime, and 857 are multiple inventors which means they filled out at least two

74

Strictly speaking, the GEIA even allows us to be quite specific about the functional form of our regressions. The compensation is proportional to patent value, to the inventor’s share in the invention, and to another factor measuring the inventor’s contribution to the service invention. The latter depends on the economic exploitability of the service invention, the duties and position of the employee in the company, and also on the degree of involvement of the company in the creation of the service invention.

136

questionnaires. The remaining 8,357 inventors live in Germany and are represented with one patented invention in the dataset. The questionnaire was mailed to the identified inventors. As addressee we chose the first inventor listed on the patent document. In cases where a verification of the inventor’s address had not been possible the second inventor was chosen. If the address of the next inventor could not be verified, we proceeded until an address definitely turned out to be correct or a new address could be assigned to the inventor. In cases where the invention had been made by a single inventor or verification of the addresses had not been possible, we chose the first inventor (the only inventor) mentioned on the patent document. The selected inventors were provided with a cover letter together with the questionnaire. The letter also contained a link leading to a web questionnaire in order to give the inventors the possibility to choose between the paper-based and the web-based questionnaire. To date, we received 3,346 responses, resulting in a response rate of 32%.75 The questionnaire is divided into six sections: section A contains personal information about the inventors, section B contains information on their educational backgrounds. Section C covers data on employment and mobility of the inventors. Section D is about the invention process (collaborations, important sources of knowledge). Section E contains information on the inventors’ rewards as well as the German Employees’ Inventions Act. This section will be most important for the following analysis. Section F finally deals with the value of the patents. We merged the data from the questionnaire with bibliographic and procedural information on the respective patents obtained from the online EPOLINE database provided by the EPO. The EPOLINE database contains information on all published EP patent applications as well as on all published PCT applications since the foundation of the EPO in 1978. The dataset is an equivalent of the EPOLINE data as of March 31st, 2003 and covers over 1,200,000 patent files with application dates ranging from June 1st, 1978 to July 25th, 2002.

75

We tested whether inventors who answered the questionnaire early differed significantly from inventors who answered late. The first 10% of respondents were considered early respondents whereas the last 10% were the late respondents in this analysis. Each of the two groups contained about 300 inventors. The most important dependent and explanatory variables were tested for differences: the value of the surveyed patent, the value of the patent family as well as the strategic value of the patent, additionally, the compensation for the surveyed patent and the compensation for all patents as share of annual income, finally, the inventor’s age as well as the number of employees of the applicant. Results show no significant differences (at the 10% level) between the two groups.

137

4.4.2 Variables

From the datasets described above, we generated a number of variables which are used in our empirical analysis. We briefly describe them here. patshare - share of salary received as inventor compensation for the surveyed patent. In the

survey, we asked the respondents which share of their fixed salary they had received for the patent in question. We employ an Ordered Probit estimator to relate this variable to the exogenous regressors described below. education – the questionnaire asked respondents to indicate their terminal degree. In order to

simplify the analysis, we aggregate our education variables to three groups: secondary school or vocational training, vocational academy or university studies, doctoral or postdoctoral studies. age – the age of the inventor at the time of the survey. invprod – inventor productivity adjusted for age. We divide the total number of patent

applications and inventions kept secret by age minus 25. One way of justifying this measure would be the assumption that inventors became active at the age of 25 and continued to work with constant productivity. Obviously, this is a stark simplification, but it will serve to generate a first version of an age-corrected productivity figure. mainfield – main technical field. This variable aggregates the technical fields to which the

inventions belong to. firmsize – number of employees. We obtained the firm size variable from the questionnaire.

We use log(number of employees) in our regressions. inventors – the number of inventors. The amount of inventor compensation received by any

of the inventors depends on the total number of inventors. Since we do not have exact information on the contributions made by the inventors, we use this variable to control for differences between inventions coming from teams of differential size. patvalue – the monetary value of the patent. A central question in our survey asked

respondents to indicate the value interval for their patent. The intervals were less than 30,000 €, between 30,000 € and 100,000 €, between 100,000 € and 300,000 €, between 300,000 € and 1 million €, between 1 and 3 million €, between 3 and 10 million €, and above 10 million €. We generate a set of 7 dummy variables from this group and include them (with the exception of the first) in the regressions.

138

stratvalue – strategic value of the patent. This variable indicates if the patent belongs to (a)

the top decile of patents in this industry, (b) the top quartile, but not the top decile, (c) the top fifty percent but not the top quartile, and (d) the lower half of the patents in the industry. We maintain three of the four dummy variables in the regressions. Since these are likely to be collinear with the PATVALUE dummies, we use these measures as alternatives, not in conjunction. originality and generality. These measures were first proposed and computed by Trajtenberg

et al. (1997) for US patents. The GENERALITY index is based on the assumption that patents which cover basic research will be cited in a broader range of technical fields. Therefore, it is assumed that the more widely a patent is cited from other patents outside its “home” patent class, the more important is the advance in knowledge of the patented invention (Mowery/Ziedonis, 2002). GENERALITY is defined as follows:

Generality i

Ni § Nciting ik 1  ¦ ¨¨ k 1 © Nciting i

· ¸¸ ¹

2

where for each patent i, k indexes the patent classes and Ni is the number of different classes to which the citing patents belong. Ncitingik denotes the number of citations patent i received in technical class k. Ncitingi stands for the total number of citations received by patent i. The index can only be calculated for patents which have received at least one citation. Otherwise the index is assumed to be zero. Generalityi ranges from [0; 1]. Higher values represent more generality and therefore less concentration (Henderson et al. 1998). The ORIGINALITY index also provides a measure for the extent to which a certain technology is spread across technological fields. Contrary to GENERALITY which is a forward-looking measure, ORIGINALITY is a backward-looking measure including cited patents instead of citing patents with regard to the underlying patent. ORIGINALITY is defined as follows:

Originalityi

Ni § Ncited ik 1  ¦ ¨¨ k 1 © Ncited i

· ¸¸ ¹

2

where for each patent i (analog GENERALITY index), k indexes the patent classes and Ni is the number of different classes to which the cited patents belong. Originalityi also ranges from [0; 1] (Trajtenberg et al. 1997).

139

PCT – a dummy variable indicating that a PCT application had been filed for this patent. citations – citations received within 5 years following the publication of the search report.

These measures were obtained from a citation database maintained at the Institute for Innovation Research. famsize – size of patent family. We measure the size of the overall international patent family

by computing the number of equivalent patents in existence. This number is obtained from the ESPACE server maintained by the European Patent Office. mobility variables – in order to test if mobile inventors (and patents based on inventions

made by mobile inventors) differ from suitable control groups. We use dummy variables indicating that the inventor either (a) did not change to another employer after having made the invention (reference group) or (b) changed employers once or (c) changed employers twice or (d) changed employers more than twice since the date of the invention. city variables – two dummies indicating if the invention has been made in a city with more

than 1 million inhabitants or in a city with between 500,000 and 1 million inhabitants. The reference group are inventions made in rural areas or cities with fewer than 500,000 inhabitants. invention context variables – these variables reflect characteristics of the invention process,

in particular, (a) if the invention came about as the planned result of an R&D project, (b) whether it was an expected by-product of such an R&D process, (c) whether it was an unexpected by-product, or (d) whether the invention was made during the leisure time of the inventor. The reference group is given by inventions that were the product of a non-R&D process, for instance, inventions made in production or other functions of the firm.

4.5 Survey Evidence - Descriptive Statistics The sample used for the multivariate analysis contains data from questionnaires received from 1,983 inventors. These are considerably fewer observations than in the overall sample – the reduction is due to the fact that some inventors were independent inventors (266 cases) and that some variables we need for the analysis are missing (1,097 cases). Table 4.1 presents summary statistics, i.e. mean values and standard errors for the variables described before. The compensation for the surveyed patent (as share of annual gross income) ranges from 0 to 100 at an average of 1.8. 100% means that inventors double their annual income due to 140

inventors’ compensation. Payment for all patents of the surveyed inventors ranges from 0 to 500 at an average of 8.3. Over all, 18 inventors receive more than their annual income due to compensation for all of their patents. The distributions of these variables are depicted in Figure 4.1 and Figure 4.2. It is clear from these graphs that inventor compensation has a rightskew distribution – most inventors receive no or very small compensations for their inventions, while few inventors can add substantial sums to their gross salary.

100% 90%

relative frequency

80%

77.9%

70% 60% 50% 40% 30% 20% 7.7%

10%

8.3% 3.8%

1.4%

0.7%

0.1%

0.2%

> 10 - 20%

> 20 - 50%

> 50 - 75%

> 75%

0% 0 - 1%

> 1 - 2%

> 2 - 5%

> 5 - 10%

share of salary

Figure 4.1: Share of salary received as inventor compensation for this patent (N = 1,983)

60%

relative frequency

50%

40%

39.2%

30% 21.3% 20% 13.7%

11.5% 7.7%

10%

4.7% 0.2%

0.9%

0.8%

> 75% 100%

> 100%

0% 0 - 1%

> 1 - 2%

> 2 - 5%

> 5 - 10% > 10 - 20% > 20 - 50% > 50 - 75%

share of salary

Figure 4.2: Share of salary received as inventor compensation for all patents, multiple inventors are included once (N = 1,800) 141

The inventors are characterized by a high educational level. On average, 50% of the inventors in the sample earned a university degree; another 38% undertook doctoral or postdoctoral studies. At the time of the survey, the inventors were aged between 32 and 76 at an average of 53 years. The inventors’ productivity ranges from 0.1 to 32.3 patents per year of inventive activity, with a mean of 1.0. This result confirms previous findings by Lotka (1926), who found that the productivity of inventors follows a highly skew distribution. Almost 60% of all patents were assigned to the chemical/pharmaceutical industry and to mechanical engineering. On average, patents were held by companies employing 52,278 employees. The number of employees ranges between 1 and 500,000 with a standard deviation amounting to 97,340. The median of the monetary patent value, ranging from “< 30,000 Euro” to “more than 10 million Euro”, falls in the third category “100,000 to 300,000 Euro”. The strategic patent value has its mean at “the patent belongs to the top 50% but not top 25% of the patents within the technological field”. The number of citations received within 5 years after publication of the search report ranges from 0 to 13 at an average of 0.5 citations.

Variable Share of salary received as inventor compensation for the surveyed patent Share of salary received as inventor compensation for all patents1 Education Lower secondary school Upper secondary school Vocational training Trade and technical school University studies Vocational academy Doctoral/postdoctoral studies Age of the inventor at the time of the survey Inventor productivity Main technical field Electricity/electronics Instruments Chemicals/pharmaceuticals Process engineering Mechanical engineering Consumer goods/civil engineering 1: Multiple inventors are included once (N = 1,800) 2: Median

Table 4.1: Descriptive statistics (N = 1,983)

142

Mean 1.76

7.94

0.01 0.01 0.06 0.03 0.50 0.02 0.38 52.90 1.02 0.15 0.09 0.27 0.19 0.24 0.07

S.D. 5.26

Min. 0

Max. 100

25.70

0

500

9.32 1.61

0 0 0 0 0 0 0 32 0.1

1 1 1 1 1 1 1 76 32.26

0 0 0 0 0 0

1 1 1 1 1 1

Variable Number of employees Number of inventors Monetary value of the patent Strategic value of the patent Originality Generality Oppositions received PCT application filed Citations received within 5 years Job mobility Inventor did not change the employer Inventor changed employer once Inventor changed employer twice Inventor changed employer three times Inventor changed employer more than three times Environment of the invention More than 1 million inhabitants 500,000 to 1 million inhabitants Less than 500,000 inhabitants Invention process R&D project / planned result R&D project / expected by-product R&D project / unexpected by-product Invention arose during normal job / not R&D Invention arose during leisure time

Mean 52,278 2.49 32 32 0.04 0.01 0.11 0.27 0.54

S.D. 97,340 1.65

Min. 1 1 1 1 0 0 0 0 0

Max. 550,000 15 10 4 0.67 0.63 1 1 13

0.83 0.11 0.04 0.01 0.00

0 0 0 0 0

1 1 1 1 1

0.10 0.12 0.78

0 0 0

1 1 1

0.29 0.18 0.18 0.29

0 0 0 0

1 1 1 1

0.05

0

1

0.14 0.07

1.14

1: Multiple inventors are included once (N = 1,800) 2: Median

Table 4.1 (continued): Descriptive statistics (N = 1,983)

Table 4.2 to Table 4.5 summarize the univariate or bivariate relations between the share of compensation received for the surveyed patent and a number of exogenous variables, i.e., inventors’ age and education, firm size and number of inventors, monetary patent value, and strategic patent value. Table 4.6 tabulates the average values of a number of variables by technical field.

143

As to inventor age and education, Table 4.2 suggests that there are almost monotonic relationships between these variables and inventor compensation for the surveyed patent.76 With greater educational attainment, the compensation share is decreasing. Presumably, this reflects the impact of the rank of the individual within the corporation (H3). As age increases, inventors tend to earn a higher share as compensation for the surveyed patent. This may very well reflect selection processes – productive inventors are retained in R&D, so that over time, a positive correlation between value of a patent and inventor age emerges. Note that the effect must be strong, since it even compensates the base effect in our dependent variable. As inventors get older, their base salary is presumably increasing due to seniority effects. If this presumption is correct, the inclusion of patent value in our multivariate regressions should render the age variable insignificant. In our value regression, the age variable should have a large positive coefficient. As we will see later, both predictions are actually born out.

Age (groups)

31 to 40 41 to 50 51 to 60 61 to 70 Total

Secondary school/ vocational training 1.25 (14) 2.56 (54) 2.70 (79) 5.32 (64) 3.36 (211)

Educational achievement (groups) Vocational academy/ Doctoral/post doctoral university studies studies 1.46 0.70 (103) (39) 1.68 1.44 (380) (341) 1.81 1.46 (263) (202) 1.80 1.23 (282) (162) 1.72 1.36 (1,028) (744)

Total 1.25 (156) 1.63 (775) 1.81 (544) 2.06 (508) 1.76 (1,983)

Note: In a bivariate ANOVA, the effect of education is highly significant (F = 11.68, p = 0.000), whereas age effects are not significant (F = 0.93, p = 0.423)

Table 4.2: Inventor compensation by age and education

A typical finding in labor economics suggests that wages in large firms are higher than those in smaller firms. Since we cannot control for the level of gross wages, there is some ambiguity associated with the tabulation of the compensation share variable. If the compensation share is also a positive function of firm size, we would expect the compensation variable to rise or be constant as firm size increases.

76

Similar results emerge for the overall compensation (inventor compensation for all inventions divided by the gross salary before compensation payments).

144

Firm size in number of employees (groups) less than 250 251 to 1,500 1,501 to 10,000 more than 10,000 Total

Number of inventors (groups) 1 2.44 (97) 2.70 (180) 2.24 (183) 1.75 (227) 2.23 (687)

2 2.56 (54) 2.33 (118) 1.45 (143) 1.46 (202) 1.77 (517)

3 0.82 (30) 1.38 (61) 1.44 (99) 1.46 (161) 1.39 (351)

4 and more 0.91 (21) 1.42 (55) 2.36 (99) 0.91 (253) 1.31 (428)

Total 2.07 (202) 2.23 (414) 1.90 (524) 1.37 (843) 1.76 (1,983)

Note: In a bivariate ANOVA, the effect of the size of inventor teams is significant at the 10% level (F = 2.42, p = 0.064), whereas firm size effects are not significant (F = 1.97, p = 0.116)

Table 4.3: Inventor compensation by firm size and number of inventors

The descriptive statistics in Table 4.3 do not confirm that view. At best, we find an inversely U-shaped relationship. It seems clear that the compensation shares in larger firms are smaller. However, note that this may reflect differences in the organization of R&D – inventor teams in large firms may very well have more members, thus reducing each inventor’s share. In the multivariate regression, we will control for such effects. Should firm size not have a statistically significant impact, then we would conclude that inventor compensation (in absolute terms) is depending on firm size just as gross wages are. The relationship between compensation and the number of inventors is more straightforward – as the invention team gets larger, the average compensation share for each inventor is reduced.

Patent value

less than 30,000 € 30,000 to 100,000 € 100,000 to 300,000 € 300,000 to 1 million € 1 to 3 million € 3 to 10 million € more than 10 million € Total

Compensation for this patent (share of gross annual income) Number of observations Share of Obs. 190 9.6% 381 19.2% 449 22.6% 433 21.8% 263 13.3% 162 8.2% 105 5.3% 1,983 100.0%

Mean 0.71 1.24 1.71 1.63 2.05 2.80 3.97 1.76

Note: In a univariate ANOVA, the effect of the monetary patent value is highly significant (F = 6.33, p = 0.000).

Table 4.4: Inventor compensation by monetary patent value

145

Table 4.4 displays the relationship between the monetary value of the patent (as indicated by the inventor) and the inventor’s compensation (again for the patent under consideration). The compensation share is (almost) monotonically increasing with patent value.

Strategic importance of patent

top 10 percent top 25 percent top 50 percent lower 50 percent Total Note:

Compensation for this patent (share of gross annual income) Number of observations Share of Obs. 332 16.7% 362 18.3% 452 22.8% 837 42.2% 1,983 100.0%

Mean 2.70 2.71 1.72 1.00 1.76

In a univariate ANOVA, the effect of the strategic patent value is highly significant (F = 13.64, p = 0.000).

Table 4.5: Inventor compensation by strategic patent value

From Table 4.5 we can see clear evidence that the other “relevance variable” – the strategic and economic importance of the patent – has a plausible and statistically significant association with inventor compensation. This finding is again in accordance with the Guidelines for Remuneration as well as § 9 (2) ArbNErfG, determining that the economic exploitability of an invention determines the amount of payment. Note that the first group of patents – those ranked among the top 10 percent in strategic and economic importance – account for 16.8% percent of the observations. That simply reflects the fact that our stratified sampling approach has led to an oversampling of valuable and strategically important patents. Table 4.6 summarizes mean values of a number of regressors by technical field. A brief inspection of this table shows that the compensation share is strongly affected by the number of inventors. While patents in chemicals and pharmaceuticals are the most highly cited and account for the largest patent families, the average number of inventors per patent is also relatively large. The average compensation share per patent in chemicals and pharmaceuticals is therefore the lowest of all technical fields. Nonetheless, the table also yields the puzzling result that overall compensation shares are the highest in the technical field of consumer goods and civil engineering. This table suggests that there are significant differences in inventor compensation across technical fields. But again, since various variables may have countervailing effects, the technical field impact needs to be considered in the multivariate setting. Before we turn to our multivariate results, we briefly comment on our qualitative survey responses.

146

Compensation for this patent

Electricity/electronics Instruments Chemicals/pharmaceuticals Process engineering Mechanical engineering Consumer goods / civil eng. Total

1.39 1.41 1.59 2.09 1.94 2.19 1.76

Compensation for all patents

6.48 6.23 6.78 9.40 9.54 13.58 8.31

No. of inventors

2.19 2.01 3.29 2.37 2.23 1.83 2.49

No. of citations received within 5 years 0.48 0.59 0.73 0.51 0.41 0.34 0.54

Size of patent family

4.42 4.29 7.26 5.06 4.70 5.01 5.40

Note: In a univariate ANOVA, the effect of the main technical field turned out to be highly significant (F = 42.03, p = 0.000).

Table 4.6: Means of compensation for this patent, compensation for all patents, number of inventors, number of citations received, and size of the patent family by main technological field (N = 1,983)

To learn more about the motivating or discouraging effect of the German Employees’ Inventions Act, the inventors were asked to give their opinion concerning the underlying legal regulations. The answers were divided into three groups according to their attitude towards legal regulations. Figure 4.3 shows that 59.5% of the inventors believe the Employees’ Invention Act to be largely motivating, whereas 28.3% assume a negative effect on their motivation. The remainder, a group of 12.2% of the surveyed inventors, do not attach much importance to the legal regulations concerning their inventive performance.

12.2%

N = 1,546 positive effect negative effect

28.3% 59.5%

neutral effect

Figure 4.3: Effect of the German Employees’ Inventions Act on incentives for innovation

147

The first group contains 920 inventors, assigning an overall positive effect to the legal regulations. Figure 4.4 shows frequencies of the incentive drivers mentioned by the inventors in the first group. Financial incentives turn out to be second to none, mentioned by 57.2% of the sub-sample. The advantage of well-defined legal provisions (18.0%) and the acknowledgement of inventive performance (16.6%) range far behind, in the second and third place. Also important for the employee-inventors is the employer’s support concerning formalities of the patent application (3.6%) as well as the absorption of costs and risks by the employer (9.3%).

N = 920 18.0%

legal regulation 2.3%

security of employment

57.2%

financial incentive 16.6%

acknowledgement 1.6%

chances for advancement

4.7%

contribution to firm performance

3.0%

realization of the invention

9.3%

expenses and risk taken by employer

3.6%

application filed by employer no co-inventorship of superiors

0.1% 0%

10%

20%

30%

40%

50%

60%

70%

80%

Figure 4.4: Incentives emerging from the German Employees’ Inventive Act (sub-sample of inventors who assign a positive effect to motivation due to the legal regulations)

Inventors who regard the German legislation concerning employee inventions as discouraging for the invention process represent the second sub-sample (n2 = 437). Figure 4.5 displays that “compensation too small” is the most frequently mentioned disincentive (33.6%). One third of the inventors in the second sub-sample consider compensation as too low, compared to their inventive performance. Almost one third (32.0%) complain about the lack of transparency concerning the determination of an appropriate compensation and about the intense influence capability of the employer on its calculation. 15.3% of the respondents mention delays in the payment of compensation (or even no payment) as non-satisfying.

148

33.6%

compensation too small

N = 437

15.3%

delayed payment 0.7%

team inv. decreases compensation

6.2%

tax regulations

8.7%

burden of administration 1.1%

unequal treatm. of internat. inventors

32.0%

lack of transparency 6.4%

lack of influence on exploitation strategy

3,0%

lack of information conflicts with the employer

5.9%

conflicts with colleagues

5.7% 2.1%

danger of loosing comp.

0.7%

inventor not mentioned

1.8%

co-inventorship of superiors 0%

10%

20%

30%

40%

50%

60%

70%

80%

Figure 4.5: Disincentives emerging from the German Employees’ Inventive Act (sub-sample of inventors who assign a negative effect to motivation due to the legal regulations)

Due to a decision of the Federal Court of Justice in November 198977, tax benefits for employee-inventor compensations have been cancelled. Therefore, it is not surprising that 6% of the second sub-sample complain about tax regulations. 9% complain about the additional burden of administration necessitated. Also mentioned by the inventors are conflicts with the employer (6%) as well as conflicts between inventors among themselves (6%). The particular problem concerning inventor - employer-conflicts is that inventors do not want to jeopardize their careers by contesting their inventor awards in court or by otherwise turning against their employers. Inventors come into conflicts with colleagues due to enviousness, resulting in an impairment of team work as well as in an interference of communication between colleagues. Results even show that inventors hinder a sequential or substitutional invention not deriving from them, in order not to loose the compensation granted for their earlier invention (2%).

77

Bundesverfassungsgerichts-Beschluss vom 29.11.1989 (1 BvR 1402/87, 1 BvR 1528/87) BStBl. 1990 II p. 479.

149

Finally, the inventors reported a phenomenon, already observed by Staudt et al. (1992): the co-inventorship of superiors (1.8%). Superiors are mentioned as a co-inventor, not due to their inventive performance or participation in the inventive process, but due to their position within the firm. Given the notoriety that this phenomenon has received in the literature, our results suggest that its importance may have been overstated considerably.

4.6 Multivariate Analysis Our multivariate analysis proceeds in two steps. First, we try to determine how our variables are related to the (presumably) most important determinant of inventor compensation – the patent’s value. We use the ordinal information from our survey (see Table 4.4) and employ an ordered probit framework for the analysis. Our second step – the analysis of the compensation share variable – also treats the data as ordinal. We observe considerable bunching around particular integer values (0, 1, 2, 5, 10, 15, ...) in our data so that a transformation to an ordinal scale appears appropriate.78 The first part of the analysis confirms earlier results which suggest that the value of patents is highly correlated with a number of indicator variables. We consider the results in Table 5.7, column (3) for the overall value specification first. Citations, legal challenges (opposition) and the size of the patent family are (as expected) positively associated with patent value. Somewhat unexpectedly, two other R&D process variables turn out to have a significant impact. First, patented inventions that are the planned product of R&D projects are more valuable than unplanned results or mere by-products of R&D. This result may reflect a selection effect – firms will actively try to develop ideas in R&D projects, if they expect the project to yield valuable results. This interpretation is strengthened by another result – the more inventors are involved in the invention, the more valuable it tends to be. Again, choosing relatively large teams is likely to reflect a company’s assessment that it should try to achieve the invention quickly – presumably, because it is a valuable invention. The surprise lies in the second R&D process variable with a positive coefficient – inventions made during the inventor’s leisure time are considerably more valuable than other types of inventions. This result may reflect two very different phenomena – first, taking the positive coefficient at face value, it may indeed be the case that leisure time provides the optimal environment for creative break-through. On the other hand, the result may involve strategic

78

However, it turns out that results from a Tobit-type analysis with a metric dependent variable are quite similar to the ones described here.

150

behavior on the part of inventors who wish to enhance their contribution to the inventive process. Social desirability may play a big role in generating this result, and we will investigate it in more detail in the future. A final comment on the value regression concerns the technical field dummy variables. In column (3), they do not contribute jointly any more to the explanation of patent value. This appears to be due to the inclusion of the R&D context variables in column (3). We now turn to the inventor compensation regressions in columns (4), (5) and (6). Our expectation is that the results should reflect strongly the legal provisions of the German Employees’ Inventions Act. Indeed, in all specifications, the dominant determinant of compensation is the patent’s value. The coefficients of the dummy variables are increasing as the value of the patent increases, and they are highly significant throughout. Moreover, the results are very stable as we include more variables. The number of inventors has the expected negative coefficient which is again highly significant in all specifications. Interestingly, inventor productivity and educational attainment carry a negative sign. This result is consistent with the view that these variables proxy for the inventor’s rank in the organization which should be negatively associated with the level of compensation for service inventions. Firm size and the technical field to which the invention belongs have no impact whatsoever. Moreover, the value correlates opposition, citations and family size do not have any impact, nor do any of the other variables in that group. Apparently, the inclusion of the value dummy variables leaves little explanatory power for these variables. Similarly, the variables describing the context of the invention have no explanatory power.

151

ORDERED PROBIT ON PATENT VALUE (1) (2) (3)

ORDERED PROBIT ON INVENTOR COMPENSATION (4) (5) (6)

Patent value: 30,000 - 100,000 Euro

0.0843** (0.0405) -0.0035 (0.0110) -0.1684** (0.0849) -0.1194 (0.0948) -0.0901 (0.0774)

0.0773* (0.0410) -0.0023 (0.0111) -0.1620* (0.0872) -0.1151 (0.0947) -0.0950 (0.0782)

0.4247*** (0.1522) 0.6332*** (0.1488) 0.7554*** (0.1465) 0.8835*** (0.1551) 1.1410*** (0.1662) 1.4803*** (0.1779) -0.2000*** (0.0542) -0.0109 (0.0141) -0.0905 (0.1163) 0.0511 (0.1240) 0.0284 (0.1056)

-0.0842 (0.0795)

-0.0796 (0.0800)

0.0597 (0.0993)

0.0363 (0.0999)

0.0494 (0.1014)

-0.1476 (0.1031)

-0.1649 (0.1044)

0.0669 (0.1443)

0.0598 (0.1440)

0.0645 (0.1454)

0.1097** (0.0536)

0.1026* (0.0537)

-0.3797*** -0.3759*** -0.3815*** (0.0802) (0.0804) (0.0805)

0.0086 (0.0742)

0.0145 (0.0745)

-0.2234** (0.0994)

100,000 - 300,000 Euro 300,000 - 1 Mio. Euro 1 - 3 Mio. Euro 3 - 10 Mio. Euro more than 10 Mio. Euro ln (number of inventors)

0.1222*** (0.0405) ln (number of employees) -0.0070 (0.0108) electricity/electronics -0.2562*** (0.0827) instruments -0.1947** (0.0931) process engineering -0.1540** (0.0764) mechanical engineering/machinery -0.1617** (0.0778) consumer goods/civil engineering -0.2021** (0.1009) ln (1+inventor productivity) 0.1089** (0.0540) vocational academy/ university studies 0.0133 (0.0735) doctoral/postdoctoral studies 0.0708 (0.0844) ln (age of the inventor) 0.2154 (0.1334) Log Likelihood -3620.573 Pseudo R-squared 0.0068 Chi-squared (df) 47.75 (11) Observations 1983

0.0384 0.0344 -0.5049*** (0.0847) (0.0857) (0.1187) 0.2808** 0.2376* 0.1401 (0.1389) (0.1417) (0.1779) -3591.251 -3581.022 -1550.377 0.0149 0.0177 0.0580 100.61 (20) 115.06 (26) 177.40 (17) 1983 1983 1983

0.4201*** (0.1528) 0.6214*** (0.1487) 0.7532*** (0.1466) 0.8864*** (0.1552) 1.1350*** (0.1663) 1.4957*** (0.1803) -0.1877*** (0.0545) -0.0120 (0.0141) -0.1159 (0.1176) 0.0303 (0.1260) 0.0173 (0.1053)

0.4332*** (0.1522) 0.6282*** (0.1485) 0.7668*** (0.1465) 0.8995*** (0.1551) 1.1401*** (0.1671) 1.4945*** (0.1802) -0.1971*** (0.0555) -0.0144 (0.0142) -0.1257 (0.1208) 0.0351 (0.1280) 0.0376 (0.1059)

-0.2277** (0.0999)

-0.2367** (0.1006)

-0.4965*** (0.1192) 0.1288 (0.1813) -1547.776 0.0596 184.34 (26) 1983

-0.5109*** (0.1211) 0.1386 (0.1833) -1545.692 0.0609 190.23 (32) 1983

Robust standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%

Table 4.7: Multivariate analysis of patent value and inventor compensation (N = 1,983)

152

measure of originality measure of generality oppositions received PCT application filed

ORDERED PROBIT ON PATENT VALUE (1) (2) (3) -0.0128 -0.0008 (0.1626) (0.1649) -0.5109 -0.5535 (0.4055) (0.4060) 0.2473*** 0.2448*** (0.0724) (0.0730) -0.0135 -0.0071 (0.0584) (0.0589)

citations received within 5 years size of patent family changed employer once changed employer twice changed employer more than twice

0.0665*** (0.0223) 0.0346*** (0.0073) 0.1018 (0.0714) 0.1951 (0.1239)

0.0669*** (0.0221) 0.0334*** (0.0074) 0.0919 (0.0713) 0.1814 (0.1253)

0.0005 (0.0294) -0.0098 (0.0094) -0.0214 (0.1076) -0.0538 (0.1565)

0.0014 (0.0295) -0.0098 (0.0094) -0.0187 (0.1085) -0.0522 (0.1566)

-0.2275 (0.2005)

-0.2277 (0.1978)

-0.4272 (0.3171)

-0.4191 (0.3192)

city with more than 1 mio. inhabitants city with 500.000 to 1 mio. inhabitants R&D project, planned result R&D project, expected by-product R&D project, unexpected by-product invention arose during leisure time Log Likelihood Pseudo R-squared Chi-squared (df) Observations

-3620.573 0.0068 47.75 (11) 1983

ORDERED PROBIT ON INVENTOR COMPENSATION (4) (5) (6) -0.2514 -0.2304 (0.2388) (0.2383) -0.2370 -0.2330 (0.5036) (0.5012) -0.0092 -0.0091 (0.1062) (0.1065) 0.0574 0.0634 (0.0779) (0.0781)

0.0449 (0.0834)

0.0919 (0.1065)

0.0076 (0.0752)

0.0419 (0.0947)

0.1262** (0.0634)

0.1068 (0.0833)

0.0118 (0.0687)

-0.0349 (0.0948)

-0.1359* (0.0734)

0.0866 (0.0967)

0.2773*** -0.0480 (0.1037) (0.1613) -3591.251 -3581.022 -1550.377 -1547.776 -1545.692 0.0149 0.0177 0.0580 0.0596 0.0609 100.61 (20) 115.06 (26) 177.40 (17) 184.34 (26) 190.23 (32) 1983 1983 1983 1983 1983

Robust standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%

Table 4.7 continued: Multivariate analysis of patent value and inventor compensation (N = 1,983)

The remarkable picture emerging from the compensation regression is that those variables which should have an impact due to the text of the law do indeed have an impact. While that 153

was expected, it seems remarkable that technical field and other variables cannot develop any additional explanatory power. It seems that on average, the GEIA is indeed applied fairly consistently across different industries and technical fields. This statement leaves considerable room for deviations from the average, but it is highly unusual that a set of control variables such as those for the technical field of the invention does not have any statistical role in the regression.

4.7 Conclusions This paper has discussed in some detail differences in national legal provisions dealing with the compensation that inventors are entitled to for their service inventions. Germany assumes an unusual role in this comparison, since inventor compensation is regulated to a level of detail that is not found in other countries. The extent of compensation can be considerable. In our sample, the inventors report that they receive on average about 8 percent of their gross salaries as compensation for service inventions. Our multivariate analysis yields the surprising result that the law appears to be applied very consistently across different technical fields. We find that the patent’s value, the number of inventors and variables associated with the inventor’s position in the company have the expected impact. Moreover, by comparing these results to those of a value regression, we can assure ourselves that the lack of explanatory power of other variables is not due to measurement problems. Taken together, there is reason to believe that inventor compensation is largely a meritocratic system. The qualitative results from our survey confirm that view to some degree. The majority of inventors view the compensation system positively. Yet, there appear to be areas in which an improvement or reform is necessary. We will consider these areas in more detail in subsequent research.

154

Chapter 5 5

Summary of the Results and Outlook

This dissertation thesis analyzed three key aspects of inventive activity: inventor productivity, the mobility of inventors, and incentives to invent. The thesis provides new results for the three issues separately, but it also improves on the current literature by pointing out connections between these topics. The literature on productivity of scientists and engineers reveals that productivity is highly concentrated within a small number of researchers. A motivation system that treats all inventors equally, therefore leads to a decrease in the motivation of the key inventors. To design an efficient incentive system, firms have to identify the most productive inventors. Therefore, the purpose of the first paper was to create an appropriate productivity measure that controls for different effects, e.g., an increasing patenting activity over time and across different technical fields and for organizational differences of large firms compared to smaller firms. To reach this goal, four different measures of inventive activity were compared: whole patent counts, fractional patent counts, citation counts, and fractional citation counts. Results revealed that patent counts are predominantly determined by decisions made by the firm. R&D management, for instance, holds authority over the research projects and decides whether to file a patent application for an invention or not. Consequently, patent counts are a rather poor productivity measure. Citation counts, in contrast, turned out to be a more appropriate measure for inventor productivity, since citation counts are less affected by characteristics of the applicant. The multivariate analysis with respect to output quality (measured as citation counts) showed that the level of education had a strong influence on inventive output. Inventors whose highest educational level was a university degree or a doctoral degree received more citations. As expected, making use of external sources of knowledge also increased the output quality. Finally, the environment of the inventors had an important impact on inventive output. Whereas earlier research looked either at inventor related determinants or at environment related determinants of productivity, this study is the first to integrate both types of determinants. 155

Furthermore, the first paper aimed at providing a deeper insight into the relationship between age and inventor productivity. Making use of the time structure of the data it was possible to trace inventor productivity over time. The results of a fixed effects panel regression estimation revealed that there is a strong relationship between age and performance. For longterm inventors who were observable for five or six five-year periods the relationship between productivity and age is inverted u-shaped and has its maximum at an age of about 45 years. Controlling for an increasing number of citations over time leads to a downward correction of the inventors’ productivity. Additionally, the turning point of productivity shifts to the age of 30 to 34 years. Beyond, the analysis provides first evidence that capable inventors are promoted into management positions and, hence, kept from inventing. This has important implications for R&D management. Assuming a maximum productivity of an inventor at the age of about 35 to 40, firms weaken their own competitive ability in the event that key inventors switch to management positions before that age. These results call for further research on inventors’ career paths within firms, especially on “dual ladder” career systems which provide career opportunities for engineers in R&D. The second analysis dealt with the causality between inventor productivity and inventor mobility. Whereas previous research on inventors implicitly assumed causality in one direction, this is the first study of inventors which ex-ante allowed for a simultaneous relationship. The results also showed that the level of education had no influence on output quantity. Making use of external sources of knowledge increases productivity. Finally, firm size has a positive impact on productivity. Firm size also influences inventor mobility, although negatively. Furthermore, the temporal concentration of inventive activity, and the inventive environment are major determinants of mobility. Whereas the number of moves decreases with the duration of inventive activity, it is higher in large cities compared to rural areas. Productivity and mobility indeed turned out to be simultaneously related to each other: whereas mobility increased productivity, an increase in productivity decreased the number of moves. The latter effect may arise due to non-compete agreements between employer and employee or due to firms constituting efficient incentive systems for their inventors. The finding that mobile inventors are more productive than immobile inventors has important policy implications. First of all, the labor market literature proposes that moves occur to find a better employer-employee match. Better match quality finally leads to a higher productivity of the employees (Topel/Ward 1992). One could also argue that a change of the employer leads to an increase in experience or to knowledge spillovers from colleagues (Song et al. 2003). Especially the latter argument has an important implication for the educational system of engineers and scientists. To enable knowledge transfer and gain in experience of the

156

individual employee, the educational system should foster mobility at an early stage of the employees’ career, e.g., by promoting temporal stays abroad during university studies. The additional quasi-experiment confirmed a positive effect of a particular move on output quality. Over time the differences caused by the move decrease and finally disappear. A possible explanation is that the inventors make special efforts during their first years to impress their employer or to gain respect from their colleagues. Another explanation may be that inventors leave R&D for another job, e.g., in sales. Intra-firm mobility makes the inventor invisible in terms of patents and leads to a decrease in his mean output over time. A dummy regression controlling for multiple movers finally revealed that the effect size of the multiple move dummy is quite large. Possibly, multiple movers collect “move-experience” which enables them to settle in faster. Thus, also the importance of gain in experience due to mobility - which was already discussed in the introduction - proved true. Consequently, R&D personnel should be offered an environment that promotes mobility, for instance, again by providing the opportunity of temporal stays abroad or by interchanges with co-operation partners. Based on the results of the first two papers, another issue has to be considered which has already been mentioned in the introduction: patent documents provide an important source of information for firms to identify valuable patents and also to identify high performing inventors. The number of patents an inventor is responsible for and the number of citations the inventor’s patents received from subsequent patents are a good proxy for the productivity of an inventor. Reliable citation counts are only available after five to ten years after the application date which makes them unattractive for the labor market. By contrast, patents are published 18 months after the priority date which turns them into a valuable signal for ingenuity. Since patent applications are published in publicly available databases, information on inventors is actually available at low costs. From the point of view of a firm this “open job market” poses severe threats to loose key inventors who received a job offer from a competitor. Firms would rather like to keep information on inventors secret. However, due to legal regulations this is not possible. Consequently, firms have to undertake special efforts, e.g., they have to provide appropriate motivation and incentive systems, to increase the commitment of important inventors to the firm. Inventors probably take advantage of this legal regulation since they receive a compensation for their merits. On the part of the national economy, this “open job market” has the advantage of promoting job mobility, leading to a better match quality between the employee and the new employer. A better match quality in turn leads to a higher productivity of the employees and consequently to an increase of social welfare. Finally, to learn more about the importance of incentive systems for inventive activity, the last paper aimed at providing a deeper insight into inventor compensation related issues, in particular, the features of the German Employees’ Invention Act were discussed. Relevant 157

literature revealed that one of the most important motives of inventing is financial gain (Macdonald 1984). Germany is one of the few countries in which the monetary compensation for inventors is not only determined by negotiations between employer and employee inventor, but also by relatively precise legal provisions, i.e., the Employees’ Invention Act. Results showed that almost two-thirds of the inventors believed the Employees’ Invention Act to be largely motivating, whereas almost one third assumed a negative effect on their motivation. The financial incentives created by the Employees’ Invention Act were mentioned most frequently. The advantage of well-defined legal provisions and the acknowledgement of inventive performance also turned out to be important for the inventors’ motivation. Almost one-third complained about the lack of transparency concerning the determination of an appropriate compensation and about the potentially powerful influence of the employer on its calculation. “Compensation is too small” was also one of the most frequently mentioned disincentives. Overall, transparency and fairness of the compensation seemed to be important determinants of an efficient motivation and incentive system for inventors. This means that the GEIA is in principle appropriate to compensate inventors for their merits and to work as an efficient incentive system. Since results show that inventors are to some extent dissatisfied with the implementation of the legal regulations within their firm there is still room for improvement in firms to exploit the full capacity of the GEIA to motivate inventors. Apart from these findings, this research also opens up interesting opportunities for future studies. In particular, the life cycle of inventors or of inventive activity should be analyzed in more detail. The data underlying the first and the second study described above possess a time series structure which had in part been neglected. The temporal information contained in the data can be used more extensively to trace the variation in productivity or overall inventive activity of researchers over time. Furthermore, intra-firm mobility should be analyzed. In particular, the shift from R&D to other jobs such as, for instance, administration should be discussed. Further research should also address multiple movements. Possibly, “moveexperience” impacts the performance of an inventor starting a new job. Personal interviews with inventors will help to shed more light into these questions. Additionally, further research aims at finding better instruments for mobility to use in the Two-Stage Least Squares (2SLS) regression estimation. In particular, moves that are determined by exogenous variables, for instance, the insolvency of a firm should improve the results of the 2SLS regression estimation. Furthermore, the labor market literature points at the importance of the employee’s wage for the probability to observe a move (see, e.g., Jovanovic 1979, McDonald 1988). In particular, workers who earn higher wages are less likely to quit (McDonald 1988). To improve the results, at least a proxy for the wages of the inventors is needed to control for a varying wage in the regression model.

158

Finally, inventor teams have to be examined more closely. Knowledge transfer within teams should also have an impact on productivity, as well as on mobility. In particular, research collaboration resulting in cross-firm teams should impact the mobility of inventors. A research project financed by the German Research Foundation (DFG) will allow for extending the mobility project to all inventors listed on EP patent documents. Within the next months a matching procedure will be developed applicable to match inventors listed on approximately 1,600,000 EP patents. The matched inventor data can be used to analyze inventor team performance or to compare inventor productivity and mobility between different industries or countries. Additionally, research can be extended to inter-regional mobility as well as to cross-country mobility. The overall objective of this research is to contribute to a better understanding of the inventor mobility phenomenon to contribute to improvements in the management of human resources in the research and development functions of firms.

159

Bibliography Abadie, A. (2003). Semiparametric Difference-in-Differences Estimators, NBER and Harvard University Working Paper, Cambridge, Mass. Allen, T. J. (1977). Managing the Flow of Technology. Cambridge, Mass.: MIT University Press. Allen, T.J. / Katz, R. (1985). The Dual Ladder: Motivational Solution or Managerial Delusion?, MIT and Sloan Management School Working Paper No. 1692-85, Boston. Allison, P.D. / Stewart, K.A. (1974). Productivity Differences among Scientists. Evidence for Accumulative Advantage, American Sociological Review 39: 596-606. Althauser, R.P. (1989). Internal Labor Markets, Annual Review of Sociology 15: 143-161. Amesse, F. / Desranleau, C. / Etemad, H. / Fortier, Y. / Seguin-Dulude, L. (1991). The Individual Inventor and the Role of Entrepreneurship: A Survey of the Canadian Evidence, Research Policy 20 (1): 13-27. Arora, A. / David, P.A. / Gambardella, A. (1998). Reputation and Competence in Publicly Funded Science: Estimating the Effects on Research Group Productivity, Annales D’Économie et de Statistique 49/50: 163-198. Arrow, K. J. (1962). Economic welfare and the allocation of resources for invention, in: R. R. Nelson (Ed.): The Rate and Direction of Inventive Activity: Economic and Social Factors, Vol. 13 of NBER Special Conference Series, Princeton University Press, New Jersey: 609-625. Barney, J.B. (1986). Strategic Factor Markets: Expectations, Luck, and Business Strategy, in: Management Science 42: 1231-1241. Barney, J.B. (1991). Firm Resources and Sustained Competitive Advantage, in: Journal of Management 17 (1): 99-120.

161

Becker, G.S. (1962). Investment in Human Capital: A Theoretical Analysis, The Journal of Political Economy 70 (5): 9-49. Becker, G.S. (1964). Human Capital, New York, NBER. Blind, K. / Edler, J. / Frietsch, R. / Schmoch, U. (2003). Erfindungen kontra Patente – Schwerpunktstudie

zur

technologischen

Leistungsfähigkeit

Deutschlands,

Fraunhofer Institut für Systemtechnik und Innovationsforschung, Endbericht, Karlsruhe, December 2003. Bound, J. / Cummins, C. / Griliches, Z. / Hall, B.H. / Jaffe, A.B. (1984). Who Does R&D and Who Patents?, in: Z. Griliches (Ed.): R&D, Patents and Productivity, Chicago, London: 21-54. Breusch, T.S. / Pagan, A.R. (1979). A Simple Test for Heteroscedasticity and Random Coefficient Variation, Econometrica 50: 987-1007. Brockhoff, K. (1997). Ist die kollektive Regelung einer Vergütung von Arbeitnehmererfindungen wirksam und nötig?, Zeitschrift für Betriebswirtschaft (ZfB) 67 (7): 677687. Brockhoff, K. (1998). Patentierung von Hochschullehrererfindungen, in: N. Franke, C.-F. von Braun (Eds.): Innovationsforschung und Technologiemanagement. Konzepte, Strategien, Fallbeispiele, Berlin: 49-62. Bruland, K. / Mowery, D. (2004). Innovation Through Time, UC Berkeley und NBER Discussion Paper, Berkeley. Card, D. / Krueger, A.B. (1994). Minimum Wages and Employment: A Case Study of the Fast Food Industry in New Jersey and Pennsylvania, American Economic Review 84: 772-793. Cohen, W. M. / Levinthal, D.A. (1989). Innovation and Learning: The Two Faces of R&D, The Economic Journal 99 (397): 569-596. Cohen, W. M. / Levinthal, D.A. (1990). Abvsorptive Capacity: A New Perspective on Learning and Innovation, Administrative Science Quarterly 35: 128-152. Cohen, W.M. / Nelson, R.R. / Walsh, J.P. (2000). Protecting Their Intellectual Assets: Appropriability Conditions and why U.S. Manufacturing Firms Patent (or Not), NBER Working Paper No. 7552, Cambridge, M.A.. 162

Cole, S. (1979). Age and Scientific Performance, American Journal of Sociology 84: 958-977. Cole, J.R. / Cole S. (1973). Social Stratification in Science. Chicago University Press. Cole, J.R. / Cole S. (1967). Scientific Output and Recognition: A Study in the Operation in the Reward System in Science, American Sociological Review 32 (6): 377-390. Cook, T.D. / Campbell, D.T. (1979). Quasi-Experimentation – Design & Analysis Issues for Field Settings, Houghton Mifflin Company. Cook, R.D. / Weisberg, S. (1983). Diagnostics for Heteroscedasticity in Regression, Biometrika 70: 1-10. Dalton, G.W. / Thompson, P.H. (1971). Accelerating Obsolescence of Older Engineers, Harvard Business Review 49: 57-67. Dahlin, K. / Taylor, M. / Fichman, M. (2004). Today’s Edisons or Weekend Hobbyists: Technical Merit and Success of Inventions by Independent Inventors, Research Policy 33: 1167-1183. Darwin, C. (1859). On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life, London. David, P.A. (1994). Positive Feedbacks and Research Productivity in Science: Reopening Another Black Box”, in: Granstrand, O. (Ed.): Economics of Technology, Amterdam. Denison, E.F. (1964). Measuring the Contribution of Education, The Residual Factor and Economic Growth, OECD: 13-55. De Solla Price, D.J. (1965). Networks of Scientific Papers, Science 149(3683): 510-515. De Solla Price, D.J. (1976). A General Theory of Bibliometric and other Cumulative Advantage Processes, Journal of the American Society for Information Science 27 (5): 292-306. Deyle, H.-G. / Grupp, H. (2005). Commuters and the regional assignment of innovative activities: A methodological patent study of German districts, Research Policy 34 (2): 221-234. Doeringer, P.B. / Piore, M. (1971). Internal Labor Markets and Manpower Analysis. Lexington. 163

Dosi, G. (1988). Technological Change and Economic Theory. Printer Publishers, New York. DPMA (2002). Patents - An information brochure of the German Patent and Trade Mark Office, Berlin. Ernst, H. / Leptien, C. / Vitt, J. (2000). Inventors are Not Alike: The Distribution of Patenting Output Among Industrial R&D Personnel, IEEE Transactions on Engineering Management 47 (2): 184-199. Evers-Koelman, I. / Fischer, M. / Nijkamp, P. (1988). Results of Cross-National Comparisons of Regional Labour Markets in 15 countries, in: Fischer, M. / Nijkamp, P. (Eds.): Regional Labour Markets, Amsterdam. Fleming, L. / Marx, M. (2005). Non-Competes and Regional Inventor Mobility, Discussion Paper, Harvard Business School, Boston. Foray, Dominique (2004). The Economics of Knowledge, Cambridge: MIT Press. Freeman, C. (1982). The Economics of Industrial Innovation, 2nd Edition, Cambridge, Mass.. Freeman, C. (1991). Networks of Innovators: A Synthesis of Research Issues, Research Policy 20 (5): 499-514. Freeman, R.B. (1978). Job Satisfaction as an Economic Variable, The American Economic Review 68 (2): 135-141. Gaul, D. (1988). Der erfassbare betriebliche Nutzen als Grundlage für Erfindungsberechnung, GRUR 4: 254-264. German Patent and Trademark Office (GPTO) (2003). Annual Report 2002, Munich. Gersbach, H. / Schmutzler, A. (2003). Endogenous Technical Spillovers: Causes and Consequences, Journal of Economics and Management Strategy 12 (2): 179-205. Gilfillan, S.C. (1935). The Sociology of Invention, Chicago. Giummo, J. (2003). Should All Patentable Inventions Receive Equal Protection? Identifying the Sources of Heterogeneity in Patent Value, Discussion Paper, University of California, Berkeley.

164

Giuri P. / Mariani M. / Brusoni S. / Crespi G. / Francoz D. / Gambardella A. / Garcia-Fontes W. / Geuna A. / Gonzales R. / Harhoff D. / Hoisl K. / Lebas C. / Luzzi A. / Magazzini L. / Nesta L. / Nomaler O. / Palomeras N. / Patel P. / Romanelli M. / Verspagen B. (2005). Everything you Always Wanted to Know About Inventors (But Never Asked): Evidence from the PatVal-EU Survey, LEM Working Paper No. 2005/20, Sant'Anna School of Advanced Studies, Pisa.. Globe, S. / Levy, G. / Schwartz, C. (1973). Key Factors and Events in the Innovation Process, Research Management 16 (4): 8-15. Goldberg, A.I. / Shenhav, Y.A. (1984). R&D Career Paths: Their Relation to Work Goals and Productivity, IEEE Transactions on Engineering Management 31 (3): 111-117. Granovetter, M. (1974). Getting a Job: A Study of Contacts and Careers, Cambridge, Mass. Granovetter, M. (1983). The Strength of Weak Ties: A Network Theory Revisited, Sociological Theory, 1: 201-233. Green, W. (1997). Econometric Analysis, Third Edition, London u.a. Green, W. (2001). Estimating Econometric Models with Fixed Effects, Stern School of Business, New York University Working Paper. Griliches, Z. (1970). Notes on the Role of Education in Production Functions and Growth Accounting, L. Hansen (Ed.): Education, Income, and Human Capital, Cambridge, NBER: 71-115. Griliches, Z. (1990). Patent Statistics as Economic Indicators: A Survey, Journal of Economic Literature 28: 1661-1707. Griliches, Z. / Pakes, A. / Hall, B.H. (1987). The Value of Patents as Indicatiors of Inventive Activity, in: P. Dasgupta and P. Stoneman (Eds.): Economic Policy and Technical Performance, Cambridge, Mass.: 97-124. Groot, W. / Verberne, M. (1997). Aging, Job Mobility, and Compensation, Oxford Economic Papers 49 (3): 380-403. Guilford, J.P. (1968). Intelligence Has Three Facets, Science 160 (3828): 615-620. Grossman, S. / Hart, O. (1983). An Analysis of the Principal-Agent Problem, Econometrica 51: 7-45. 165

Hagstrom, W.O. (1968). Departmental Prestige and Scientific Productivity. Discussion Paper presented at the 63rd Annual Meeting of the American Sociological Association, Lexington Books. Hall, B.H. (2004). Exploring the Patent Explosion, NBER Working Paper No. 10605, Cambridge, Mass. Hall, B.H. / Mairesse, J. / Turner, L. (2005). Identifying Age, Cohort and Period Effects in Scientific Research Productivity: Discussion and Illustration Using Simulated and Actual Data on French Physicist, paper presented at the Keith Pavitt Memorial Conference, SPRU, November 2003, revised June 2005. Harhoff, D. (2006). Patent Quantity and Quality: Trends and Policy Implications, in: Kahin, B. / Foray, D. (Eds.): Advancing Knowledge and the Knowledge Economy, Cambridge, Mass. Harhoff, D. (2005). Innovationen und Wettbewerbspolitik – Ansätze zur ökonomischen Analyse des Patentsystems, in: Monopolkommission (Ed.): Zukunft der Wettbewerbspolitik, Baden-Baden: Nomos (forthcoming). Harhoff, D. / Hall, B.H. (2003). Intellectual Property Strategy in the Global Cosmetics Industry, unpublished manuscript, University of Munich. Harhoff, D. / Hoisl, K. (2005). Institutionalized Incentives for Ingenuity – Patent Value and the German Employees’ Inventions Act, Working Paper, Munich. Harhoff, D. / Kane, T.J. (1997). Is the German Apprenticeship System a Panacea for the U.S. Labor Market? Journal of Population Economics 10: 171-196. Harhoff, D. / Reitzig, M. (2001). Strategien zur Gewinnmaximierung bei der Anmeldung von Patenten, Zeitschrift für Betriebswirtschaft (ZfB) 71 (5): 509-529. Harhoff, D. / Scherer, F.M. / Vopel, K. (1999). Citations, Family Size, Opposition and the Value of Patent Rights, Discussion Paper No. 97-27, Zentrum für Europäische Wirtschaftsforschung (ZEW), Mannheim. Harhoff, D. / Wagner, S. (2005). Modeling the Duration of Patent Examination at the European Patent Office, CEPR Discussion Paper No. 5283, London. Hausman, J.A. (1978). Specification Tests in Econometrics, Econometrica 46: 1251-1271.

166

Heimbach, K. J. (1992). Law and Practice of Employee Inventions in Germany, WIPO Symposium of Employee Inventions, Bucharest: 43-54. Idson, T.L. / Oi, W.Y. (1999). Workers are More Productive in Large Firms, American Economic Review 89 (2): 104-108. Jaffe, A.B. (1989). Real Effects of Academic Research, in American Academic Review 79 (5): 957-970. Johnson, J.M. / Regets, M. (1998). International Mobility of Scientists and Engineers to the US – Brain Drain or Brain Circulation? NSF Issue Brief 98-316, June, 1998. Jones, B.F. (2005). Age and Great Invention, NBER Working Paper No. 11359, Cambridge, M.A.. Jovanovic, B. (1979). Job-Matching and the Theory of Turnover, Journal of Political Economy 87 (5): 972-990. Kesten, R. (1996). Innovationen durch eigene Mitarbeiter – Betriebswirtschaftliche Aspekte zur monetären Beurteilung von Diensterfindungen nach dem Gesetz für Arbeitnehmererfindungen, Zeitschrift für Betriebswirtschaft (ZfB) 66 (6): 651-673. Kim, J. / Lee, S.J. / Marschke, G. (2004). Research Scientist Productivity and Firm Size: Evidence from Panel Data on Inventors, Working Paper. Kim, J. / Marschke, G. (2005). Labor Mobility of Scientists, Technological Diffusion, and the Firm’s Patenting Decision, Rand Journal of Economics 36 (2): 298-317. Kline, R.C. (1992). Law and Practice of Employment Inventions in the United States of America, WIPO Symposium of Employee Inventions, Bucharest: 65-111. Krueger, A.B. (1990). Workers’ Compensation Insurance and the Duration of Workplace Injuries, NBER Working Paper No. 3253, Cambridge, Mass. Kurz, P. (1997). Geschichte des Arbeitnehmererfindungsrechts, Aachen. Kwok, V. / Leland, H. (1982). An Economic Model of the Brain Drain, The American Economic Review 72 (1): 91-100.

167

Lanjouw, J.O. / Pakes, A. / Putnam, J. (1998). How to Count Patents and Value Intellectual Property. The Use of Patent and Renewal and Application Data, The Journal of Industrial Economics 46 (4): 405-433. Lanjouw, J.O. / Schankerman, M. (1999). The Quality of Ideas. Measuring Innovation with Multiple Indicators, Working Paper No. 7345, NBER, Cambridge, Mass.. Lanjouw, J.O. / Schankerman, M. (2001). Characteristics of Patent Litigation. A Window on Competition, Rand Journal of Economics 32 (1): 129-151. Lanjouw, J.O. and Schankerman, M. (2004). Patent Quality and Research Productivity: Measuring Innovation with Multiple Indicators, Economic Journal 114 (495): 441-465. Lazear, E. (1979). Why is there Mandatory Retirement?, Journal of Political Economy 87: 1261-1284. Lehman, H.C. (1966). The Most Creative Years of Engineers and Other Technologists, Journal of Genetic Psychology 108: 263-277. Leptien, C. (1996). Anreizsysteme in Forschung und Entwicklung, Wiesbaden. Litter, D.A. / Pearson, A.W. (1979). Rewarding the Employee Inventor and the Patents Act (1977), R&D Management 10 (1): 29-32. Liu, P.-W. (1986). Human Capital, Job Matching and Earnings Growth Between Jobs: An Empirical Analysis, Applied Economics 18: 1135-1147. Los, B. / Verspagen, B. (2003). Technology Spillovers and Their Impact on Productivity, Working Paper. Lotka, A.J. (1926). The Frequency Distribution of Scientific Productivity, Journal of the Washington Academy of Science 16 (2): 317-323. MacDonald, G.M. (1988). Job Mobility in Market Equilibrium, Review of Economic Studies 55 (1): 153-168. MacDonald, S. (1984). The Patent System and the Individual Inventor, The Inventor 24 (1): 25-29.

168

Machlup, F. (1958). An Economic Review of the Patent System. Study of the Subcommittee on Patents, Trademarks, and Copyrights of the Committee on the Judiciary, U.S. Senate, study no. 15. Manly, D.G. (1978). Inventors, Innovators, Compensation and the Law, Research Management 21: 29-32. Mansfield, E. (1986). Patents and Innovation: An Empirical Study, Management Science 32 (2): 173-181. Marshall, A. (1964). Principles of Economics, London. Merges, R.P. (1999). The Law and Economics of Employee Inventions, Harvard Journal of Law & Technology 13 (1): 1-53. Merton, R.K. (1968). The Matthew Effect in Science, Science 159 (January): 56-63. Meyer, B.D. (1995). Natural and Quasi-Experiments in Economics, Journal of Business & Economic Statistics 13 (2): 151-162. Montgomery, J.D. (1991). Social Networks and Labor-Market Outcomes: Toward an Economic Analysis, The American Economic Review 81 (5): 1408-1418. Moon, S. (2005). How Does the Organization of Research Impact Research Outcomes? Evidence from Scientific Publications and Patenting Behavior, Discussion Paper, Kellogg School of Management. Mowery, D.C. / Ziedonis, A.A. (2002). Academic Patent Quality and Quantity before and after the Bayh-Dole Act in the United States, in Research Policy 31: 399-418. Mullahy, J. / Sindelar, J.L. (1996). Employment, Unemployment, and Problem Drinking, Journal of Health Economics 15: 409-434. Narin, F. (2000). Tech-Line Background Paper, in: J. Tidd (Ed.): Measuring Strategic Competence, Technology Management Series, London. Narin, F. / Breitzman, A. (1995). Inventive Productivity, Research Policy 24(4): 507-519. Oberg, W. (1960). Age and Achievement – and the Technical Man, Personnel Psychology 12: 245-259.

169

OECD (1994). Using Patent Data as Science and Technological Indicators – Patent Manual 1994, Paris. Orkin, N. (1984). Rewarding Employee Invention. Time for Change, Harvard Business Review 62 (1): 56-57. Pakes, A. (1986). Patents as Options. Some Estimates of the Value of Holding European Patent Stocks, Econometrica 54 (4): 755-784. Parsons, D.O. (1972). Specific Human Capital: An Application to Quit Rates and Layoff Rates, The Journal of Political Economy 80 (6): 1120-1143. Pelz, D.C. / Andrews, F.M. (1966). Scientists in Organizations: Productive Climates for Research and Development, New York. Rebel, D. (1993). Handbuch Gewerbliche Schutzrechte – Übersichten und Strategien: Europa - USA - Japan, Wiesbaden. Reiersol, O. (1945). Confluence Analysis by Means of Instrumental Sets of Variables, Arkiv for Mathematik, Astronomi och Fysik 32: 1-119. Reitzle, H. / Butenschön, A. / Bergmann, J. (2000). Act on Employees’ Inventions, 2nd edition, Weinheim. Ritti, R.R. (1971). The Engineer in the Industrial Corporation, New York. Rossman, J. (1931). The Motives of Inventors, The Quarterly Journal of Economics 45 (3): 522-528. Ruud, P.A. (2000). An Introduction to Classical Econometric Theory, New York. Sargan, J.D. (1958). The Estimation of Economic Relationships Using Instrumental Variables, Econometrica 26 (3): 393-415. Sauter, S. (2000). Non-Compete Agreements and the Discharged Employee”, Employee Rights Quarterly 1 (2): 28-41. Savitsky, T.R. (1991). Compensation Practices for Employment Inventions, Journal of the Patent and Trademark Office Society, 9/91: 645-679.

170

Schankerman, M. / Pakes, A. (1986). Estimates of the Value of Patent Rights in European Countries - During the Post-1950 Period, Economic Journal 96(384): 1052-1076. Scherer, F. M. / Harhoff, D. / Kukies, J. (2000). Uncertainty and the Size Distribution of Rewards from Technological Innovation, Journal of Evolutionary Economics 10 (2): 175-200. Schmeisser, W. (1986). Systematische Erfindungsförderung als Unternehmensaufgabe, Berlin. Schmookler, J. (1957). Inventors Past and Present, The Review of Economics and Statistics 39 (3): 321-333. Schneider, W. / Stahl, K. / Struyk, R. (1985). Residential Mobility in the United States and the Federal Republic of Germany, in: Stahl, K. / Struyk, R. (Eds.): U.S. and West German Housing Markets, Washington D.C. Sher, I.H. / Garfield, E. (1966). New Tools for Improving and Evaluating the Effectiveness of Research, in: M.C. Yovits / D.M. Gilford / R.H. Wilcox / E. Staveley / H.D. Lemer, (Eds.): Research Program Effectiveness, New York: 135-146. Shockley, W. (1957). On the Statistics of Individual Variations of Productivity in Research Laboratories, Proceedings of the Institute of Radio Engineers 45: 279-290. Solon, G. (1985). Work Incentive Effects of Taxing Unemployment Benefits, Econometrica 53: 295-306.Afuah, A. (2000). How Much Do Your Co-opetitors’ Capabilities Matter in the Face of Technological Change?, Strategic Management Journal 21 (Special Issue): 387-404Clark, A. / Georgellis, Y. / Sanfey, P. (1998). Job Satisfaction, Wage Change and Quits: Evidence from Germany, Research in Labor Economics 17(1): 95-121.Krugman, P. (1991). Geography and Trade, Cambridge. Song, J. / Almeida, P. / Wu, G. (2001). Learning-by-Hiring: When is Mobility Useful? Discussion Paper, Columbia University. Song, J. / Almeida / P. Wu, G. (2003). Learning-by-Hiring: When is Mobility More Likely to Facilitate Interfirm Knowledge Transfer?, Management Science 49 (4): 351-365 Spence, M. (1973). Job Market Signalling, The Quarterly Journal of Economics 87 (3): 355374.

171

Spilerman, S. (1970). The Causes of Radical Disturbances: A Comparison of Alternative Explanations, American Sociological Review 35 (8): 627-649. Staudt, E. / Bock, J. / Mühlemeyer, P. / Kriegesmann, B. (1990). Anreizsysteme als Instrument des betrieblichen Innovationsmanagements – Ergebnisse einer empirischen Untersuchung im F+E-Bereich, Zeitschrift für Betriebswirtschaft (ZfB) 60 (11): 1183-1204. Staudt, E. / Bock, J. / Mühlemeyer, P. / Kriegesmann, B. (1992). Der Arbeitnehmererfinder im

betrieblichen

Innovationsprozess,

Zeitschrift

für

Betriebswirtschaftliche

Forschung (zfbf) 44 (2): 111-130. Stephan, P. (1996). The Economics of Science, Journal of Economic Literature 34: 11991235. Timpone, R.J. (2003). Concerns with Endogeneity in Statistical Analysis: Modeling the Interdependence Between Economic Ties and Conflict, in: Mansfield, E. / Pollins, B. (Eds.): New Perspectives on Economic Exchange and Armed Conflict, Ann Arbor: 289-309. Topel, R.H. / Ward, P.W. (1992). Job Mobility and the Careers of Young Men, The Quarterly Journal of Economics 107 (2): 439-479. Trajtenberg, M. (1990). A Penny for Your Quotes. Patent Citations and the Value of Innovations, Rand Journal of Economics 21 (1): 172-187. Trajtenberg, M. (2005). Recombinant Ideas: The Mobility of Inventors and the Productivity of Research, CEPR-Conference, Munich, May 26-28, 2005. Trajtenberg, M. / Henderson, R. / Jaffe, A. (1997). University versus Corporate Patents: A Window on the Basicness of Invention, Economics of Innovation and New Technology 5: 19-50. Tuma, N.B. (1976). Rewards, Resources, and the Rate of Mobility: A Nonstationary Multivariate Stochastic Model, American Sociological Review 41 (2): 338-360. Turner, L. / Mairesse, J. (2003). Individual Productivity Differences in Scientific Research: An Econometric Study of the Publications of French Physicists, Discussion Paper.

172

Vincent, H.F. / Mirakhor, A. (1972). Relationship between Productivity, Satisfaction, Ability, Age, and Salary in a Military R&D Organization, IEEE Transactions on Engineering Management 19 (1): 4-52. Von Hippel, E. (1988). The Sources of Innovation, Oxford. Widerstedt, B. (1998). Moving or Staying? Job Mobility as a Sorting Process. UMEA University Dissertation. Wooldridge, J.M. (1999). Introductory Econometrics – A Modern Approach, New York et al.. Wooldridge, J.M. (2001). Econometric Analysis of Cross Section and Panel Data, Cambridge, Mass. Yano, M. (1992). Law and Practice of Employee Inventions in Japan, WIPO Symposium of Employee Inventions, Bucharest: 165-182. Zucker, L.G. / Darby, M.R. / Torero, M. (2002). Labor Mobility from Academe to Commerce, Journal of Labor Economics 20 (3): 629-660. Zuckerman, H. / Merton, R.K. (1972). Age, Aging, and Age Structure in Science, in: R.K. Merton (Ed.), The Sociology of Science: Theoretical and Empirical Investigation, Chicago, 497-527.

173

175

LAST NAME

x x x x x x x x x x x x x x x x

% OF MATCH

100% 100% 100% 100% 100% 100% 99% 99% 99% 97% 97% 97% 95% 95% 80% 80%

x x

x

x

x

x

x

FIRST NAME

x x x

x x x

STREET

x x x

STREET (PART)

x x x

STREET_2

x x x x x x

x x x

CITY

x x x

CITY (PART)

x x x

CITY_2

x x

APPLICANT

x INVENTORS QUESTIONNED IN THE PATVAL SURVEY INVENTORS WHO FILLED OUT TWO OR MORE QUESTIONNAIRES SUM

x

x

x

x

x

FIRST NAME (PART)

Match of Addresses (Part 1)

Annex 1

4,185 0 3,049 297 29,971

0

76

2,200

6,769

19,487

FREQ.

175

13.97 0.00 10.17 0.99 100.00

0.00

0.25

7.34

2.26

65.02

%

Annex 2 Match of Addresses (Part 2)

PatVal Data ƒ

The PatVal data include information on 3,346 granted EP patents with priority date between 1993 and 1997. Since 288 inventors filled out two or more questionnaires (five at the most), the dataset contains 3,049 different inventors who form the basis for the following search procedure.

Search Database ƒ

Use of the EPOLINE patent database of the European Patent Office (EPO). The dataset is a counterpart of the EPOLINE database as of March 1st, 2003 and covers over 1,260,000 patent files with application dates ranging from June 1st, 1978 to July 25th, 2002. Data on inventor addresses were available until 1999. The search procedure aims at identifying all EP patents of each inventor contained in the PatVal data.

ƒ

In order to do this, co-inventors were split. For each inventor duplicate records were created, leading to 2,436,260 inventor files. In order to limit the search effort, last names not included in the PatVal data were removed from the dataset. Due to this procedure, the inventor files could be reduced from 2,436,260 to 231,681 inventor files.

Database Query

The query was carried out using MYSQL version 4. The MYSQL-control center was applied as SQL-Interface. All Java classes were constructed with Eclipse.

177

Adjustment before Search

Both datasets, the PatVal data (including 3,049 inventor files) and the EPOLINE data (including 231,681 inventor files), were standardized as follows: ƒ

ß ĺ

ss

ƒ

ä ĺ

ae

ƒ

ö ĺ

oe

ƒ

ü ĺ

ue

ƒ

removal of “;“ and “,“

ƒ

use of small letters

ƒ

harmonization of the names’ spelling, of street names or city names

ƒ

split of last name, first name and title

ƒ

split of zip code and city name

ƒ

delete of information concerning post office boxes from address data

ƒ

delete “c/o …“ from address data

(Meier, Hans-Juergen ĺ meier, hans-juergen)

Search Procedure ƒ

removal of „str., strasse“ from street names leopoldstrasse / leopoldstr. / leopold-strasse / leopold-str.

ƒ ƒ

ƒ

ĺ

siemens

microsoft deuschland GmbH

ĺ

microsoft deutschland

ĺ

frankfurt

delete of city add-ons ĺ

poelten

stefan jr.

ĺ

stefan

stefan I.

ĺ

stefan

st. poelten

ƒ

leopold

removal of corporate form identifiers siemens AG

frankfurt am main ƒ

ĺ

removal of house numbers

delete of name add-ons

additional search for parts of a first name, e.g., hans juergen, hans-juergen, hansjuergen, hans, juergen

ƒ

additional search for parts of a city name, e.g., bergisch gladbach, bergisch-gladbach, bergischgladbach, bergisch, gladbach)

178

Unavoidable Mistakes Resulting from the Search Procedure ƒ

brothers jointly owning a firm who assign the firm’s address as the inventors’ address ĺ last name, street, and city match completely ĺ a search procedure not accounting for the first name results in an additional inventor in the dataset

ƒ

married couple with matching last name and address ĺ last name and home address match completely ĺ a search procedure not accounting for the first name results in an additional inventor in the dataset

ƒ

inventors with common last names employed with large firms (e.g. SIEMENS), assigning the firm’s address or the address of a patent agency as inventor’s address ĺ last name (probably even the first name), street, city, and applicant match completely ĺ in case the first names do not match, the search procedure results in an additional inventor in the dataset. In case the first names do match, information including IPC classes as well as the names of the co-inventors have to be employed to distinguish between PatVal and non-PatVal inventors

ƒ

further mistakes are misspellings, e.g., missing hyphens hansjuergen/hans-juergen), ph vs. f. (rudolph/rudolf), v vs. f (detlev/detlef) or inconsistent abbreviations of first names (hans peter/hans p.) ĺ due to the additional search for parts of the inventor names, the search procedure detects misspelled inventor names. Unfortunately, the detected patents are not assigned to one but to a number of inventors, resulting in a biased frequency measure:

APPLICATION NUMER EP19890904071 EP19900907047 EP19900908936 EP19900910736 EP19910912570 EP19910903346 EP19910905775

LAST NAME

FIRST NAME

FREQUENCY

reinartz reinartz reinartz reinartz reinartz reinartz reinartz

hansdieter hans-dieter hans dieter hansdieter hansditer hans dieter

1 1 1 1 1 1 1

ADJUSTED FREQUENCY 7 7 7 7 7 7 7

To minimize the described mistakes, the matching of the inventor files was subsequently checked manually.

179

Annex 3 Questionnaire

Fragebogen

Wenn in diesem Fragebogen auf den Erfinder Bezug genommen wird, dann sind gleichermaßen männliche und weibliche Erfinder angesprochen. Um den Fragebogen nicht unnötig komplex zu gestalten, verwenden wir keine Ausdrücke wie „ErfinderIn“ oder ähnliches.

Teil A:

Persönliche Informationen

A.1.

Geburtsland

_______________________________

A.2

Geburtsjahr

_______________________________

A.3

Wohnort (zum Zeitpunkt der Erfindung)

_______________________________

A.4

Geschlecht

ˆ Männlich

ˆ Weiblich

181

Teil B:

B.1

Ausbildung

Bitte geben Sie an, welche der folgenden Schul- und Berufsausbildungsabschlüsse Sie zum Zeitpunkt der Erfindung bereits erworben hatten. (Mehrfachnennungen möglich)

ˆ Hauptschule oder Realschule

ˆ Hochschulabschluss / Fachhoch-

schulabschluss ˆ Abitur

ˆ Berufsakademie

ˆ Lehre

ˆ Promotion oder Habilitation

ˆ Meister

ˆ Sonstiges ___________________

Die folgenden Fragen beziehen sich auf den höchsten Bildungsabschluss, den Sie bis zum Zeitpunkt der Erfindung erworben hatten.

B.2a

In welchem Jahr hatten Sie diesen Abschluss erworben? _________________

B.2b In welchem Land hatten Sie diesen Abschluss erworben? _________________ B.2c

Falls Sie diesen Abschluss an einer Hochschule oder Fachhochschule erworben hatten, geben Sie bitte an, welchem Fachbereich dieser Abschluss zuzurechnen ist. (z.B. Maschinenbau, Biochemie) ______________________________________

B.3

Haben Sie als Erwerbstätiger oder Student einen längeren Auslandsaufenthalt in einem nicht-deutschsprachigen Land verbracht? (Falls Sie bereits mehr als einmal längere Zeit im Ausland verbracht haben, beziehen Sie sich bitte auf den längsten Aufenthalt.)

ˆ 3-6 Monate

ˆ 6-12 Monate ˆ 1-2 Jahre

ˆ > 2 Jahre

kein längerer Auslandsaufenthalt

Falls ja, in welchem Land?

182

____________________

Teil C:

C.1

Beschäftigung und Mobilität

Welchen Erwerbsstatus hatten Sie zum Zeitpunkt der Erfindung?

ˆ Arbeitnehmer (bitte geben Sie den Namen Ihres Arbeitgebers an)

_______________________________________________________________ ˆ Selbständig (bitte geben Sie den Namen Ihres Geschäftsbetriebs/Unternehmens an)

_______________________________________________________________ ˆ Beamter (bitte geben Sie den Namen der Behörde an, bei der Sie beschäftigt waren

bzw. die Universität, falls Sie Professor oder wissenschaftlicher Mitarbeiter waren) _______________________________________________________________ ˆ In Ausbildung (bitte geben Sie den Namen der Ausbildungsstätte an)

_______________________________________________________________ ˆ Sonstige (bitte angeben) ________________________________________

C.2

Welche der folgenden Beschreibungen kommt dem Typ der unter C.1 genannten Organisation am nächsten? ˆ Unternehmen mit ca. __________________ Mitarbeitern ˆ Staatliches Forschungsinstitut

ˆ Universität, Hochschule, sonstige

Ausbildungseinrichtung ˆ Krankenhaus, Stiftung, privates

ˆ Sonstige staatliche Organisation

Forschungsinstitut ˆ Sonstiges (bitte angeben)

_________________________________

C.3

War die in C.1 genannte Organisation der Anmelder oder einer der Anmelder des Patents?

C.4

ˆ Ja

ˆ Nein

In welchem Jahr hatte Ihre Tätigkeit bei der in C.1 genannten Organisation begonnen? _______________

183

Beschäftigung vor der Erfindung

C.5

Welchen Erwerbsstatus hatten Sie vor Ihrer Erfindung?

ˆ Arbeitnehmer (bitte geben Sie den Namen Ihres Arbeitgebers an)

_______________________________________________________________ ˆ Selbständig (bitte geben Sie den Namen Ihres Geschäftsbetriebs/Unternehmens an)

_______________________________________________________________ ˆ Beamter (bitte geben Sie den Namen der Behörde an, bei der Sie beschäftigt waren

bzw. die Universität, falls Sie Professor oder wissenschaftlicher Mitarbeiter waren) _______________________________________________________________ ˆ In Ausbildung (bitte geben Sie den Namen der Ausbildungsstätte an)

_______________________________________________________________ ˆ Sonstige (bitte angeben) ________________________________________

C.6

Welche der folgenden Beschreibungen kommt dem Typ der unte C.5 genannten Organisation am nächsten? ˆ Unternehmen mit ca. __________________ Mitarbeitern ˆ Staatliches Forschungsinstitut

ˆ Universität, Hochschule, sonstige

Ausbildungseinrichtung ˆ Krankenhaus, Stiftung, privates

ˆ Sonstige staatliche Organisation

Forschungsinstitut ˆ Sonstiges (bitte angeben)

_________________________________

C.7

In welchem Jahr hatte Ihre Tätigkeit bei der in C.5 genannten Organisation begonnen? _______________

C.8

Handelt es sich bei Ihrem früheren Arbeitgeber um ein Unternehmen, das in derselben Industrie tätig ist wie Ihr Arbeitgeber zur Zeit der Erfindung?

184

ˆ Ja

ˆ Nein

Beschäftigung nach der Erfindung

C.9

Wie oft haben Sie Ihren Arbeitgeber nach dem in C.1 genannten Arbeitgeber gewechselt?

ˆ Ich habe meinen Arbeitgeber nicht gewechselt

ˆ1

ˆ2

ˆ3

o weiter mit Teil D

ˆ mehr als 3 mal

C.10 Welchen Erwerbsstatus hatten Sie direkt nach Ihrer Erfindung? (Falls Sie Ihren

Arbeitgeber mehr als einmal gewechselt haben, bitte geben Sie den Erwerbsstatus direkt nach der Erfindung an.)

ˆ Arbeitnehmer (bitte geben Sie den Namen Ihres Arbeitgebers an)

_______________________________________________________________ ˆ Selbständig (bitte geben Sie den Namen Ihres Geschäftsbetriebs/Unternehmens an)

_______________________________________________________________ ˆ Beamter (bitte geben Sie den Namen der Behörde an, bei der Sie beschäftigt sind

bzw. die Universität, falls Sie Professor oder wissenschaftlicher Mitarbeiter sind) _______________________________________________________________ ˆ In Ausbildung (bitte geben Sie den Namen der Ausbildungsstätte an)

_______________________________________________________________ ˆ Sonstige (bitte angeben) ________________________________________

185

C.11 Welche der folgenden Beschreibungen kommt dem Typ der unter C.10 genannten

Organisation am nächsten? ˆ Unternehmen mit ca. __________________ Mitarbeitern ˆ Staatliches Forschungsinstitut

ˆ Universität, Hochschule, sonstige

ˆ Krankenhaus, Stiftung, privates

ˆ Sonstige staatliche Organisation

Ausbildungseinrichtung

Forschungsinstitut ˆ Sonstiges (bitte angeben)

_________________________________

C.12 In welchem Jahr hatte Ihre Tätigkeit bei der in C.10 genannten Organisation

begonnen? _______________

Derzeitige Beschäftigung

C.13 Handelt es sich bei Ihrem jetzigen Arbeitgeber um ein Unternehmen, das in derselben

Industrie tätig ist wie Ihr Arbeitgeber zur Zeit der Erfindung?

Teil D:

ˆ Ja

ˆ Nein

Erfindungsprozess

Falls außer Ihnen keine weiteren Erfinder existieren o weiter mit Frage D.4

D.1

Waren einer oder mehrere der im Patent genannten Erfinder bei einer anderen als der von Ihnen in C.1 genannten Organisation beschäftigt?

ˆ Ja

186

ˆ Nein

o weiter mit Frage D.4

D.2

Bitte geben Sie an, in welchem Typ von Organisation Ihre Miterfinder tätig waren? (Mehrfachnennungen möglich) ˆ Unternehmen mit ca. __________________ Mitarbeitern ˆ Staatliches Forschungsinstitut

ˆ Universität, Hochschule, sonstige

Ausbildungseinrichtung ˆ Krankenhaus, Stiftung, privates

ˆ Sonstige staatliche Organisation

Forschungsinstitut ˆ Sonstiges (bitte angeben)

_________________________________

D.3

War einer Ihrer Miterfinder bei einer Organisation tätig, die nicht zu den Anmeldern ˆ Ja

des Patents gehört?

D.4

War

die

Erfindung

das

Ergebnis

ˆ Nein

einer

formalen

oder

informalen

Forschungskooperation Ihres Arbeitgebers/Ihres eigenen Unternehmens mit einem anderen Partner? (Formal bezeichnet eine vertraglich geregelte Kooperation zwischen zwei Parteien)

ˆ Ja

ˆ Nein

Bitte nennen Sie den Namen des Partners sowie das Ziel der Kooperation. Bitte geben Sie auch Kooperationen mit anderen Anmeldern des Patents an. (Ziele einer Kooperation könnten beispielsweise Kostenteilung oder Erwerb von Know-how sein)

Namen der Partner

Ziel der Kooperation

Formal Informal

_______________________ _______________________

ˆ

ˆ

_______________________ _______________________

ˆ

ˆ

_______________________ _______________________

ˆ

ˆ

_______________________ _______________________

ˆ

ˆ

187

Während des Erfindungsprozesses kommunizieren Erfinder häufig. Wir möchten mehr über die Bedeutung verschiedener Informationsquellen und über den Kommunika-tionsprozess selbst erfahren. Bitte beantworten Sie daher die folgenden Fragen.

D.5

Waren Interaktionen (Diskussionen, Meetings, etc.) mit den folgenden Personen (abgesehen von den Miterfindern) während der Forschung, die zu der patentierten Erfindung geführt hat, besonders wichtig?

völlig unwichtig

188

teils teils

sehr wichtig

keine Interaktion

1

2

3

4

5

Personen, die in der selben Organisation arbeiteten und deren ständiger Arbeitsplatz weniger als eine Stunde entfernt lag.

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Personen, die in der selben Organisation arbeiteten und deren ständiger Arbeitsplatz mehr als eine Stunde entfernt lag.

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Personen, die nicht in der selben Organisation arbeiteten und deren ständiger Arbeitsplatz weniger als eine Stunde entfernt lag.

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Personen, die nicht in der selben Organisation arbeiteten und deren ständiger Arbeitsplatz mehr als eine Stunde entfernt lag.

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

D.6

Bitte beurteilen Sie die Bedeutung der folgenden Wissensquellen für die Entstehung der Erfindung.

völlig unwichtig

teils teils

sehr wichtig

Wissensquelle nicht genutzt

1

2

3

4

5

Universitäre Forschungseinrichtungen und Fakultäten

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Ausseruniversitäre öffentliche Forschungslabors

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Teilnahme an technischen Konferenzen, Workshops o.ä.

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Wissenschaftliche Literatur

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Andere Patentschriften

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Kunden / Nutzer

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Zulieferer

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Wettbewerber

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Sonstige wichtige Quellen (bitte angeben) ________________________

D.7

Wir sind daran interessiert, in welcher Region oder Stadt die Erfindung gemacht wurde. Bitte geben Sie die Postleitzahl und den Namen des betreffenden Ortes/der betreffenden Stadt sowie die Region/das Land an.

Postleitzahl Kreis

______________

Stadt _________________

______________

189

D.8

Bitte geben Sie weiterhin an, in welchem Umfeld (städtisch oder ländlich) die Erfindung gemacht wurde.

Die Erfindung entstand...

ˆ in einer Stadt mit mehr als 1 Mio. Einwohnern ˆ in einer Stadt mit 500.000 bis 1 Mio. Einwohnern ˆ in einer Stadt mit 100.000 bis 500.000 Einwohnern ˆ in einer Stadt mit 50.000 bis 100.000 Einwohnern ˆ in einer Stadt mit 10.000 bis 50.000 Einwohnern ˆ in einer Stadt mit weniger als 10.000 Einwohnern ˆ in einem ländlichen Gebiet

D.9

Bitte beschreiben Sie, wie es zu Ihrer Erfindung gekommen ist.

ˆ Die Erfindung entstand während eines F&E-Projekts und ... ˆ

war das geplante Ergebnis dieses F&E-Projekts

ˆ

war ein erwartetes Nebenprodukt dieses F&E-Projekts.

ˆ

war ein unerwartetes Nebenprodukt dieses F&E-Projekts

ˆ Die Idee zu dieser Erfindung entstand im Rahmen Ihrer normalen Tätigkeit außerhalb der F&E ...

ˆ

ˆ

und wurde in einem F&E-Projekt weiterentwickelt.

ˆ

eine Weiterentwicklung in einem F&E-Projekt war nicht erforderlich.

Die Idee zu dieser Erfindung entstand außerhalb des Unternehmens/in Ihrer Freizeit ... ˆ

und wurde in einem F&E-Projekt weiterentwickelt.

ˆ

eine Weiterentwicklung in einem F&E-Projekt war nicht erforderlich.

ˆ Sonstiges (bitte angeben)

190

____________________________________

D.10 Wieviele Personen-Monate wurden für die Forschungsarbeiten, die zu diesem Patent

geführt hat, aufgewendet?

ˆ Weniger als 1 Personen-Monat

ˆ 13-24 Personen-Monate

ˆ 1-3 Personen-Monate

ˆ 25-48 Personen-Monate

ˆ 4-6 Personen-Monate

ˆ 49-72 Personen-Monate

ˆ 7-12 Personen-Monate

ˆ Mehr als 72 Personen-Monate

D.11 Bitte geben Sie eine Schätzung der Gesamtkosten (in Euro) für die Forschungsarbeiten

an, die zu diesem Patent geführt hat? (Bitte schätzen Sie die Kosten der Erfindung ohne Amtsgebühren oder sonstige die Patentanmeldung betreffenden Kosten) _______________________________________

D.12 Welche der folgenden Finanzierungsarten beschreibt am besten die Finanzierung der

Forschungsarbeiten, die zu diesem Patent geführt hat? (Mehrfachantworten möglich) ˆ

Interne Geldmittel des Patentanmelders (inklusive Tochtergesellschaften)

ˆ

Geldmittel von anderen unverbundenen Organisationen, die am Projekt beteiligt waren

ˆ

Geldmittel von Finanzintermediären jeglicher Art (Banken, weitere Finanzinstitutionen, etc.)

ˆ

Staatliche Forschungsprogramme oder sonstige staatliche Geldmittel

ˆ

Sonstige (bitte angeben)

_____________________________________

191

D.13

Aus welchem Grund wurde die Erfindung so patentiert wie sie war, statt sie mit Hilfe zusätzlicher Ressourcen weiterzuentwickeln? ˆ

Die Erfindung ist gut genug wie sie ist

ˆ

Das gesetzte Ziel der Erfindung war erreicht

ˆ

Weitere Verbesserungen hätten nur mit Hilfe von Ressourcen erreicht werden können, die das Budget überschritten hätten

ˆ

Weitere Verbesserungen wären mit vorhandenen technischen Möglichkeiten nicht möglich gewesen

ˆ

Weitere Verbesserungen hätten zu einer weiteren Erfindung geführt, die zusätzlich hätte patentiert werden können

ˆ

Die Erfindung musste schnell patentiert werden, da andere Erfinder, Forschungsgruppen oder Unternehmen an einer Erfindung im selben Bereich gearbeitet haben

D.14 Baut diese Erfindung grundlegend auf einer anderen Ihnen bekannten Erfindung auf?

ˆ Ja

ˆ Nein

ˆ Unbekannt

o Nein oder unbekannt: weiter mit Frage D.16 D.15 Wurde diese andere Erfindung in der selben Organisation gemacht?

ˆ Ja

ˆ Nein

ˆ Unbekannt

Wir definieren eine „Patentfamilie“ als eine Gruppe von Patenten, die hinsichtlich Ihres Wertes oder technisch wesentlich aufeinander aufbauen.

D.16 Ist dieses Patent Teil einer “Patentfamilie”?

ˆ Ja

ˆ Nein

ˆ Unbekannt

o Nein oder unbekannt: weiter mit Teil E D.17 Aus wievielen Patenten besteht diese Patentfamilie?

ˆ 1-2

192

ˆ 3-5

ˆ 6-10

ˆ 11-20

ˆ > 20

ˆ Unbekannt

D.18 Wieviele Personen-Monate wurden für die Forschung, die zur gesamten Patentfamilie

geführt hat, aufgewendet?

Weniger als 1 Personen-Monat

25-48 Personen-Monate

1-3 Personen-Monate

49-72 Personen-Monate

4-6 Personen-Monate

73-96 Personen-Monate

7-12 Personen-Monate

97-120 Personen-Monate

13-24 Personen-Monate

Mehr als 120 Personen-Monate

D.19 Bitte geben Sie eine Schätzung der Gesamtkosten (in Euro) für die Forschung an, die

zur gesamten Patentfamilie geführt hat? (Bitte schätzen Sie die Kosten der Erfindung ohne Amtsgebühren oder sonstige die Patentanmeldung betreffenden Kosten) _______________________________________

Teil E:

E.1

Erfindervergütung

Waren Sie zur Zeit dieser Erfindung Arbeitnehmererfinder im privaten oder öffentlichen Dienst oder waren Sie freier Erfinder?

ˆ Arbeitnehmererfinder

ˆ Freier Erfinder

o weiter mit Frage E.6

E.2

Wurde Ihre Erfindung von Ihrem Arbeitgeber unbeschränkt oder beschränkt in Anspruch genommen?

ˆ Unbeschränkte

Inanspruchnahme

ˆ Beschränkte

Inanspruchnahme

ˆ Nichtinanspruchnahme

o weiter mit Frage E.6

193

E.3

Hätten Sie diese Erfindung im Falle einer Nichtinanspruchnahme selbst als Patent angemeldet?

ˆ Ja

ˆ Nein

Laut § 16 (1) des Gesetzes über Arbeitnehmererfindungen (ArbNErfG) steht Ihnen das Recht zu, eine Übertragung eines Schutzrechts auf Sie zu verlangen, falls das betreffende Schutzrecht seitens Ihres Arbeitgebers nicht aufrechterhalten werden soll.

E.4

Würden Sie dieses Patent im Falle einer Aufgabe des Patents durch Ihren Arbeitgeber selbst aufrechterhalten? ˆ Ja

E.5a

ˆ Nein

Welchen Anteil Ihres Brutto-Jahreseinkommens macht die Ihnen für diese Erfindung gewährte Erfindervergütung aus?

_____________%

E.5b Welchen Anteil Ihres Brutto-Jahreseinkommens machen die Ihnen für alle

Erfindungen gewährten Erfindervergütungen aus?

E.6

_____________%

Ist das Gesetz über Arbeitnehmererfindungen Ihrer Meinung nach als Schutzgesetz zugunsten des angestellten Erfinders geeignet? (1=völlig ungeeignet;

5= völlig

geeignet)

ˆ1

E.7

ˆ2

ˆ3

ˆ4

Fördert oder hemmt das Gesetz über Arbeitnehmererfindungen Ihrer Meinung nach die Motivation von Arbeitnehmererfindern?

ˆ Fördert die Motivation

194

ˆ5

ˆ Hemmt die Motivation

E.8

Bitte beschreiben Sie kurz, warum Sie für das Gesetz über Arbeit-nehmererfindungen eine motivationsfördernde oder motivationshemmende Wirkung vermuten. _____________________________________________________________________ _____________________________________________________________________ _____________________________________________________________________

E.9

Haben Sie über die Erfindervergütung hinaus eine permanente Erhöhung Ihres Einkommens erhalten, die Ihnen ausdrücklich aufgrund der Erstellung dieser ˆ Ja

Erfindung angeboten wurde?

ˆ Nein

Unternehmen können in vielfältiger Weise Anreize für Erfinder schaffen. Wir interessieren uns für Ihre Einschätzung dieser Anreize.

E.10

Wie wichtig sind für Sie die folgenden Anreize für Ihre erfinderische Tätigkeit?

völlig unwichtig

teils teils

sehr wichtig

1

2

3

4

5

Erfindervergütung nach dem Arbeitnehmererfindungsgesetz

ˆ

ˆ

ˆ

ˆ

ˆ

Lohn-/Gehaltserhöhung

ˆ

ˆ

ˆ

ˆ

ˆ

Beförderung

ˆ

ˆ

ˆ

ˆ

ˆ

Anerkennung im Unternehmen

ˆ

ˆ

ˆ

ˆ

ˆ

Zufriedenheit über technische Machbarkeit

ˆ

ˆ

ˆ

ˆ

ˆ

Zufriedenheit über einen positiven Beitrag zum Unternehmensergebnis

ˆ

ˆ

ˆ

ˆ

ˆ

Verbesserung der eigenen Arbeitsbedingungen

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Sonstiges (bitte angeben): ____________________________

195

E.11

Bei wievielen Europäischen Patentanmeldungen (inklusive diesem Patent) waren Sie bisher als Erfinder mitaufgeführt?

E.12

__________ Patentanmeldungen

Wieviele Erfindungen haben Sie in Ihrer bisherigen Laufbahn gemacht, die nicht zur Patentanmeldung kamen, sondern durch Geheimhaltung geschützt wurden ? __________ Erfindungen

Teil F:

Wert des Patentes

Abschließend interessieren wir uns noch für die Bedeutung Ihrer patentierten Erfindung.

F.1

Wie hoch schätzen Sie den wirtschaftlichen und strategischen Wert dieses Patentes im Vergleich zu anderen Patenten der gleichen Branche oder des gleichen Technologiefeldes ein?

ˆ Das Patent gehört zu den wichtigsten 10% aller Patente in dieser Branche ˆ Das Patent gehört zu den wichtigsten 25%, aber nicht zu den wichtigsten 10%

aller Patente in dieser Branche ˆ Das Patent gehört zu den wichtigsten 50%, aber nicht zu den wichtigsten 25%

aller Patente in dieser Branche ˆ Das Patent gehört nicht zu den wichtigsten 50% aller Patente in dieser Branche

F.2

Wurde das Patent vom Eigentümer/Anmelder jemals kommerziell genutzt?

ˆ Ja

F.3

ˆ Nein

ˆ Bisher noch nicht, Möglichkeiten werden noch geprüft

Hat der Anmelder/Eigentümer dieses Patent an eine unabhängige dritte Partei lizenziert?

ˆ Ja

196

ˆ Nein

ˆ Nein, das Patent soll aber lizenziert werden

F.4a

Wurde dieses Patent durch einen Lizenznehmer zur Gründung eines neuen ˆ Ja

Unternehmens genutzt?

F.4b

ˆ Unbekannt

Wurde dieses Patent durch Sie oder einen Ihrer Miterfinder zur Gründung eines neuen ˆ Ja

Unternehmens genutzt?

F.5

ˆ Nein

ˆ Nein

Wie wichtig waren die folgenden Ziele für die Entscheidung, die Erfindung zu patentieren?

völlig unwichtig

teils teils

sehr wichtig

1

2

3

4

5

Eigene wirtschaftliche Nutzung

ˆ

ˆ

ˆ

ˆ

ˆ

Lizenzierung

ˆ

ˆ

ˆ

ˆ

ˆ

Lizenzierung im Austausch für Lizenzrechte anderer (Kreuzlizenzierung)

ˆ

ˆ

ˆ

ˆ

ˆ

Schutz von eigenen Produkten und Prozessen vor Imitation

ˆ

ˆ

ˆ

ˆ

ˆ

Blockadepatent

ˆ

ˆ

ˆ

ˆ

ˆ

Reputation

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

Sonstige Gründe (bitte angeben) _______________________________

F.6a1 Wurde dieses Patent jemals von einer dritten Partei verletzt? ˆ Ja

ˆ Nein

ˆ Weiß nicht

197

F.6a2 Gab es jemals rechtliche Auseinandersetzungen, die dieses Patent betrafen? ˆ Ja, und zwar Einspruchs-/Beschwerdeverfahren am Europäischen Patentamt ˆ Ja, und zwar Nichtigkeitsverfahren am Bundespatentgericht ˆ Ja, und zwar eine gerichtliche Klage gegen ein anderes Unternehmen oder eine

Person wegen Verletzung des Patents ˆ Nein ˆ Weiß nicht

F.6b

Wurde dieses Patent bis heute in Deutschland aufrechterhalten?

ˆ Ja ˆ Nein, das Patent wurde nur bis zum Jahr 19 __ / 200_ (bitte angeben)

aufrechterhalten.

F.7

Abschließend bitten wir Sie um Ihre Mitarbeit in einem Gedankenexperiment: Bitte stellen Sie sich vor, der Eigentümer des Patentes hätte am Tag der Erteilung bereits die Informationen gehabt, die heute über den Wert des Patentes vorliegen. Wenn der schärfste Wettbewerber des Eigentümers an einem Erwerb des Patentes interessiert gewesen wäre, welchen Preis hätte der Eigentümer mindestens fordern sollen?

198

ˆ Weniger als € 30.000

ˆ € 3 Millionen bis € 10 Millionen

ˆ € 30.000 bis € 100.000

ˆ € 10 Millionen bis € 30 Millionen

ˆ € 100.000 bis € 300.000

ˆ € 30 Millionen bis € 100 Millionen

ˆ € 300.000 bis € 1 Million

ˆ € 100 Millionen bis € 300 Millionen

ˆ € 1 Million bis € 3 Millionen

ˆ Mehr als € 300 Millionen

Falls dieses Patent nicht Teil einer Patentfamilie ist (wenn Sie Frage D.16 mit Nein oder Unbekannt beantwortet haben), überspringen Sie bitte Frage F.8 und fahren Sie direkt mit „Bemerkungen“ am Ende des Fragebogens fort.

F.8

Sie haben bereits in Frage F.7 einen geschätzten Wert für das vorliegende Patent angegeben. Bitte geben Sie nun noch eine Schätzung für den Wert der gesamten Patentfamilie an.

ˆ Weniger als € 30.000

ˆ € 10 Millionen bis € 30 Millionen

ˆ € 30.000 bis € 100.000

ˆ € 30 Millionen bis € 100 Millionen

ˆ € 100.000 bis € 300.000

ˆ € 100 Millionen bis € 300 Millionen

ˆ € 300.000 bis € 1 Million

ˆ € 300 Millionen bis € 1 Milliarde

ˆ € 1 Million bis € 3 Millionen

ˆ € 1 Milliarde bis € 3 Milliarden

ˆ € 3 Millionen bis € 10 Millionen

ˆ Mehr als € 3 Milliarden

Bemerkungen:

Wir bedanken uns für Ihre Unterstützung. Wenn Sie daran interessiert sind, einen persönlichen Abschlussbericht dieses Forschungsprojektes zu erhalten, geben Sie bitte Ihre EMail-Adresse oder Postanschrift an: E-mail:

___________________________________

Name:

___________________________________

Adresse:

___________________________________ ___________________________________

199

Erratum A Study of Inventors

An incorrect content was temporarily assigned to this book. Since it has been corrected. The correct content for this book is "[A Study of Inventors]" ([10.1007/978-3-8350-9492-5]). The previously miss assigned contents belonged to [Kundenbindung und Kundenwert] and its correct content is available under [http://link.springer.com/book/10.1007/978-3-8350-9451-2] We apologize for this error.

The online version of the book can be found at: http://dx.doi.org/ 10.1007/978-3-8350-9492-5

K. Hoisl, A Study of Inventors, DOI 10.1007/978-3-8350-9492-5_6 © Deutscher Universitäts-Verlag | GWV Fachverlage GmbH, Wiesbaden 2007

200